Skip to content

Replace hash groupby internals with HashCSR - #24050

Draft
PointKernel wants to merge 7 commits into
NVIDIA:mainfrom
PointKernel:hashcsr-groupby
Draft

Replace hash groupby internals with HashCSR#24050
PointKernel wants to merge 7 commits into
NVIDIA:mainfrom
PointKernel:hashcsr-groupby

Conversation

@PointKernel

@PointKernel PointKernel commented Sep 9, 2026 •
edited
Loading

Copy link
Copy Markdown
Member

Description

This PR replaces the hash groupby internals with a HashCSR build followed by CUB reductions, in the spirit of #23640 for hash join. Rows are grouped by a single probe pass that gives every row a slot and a rank, the occupied slots are compacted into groups, and a fill pass produces the grouped row order. For inputs of two million rows or more with keys up to 32 bytes, the table is sized from a sampled estimate of the number of distinct keys rather than the row count, with a bounded probe that falls back to a slot per row when the estimate is short; this keeps the table cache-resident and roughly halves peak memory on low- and mid-cardinality inputs.

Aggregations are DeviceSegmentedReduce calls over the grouped order. Groups averaging under 128 rows are packed several per block through the dispatch's segment-size hint (one thread or one 8-lane sub-warp per group), larger groups get a block each, and groups longer than one chunk are reduced in two levels. Nullable aggregations carry the group validity in the accumulator, so a result and its null mask come out of one pass, and the sums behind MEAN, M2, VARIANCE and STD are fused into one reduction.

This removes the shared-memory aggregation kernel, the block-local mapping kernel, and the dense and sparse global-memory atomic paths (about 600 lines net). All dispatch is on the host, and because reductions no longer need atomics, decimal128 MIN/MAX and fixed-point SUM_OVERFLOW now take the hash path instead of the sort-based one.

Checklist

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

@PointKernel
PointKernel requested review from a team as code owners September 9, 2026 02:34
@PointKernel
PointKernel requested a review from ttnghia September 9, 2026 02:34
@PointKernel PointKernel added libcudf Affects libcudf (C++/CUDA) code. Performance Performance related issue improvement Improvement / enhancement to an existing function non-breaking Non-breaking change labels Sep 9, 2026
@PointKernel
PointKernel requested a review from lamarrr September 9, 2026 02:34
@github-actions github-actions Bot added the CMake CMake build issue label Sep 9, 2026
@PointKernel
PointKernel marked this pull request as draft September 9, 2026 02:35
@copy-pr-bot

copy-pr-bot Bot commented Sep 9, 2026

Copy link
Copy Markdown

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 PointKernel added the 2 - In Progress Currently a work in progress label Sep 9, 2026
@coderabbitai

coderabbitai Bot commented Sep 9, 2026 •
edited
Loading

Copy link
Copy Markdown

Review Change StackReview Change Stack

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: dc27a7c7-1ed1-40e6-83b0-4895912f9e81

📥 Commits

Reviewing files that changed from the base of the PR and between e44e3cf and 67c719a.

📒 Files selected for processing (24)
  • cpp/CMakeLists.txt
  • cpp/src/groupby/hash/compute_global_memory_aggs.cu
  • cpp/src/groupby/hash/compute_global_memory_aggs.cuh
  • cpp/src/groupby/hash/compute_global_memory_aggs.hpp
  • cpp/src/groupby/hash/compute_global_memory_aggs_null.cu
  • cpp/src/groupby/hash/compute_groupby.cu
  • cpp/src/groupby/hash/compute_mapping_indices.cu
  • cpp/src/groupby/hash/compute_mapping_indices.cuh
  • cpp/src/groupby/hash/compute_mapping_indices.hpp
  • cpp/src/groupby/hash/compute_mapping_indices_null.cu
  • cpp/src/groupby/hash/compute_shared_memory_aggs.cu
  • cpp/src/groupby/hash/compute_shared_memory_aggs.hpp
  • cpp/src/groupby/hash/compute_single_pass_aggs.cu
  • cpp/src/groupby/hash/compute_single_pass_aggs.cuh
  • cpp/src/groupby/hash/compute_single_pass_aggs.hpp
  • cpp/src/groupby/hash/compute_single_pass_aggs_null.cu
  • cpp/src/groupby/hash/global_memory_aggregator.cuh
  • cpp/src/groupby/hash/groupby.cu
  • cpp/src/groupby/hash/hash_csr_kernels.cuh
  • cpp/src/groupby/hash/helpers.cuh
  • cpp/src/groupby/hash/output_utils.cu
  • cpp/src/groupby/hash/output_utils.hpp
  • cpp/src/groupby/hash/shared_memory_aggregator.cuh
  • cpp/src/groupby/hash/single_pass_functors.cuh
💤 Files with no reviewable changes (17)
  • cpp/src/groupby/hash/compute_mapping_indices_null.cu
  • cpp/src/groupby/hash/compute_global_memory_aggs.cuh
  • cpp/src/groupby/hash/compute_global_memory_aggs.cu
  • cpp/src/groupby/hash/compute_global_memory_aggs.hpp
  • cpp/src/groupby/hash/shared_memory_aggregator.cuh
  • cpp/src/groupby/hash/global_memory_aggregator.cuh
  • cpp/src/groupby/hash/compute_global_memory_aggs_null.cu
  • cpp/src/groupby/hash/compute_mapping_indices.hpp
  • cpp/src/groupby/hash/compute_mapping_indices.cuh
  • cpp/src/groupby/hash/compute_shared_memory_aggs.cu
  • cpp/CMakeLists.txt
  • cpp/src/groupby/hash/compute_shared_memory_aggs.hpp
  • cpp/src/groupby/hash/compute_single_pass_aggs.cuh
  • cpp/src/groupby/hash/compute_mapping_indices.cu
  • cpp/src/groupby/hash/output_utils.hpp
  • cpp/src/groupby/hash/compute_single_pass_aggs_null.cu
  • cpp/src/groupby/hash/output_utils.cu

Included review availability: Your plan provides up to 12 included reviews per hour; 11 remain after this review.

📝 Summary

Summary by CodeRabbit

  • New Features
    • Updated hash-based groupby processing with improved grouped-row handling and key organization.
    • Expanded single-pass aggregation support, including counts, sums, products, minimums, maximums, argument-based aggregates, and sum-of-squares.
    • Improved handling of nullable data and aggregation result validity.
    • Added support for combining compatible consecutive aggregations during groupby processing.
  • Bug Fixes
    • Added validation for groupby capacity, input shapes, and supported aggregation types.

Walkthrough

Hash groupby now uses HashCSR to create grouped rows and a new single-pass aggregation pipeline. Obsolete cuco-based grouping, shared-memory, global-memory, mapping-index, and sparse-output implementations are removed.

Changes

Hash groupby migration

Layer / File(s) Summary
HashCSR grouping pipeline
cpp/src/groupby/hash/hash_csr_kernels.cuh, cpp/src/groupby/hash/compute_groupby.cu		 HashCSR builds grouped keys, group offsets, representative rows, and grouped input rows for aggregation.
Grouped-row aggregation pipeline
cpp/src/groupby/hash/compute_single_pass_aggs.hpp, cpp/src/groupby/hash/compute_single_pass_aggs.cu		 Single-pass aggregation now uses grouped rows, adaptive chunking, typed reductions, null handling, fused additive operations, and a column-based API.
Aggregation capability wiring
cpp/src/groupby/hash/groupby.cu		 Hash groupby eligibility now delegates to is_single_pass_agg_supported.
Legacy hash paths removal
cpp/CMakeLists.txt, cpp/src/groupby/hash/*		 Obsolete cuco-based grouping, shared-memory and global-memory aggregation, mapping-index, sparse-output, and key-transformation code is deleted.

Priority: ➖ Normal

Estimated code review effort: 4 (Complex) | ~60 minutes

Merge Risk: ⚪ Minimal · up to 67c71

No concrete current-head issue remains from the finalized findings; the PR is mergeable after normal checks.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name Status Explanation Resolution
Docstring Coverage ⚠️ Warning Docstring coverage is 5.56% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 36 functions across 4 files. (3 skipped: 3… Write docstrings for the functions missing them to satisfy the coverage threshold.
✅ Passed checks (4 passed)
Check name Status Explanation
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.
Title check ✅ Passed The title clearly and concisely summarizes the primary change: replacing the hash groupby internals with HashCSR.
Description check ✅ Passed The description directly explains the HashCSR groupby implementation, segmented reductions, removed aggregation paths, and resulting behavior changes.
Full details: Docstring Coverage

Explanation

Docstring coverage is 5.56% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 36 functions across 4 files. (3 skipped: 3 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.

@PointKernel

PointKernel commented Sep 9, 2026

Copy link
Copy Markdown
Member Author

Groupby benchmarks vs main, RTX PRO 6000 (sm_120), CUDA 12.9, idle GPU. nvbench_compare.py output, status relative to measured noise. Extra axes added for 1 to 8 groups, 200K to 2M groups, and 32 to 200 rows per group.

benchmark faster slower same
groupby_max_cardinality (20M rows, MAX) 56 0 0
groupby_max_cardinality, 1 to 8 groups 0 6 2
groupby_max_cardinality, 200K to 2M groups 4 2 0
complex_int_keys 20 0 0
complex_mixed_keys 36 4 0
groupby_max 35 1 0
groupby_struct_keys 18 0 0
groupby_m2_var_std 42 6 0
groupby_m2_var_std, 32 to 200 rows per group 13 7 0
sum 6 2 0
no_requests 3 0 1
total 233 28 3

Slower cases: 4 to 8 groups with 8 aggregations, variance on ~20-row groups with 50% nulls, a single sum at 100M rows, and string keys at 84K groups without nulls. Peak memory is unchanged for distinct keys and 15 to 40% higher when aggregating. decimal128 MIN/MAX moves from the sort path to hash (12x to 91x).

