feat[array]: Patched array remove GPU transpose and wire up sparse#8127
Draft
joseph-isaacs wants to merge 6 commits into
Draft
feat[array]: Patched array remove GPU transpose and wire up sparse#8127joseph-isaacs wants to merge 6 commits into
joseph-isaacs wants to merge 6 commits into
Conversation
joseph-isaacs
changed the title
joseph-isaacs
force-pushed
the
claude/add-transpose-metadata-8iZD0
branch
from
da1a0a9 to
9a73899
Compare
joseph-isaacs
added
changelog/feature
github-actions
Bot
removed
the
action/benchmark
Remove the eager GPU lane transpose from the Patched array. Patches are now kept in their natural sorted layout with one chunk offset per 1024-element chunk, mirroring the Patches helper, and execute/slice/take/compare/filter reuse the existing untransposed patch machinery. The transpose count (n_lanes) is retained as metadata so the data-parallel GPU transpose can be re-added in the future without a format change. Signed-off-by: Claude <noreply@anthropic.com>
Add SparsePatchedPlugin which, when the experimental Patched encoding is enabled, deserializes a primitive Sparse array as a Patched array over a constant fill. A Sparse array is logically patches on top of a constant, so this reuses the same externalization shim already used for BitPacked and ALP. Non-primitive (bool/varbin/struct/fixed-size-list) and nullable-patch sparse arrays remain Sparse, since Patched only represents primitive inners with non-null patch values. Signed-off-by: Claude <noreply@anthropic.com>
Microbenchmark isolating loop ordering for patched bitpacked decode: unpack every 1024-block then scatter all patches, versus unpacking one block and patching it while still hot in cache. Both do identical work, so any delta is cache locality. Fused wins 20-40% once the output exceeds L2, peaking around moderate patch density. Signed-off-by: Claude <noreply@anthropic.com>
Signed-off-by: Joe Isaacs <joe.isaacs@live.co.uk>
joseph-isaacs
changed the title
joseph-isaacs
force-pushed
the
claude/add-transpose-metadata-8iZD0
branch
from
9a73899 to
b38b783
Compare
Contributor
Polar Signals Profiling ResultsLatest Run
Previous Runs (3)
Powered by Polar Signals Cloud |
Contributor
Benchmarks: TPC-H SF=1 on NVMEVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (0.926x ➖, 7↑ 0↓)
datafusion / vortex-compact (0.946x ➖, 1↑ 0↓)
datafusion / parquet (0.922x ➖, 6↑ 0↓)
datafusion / arrow (0.913x ➖, 7↑ 0↓)
duckdb / vortex-file-compressed (0.991x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.002x ➖, 0↑ 0↓)
duckdb / parquet (0.980x ➖, 2↑ 1↓)
duckdb / duckdb (0.998x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: TPC-H SF=1 on NVMEFile Size Changes (3 files changed, -0.0% overall, 0↑ 3↓)
Totals:
|
Contributor
Benchmarks: FineWeb NVMeVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (1.024x ➖, 0↑ 0↓)
datafusion / vortex-compact (1.038x ➖, 0↑ 2↓)
datafusion / parquet (1.003x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (0.966x ➖, 2↑ 0↓)
duckdb / vortex-compact (1.002x ➖, 0↑ 1↓)
duckdb / parquet (0.996x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: FineWeb NVMeFile Size Changes (1 files changed, -0.0% overall, 0↑ 1↓)
Totals:
|
Contributor
Benchmarks: PolarSignals ProfilingVortex (geomean): 0.822x ✅ datafusion / vortex-file-compressed (0.822x ✅, 9↑ 0↓)
|
Contributor
File Sizes: PolarSignals ProfilingNo file size changes detected. |
joseph-isaacs
added
the
action/benchmark
github-actions
Bot
removed
the
action/benchmark
Contributor
Benchmarks: TPC-DS SF=1 on NVMEVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (0.994x ➖, 3↑ 1↓)
datafusion / vortex-compact (0.998x ➖, 3↑ 2↓)
datafusion / parquet (1.009x ➖, 2↑ 2↓)
duckdb / vortex-file-compressed (1.003x ➖, 1↑ 3↓)
duckdb / vortex-compact (1.004x ➖, 3↑ 4↓)
duckdb / parquet (1.016x ➖, 1↑ 1↓)
duckdb / duckdb (1.004x ➖, 1↑ 1↓)
Full attributed analysis
|
Contributor
File Sizes: TPC-DS SF=1 on NVMEFile Size Changes (11 files changed, -0.5% overall, 0↑ 11↓)