groupby_max_cardinality (20M rows, 3 int32 key columns, MAX): 56 faster, 0 slower, 0 same
T num_rows num_aggregations cardinality api Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
I32 20000000 1 20 normal 2.791 ms 5.37% 1.044 ms 0.71% -1746.657 us -62.59% 🟢 FAST
I32 20000000 2 20 normal 3.386 ms 76.79% 1.204 ms 0.55% -2181.493 us -64.43% 🟢 FAST
I32 20000000 4 20 normal 2.563 ms 1.18% 1.532 ms 0.47% -1030.455 us -40.21% 🟢 FAST
I32 20000000 8 20 normal 2.956 ms 1.20% 2.182 ms 0.44% -773.879 us -26.18% 🟢 FAST
I32 20000000 1 50 normal 2.316 ms 1.02% 1.056 ms 0.77% -1259.631 us -54.40% 🟢 FAST
I32 20000000 2 50 normal 2.514 ms 2.42% 1.225 ms 0.76% -1288.773 us -51.26% 🟢 FAST
I32 20000000 4 50 normal 2.788 ms 2.14% 1.567 ms 0.50% -1220.241 us -43.77% 🟢 FAST
I32 20000000 8 50 normal 3.216 ms 1.96% 2.230 ms 0.30% -985.850 us -30.65% 🟢 FAST
I32 20000000 1 100 normal 2.522 ms 2.28% 1.051 ms 0.43% -1470.992 us -58.33% 🟢 FAST
I32 20000000 2 100 normal 2.692 ms 2.12% 1.223 ms 0.40% -1469.086 us -54.58% 🟢 FAST
I32 20000000 4 100 normal 2.949 ms 2.01% 1.571 ms 0.57% -1378.581 us -46.74% 🟢 FAST
I32 20000000 8 100 normal 3.413 ms 1.83% 2.248 ms 0.45% -1164.403 us -34.12% 🟢 FAST
I32 20000000 1 1000 normal 4.116 ms 1.90% 1.054 ms 0.49% -3062.604 us -74.40% 🟢 FAST
I32 20000000 2 1000 normal 4.601 ms 3.19% 1.228 ms 0.43% -3372.791 us -73.31% 🟢 FAST
I32 20000000 4 1000 normal 8.817 ms 0.77% 1.558 ms 0.32% -7259.532 us -82.33% 🟢 FAST
I32 20000000 8 1000 normal 13.956 ms 0.59% 2.223 ms 0.27% -11732.861 us -84.07% 🟢 FAST
I32 20000000 1 10000 normal 4.036 ms 1.57% 1.075 ms 0.45% -2961.148 us -73.37% 🟢 FAST
I32 20000000 2 10000 normal 4.278 ms 1.57% 1.243 ms 0.44% -3035.113 us -70.95% 🟢 FAST
I32 20000000 4 10000 normal 4.949 ms 1.20% 1.568 ms 0.32% -3381.634 us -68.32% 🟢 FAST
I32 20000000 8 10000 normal 7.887 ms 37.03% 2.228 ms 0.20% -5658.653 us -71.75% 🟢 FAST
I32 20000000 1 100000 normal 4.842 ms 6.23% 1.244 ms 0.30% -3598.867 us -74.32% 🟢 FAST
I32 20000000 2 100000 normal 5.142 ms 5.85% 1.450 ms 0.25% -3691.873 us -71.80% 🟢 FAST
I32 20000000 4 100000 normal 5.918 ms 46.94% 1.846 ms 0.49% -4071.837 us -68.80% 🟢 FAST
I32 20000000 8 100000 normal 6.172 ms 5.02% 2.644 ms 0.13% -3527.984 us -57.16% 🟢 FAST
I32 20000000 1 1000000 normal 12.197 ms 0.10% 2.870 ms 0.19% -9326.107 us -76.46% 🟢 FAST
I32 20000000 2 1000000 normal 17.117 ms 6.04% 3.183 ms 0.17% -13933.926 us -81.40% 🟢 FAST
I32 20000000 4 1000000 normal 15.483 ms 6.95% 3.827 ms 0.50% -11656.545 us -75.29% 🟢 FAST
I32 20000000 8 1000000 normal 9.383 ms 47.69% 5.110 ms 0.14% -4273.653 us -45.55% 🟢 FAST
decimal128 20000000 1 20 normal 96.664 ms 0.07% 1.260 ms 0.76% -95404.035 us -98.70% 🟢 FAST
decimal128 20000000 2 20 normal 97.978 ms 0.21% 1.622 ms 0.35% -96356.132 us -98.34% 🟢 FAST
decimal128 20000000 4 20 normal 98.065 ms 0.34% 2.363 ms 0.41% -95701.571 us -97.59% 🟢 FAST
decimal128 20000000 8 20 normal 104.450 ms 0.21% 3.891 ms 0.18% -100558.573 us -96.27% 🟢 FAST
decimal128 20000000 1 50 normal 100.640 ms 0.05% 1.255 ms 0.50% -99384.552 us -98.75% 🟢 FAST
decimal128 20000000 2 50 normal 101.800 ms 0.25% 1.647 ms 0.25% -100153.816 us -98.38% 🟢 FAST
decimal128 20000000 4 50 normal 103.609 ms 0.06% 2.417 ms 0.20% -101191.207 us -97.67% 🟢 FAST
decimal128 20000000 8 50 normal 108.919 ms 0.03% 3.955 ms 0.14% -104964.937 us -96.37% 🟢 FAST
decimal128 20000000 1 100 normal 104.253 ms 0.27% 1.261 ms 0.33% -102992.462 us -98.79% 🟢 FAST
decimal128 20000000 2 100 normal 150.131 ms 9.14% 1.658 ms 0.35% -148473.577 us -98.90% 🟢 FAST
decimal128 20000000 4 100 normal 115.670 ms 19.20% 2.441 ms 0.23% -113228.626 us -97.89% 🟢 FAST
decimal128 20000000 8 100 normal 106.146 ms 0.17% 3.999 ms 0.16% -102147.753 us -96.23% 🟢 FAST
decimal128 20000000 1 1000 normal 106.161 ms 0.14% 1.274 ms 0.31% -104887.785 us -98.80% 🟢 FAST
decimal128 20000000 2 1000 normal 107.257 ms 0.12% 1.667 ms 0.26% -105589.245 us -98.45% 🟢 FAST
decimal128 20000000 4 1000 normal 109.292 ms 0.15% 2.447 ms 0.23% -106845.040 us -97.76% 🟢 FAST
decimal128 20000000 8 1000 normal 113.612 ms 0.13% 3.998 ms 0.16% -109614.494 us -96.48% 🟢 FAST
decimal128 20000000 1 10000 normal 108.298 ms 0.12% 1.298 ms 0.42% -106999.987 us -98.80% 🟢 FAST
decimal128 20000000 2 10000 normal 109.306 ms 0.11% 1.676 ms 0.28% -107629.348 us -98.47% 🟢 FAST
decimal128 20000000 4 10000 normal 111.313 ms 0.10% 2.432 ms 0.21% -108881.649 us -97.82% 🟢 FAST
decimal128 20000000 8 10000 normal 115.562 ms 0.10% 3.948 ms 0.50% -111614.763 us -96.58% 🟢 FAST
decimal128 20000000 1 100000 normal 100.551 ms 0.06% 1.420 ms 0.30% -99130.772 us -98.59% 🟢 FAST
decimal128 20000000 2 100000 normal 101.673 ms 0.13% 1.768 ms 0.29% -99904.943 us -98.26% 🟢 FAST
decimal128 20000000 4 100000 normal 103.683 ms 0.15% 2.462 ms 0.34% -101221.837 us -97.63% 🟢 FAST
decimal128 20000000 8 100000 normal 107.909 ms 0.12% 3.861 ms 0.50% -104048.254 us -96.42% 🟢 FAST
decimal128 20000000 1 1000000 normal 90.185 ms 0.23% 3.216 ms 0.18% -86969.063 us -96.43% 🟢 FAST
decimal128 20000000 2 1000000 normal 91.259 ms 0.15% 3.881 ms 0.14% -87377.666 us -95.75% 🟢 FAST
decimal128 20000000 4 1000000 normal 93.385 ms 0.13% 5.214 ms 0.11% -88170.706 us -94.42% 🟢 FAST
decimal128 20000000 8 1000000 normal 97.414 ms 0.19% 7.894 ms 0.07% -89520.165 us -91.90% 🟢 FAST
groupby_max_cardinality, extra axis: 1 to 8 groups: 0 faster, 6 slower, 2 same
T num_rows num_aggregations cardinality api Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
I32 20000000 1 1 normal 1.029 ms 0.95% 1.030 ms 0.61% 1.293 us 0.13% 🔵 SAME
I32 20000000 8 1 normal 1.661 ms 0.96% 1.741 ms 0.35% 80.285 us 4.83% 🔴 SLOW
I32 20000000 1 2 normal 1.024 ms 0.95% 1.027 ms 0.43% 2.741 us 0.27% 🔵 SAME
I32 20000000 8 2 normal 1.690 ms 1.23% 1.830 ms 0.25% 139.725 us 8.27% 🔴 SLOW
I32 20000000 1 4 normal 1.024 ms 0.95% 1.214 ms 0.95% 189.459 us 18.50% 🔴 SLOW
I32 20000000 8 4 normal 1.673 ms 1.62% 2.157 ms 1.00% 484.319 us 28.96% 🔴 SLOW
I32 20000000 1 8 normal 1.023 ms 0.88% 1.128 ms 0.44% 104.572 us 10.22% 🔴 SLOW
I32 20000000 8 8 normal 1.718 ms 1.82% 2.116 ms 0.24% 398.483 us 23.20% 🔴 SLOW
groupby_max_cardinality, extra axis: 200K to 2M groups: 4 faster, 2 slower, 0 same
T num_rows num_aggregations cardinality api Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
I32 20000000 1 200000 normal 2.769 ms 0.32% 1.517 ms 0.25% -1252.467 us -45.22% 🟢 FAST
I32 20000000 8 200000 normal 4.219 ms 0.13% 3.668 ms 0.10% -550.670 us -13.05% 🟢 FAST
I32 20000000 1 500000 normal 3.397 ms 0.50% 2.214 ms 0.17% -1183.015 us -34.83% 🟢 FAST
I32 20000000 8 500000 normal 4.415 ms 0.15% 4.429 ms 0.14% 14.168 us 0.32% 🔴 SLOW
I32 20000000 1 2000000 normal 4.314 ms 0.24% 3.421 ms 0.14% -893.283 us -20.71% 🟢 FAST
I32 20000000 8 2000000 normal 5.271 ms 0.15% 5.689 ms 0.24% 417.840 us 7.93% 🔴 SLOW
complex_int_keys: 20 faster, 0 slower, 0 same
num_cols num_rows value_key_ratio null_probability Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
1 2^24 20 0 9.076 ms 15.44% 2.315 ms 0.23% -6760.679 us -74.49% 🟢 FAST
2 2^24 20 0 9.510 ms 14.72% 2.440 ms 0.27% -7070.671 us -74.35% 🟢 FAST
4 2^24 20 0 10.254 ms 13.44% 2.798 ms 0.19% -7456.233 us -72.71% 🟢 FAST
8 2^24 20 0 12.719 ms 9.46% 3.704 ms 0.17% -9014.795 us -70.88% 🟢 FAST
16 2^24 20 0 19.306 ms 5.88% 5.632 ms 0.50% -13673.658 us -70.83% 🟢 FAST
1 2^24 200 0 6.834 ms 21.59% 1.008 ms 0.44% -5825.521 us -85.25% 🟢 FAST
2 2^24 200 0 7.028 ms 20.59% 1.104 ms 0.50% -5924.331 us -84.29% 🟢 FAST
4 2^24 200 0 8.547 ms 16.40% 1.282 ms 0.35% -7265.254 us -85.00% 🟢 FAST
8 2^24 200 0 10.995 ms 13.41% 1.644 ms 0.44% -9350.441 us -85.05% 🟢 FAST
16 2^24 200 0 16.202 ms 7.23% 2.473 ms 0.23% -13729.231 us -84.74% 🟢 FAST
1 2^24 20 0.5 8.177 ms 44.51% 1.321 ms 0.36% -6855.913 us -83.85% 🟢 FAST
2 2^24 20 0.5 6.776 ms 33.91% 1.500 ms 0.64% -5276.160 us -77.87% 🟢 FAST
4 2^24 20 0.5 7.657 ms 3.37% 1.851 ms 0.37% -5805.835 us -75.82% 🟢 FAST
8 2^24 20 0.5 10.229 ms 28.40% 2.641 ms 0.56% -7588.064 us -74.18% 🟢 FAST
16 2^24 20 0.5 18.248 ms 0.23% 4.282 ms 0.24% -13966.260 us -76.54% 🟢 FAST
1 2^24 200 0.5 6.255 ms 41.95% 931.814 us 0.58% -5322.754 us -85.10% 🟢 FAST
2 2^24 200 0.5 5.382 ms 0.14% 1.124 ms 0.57% -4258.016 us -79.12% 🟢 FAST
4 2^24 200 0.5 4.811 ms 49.69% 1.396 ms 0.50% -3414.811 us -70.98% 🟢 FAST
8 2^24 200 0.5 4.709 ms 0.24% 1.955 ms 0.34% -2754.041 us -58.49% 🟢 FAST
16 2^24 200 0.5 8.314 ms 0.17% 3.148 ms 0.24% -5166.338 us -62.14% 🟢 FAST
complex_mixed_keys: 36 faster, 4 slower, 0 same
num_cols num_rows value_key_ratio null_probability Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
1 2^18 20 0 1.488 ms 4.52% 111.094 us 4.43% -1377.405 us -92.54% 🟢 FAST
2 2^18 20 0 3.066 ms 98.42% 208.925 us 3.16% -2857.116 us -93.19% 🟢 FAST
3 2^18 20 0 2.031 ms 2.03% 221.668 us 2.95% -1809.598 us -89.09% 🟢 FAST
4 2^18 20 0 6.931 ms 0.08% 2.698 ms 0.36% -4233.616 us -61.08% 🟢 FAST
5 2^18 20 0 7.525 ms 0.24% 2.796 ms 0.43% -4728.286 us -62.84% 🟢 FAST
1 2^24 20 0 5.840 ms 54.06% 2.317 ms 0.16% -3522.190 us -60.32% 🟢 FAST
2 2^24 20 0 8.812 ms 0.28% 7.020 ms 0.12% -1792.735 us -20.34% 🟢 FAST
3 2^24 20 0 10.362 ms 2.00% 9.228 ms 0.14% -1134.130 us -10.94% 🟢 FAST
4 2^24 20 0 47.576 ms 0.24% 43.615 ms 0.16% -3961.467 us -8.33% 🟢 FAST
5 2^24 20 0 54.229 ms 0.21% 48.281 ms 0.19% -5947.396 us -10.97% 🟢 FAST
1 2^18 200 0 1.243 ms 1.26% 107.001 us 4.80% -1136.415 us -91.39% 🟢 FAST
2 2^18 200 0 1.029 ms 79.08% 208.716 us 2.74% -819.853 us -79.71% 🟢 FAST
3 2^18 200 0 302.284 us 2.57% 222.015 us 2.67% -80.269 us -26.55% 🟢 FAST
4 2^18 200 0 3.134 ms 0.25% 2.691 ms 0.36% -443.601 us -14.15% 🟢 FAST
5 2^18 200 0 3.295 ms 0.34% 2.807 ms 0.38% -487.904 us -14.81% 🟢 FAST
1 2^24 200 0 1.909 ms 0.43% 1.008 ms 0.36% -901.245 us -47.21% 🟢 FAST
2 2^24 200 0 6.711 ms 0.50% 7.014 ms 0.14% 303.230 us 4.52% 🔴 SLOW
3 2^24 200 0 8.378 ms 0.24% 9.235 ms 0.13% 857.363 us 10.23% 🔴 SLOW
4 2^24 200 0 42.088 ms 0.13% 43.661 ms 0.09% 1.573 ms 3.74% 🔴 SLOW
5 2^24 200 0 48.008 ms 0.15% 48.171 ms 0.25% 162.300 us 0.34% 🔴 SLOW
1 2^18 20 0.5 217.288 us 2.92% 184.382 us 2.87% -32.906 us -15.14% 🟢 FAST
2 2^18 20 0.5 348.870 us 3.46% 313.863 us 2.40% -35.007 us -10.03% 🟢 FAST
3 2^18 20 0.5 348.334 us 2.83% 321.784 us 4.22% -26.549 us -7.62% 🟢 FAST
4 2^18 20 0.5 630.493 us 3.01% 554.922 us 4.92% -75.570 us -11.99% 🟢 FAST
5 2^18 20 0.5 1.178 ms 2.93% 1.085 ms 1.49% -92.371 us -7.84% 🟢 FAST
1 2^24 20 0.5 2.434 ms 0.69% 1.338 ms 0.43% -1096.389 us -45.04% 🟢 FAST
2 2^24 20 0.5 3.303 ms 0.43% 2.326 ms 0.33% -977.619 us -29.59% 🟢 FAST
3 2^24 20 0.5 2.687 ms 0.46% 1.918 ms 0.42% -769.494 us -28.63% 🟢 FAST
4 2^24 20 0.5 3.081 ms 0.36% 2.800 ms 0.45% -280.824 us -9.11% 🟢 FAST
5 2^24 20 0.5 3.437 ms 0.45% 3.135 ms 0.63% -302.283 us -8.79% 🟢 FAST
1 2^18 200 0.5 216.153 us 2.52% 173.612 us 3.61% -42.540 us -19.68% 🟢 FAST
2 2^18 200 0.5 347.703 us 2.26% 314.724 us 2.52% -32.979 us -9.48% 🟢 FAST
3 2^18 200 0.5 350.044 us 2.11% 321.588 us 2.12% -28.456 us -8.13% 🟢 FAST
4 2^18 200 0.5 634.212 us 2.17% 551.958 us 1.72% -82.254 us -12.97% 🟢 FAST
5 2^18 200 0.5 1.191 ms 1.56% 1.084 ms 2.02% -107.809 us -9.05% 🟢 FAST
1 2^24 200 0.5 1.881 ms 0.78% 945.177 us 0.56% -936.180 us -49.76% 🟢 FAST
2 2^24 200 0.5 3.318 ms 0.53% 2.323 ms 0.50% -994.876 us -29.98% 🟢 FAST
3 2^24 200 0.5 2.671 ms 0.73% 1.912 ms 0.35% -758.837 us -28.41% 🟢 FAST
4 2^24 200 0.5 3.072 ms 0.41% 2.800 ms 0.50% -272.372 us -8.87% 🟢 FAST
5 2^24 200 0.5 3.438 ms 0.60% 3.115 ms 0.39% -322.181 us -9.37% 🟢 FAST
groupby_max: 35 faster, 1 slower, 0 same
T cardinality num_rows null_probability num_aggregations Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
I32 0 2^24 0 1 4.942 ms 1.44% 3.245 ms 0.18% -1697.169 us -34.34% 🟢 FAST
I32 0 2^24 0.1 1 5.679 ms 1.44% 3.453 ms 0.44% -2225.999 us -39.20% 🟢 FAST
I32 0 2^24 0.9 1 5.316 ms 1.57% 3.261 ms 0.13% -2054.982 us -38.66% 🟢 FAST
I32 0 2^24 0 2 5.640 ms 1.41% 3.511 ms 0.14% -2128.244 us -37.74% 🟢 FAST
I32 0 2^24 0.1 2 6.416 ms 1.28% 3.993 ms 0.18% -2423.081 us -37.76% 🟢 FAST
I32 0 2^24 0.9 2 5.681 ms 1.41% 3.689 ms 0.25% -1991.790 us -35.06% 🟢 FAST
I32 0 2^24 0 4 5.456 ms 1.37% 4.126 ms 0.14% -1330.495 us -24.38% 🟢 FAST
I32 0 2^24 0.1 4 6.130 ms 1.29% 4.839 ms 0.14% -1290.230 us -21.05% 🟢 FAST
I32 0 2^24 0.9 4 5.406 ms 1.52% 4.246 ms 0.12% -1160.477 us -21.46% 🟢 FAST
I32 0 2^24 0 8 6.265 ms 1.17% 5.138 ms 0.23% -1126.978 us -17.99% 🟢 FAST
I32 0 2^24 0.1 8 9.491 ms 42.20% 6.816 ms 0.50% -2674.488 us -28.18% 🟢 FAST
I32 0 2^24 0.9 8 23.600 ms 9.31% 5.566 ms 0.19% -18033.979 us -76.42% 🟢 FAST
I32 0 2^24 0 16 24.514 ms 7.97% 7.262 ms 0.09% -17251.473 us -70.38% 🟢 FAST
I32 0 2^24 0.1 16 29.784 ms 0.08% 10.797 ms 0.18% -18987.094 us -63.75% 🟢 FAST
I32 0 2^24 0.9 16 10.147 ms 45.33% 8.187 ms 0.28% -1960.384 us -19.32% 🟢 FAST
I32 0 2^24 0 32 16.349 ms 36.36% 11.547 ms 0.44% -4802.145 us -29.37% 🟢 FAST
I32 0 2^24 0.1 32 39.074 ms 0.22% 17.935 ms 0.14% -21139.174 us -54.10% 🟢 FAST
I32 0 2^24 0.9 32 27.693 ms 7.55% 13.383 ms 0.29% -14310.023 us -51.67% 🟢 FAST
F64 0 2^24 0 1 18.115 ms 5.34% 3.428 ms 0.15% -14687.233 us -81.08% 🟢 FAST
F64 0 2^24 0.1 1 24.572 ms 0.08% 3.598 ms 0.15% -20973.979 us -85.36% 🟢 FAST
F64 0 2^24 0.9 1 24.051 ms 0.16% 3.357 ms 0.15% -20693.849 us -86.04% 🟢 FAST
F64 0 2^24 0 2 19.064 ms 5.09% 3.884 ms 0.10% -15179.579 us -79.63% 🟢 FAST
F64 0 2^24 0.1 2 25.534 ms 0.16% 4.198 ms 0.18% -21336.341 us -83.56% 🟢 FAST
F64 0 2^24 0.9 2 8.643 ms 49.44% 3.766 ms 0.15% -4877.531 us -56.43% 🟢 FAST
F64 0 2^24 0 4 8.407 ms 0.20% 4.789 ms 0.09% -3617.717 us -43.03% 🟢 FAST
F64 0 2^24 0.1 4 24.119 ms 8.82% 5.430 ms 0.49% -18688.773 us -77.49% 🟢 FAST
F64 0 2^24 0.9 4 6.462 ms 0.46% 4.601 ms 0.15% -1861.694 us -28.81% 🟢 FAST
F64 0 2^24 0 8 9.161 ms 0.93% 6.609 ms 0.07% -2552.154 us -27.86% 🟢 FAST
F64 0 2^24 0.1 8 10.120 ms 0.17% 7.889 ms 0.22% -2230.806 us -22.04% 🟢 FAST
F64 0 2^24 0.9 8 7.059 ms 0.33% 6.252 ms 0.20% -806.400 us -11.42% 🟢 FAST
F64 0 2^24 0 16 14.859 ms 34.85% 10.229 ms 0.05% -4630.004 us -31.16% 🟢 FAST
F64 0 2^24 0.1 16 16.032 ms 0.23% 12.939 ms 0.11% -3092.814 us -19.29% 🟢 FAST
F64 0 2^24 0.9 16 9.632 ms 0.20% 9.604 ms 0.15% -27.690 us -0.29% 🟢 FAST
F64 0 2^24 0 32 22.225 ms 0.07% 17.456 ms 0.04% -4769.335 us -21.46% 🟢 FAST
F64 0 2^24 0.1 32 20.454 ms 12.78% 22.585 ms 0.06% 2.132 ms 10.42% 🔴 SLOW
F64 0 2^24 0.9 32 18.026 ms 0.08% 16.927 ms 0.19% -1098.664 us -6.09% 🟢 FAST
groupby_struct_keys: 18 faster, 0 slower, 0 same
NumRows Depth Nulls Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
2^10 0 0 115.844 us 4.58% 100.942 us 4.87% -14.902 us -12.86% 🟢 FAST
2^16 0 0 124.781 us 3.39% 105.802 us 4.66% -18.978 us -15.21% 🟢 FAST
2^20 0 0 188.893 us 2.01% 147.544 us 3.09% -41.349 us -21.89% 🟢 FAST
2^10 1 0 260.841 us 2.59% 246.434 us 3.18% -14.407 us -5.52% 🟢 FAST
2^16 1 0 277.128 us 2.87% 257.600 us 3.24% -19.528 us -7.05% 🟢 FAST
2^20 1 0 457.183 us 1.55% 363.042 us 2.10% -94.141 us -20.59% 🟢 FAST
2^10 8 0 983.800 us 1.78% 929.609 us 1.78% -54.191 us -5.51% 🟢 FAST
2^16 8 0 1.007 ms 1.67% 935.984 us 1.57% -70.833 us -7.04% 🟢 FAST
2^20 8 0 1.219 ms 1.38% 1.058 ms 5.17% -160.217 us -13.15% 🟢 FAST
2^10 0 1 158.701 us 3.00% 144.977 us 5.43% -13.725 us -8.65% 🟢 FAST
2^16 0 1 163.521 us 3.31% 150.360 us 3.36% -13.162 us -8.05% 🟢 FAST
2^20 0 1 235.123 us 2.46% 195.138 us 3.86% -39.985 us -17.01% 🟢 FAST
2^10 1 1 259.573 us 2.71% 249.130 us 3.71% -10.443 us -4.02% 🟢 FAST
2^16 1 1 282.175 us 2.81% 255.303 us 2.81% -26.872 us -9.52% 🟢 FAST
2^20 1 1 454.137 us 1.63% 379.451 us 1.69% -74.686 us -16.45% 🟢 FAST
2^10 8 1 986.536 us 1.81% 935.062 us 1.75% -51.474 us -5.22% 🟢 FAST
2^16 8 1 1.017 ms 1.70% 953.532 us 1.64% -63.725 us -6.26% 🟢 FAST
2^20 8 1 1.230 ms 1.46% 1.071 ms 1.64% -158.960 us -12.92% 🟢 FAST
groupby_m2_var_std: 42 faster, 6 slower, 0 same
T U value_key_ratio num_rows null_probability num_aggs Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
I32 12 20 100000 0 1 183.363 us 2.48% 155.387 us 3.42% -27.975 us -15.26% 🟢 FAST
I32 12 100 100000 0 1 171.177 us 2.71% 140.297 us 3.75% -30.880 us -18.04% 🟢 FAST
I32 12 20 10000000 0 1 1.251 ms 0.63% 1.103 ms 0.77% -148.281 us -11.85% 🟢 FAST
I32 12 100 10000000 0 1 1.130 ms 0.53% 659.877 us 0.67% -469.862 us -41.59% 🟢 FAST
I32 12 20 100000 0.5 1 184.286 us 2.75% 160.646 us 3.22% -23.640 us -12.83% 🟢 FAST
I32 12 100 100000 0.5 1 171.166 us 2.83% 140.778 us 3.71% -30.388 us -17.75% 🟢 FAST
I32 12 20 10000000 0.5 1 1.155 ms 0.57% 1.076 ms 0.44% -78.345 us -6.79% 🟢 FAST
I32 12 100 10000000 0.5 1 1.020 ms 0.66% 668.093 us 0.72% -351.653 us -34.48% 🟢 FAST
I32 12 20 100000 0 10 705.481 us 1.62% 644.848 us 1.62% -60.633 us -8.59% 🟢 FAST
I32 12 100 100000 0 10 610.580 us 2.03% 491.392 us 1.82% -119.188 us -19.52% 🟢 FAST
I32 12 20 10000000 0 10 4.799 ms 0.35% 3.743 ms 0.31% -1056.049 us -22.01% 🟢 FAST
I32 12 100 10000000 0 10 4.498 ms 0.18% 2.086 ms 0.37% -2412.348 us -53.63% 🟢 FAST
I32 12 20 100000 0.5 10 742.170 us 1.68% 679.715 us 1.29% -62.455 us -8.42% 🟢 FAST
I32 12 100 100000 0.5 10 646.907 us 1.55% 502.046 us 6.83% -144.861 us -22.39% 🟢 FAST
I32 12 20 10000000 0.5 10 3.808 ms 0.38% 4.129 ms 0.31% 320.368 us 8.41% 🔴 SLOW
I32 12 100 10000000 0.5 10 3.464 ms 0.28% 2.183 ms 0.59% -1280.295 us -36.96% 🟢 FAST
I32 12 20 100000 0 50 2.989 ms 1.01% 2.899 ms 6.78% -89.967 us -3.01% 🟢 FAST
I32 12 100 100000 0 50 2.567 ms 2.10% 2.071 ms 1.32% -495.889 us -19.32% 🟢 FAST
I32 12 20 10000000 0 50 20.493 ms 0.15% 15.367 ms 0.12% -5125.514 us -25.01% 🟢 FAST
I32 12 100 10000000 0 50 19.486 ms 0.17% 8.398 ms 0.24% -11087.416 us -56.90% 🟢 FAST
I32 12 20 100000 0.5 50 3.170 ms 1.43% 2.975 ms 0.99% -195.201 us -6.16% 🟢 FAST
I32 12 100 100000 0.5 50 2.688 ms 1.54% 2.084 ms 2.85% -604.143 us -22.47% 🟢 FAST
I32 12 20 10000000 0.5 50 15.507 ms 0.26% 17.850 ms 0.22% 2.342 ms 15.11% 🔴 SLOW
I32 12 100 10000000 0.5 50 14.263 ms 0.22% 8.874 ms 0.26% -5388.641 us -37.78% 🟢 FAST
F64 12 20 100000 0 1 187.617 us 2.93% 164.240 us 2.83% -23.376 us -12.46% 🟢 FAST
F64 12 100 100000 0 1 177.076 us 3.24% 146.659 us 4.12% -30.417 us -17.18% 🟢 FAST
F64 12 20 10000000 0 1 1.270 ms 0.55% 1.124 ms 0.31% -145.140 us -11.43% 🟢 FAST
F64 12 100 10000000 0 1 1.137 ms 0.73% 862.337 us 0.56% -274.326 us -24.13% 🟢 FAST
F64 12 20 100000 0.5 1 194.707 us 3.46% 166.647 us 2.75% -28.061 us -14.41% 🟢 FAST
F64 12 100 100000 0.5 1 181.488 us 3.03% 145.601 us 3.52% -35.887 us -19.77% 🟢 FAST
F64 12 20 10000000 0.5 1 1.166 ms 0.64% 1.156 ms 0.48% -9.959 us -0.85% 🟢 FAST
F64 12 100 10000000 0.5 1 1.029 ms 0.72% 865.218 us 1.56% -163.831 us -15.92% 🟢 FAST
F64 12 20 100000 0 10 792.794 us 1.94% 735.731 us 9.73% -57.063 us -7.20% 🟢 FAST
F64 12 100 100000 0 10 705.862 us 2.09% 550.068 us 8.64% -155.794 us -22.07% 🟢 FAST
F64 12 20 10000000 0 10 4.953 ms 0.36% 4.598 ms 0.50% -355.113 us -7.17% 🟢 FAST
F64 12 100 10000000 0 10 4.595 ms 0.20% 4.160 ms 0.19% -434.654 us -9.46% 🟢 FAST
F64 12 20 100000 0.5 10 823.053 us 1.80% 754.872 us 1.30% -68.181 us -8.28% 🟢 FAST
F64 12 100 100000 0.5 10 730.991 us 2.26% 537.926 us 2.21% -193.065 us -26.41% 🟢 FAST
F64 12 20 10000000 0.5 10 3.919 ms 0.24% 5.019 ms 0.22% 1.099 ms 28.05% 🔴 SLOW
F64 12 100 10000000 0.5 10 3.568 ms 0.50% 4.150 ms 0.26% 581.549 us 16.30% 🔴 SLOW
F64 12 20 100000 0 50 3.451 ms 2.00% 3.070 ms 1.17% -381.384 us -11.05% 🟢 FAST
F64 12 100 100000 0 50 2.984 ms 1.94% 2.247 ms 1.25% -737.458 us -24.71% 🟢 FAST
F64 12 20 10000000 0 50 21.333 ms 0.25% 19.921 ms 0.08% -1412.628 us -6.62% 🟢 FAST
F64 12 100 10000000 0 50 19.943 ms 0.19% 18.773 ms 0.17% -1170.184 us -5.87% 🟢 FAST
F64 12 20 100000 0.5 50 3.632 ms 1.46% 3.332 ms 0.78% -300.128 us -8.26% 🟢 FAST
F64 12 100 100000 0.5 50 3.122 ms 1.00% 2.267 ms 1.30% -855.265 us -27.40% 🟢 FAST
F64 12 20 10000000 0.5 50 16.086 ms 0.30% 21.946 ms 0.13% 5.860 ms 36.43% 🔴 SLOW
F64 12 100 10000000 0.5 50 14.747 ms 0.22% 18.755 ms 0.12% 4.008 ms 27.18% 🔴 SLOW
groupby_m2_var_std, extra axis: 32 to 200 rows per group: 13 faster, 7 slower, 0 same
T U value_key_ratio num_rows null_probability num_aggs Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
I32 12 32 10000000 0 10 4.575 ms 0.25% 3.410 ms 0.28% -1165.055 us -25.47% 🟢 FAST
I32 12 50 10000000 0 10 4.521 ms 0.25% 3.322 ms 0.31% -1198.886 us -26.52% 🟢 FAST
I32 12 64 10000000 0 10 4.523 ms 0.26% 3.281 ms 0.29% -1242.304 us -27.47% 🟢 FAST
I32 12 100 10000000 0 10 4.505 ms 0.22% 2.095 ms 0.41% -2410.290 us -53.50% 🟢 FAST
I32 12 200 10000000 0 10 4.578 ms 0.30% 1.638 ms 0.45% -2940.638 us -64.23% 🟢 FAST
I32 12 32 10000000 0.5 10 3.586 ms 0.23% 3.868 ms 0.24% 282.178 us 7.87% 🔴 SLOW
I32 12 50 10000000 0.5 10 3.519 ms 0.50% 3.777 ms 0.19% 257.415 us 7.31% 🔴 SLOW
I32 12 64 10000000 0.5 10 3.496 ms 0.24% 3.729 ms 0.20% 232.809 us 6.66% 🔴 SLOW
I32 12 100 10000000 0.5 10 3.478 ms 0.34% 2.180 ms 0.42% -1298.252 us -37.33% 🟢 FAST
I32 12 200 10000000 0.5 10 3.542 ms 1.02% 1.664 ms 0.42% -1878.482 us -53.03% 🟢 FAST
F64 12 32 10000000 0 10 4.703 ms 0.19% 4.259 ms 0.25% -443.920 us -9.44% 🟢 FAST
F64 12 50 10000000 0 10 4.649 ms 0.50% 4.169 ms 0.50% -479.602 us -10.32% 🟢 FAST
F64 12 64 10000000 0 10 4.648 ms 0.24% 4.126 ms 0.24% -521.764 us -11.23% 🟢 FAST
F64 12 100 10000000 0 10 4.618 ms 0.19% 4.158 ms 0.22% -460.467 us -9.97% 🟢 FAST
F64 12 200 10000000 0 10 4.710 ms 0.19% 2.774 ms 0.31% -1936.038 us -41.11% 🟢 FAST
F64 12 32 10000000 0.5 10 3.690 ms 0.36% 4.666 ms 0.23% 975.751 us 26.44% 🔴 SLOW
F64 12 50 10000000 0.5 10 3.622 ms 0.26% 4.584 ms 0.24% 961.694 us 26.55% 🔴 SLOW
F64 12 64 10000000 0.5 10 3.613 ms 0.31% 4.557 ms 0.17% 943.781 us 26.12% 🔴 SLOW
F64 12 100 10000000 0.5 10 3.594 ms 0.28% 4.154 ms 0.20% 559.553 us 15.57% 🔴 SLOW
F64 12 200 10000000 0.5 10 3.607 ms 0.40% 2.755 ms 0.38% -852.190 us -23.63% 🟢 FAST
sum: 6 faster, 2 slower, 0 same
T num_rows Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
I64 100000 136.001 us 3.10% 113.854 us 3.90% -22.147 us -16.28% 🟢 FAST
I64 1000000 203.104 us 2.09% 136.065 us 4.53% -67.038 us -33.01% 🟢 FAST
I64 10000000 793.297 us 1.06% 562.268 us 0.64% -231.029 us -29.12% 🟢 FAST
I64 100000000 5.539 ms 0.31% 5.897 ms 0.16% 357.312 us 6.45% 🔴 SLOW
decimal64 100000 140.317 us 4.59% 116.888 us 3.86% -23.428 us -16.70% 🟢 FAST
decimal64 1000000 352.257 us 2.99% 167.914 us 2.31% -184.342 us -52.33% 🟢 FAST
decimal64 10000000 2.381 ms 0.50% 1.713 ms 0.19% -667.998 us -28.06% 🟢 FAST
decimal64 100000000 26.328 ms 0.15% 30.495 ms 0.07% 4.167 ms 15.83% 🔴 SLOW
no_requests: 3 faster, 0 slower, 1 same
num_rows Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
100000 47.422 us 7.48% 45.536 us 6.74% -1.887 us -3.98% 🔵 SAME
1000000 93.446 us 2.72% 56.542 us 4.89% -36.904 us -39.49% 🟢 FAST
10000000 340.539 us 0.90% 211.472 us 1.43% -129.068 us -37.90% 🟢 FAST
100000000 2.842 ms 0.28% 1.835 ms 0.22% -1007.541 us -35.45% 🟢 FAST

@PointKernel

PointKernel commented Sep 9, 2026 •
edited
Loading

Copy link
Copy Markdown
Member Author

Groupby benchmarks vs main (2062f4d) at 11d3dd1, RTX PRO 6000 (sm_120), CUDA 12.9, idle GPU, nvbench_compare.py status relative to noise. Supersedes the earlier comment, whose main times for groupby_max, complex_int_keys and the 20 to 1M group configs were inflated by another job on the GPU.

benchmark faster slower same
groupby_max_cardinality, 20 to 1M groups 54 1 1
groupby_max_cardinality, 1 to 16 groups 18 2 0
groupby_max_cardinality, 2M to 5M groups 8 0 0
complex_int_keys 20 0 0
complex_mixed_keys 33 7 0
groupby_max 31 5 0
groupby_struct_keys 18 0 0
groupby_m2_var_std, default axes 48 0 0
groupby_m2_var_std, 32 to 1000 rows per group 56 0 0
sum 7 1 0
no_requests 3 0 1
total 296 16 2

Since 67c719a: hot table slots are read through L1; for 2M+ rows with keys up to 32 bytes the table is sized from a sampled distinct-key estimate (full-size rebuild if it falls short); groups averaging under 128 rows are packed several per block via the segmented-reduce size hint, replacing ReduceByKey and its label array; nullable aggregations produce the result and its null mask in one pass.

case (ms) main previous HashCSR (67c719a) now peak MB, previous HashCSR -> now
20M rows, 1 group, 1 MAX 1.053 1.033 0.682 560 -> 321
20M rows, 8 groups, 8 MAX 1.729 2.250 1.900 560 -> 321
20M rows, 100 groups, 1 MAX 1.263 1.047 0.797 560 -> 321
20M rows, 1M groups, 8 MAX 4.783 5.112 3.043 564 -> 339
VAR/STD 10M F64, 20 rows/group, 50% nulls, 10 aggs 3.920 4.978 2.597 282 -> 213
groupby_max unique keys, 90% nulls, 32 aggs 7.129 13.329 7.652 575 -> 519
no_requests 100M rows 2.826 1.826 0.918 1200 -> 401

Slower than main: 8 to 20 groups with 8 aggregations (6 to 10%), groupby_max with 90% null keys and values at 16 to 32 aggregations (up to 28%), sum decimal64 at 100M near-unique keys (13%), 3 to 5 mixed int/string/list key columns (3 to 8%). Peak memory is 1.7x below main on the 20M-row benchmarks, 3x for 100M-row distinct keys, equal for unique and wide keys.

groupby_max_cardinality, 20 to 1M groups: 54 faster, 1 slower, 1 same
T num_rows num_aggregations cardinality api Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
I32 20000000 1 20 normal 1.031 ms 0.66% 779.602 us 0.80% -251.135 us -24.36% 🟢 FAST
I32 20000000 2 20 normal 1.150 ms 2.04% 955.986 us 0.79% -194.118 us -16.88% 🟢 FAST
I32 20000000 4 20 normal 1.400 ms 1.38% 1.274 ms 0.55% -125.893 us -8.99% 🟢 FAST
I32 20000000 8 20 normal 1.798 ms 1.46% 1.918 ms 0.50% 120.078 us 6.68% 🔴 SLOW
I32 20000000 1 50 normal 1.118 ms 0.68% 785.541 us 0.74% -332.715 us -29.75% 🟢 FAST
I32 20000000 2 50 normal 1.274 ms 0.64% 960.716 us 0.52% -313.507 us -24.60% 🟢 FAST
I32 20000000 4 50 normal 1.532 ms 0.93% 1.304 ms 0.46% -228.112 us -14.89% 🟢 FAST
I32 20000000 8 50 normal 1.969 ms 1.10% 1.971 ms 0.39% 1.985 us 0.10% 🔵 SAME
I32 20000000 1 100 normal 1.263 ms 0.50% 796.551 us 0.67% -466.026 us -36.91% 🟢 FAST
I32 20000000 2 100 normal 1.425 ms 0.62% 969.108 us 0.52% -455.997 us -32.00% 🟢 FAST
I32 20000000 4 100 normal 1.708 ms 1.07% 1.325 ms 0.45% -383.385 us -22.44% 🟢 FAST
I32 20000000 8 100 normal 2.152 ms 0.88% 2.006 ms 0.31% -145.652 us -6.77% 🟢 FAST
I32 20000000 1 1000 normal 2.785 ms 2.24% 801.073 us 0.46% -1983.725 us -71.23% 🟢 FAST
I32 20000000 2 1000 normal 3.178 ms 4.05% 976.514 us 0.41% -2201.731 us -69.28% 🟢 FAST
I32 20000000 4 1000 normal 7.027 ms 0.22% 1.313 ms 0.36% -5714.319 us -81.31% 🟢 FAST
I32 20000000 8 1000 normal 11.665 ms 0.12% 1.981 ms 0.27% -9684.763 us -83.02% 🟢 FAST
I32 20000000 1 10000 normal 2.701 ms 0.77% 845.381 us 0.56% -1855.346 us -68.70% 🟢 FAST
I32 20000000 2 10000 normal 2.928 ms 0.80% 1.021 ms 0.35% -1906.260 us -65.11% 🟢 FAST
I32 20000000 4 10000 normal 3.679 ms 0.44% 1.346 ms 0.29% -2332.672 us -63.40% 🟢 FAST
I32 20000000 8 10000 normal 5.132 ms 0.20% 2.000 ms 0.21% -3131.985 us -61.02% 🟢 FAST
I32 20000000 1 100000 normal 2.730 ms 0.27% 1.090 ms 0.41% -1640.612 us -60.09% 🟢 FAST
I32 20000000 2 100000 normal 2.963 ms 0.26% 1.296 ms 0.35% -1666.912 us -56.26% 🟢 FAST
I32 20000000 4 100000 normal 3.265 ms 0.19% 1.695 ms 0.23% -1569.639 us -48.07% 🟢 FAST
I32 20000000 8 100000 normal 4.151 ms 0.22% 2.487 ms 0.16% -1663.882 us -40.09% 🟢 FAST
I32 20000000 1 1000000 normal 3.991 ms 0.32% 1.470 ms 0.50% -2521.151 us -63.17% 🟢 FAST
I32 20000000 2 1000000 normal 5.433 ms 0.50% 1.682 ms 0.25% -3751.341 us -69.05% 🟢 FAST
I32 20000000 4 1000000 normal 3.871 ms 0.16% 2.085 ms 0.23% -1785.625 us -46.13% 🟢 FAST
I32 20000000 8 1000000 normal 4.783 ms 0.50% 3.043 ms 0.29% -1739.311 us -36.37% 🟢 FAST
decimal128 20000000 1 20 normal 82.847 ms 0.19% 961.751 us 0.46% -81884.874 us -98.84% 🟢 FAST
decimal128 20000000 2 20 normal 83.951 ms 0.14% 1.342 ms 0.40% -82609.389 us -98.40% 🟢 FAST
decimal128 20000000 4 20 normal 85.020 ms 0.14% 2.085 ms 0.36% -82934.811 us -97.55% 🟢 FAST
decimal128 20000000 8 20 normal 87.788 ms 0.23% 3.568 ms 0.24% -84220.149 us -95.94% 🟢 FAST
decimal128 20000000 1 50 normal 85.050 ms 0.16% 974.236 us 0.55% -84076.147 us -98.85% 🟢 FAST
decimal128 20000000 2 50 normal 85.939 ms 0.16% 1.366 ms 0.47% -84572.696 us -98.41% 🟢 FAST
decimal128 20000000 4 50 normal 87.759 ms 0.24% 2.137 ms 0.30% -85622.365 us -97.57% 🟢 FAST
decimal128 20000000 8 50 normal 91.472 ms 0.18% 3.679 ms 0.28% -87793.150 us -95.98% 🟢 FAST
decimal128 20000000 1 100 normal 87.719 ms 0.16% 1.002 ms 0.60% -86716.389 us -98.86% 🟢 FAST
decimal128 20000000 2 100 normal 88.648 ms 0.17% 1.400 ms 0.38% -87248.214 us -98.42% 🟢 FAST
decimal128 20000000 4 100 normal 90.471 ms 0.29% 2.177 ms 0.28% -88293.286 us -97.59% 🟢 FAST
decimal128 20000000 8 100 normal 94.270 ms 0.10% 3.736 ms 0.18% -90534.373 us -96.04% 🟢 FAST
decimal128 20000000 1 1000 normal 94.525 ms 0.20% 1.028 ms 0.47% -93496.667 us -98.91% 🟢 FAST
decimal128 20000000 2 1000 normal 95.248 ms 0.13% 1.424 ms 0.36% -93824.578 us -98.51% 🟢 FAST
decimal128 20000000 4 1000 normal 97.034 ms 0.13% 2.200 ms 0.24% -94833.505 us -97.73% 🟢 FAST
decimal128 20000000 8 1000 normal 100.969 ms 0.11% 3.762 ms 0.26% -97206.930 us -96.27% 🟢 FAST
decimal128 20000000 1 10000 normal 96.014 ms 0.15% 1.068 ms 0.50% -94946.006 us -98.89% 🟢 FAST
decimal128 20000000 2 10000 normal 96.848 ms 0.13% 1.451 ms 0.32% -95396.862 us -98.50% 🟢 FAST
decimal128 20000000 4 10000 normal 98.707 ms 0.10% 2.210 ms 0.25% -96496.171 us -97.76% 🟢 FAST
decimal128 20000000 8 10000 normal 102.552 ms 0.09% 3.723 ms 0.17% -98829.071 us -96.37% 🟢 FAST
decimal128 20000000 1 100000 normal 89.228 ms 0.14% 1.260 ms 0.68% -87968.247 us -98.59% 🟢 FAST
decimal128 20000000 2 100000 normal 90.163 ms 0.14% 1.607 ms 0.33% -88555.971 us -98.22% 🟢 FAST
decimal128 20000000 4 100000 normal 92.171 ms 0.12% 2.307 ms 0.34% -89864.681 us -97.50% 🟢 FAST
decimal128 20000000 8 100000 normal 96.197 ms 0.12% 3.706 ms 0.29% -92491.419 us -96.15% 🟢 FAST
decimal128 20000000 1 1000000 normal 79.836 ms 0.16% 1.748 ms 0.30% -78088.364 us -97.81% 🟢 FAST
decimal128 20000000 2 1000000 normal 80.778 ms 0.15% 2.206 ms 0.21% -78571.430 us -97.27% 🟢 FAST
decimal128 20000000 4 1000000 normal 82.510 ms 0.12% 3.118 ms 0.20% -79391.902 us -96.22% 🟢 FAST
decimal128 20000000 8 1000000 normal 86.134 ms 0.18% 4.946 ms 0.20% -81187.663 us -94.26% 🟢 FAST
groupby_max_cardinality, 1 to 16 groups: 18 faster, 2 slower, 0 same
T num_rows num_aggregations cardinality api Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
I32 20000000 1 1 normal 1.053 ms 2.01% 681.977 us 1.17% -370.906 us -35.23% 🟢 FAST
I32 20000000 8 1 normal 1.663 ms 0.83% 1.387 ms 0.54% -276.584 us -16.63% 🟢 FAST
I32 20000000 1 2 normal 1.036 ms 0.64% 704.900 us 0.83% -331.244 us -31.97% 🟢 FAST
I32 20000000 8 2 normal 1.697 ms 1.21% 1.502 ms 0.44% -194.581 us -11.47% 🟢 FAST
I32 20000000 1 4 normal 1.038 ms 0.63% 730.812 us 0.58% -306.702 us -29.56% 🟢 FAST
I32 20000000 8 4 normal 1.674 ms 1.58% 1.624 ms 0.40% -49.968 us -2.99% 🟢 FAST
I32 20000000 1 8 normal 1.039 ms 0.59% 752.338 us 0.66% -286.844 us -27.60% 🟢 FAST
I32 20000000 8 8 normal 1.729 ms 1.83% 1.900 ms 0.50% 170.509 us 9.86% 🔴 SLOW
I32 20000000 1 16 normal 1.040 ms 0.75% 787.607 us 0.80% -252.517 us -24.28% 🟢 FAST
I32 20000000 8 16 normal 1.789 ms 1.48% 1.966 ms 0.33% 176.855 us 9.89% 🔴 SLOW
decimal128 20000000 1 1 normal 55.772 ms 0.20% 844.116 us 1.01% -54928.384 us -98.49% 🟢 FAST
decimal128 20000000 8 1 normal 61.195 ms 0.13% 2.629 ms 0.38% -58566.126 us -95.70% 🟢 FAST
decimal128 20000000 1 2 normal 67.851 ms 0.26% 870.708 us 0.62% -66980.316 us -98.72% 🟢 FAST
decimal128 20000000 8 2 normal 73.712 ms 0.17% 2.759 ms 0.27% -70952.491 us -96.26% 🟢 FAST
decimal128 20000000 1 4 normal 73.118 ms 0.17% 895.942 us 0.65% -72221.898 us -98.77% 🟢 FAST
decimal128 20000000 8 4 normal 79.045 ms 0.20% 3.006 ms 0.22% -76039.270 us -96.20% 🟢 FAST
decimal128 20000000 1 8 normal 78.290 ms 0.23% 936.371 us 1.36% -77353.527 us -98.80% 🟢 FAST
decimal128 20000000 8 8 normal 84.458 ms 0.22% 3.327 ms 0.50% -81131.019 us -96.06% 🟢 FAST
decimal128 20000000 1 16 normal 80.845 ms 0.21% 961.453 us 0.57% -79883.394 us -98.81% 🟢 FAST
decimal128 20000000 8 16 normal 87.248 ms 0.21% 3.532 ms 0.22% -83716.084 us -95.95% 🟢 FAST
groupby_max_cardinality, 2M to 5M groups: 8 faster, 0 slower, 0 same
T num_rows num_aggregations cardinality api Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
I32 20000000 1 2000000 normal 4.327 ms 0.62% 1.684 ms 2.12% -2642.909 us -61.08% 🟢 FAST
I32 20000000 8 2000000 normal 5.214 ms 0.20% 3.109 ms 0.28% -2105.287 us -40.38% 🟢 FAST
I32 20000000 1 5000000 normal 4.480 ms 0.29% 3.613 ms 0.50% -866.469 us -19.34% 🟢 FAST
I32 20000000 8 5000000 normal 5.757 ms 0.16% 5.068 ms 0.12% -688.981 us -11.97% 🟢 FAST
decimal128 20000000 1 2000000 normal 76.067 ms 0.12% 1.833 ms 0.25% -74233.576 us -97.59% 🟢 FAST
decimal128 20000000 8 2000000 normal 83.085 ms 0.17% 4.877 ms 0.13% -78208.109 us -94.13% 🟢 FAST
decimal128 20000000 1 5000000 normal 71.242 ms 0.11% 3.819 ms 0.15% -67423.531 us -94.64% 🟢 FAST
decimal128 20000000 8 5000000 normal 80.014 ms 0.10% 6.661 ms 0.50% -73352.895 us -91.68% 🟢 FAST
complex_int_keys: 20 faster, 0 slower, 0 same
num_cols num_rows value_key_ratio null_probability Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
1 2^24 20 0 3.340 ms 0.79% 1.271 ms 3.64% -2068.754 us -61.95% 🟢 FAST
2 2^24 20 0 3.735 ms 0.24% 1.428 ms 0.73% -2306.632 us -61.76% 🟢 FAST
4 2^24 20 0 4.422 ms 0.25% 1.823 ms 2.28% -2599.293 us -58.78% 🟢 FAST
8 2^24 20 0 6.194 ms 0.23% 2.929 ms 0.38% -3265.026 us -52.72% 🟢 FAST
16 2^24 20 0 11.311 ms 0.19% 5.328 ms 0.07% -5982.528 us -52.89% 🟢 FAST
1 2^24 200 0 1.864 ms 0.45% 845.591 us 0.42% -1017.966 us -54.62% 🟢 FAST
2 2^24 200 0 2.098 ms 0.45% 982.134 us 0.55% -1116.355 us -53.20% 🟢 FAST
4 2^24 200 0 2.797 ms 0.50% 1.262 ms 0.90% -1534.550 us -54.87% 🟢 FAST
8 2^24 200 0 4.981 ms 0.23% 1.807 ms 0.28% -3173.746 us -63.72% 🟢 FAST
16 2^24 200 0 8.991 ms 0.19% 2.452 ms 0.18% -6539.407 us -72.73% 🟢 FAST
1 2^24 20 0.5 2.297 ms 0.50% 787.183 us 0.77% -1509.751 us -65.73% 🟢 FAST
2 2^24 20 0.5 2.647 ms 0.37% 1.041 ms 1.95% -1605.547 us -60.66% 🟢 FAST
4 2^24 20 0.5 3.168 ms 0.48% 1.475 ms 0.59% -1693.258 us -53.44% 🟢 FAST
8 2^24 20 0.5 5.105 ms 0.24% 2.434 ms 0.34% -2670.819 us -52.32% 🟢 FAST
16 2^24 20 0.5 9.058 ms 0.50% 4.136 ms 0.73% -4922.456 us -54.34% 🟢 FAST
1 2^24 200 0.5 1.787 ms 0.34% 668.577 us 2.94% -1118.329 us -62.58% 🟢 FAST
2 2^24 200 0.5 2.047 ms 0.43% 883.099 us 2.37% -1164.214 us -56.87% 🟢 FAST
4 2^24 200 0.5 2.665 ms 0.28% 1.237 ms 0.63% -1427.572 us -53.57% 🟢 FAST
8 2^24 200 0.5 4.763 ms 0.64% 1.969 ms 0.50% -2793.983 us -58.67% 🟢 FAST
16 2^24 200 0.5 8.392 ms 0.30% 3.116 ms 0.25% -5275.838 us -62.87% 🟢 FAST
complex_mixed_keys: 33 faster, 7 slower, 0 same
num_cols num_rows value_key_ratio null_probability Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
1 2^18 20 0 154.960 us 10.20% 126.675 us 9.93% -28.285 us -18.25% 🟢 FAST
2 2^18 20 0 241.995 us 5.98% 210.609 us 5.34% -31.386 us -12.97% 🟢 FAST
3 2^18 20 0 306.846 us 2.71% 230.753 us 8.75% -76.092 us -24.80% 🟢 FAST
4 2^18 20 0 3.130 ms 0.25% 2.838 ms 0.96% -291.860 us -9.33% 🟢 FAST
5 2^18 20 0 3.293 ms 0.30% 2.947 ms 0.33% -346.748 us -10.53% 🟢 FAST
1 2^24 20 0 3.305 ms 0.32% 1.263 ms 0.46% -2041.437 us -61.77% 🟢 FAST
2 2^24 20 0 7.237 ms 1.46% 6.942 ms 0.98% -294.671 us -4.07% 🟢 FAST
3 2^24 20 0 8.622 ms 0.21% 8.986 ms 0.12% 363.891 us 4.22% 🔴 SLOW
4 2^24 20 0 42.136 ms 0.09% 44.717 ms 0.10% 2.582 ms 6.13% 🔴 SLOW
5 2^24 20 0 48.216 ms 0.32% 49.503 ms 0.14% 1.287 ms 2.67% 🔴 SLOW
1 2^18 200 0 142.923 us 8.11% 112.304 us 3.63% -30.619 us -21.42% 🟢 FAST
2 2^18 200 0 244.399 us 2.29% 207.918 us 2.72% -36.481 us -14.93% 🟢 FAST
3 2^18 200 0 303.164 us 2.32% 224.476 us 2.23% -78.689 us -25.96% 🟢 FAST
4 2^18 200 0 3.135 ms 0.26% 2.834 ms 0.30% -301.279 us -9.61% 🟢 FAST
5 2^18 200 0 3.297 ms 0.34% 2.946 ms 0.35% -350.655 us -10.64% 🟢 FAST
1 2^24 200 0 1.850 ms 0.51% 839.007 us 0.43% -1011.319 us -54.66% 🟢 FAST
2 2^24 200 0 6.689 ms 0.50% 6.778 ms 0.50% 88.615 us 1.32% 🔴 SLOW
3 2^24 200 0 8.352 ms 0.18% 9.026 ms 0.38% 674.519 us 8.08% 🔴 SLOW
4 2^24 200 0 42.063 ms 0.12% 44.750 ms 0.10% 2.687 ms 6.39% 🔴 SLOW
5 2^24 200 0 48.085 ms 0.22% 49.481 ms 0.11% 1.396 ms 2.90% 🔴 SLOW
1 2^18 20 0.5 222.503 us 3.02% 191.385 us 2.77% -31.118 us -13.99% 🟢 FAST
2 2^18 20 0.5 355.784 us 2.70% 319.730 us 4.32% -36.054 us -10.13% 🟢 FAST
3 2^18 20 0.5 356.929 us 2.90% 326.428 us 2.69% -30.501 us -8.55% 🟢 FAST
4 2^18 20 0.5 643.305 us 1.16% 560.553 us 1.74% -82.752 us -12.86% 🟢 FAST
5 2^18 20 0.5 1.201 ms 0.83% 1.101 ms 3.54% -99.766 us -8.31% 🟢 FAST
1 2^24 20 0.5 2.346 ms 0.39% 805.066 us 1.16% -1540.879 us -65.68% 🟢 FAST
2 2^24 20 0.5 3.244 ms 0.44% 2.236 ms 0.53% -1008.334 us -31.08% 🟢 FAST
3 2^24 20 0.5 2.631 ms 0.48% 1.876 ms 0.42% -754.268 us -28.67% 🟢 FAST
4 2^24 20 0.5 3.082 ms 0.43% 2.827 ms 0.29% -255.676 us -8.29% 🟢 FAST
5 2^24 20 0.5 3.458 ms 0.35% 3.209 ms 0.36% -248.629 us -7.19% 🟢 FAST
1 2^18 200 0.5 220.679 us 6.70% 179.353 us 4.78% -41.326 us -18.73% 🟢 FAST
2 2^18 200 0.5 354.712 us 1.55% 323.388 us 6.50% -31.324 us -8.83% 🟢 FAST
3 2^18 200 0.5 359.169 us 5.07% 320.675 us 2.13% -38.495 us -10.72% 🟢 FAST
4 2^18 200 0.5 648.821 us 1.07% 560.118 us 1.36% -88.702 us -13.67% 🟢 FAST
5 2^18 200 0.5 1.227 ms 4.48% 1.102 ms 1.04% -124.767 us -10.17% 🟢 FAST
1 2^24 200 0.5 1.805 ms 0.42% 675.942 us 0.85% -1128.796 us -62.55% 🟢 FAST
2 2^24 200 0.5 3.248 ms 0.68% 2.265 ms 1.98% -982.657 us -30.25% 🟢 FAST
3 2^24 200 0.5 2.626 ms 0.40% 1.882 ms 1.84% -743.581 us -28.32% 🟢 FAST
4 2^24 200 0.5 3.078 ms 0.33% 2.842 ms 1.31% -236.693 us -7.69% 🟢 FAST
5 2^24 200 0.5 3.452 ms 0.37% 3.249 ms 2.37% -203.313 us -5.89% 🟢 FAST
groupby_max: 31 faster, 5 slower, 0 same
T cardinality num_rows null_probability num_aggregations Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
I32 0 2^24 0 1 3.717 ms 1.63% 3.039 ms 0.50% -678.355 us -18.25% 🟢 FAST
I32 0 2^24 0.1 1 3.796 ms 0.26% 3.098 ms 0.55% -697.917 us -18.39% 🟢 FAST
I32 0 2^24 0.9 1 3.423 ms 0.36% 2.995 ms 0.34% -428.126 us -12.51% 🟢 FAST
I32 0 2^24 0 2 4.376 ms 0.27% 3.200 ms 0.13% -1175.587 us -26.86% 🟢 FAST
I32 0 2^24 0.1 2 4.539 ms 0.32% 3.350 ms 0.14% -1189.045 us -26.20% 🟢 FAST
I32 0 2^24 0.9 2 3.789 ms 0.31% 3.143 ms 0.23% -645.685 us -17.04% 🟢 FAST
I32 0 2^24 0 4 4.049 ms 0.29% 3.537 ms 0.14% -512.034 us -12.65% 🟢 FAST
I32 0 2^24 0.1 4 4.468 ms 0.25% 3.853 ms 0.16% -614.693 us -13.76% 🟢 FAST
I32 0 2^24 0.9 4 3.769 ms 0.48% 3.446 ms 0.26% -323.402 us -8.58% 🟢 FAST
I32 0 2^24 0 8 4.860 ms 0.22% 4.221 ms 0.13% -638.119 us -13.13% 🟢 FAST
I32 0 2^24 0.1 8 5.698 ms 0.40% 4.871 ms 0.50% -826.607 us -14.51% 🟢 FAST
I32 0 2^24 0.9 8 4.251 ms 0.50% 4.049 ms 0.17% -201.955 us -4.75% 🟢 FAST
I32 0 2^24 0 16 6.533 ms 0.21% 5.597 ms 0.10% -936.321 us -14.33% 🟢 FAST
I32 0 2^24 0.1 16 8.209 ms 0.22% 6.905 ms 0.26% -1303.738 us -15.88% 🟢 FAST
I32 0 2^24 0.9 16 5.240 ms 0.28% 5.252 ms 0.17% 11.684 us 0.22% 🔴 SLOW
I32 0 2^24 0 32 9.830 ms 0.22% 8.327 ms 0.06% -1503.168 us -15.29% 🟢 FAST
I32 0 2^24 0.1 32 13.158 ms 0.21% 10.957 ms 0.12% -2201.136 us -16.73% 🟢 FAST
I32 0 2^24 0.9 32 7.129 ms 0.16% 7.652 ms 0.16% 522.662 us 7.33% 🔴 SLOW
F64 0 2^24 0 1 3.983 ms 0.25% 3.204 ms 0.50% -778.730 us -19.55% 🟢 FAST
F64 0 2^24 0.1 1 4.099 ms 0.24% 3.160 ms 0.17% -938.696 us -22.90% 🟢 FAST
F64 0 2^24 0.9 1 3.544 ms 0.28% 3.054 ms 0.16% -490.820 us -13.85% 🟢 FAST
F64 0 2^24 0 2 4.928 ms 0.26% 3.525 ms 0.18% -1403.742 us -28.48% 🟢 FAST
F64 0 2^24 0.1 2 5.067 ms 0.29% 3.455 ms 0.53% -1612.264 us -31.82% 🟢 FAST
F64 0 2^24 0.9 2 4.007 ms 0.20% 3.280 ms 0.16% -727.242 us -18.15% 🟢 FAST
F64 0 2^24 0 4 4.889 ms 0.23% 4.179 ms 0.15% -710.446 us -14.53% 🟢 FAST
F64 0 2^24 0.1 4 5.219 ms 0.24% 4.047 ms 0.25% -1171.470 us -22.45% 🟢 FAST
F64 0 2^24 0.9 4 3.896 ms 0.32% 3.719 ms 0.16% -176.901 us -4.54% 🟢 FAST
F64 0 2^24 0 8 6.584 ms 0.16% 5.475 ms 0.50% -1109.569 us -16.85% 🟢 FAST
F64 0 2^24 0.1 8 7.200 ms 0.23% 5.236 ms 0.13% -1963.673 us -27.27% 🟢 FAST
F64 0 2^24 0.9 8 4.515 ms 0.32% 4.642 ms 0.46% 126.694 us 2.81% 🔴 SLOW
F64 0 2^24 0 16 9.982 ms 0.27% 8.065 ms 0.11% -1917.493 us -19.21% 🟢 FAST
F64 0 2^24 0.1 16 11.206 ms 0.35% 7.601 ms 0.13% -3604.794 us -32.17% 🟢 FAST
F64 0 2^24 0.9 16 5.738 ms 0.23% 6.444 ms 0.15% 705.902 us 12.30% 🔴 SLOW
F64 0 2^24 0 32 16.862 ms 0.50% 13.261 ms 0.15% -3601.064 us -21.36% 🟢 FAST
F64 0 2^24 0.1 32 19.403 ms 0.50% 12.367 ms 0.13% -7036.192 us -36.26% 🟢 FAST
F64 0 2^24 0.9 32 8.217 ms 0.50% 10.546 ms 0.30% 2.329 ms 28.34% 🔴 SLOW
groupby_struct_keys: 18 faster, 0 slower, 0 same
NumRows Depth Nulls Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
2^10 0 0 121.149 us 10.25% 105.327 us 6.52% -15.822 us -13.06% 🟢 FAST
2^16 0 0 131.008 us 5.83% 107.466 us 4.01% -23.542 us -17.97% 🟢 FAST
2^20 0 0 189.628 us 2.65% 142.875 us 2.61% -46.753 us -24.66% 🟢 FAST
2^10 1 0 261.206 us 2.33% 249.380 us 2.40% -11.827 us -4.53% 🟢 FAST
2^16 1 0 281.101 us 2.51% 257.570 us 2.63% -23.531 us -8.37% 🟢 FAST
2^20 1 0 456.862 us 3.24% 370.005 us 1.74% -86.857 us -19.01% 🟢 FAST
2^10 8 0 997.070 us 1.11% 951.140 us 0.84% -45.930 us -4.61% 🟢 FAST
2^16 8 0 1.008 ms 1.26% 963.634 us 1.14% -44.000 us -4.37% 🟢 FAST
2^20 8 0 1.227 ms 1.06% 1.078 ms 0.94% -149.286 us -12.17% 🟢 FAST
2^10 0 1 159.324 us 3.28% 149.436 us 3.43% -9.888 us -6.21% 🟢 FAST
2^16 0 1 164.664 us 2.60% 151.402 us 3.04% -13.262 us -8.05% 🟢 FAST
2^20 0 1 233.623 us 5.12% 194.250 us 2.48% -39.373 us -16.85% 🟢 FAST
2^10 1 1 260.933 us 2.39% 249.189 us 2.60% -11.744 us -4.50% 🟢 FAST
2^16 1 1 284.515 us 4.25% 258.582 us 3.18% -25.934 us -9.12% 🟢 FAST
2^20 1 1 457.147 us 2.74% 374.030 us 1.48% -83.116 us -18.18% 🟢 FAST
2^10 8 1 994.232 us 1.07% 945.874 us 2.30% -48.357 us -4.86% 🟢 FAST
2^16 8 1 1.018 ms 1.84% 964.457 us 1.18% -54.009 us -5.30% 🟢 FAST
2^20 8 1 1.230 ms 1.16% 1.074 ms 0.86% -155.961 us -12.68% 🟢 FAST
groupby_m2_var_std, default axes: 48 faster, 0 slower, 0 same
T U value_key_ratio num_rows null_probability num_aggs Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
I32 12 20 100000 0 1 186.794 us 6.27% 160.707 us 9.45% -26.086 us -13.97% 🟢 FAST
I32 12 100 100000 0 1 174.280 us 3.65% 154.582 us 6.57% -19.698 us -11.30% 🟢 FAST
I32 12 20 10000000 0 1 1.250 ms 0.64% 626.818 us 3.07% -623.454 us -49.87% 🟢 FAST
I32 12 100 10000000 0 1 1.124 ms 0.62% 499.454 us 2.80% -625.028 us -55.58% 🟢 FAST
I32 12 20 100000 0.5 1 187.310 us 2.64% 161.921 us 2.93% -25.390 us -13.55% 🟢 FAST
I32 12 100 100000 0.5 1 175.431 us 3.35% 160.142 us 2.56% -15.290 us -8.72% 🟢 FAST
I32 12 20 10000000 0.5 1 1.154 ms 0.60% 624.458 us 0.74% -529.153 us -45.87% 🟢 FAST
I32 12 100 10000000 0.5 1 1.016 ms 0.70% 505.223 us 0.90% -510.565 us -50.26% 🟢 FAST
I32 12 20 100000 0 10 714.778 us 4.04% 594.791 us 1.18% -119.988 us -16.79% 🟢 FAST
I32 12 100 100000 0 10 630.941 us 4.16% 558.821 us 1.31% -72.120 us -11.43% 🟢 FAST
I32 12 20 10000000 0 10 4.788 ms 0.50% 1.895 ms 0.92% -2893.354 us -60.43% 🟢 FAST
I32 12 100 10000000 0 10 4.478 ms 0.48% 1.430 ms 1.68% -3047.924 us -68.06% 🟢 FAST
I32 12 20 100000 0.5 10 777.527 us 8.97% 620.617 us 0.97% -156.909 us -20.18% 🟢 FAST
I32 12 100 100000 0.5 10 701.655 us 11.61% 601.678 us 1.93% -99.976 us -14.25% 🟢 FAST
I32 12 20 10000000 0.5 10 3.788 ms 0.36% 1.957 ms 0.50% -1831.277 us -48.34% 🟢 FAST
I32 12 100 10000000 0.5 10 3.448 ms 0.95% 1.510 ms 2.50% -1938.259 us -56.21% 🟢 FAST
I32 12 20 100000 0 50 3.051 ms 0.52% 2.546 ms 1.71% -504.582 us -16.54% 🟢 FAST
I32 12 100 100000 0 50 2.619 ms 1.04% 2.301 ms 0.66% -318.243 us -12.15% 🟢 FAST
I32 12 20 10000000 0 50 20.444 ms 0.17% 7.511 ms 0.27% -12933.268 us -63.26% 🟢 FAST
I32 12 100 10000000 0 50 19.458 ms 0.36% 5.510 ms 0.27% -13948.042 us -71.68% 🟢 FAST
I32 12 20 100000 0.5 50 3.199 ms 1.19% 2.666 ms 0.99% -532.914 us -16.66% 🟢 FAST
I32 12 100 100000 0.5 50 2.745 ms 0.73% 2.631 ms 1.76% -113.669 us -4.14% 🟢 FAST
I32 12 20 10000000 0.5 50 15.412 ms 0.50% 7.946 ms 0.81% -7466.478 us -48.44% 🟢 FAST
I32 12 100 10000000 0.5 50 14.144 ms 0.17% 5.840 ms 0.27% -8304.323 us -58.71% 🟢 FAST
F64 12 20 100000 0 1 191.239 us 2.39% 165.382 us 2.44% -25.857 us -13.52% 🟢 FAST
F64 12 100 100000 0 1 179.834 us 2.97% 162.255 us 2.89% -17.579 us -9.78% 🟢 FAST
F64 12 20 10000000 0 1 1.269 ms 0.53% 706.883 us 0.70% -562.409 us -44.31% 🟢 FAST
F64 12 100 10000000 0 1 1.137 ms 0.64% 580.814 us 0.72% -555.723 us -48.90% 🟢 FAST
F64 12 20 100000 0.5 1 197.215 us 3.55% 166.821 us 2.48% -30.394 us -15.41% 🟢 FAST
F64 12 100 100000 0.5 1 186.692 us 3.46% 166.977 us 2.70% -19.715 us -10.56% 🟢 FAST
F64 12 20 10000000 0.5 1 1.170 ms 0.61% 697.594 us 0.63% -472.659 us -40.39% 🟢 FAST
F64 12 100 10000000 0.5 1 1.030 ms 0.89% 547.405 us 1.05% -482.651 us -46.86% 🟢 FAST
F64 12 20 100000 0 10 809.880 us 2.05% 660.009 us 1.33% -149.871 us -18.51% 🟢 FAST
F64 12 100 100000 0 10 720.840 us 2.08% 649.401 us 1.11% -71.438 us -9.91% 🟢 FAST
F64 12 20 10000000 0 10 4.950 ms 0.29% 2.738 ms 0.34% -2212.160 us -44.69% 🟢 FAST
F64 12 100 10000000 0 10 4.593 ms 0.31% 2.296 ms 0.48% -2296.588 us -50.00% 🟢 FAST
F64 12 20 100000 0.5 10 839.168 us 1.55% 686.650 us 1.03% -152.518 us -18.17% 🟢 FAST
F64 12 100 100000 0.5 10 744.561 us 1.73% 699.412 us 2.10% -45.149 us -6.06% 🟢 FAST
F64 12 20 10000000 0.5 10 3.920 ms 0.25% 2.597 ms 0.39% -1322.853 us -33.74% 🟢 FAST
F64 12 100 10000000 0.5 10 3.561 ms 0.40% 1.995 ms 0.46% -1565.797 us -43.97% 🟢 FAST
F64 12 20 100000 0 50 3.499 ms 1.81% 2.845 ms 1.07% -654.225 us -18.70% 🟢 FAST
F64 12 100 100000 0 50 3.041 ms 1.93% 2.795 ms 5.15% -245.744 us -8.08% 🟢 FAST
F64 12 20 10000000 0 50 21.334 ms 0.44% 11.833 ms 0.28% -9500.360 us -44.53% 🟢 FAST
F64 12 100 10000000 0 50 19.940 ms 0.22% 10.011 ms 1.13% -9929.571 us -49.80% 🟢 FAST
F64 12 20 100000 0.5 50 3.685 ms 0.93% 3.025 ms 3.69% -660.328 us -17.92% 🟢 FAST
F64 12 100 100000 0.5 50 3.187 ms 0.90% 3.035 ms 1.82% -151.887 us -4.77% 🟢 FAST
F64 12 20 10000000 0.5 50 16.030 ms 0.15% 11.045 ms 0.36% -4984.915 us -31.10% 🟢 FAST
F64 12 100 10000000 0.5 50 14.717 ms 0.22% 8.427 ms 1.40% -6289.485 us -42.74% 🟢 FAST
groupby_m2_var_std, 32 to 1000 rows per group: 56 faster, 0 slower, 0 same
T U value_key_ratio num_rows null_probability num_aggs Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
I32 12 32 10000000 0 1 1.171 ms 1.38% 579.191 us 2.65% -591.686 us -50.53% 🟢 FAST
I32 12 50 10000000 0 1 1.139 ms 0.51% 545.784 us 1.97% -593.706 us -52.10% 🟢 FAST
I32 12 64 10000000 0 1 1.133 ms 0.61% 531.030 us 1.43% -601.715 us -53.12% 🟢 FAST
I32 12 100 10000000 0 1 1.122 ms 0.62% 498.915 us 2.67% -622.868 us -55.52% 🟢 FAST
I32 12 200 10000000 0 1 1.106 ms 0.57% 473.594 us 3.66% -631.966 us -57.16% 🟢 FAST
I32 12 500 10000000 0 1 1.147 ms 0.48% 426.998 us 2.15% -720.084 us -62.78% 🟢 FAST
I32 12 1000 10000000 0 1 1.186 ms 1.33% 418.391 us 2.93% -767.481 us -64.72% 🟢 FAST
I32 12 32 10000000 0.5 1 1.073 ms 0.51% 581.830 us 0.82% -491.131 us -45.77% 🟢 FAST
I32 12 50 10000000 0.5 1 1.039 ms 1.20% 549.611 us 0.91% -489.088 us -47.09% 🟢 FAST
I32 12 64 10000000 0.5 1 1.033 ms 0.66% 537.734 us 1.52% -494.897 us -47.93% 🟢 FAST
I32 12 100 10000000 0.5 1 1.017 ms 0.72% 508.862 us 2.15% -507.880 us -49.95% 🟢 FAST
I32 12 200 10000000 0.5 1 1.005 ms 0.63% 477.605 us 1.64% -527.418 us -52.48% 🟢 FAST
I32 12 500 10000000 0.5 1 1.019 ms 0.83% 422.756 us 1.43% -595.911 us -58.50% 🟢 FAST
I32 12 1000 10000000 0.5 1 1.084 ms 0.53% 413.208 us 4.21% -671.218 us -61.90% 🟢 FAST
I32 12 32 10000000 0 10 4.541 ms 0.22% 1.618 ms 0.50% -2923.001 us -64.36% 🟢 FAST
I32 12 50 10000000 0 10 4.490 ms 0.50% 1.507 ms 0.50% -2983.150 us -66.44% 🟢 FAST
I32 12 64 10000000 0 10 4.473 ms 0.28% 1.481 ms 0.50% -2991.806 us -66.89% 🟢 FAST
I32 12 100 10000000 0 10 4.487 ms 0.19% 1.424 ms 0.66% -3062.979 us -68.26% 🟢 FAST
I32 12 200 10000000 0 10 4.597 ms 0.25% 1.527 ms 0.50% -3069.931 us -66.78% 🟢 FAST
I32 12 500 10000000 0 10 4.975 ms 0.24% 1.323 ms 0.71% -3652.037 us -73.41% 🟢 FAST
I32 12 1000 10000000 0 10 5.649 ms 0.70% 1.275 ms 1.66% -4373.617 us -77.42% 🟢 FAST
I32 12 32 10000000 0.5 10 3.567 ms 0.26% 1.686 ms 0.63% -1881.070 us -52.74% 🟢 FAST
I32 12 50 10000000 0.5 10 3.493 ms 0.34% 1.587 ms 0.60% -1905.622 us -54.56% 🟢 FAST
I32 12 64 10000000 0.5 10 3.476 ms 0.39% 1.536 ms 0.58% -1939.713 us -55.81% 🟢 FAST
I32 12 100 10000000 0.5 10 3.460 ms 0.35% 1.517 ms 2.38% -1943.391 us -56.17% 🟢 FAST
I32 12 200 10000000 0.5 10 3.505 ms 0.21% 1.567 ms 2.12% -1937.713 us -55.29% 🟢 FAST
I32 12 500 10000000 0.5 10 3.674 ms 0.79% 1.277 ms 3.53% -2396.259 us -65.23% 🟢 FAST
I32 12 1000 10000000 0.5 10 3.956 ms 0.97% 1.211 ms 1.55% -2744.483 us -69.38% 🟢 FAST
F64 12 32 10000000 0 1 1.192 ms 0.51% 647.735 us 1.94% -543.949 us -45.65% 🟢 FAST
F64 12 50 10000000 0 1 1.158 ms 1.04% 630.115 us 3.45% -527.639 us -45.57% 🟢 FAST
F64 12 64 10000000 0 1 1.149 ms 0.53% 617.447 us 1.95% -531.955 us -46.28% 🟢 FAST
F64 12 100 10000000 0 1 1.137 ms 0.39% 593.743 us 1.87% -542.822 us -47.76% 🟢 FAST
F64 12 200 10000000 0 1 1.127 ms 0.61% 588.834 us 4.15% -538.386 us -47.76% 🟢 FAST
F64 12 500 10000000 0 1 1.165 ms 0.53% 498.708 us 3.11% -666.555 us -57.20% 🟢 FAST
F64 12 1000 10000000 0 1 1.283 ms 1.41% 490.997 us 6.78% -792.079 us -61.73% 🟢 FAST
F64 12 32 10000000 0.5 1 1.087 ms 0.52% 630.771 us 2.26% -456.697 us -42.00% 🟢 FAST
F64 12 50 10000000 0.5 1 1.054 ms 0.59% 594.562 us 1.95% -459.246 us -43.58% 🟢 FAST
F64 12 64 10000000 0.5 1 1.051 ms 0.70% 579.340 us 1.86% -471.831 us -44.89% 🟢 FAST
F64 12 100 10000000 0.5 1 1.033 ms 0.79% 553.515 us 1.87% -479.795 us -46.43% 🟢 FAST
F64 12 200 10000000 0.5 1 1.023 ms 0.64% 578.885 us 1.52% -443.745 us -43.39% 🟢 FAST
F64 12 500 10000000 0.5 1 1.027 ms 0.84% 488.695 us 5.23% -538.389 us -52.42% 🟢 FAST
F64 12 1000 10000000 0.5 1 1.045 ms 0.50% 460.389 us 5.87% -584.527 us -55.94% 🟢 FAST
F64 12 32 10000000 0 10 4.678 ms 0.25% 2.428 ms 2.30% -2249.312 us -48.09% 🟢 FAST
F64 12 50 10000000 0 10 4.627 ms 0.18% 2.354 ms 1.89% -2272.400 us -49.12% 🟢 FAST
F64 12 64 10000000 0 10 4.603 ms 0.22% 2.328 ms 0.53% -2275.034 us -49.43% 🟢 FAST
F64 12 100 10000000 0 10 4.623 ms 0.21% 2.302 ms 0.33% -2321.549 us -50.21% 🟢 FAST
F64 12 200 10000000 0 10 4.680 ms 0.17% 2.667 ms 0.28% -2013.148 us -43.02% 🟢 FAST
F64 12 500 10000000 0 10 5.155 ms 0.19% 2.013 ms 0.46% -3142.420 us -60.96% 🟢 FAST
F64 12 1000 10000000 0 10 5.779 ms 0.46% 1.970 ms 0.37% -3809.054 us -65.91% 🟢 FAST
F64 12 32 10000000 0.5 10 3.671 ms 0.24% 2.181 ms 0.40% -1489.885 us -40.59% 🟢 FAST
F64 12 50 10000000 0.5 10 3.614 ms 0.26% 2.045 ms 0.36% -1568.910 us -43.41% 🟢 FAST
F64 12 64 10000000 0.5 10 3.600 ms 0.31% 2.033 ms 0.43% -1567.141 us -43.53% 🟢 FAST
F64 12 100 10000000 0.5 10 3.584 ms 0.31% 2.002 ms 0.39% -1582.186 us -44.14% 🟢 FAST
F64 12 200 10000000 0.5 10 3.599 ms 0.76% 2.642 ms 0.32% -956.691 us -26.58% 🟢 FAST
F64 12 500 10000000 0.5 10 3.835 ms 2.04% 1.836 ms 0.39% -1999.688 us -52.14% 🟢 FAST
F64 12 1000 10000000 0.5 10 4.082 ms 0.24% 1.614 ms 0.48% -2467.651 us -60.45% 🟢 FAST
sum: 7 faster, 1 slower, 0 same
T num_rows Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
I64 100000 138.087 us 4.84% 118.344 us 7.06% -19.743 us -14.30% 🟢 FAST
I64 1000000 201.953 us 2.90% 144.028 us 10.34% -57.925 us -28.68% 🟢 FAST
I64 10000000 794.452 us 4.45% 532.425 us 4.01% -262.027 us -32.98% 🟢 FAST
I64 100000000 5.519 ms 0.50% 4.441 ms 0.16% -1077.376 us -19.52% 🟢 FAST
decimal64 100000 142.626 us 6.32% 126.947 us 3.13% -15.679 us -10.99% 🟢 FAST
decimal64 1000000 346.463 us 1.59% 240.369 us 1.79% -106.094 us -30.62% 🟢 FAST
decimal64 10000000 2.378 ms 0.41% 1.763 ms 0.24% -614.597 us -25.84% 🟢 FAST
decimal64 100000000 26.281 ms 0.11% 29.731 ms 0.13% 3.451 ms 13.13% 🔴 SLOW
no_requests: 3 faster, 0 slower, 1 same
num_rows Ref Time Ref Noise Cmp Time Cmp Noise Diff %Diff Status
100000 48.707 us 14.64% 48.157 us 10.98% -0.550 us -1.13% 🔵 SAME
1000000 94.691 us 9.48% 56.921 us 3.33% -37.770 us -39.89% 🟢 FAST
10000000 336.131 us 1.01% 173.423 us 2.54% -162.708 us -48.41% 🟢 FAST
100000000 2.826 ms 0.29% 917.864 us 0.61% -1907.940 us -67.52% 🟢 FAST

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

Reviewers

@ttnghia ttnghia Awaiting requested review from ttnghia ttnghia is a code owner automatically assigned from NVIDIA/cudf-cmake-codeowners
@lamarrr lamarrr Awaiting requested review from lamarrr lamarrr is a code owner automatically assigned from NVIDIA/cudf-cpp-codeowners

At least 1 approving review is required to merge this pull request.

Assignees

@PointKernel PointKernel

Labels

2 - In Progress Currently a work in progress CMake CMake build issue improvement Improvement / enhancement to an existing function libcudf Affects libcudf (C++/CUDA) code. non-breaking Non-breaking change Performance Performance related issue

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@PointKernel