Totals:
|
joseph-isaacs
added
the
action/benchmark
github-actions
Bot
removed
the
action/benchmark
Contributor
Benchmarks: FineWeb S3Verdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (1.118x ➖, 0↑ 2↓)
datafusion / vortex-compact (0.989x ➖, 1↑ 0↓)
datafusion / parquet (1.035x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (0.943x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.088x ➖, 0↑ 1↓)
duckdb / parquet (1.049x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
Benchmarks: Random AccessVortex (geomean): 0.935x ➖ unknown / unknown (0.982x ➖, 1↑ 0↓)
|
Contributor
Benchmarks: Statistical and Population GeneticsVerdict: No clear signal (low confidence) duckdb / vortex-file-compressed (1.007x ➖, 2↑ 2↓)
duckdb / vortex-compact (0.988x ➖, 0↑ 0↓)
duckdb / parquet (0.995x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: Statistical and Population GeneticsFile Size Changes (2 files changed, -6.0% overall, 1↑ 1↓)
Totals:
|
Contributor
Benchmarks: Clickbench on NVMEVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (1.003x ➖, 2↑ 0↓)
datafusion / parquet (1.028x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (1.030x ➖, 1↑ 8↓)
duckdb / parquet (0.995x ➖, 0↑ 0↓)
duckdb / duckdb (1.010x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: Clickbench on NVMEFile Size Changes (184 files changed, -0.5% overall, 40↑ 144↓)
Totals:
|
Contributor
Benchmarks: TPC-H SF=1 on S3Verdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (1.103x ➖, 0↑ 5↓)
datafusion / vortex-compact (1.084x ➖, 1↑ 4↓)
datafusion / parquet (0.894x ➖, 5↑ 3↓)
duckdb / vortex-file-compressed (1.051x ➖, 0↑ 1↓)
duckdb / vortex-compact (1.092x ➖, 0↑ 0↓)
duckdb / parquet (1.078x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
Benchmarks: TPC-H SF=10 on NVMEVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (0.999x ➖, 0↑ 0↓)
datafusion / vortex-compact (1.004x ➖, 0↑ 0↓)
datafusion / parquet (0.996x ➖, 0↑ 0↓)
datafusion / arrow (0.993x ➖, 1↑ 0↓)
duckdb / vortex-file-compressed (1.002x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.009x ➖, 0↑ 0↓)
duckdb / parquet (0.994x ➖, 0↑ 0↓)
duckdb / duckdb (1.004x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: TPC-H SF=10 on NVMEFile Size Changes (19 files changed, -0.0% overall, 1↑ 18↓)
Totals:
|
Contributor
Benchmarks: CompressionVortex (geomean): 0.995x ➖ unknown / unknown (0.997x ➖, 2↑ 2↓)
|
Contributor
Benchmarks: TPC-H SF=10 on S3Verdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (1.052x ➖, 0↑ 0↓)
datafusion / vortex-compact (1.114x ➖, 0↑ 2↓)
datafusion / parquet (1.045x ➖, 1↑ 5↓)
duckdb / vortex-file-compressed (1.135x ➖, 0↑ 1↓)
duckdb / vortex-compact (1.065x ➖, 0↑ 0↓)
duckdb / parquet (1.136x ➖, 0↑ 0↓)
Full attributed analysis
|
joseph-isaacs
added
the
action/benchmark
github-actions
Bot
removed
the
action/benchmark
Merging this PR will not alter performance
|
| Mode | Benchmark | BASE |
HEAD |
Efficiency | |
|---|---|---|---|---|---|
| ⚡ | Simulation | chunked_varbinview_opt_into_canonical[(1000, 10)] |
240.1 µs | 202.4 µs | +18.66% |
| ⚡ | Simulation | chunked_varbinview_opt_canonical_into[(100, 100)] |
412.8 µs | 374.1 µs | +10.35% |
| ❌ | Simulation | chunked_varbinview_opt_canonical_into[(1000, 10)] |
187.8 µs | 225.2 µs | -16.59% |
| 🆕 | Simulation | unpack_then_patch_bench[(65536, 9, 20)] |
N/A | 202.3 µs | N/A |
| 🆕 | Simulation | unpack_then_patch_bench[(65536, 9, 5)] |
N/A | 292.7 µs | N/A |
| 🆕 | Simulation | fused_bench[(65536, 9, 20)] |
N/A | 200.2 µs | N/A |
| 🆕 | Simulation | unpack_then_patch_bench[(65536, 9, 200)] |
N/A | 171.4 µs | N/A |
| 🆕 | Simulation | fused_bench[(65536, 9, 200)] |
N/A | 171.8 µs | N/A |
| 🆕 | Simulation | fused_bench[(65536, 9, 5)] |
N/A | 295 µs | N/A |
Tip
Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.
Comparing claude/add-transpose-metadata-8iZD0 (79363ef) with develop (a2323f1)
joseph-isaacs
added
the
action/benchmark
github-actions
Bot
removed
the
action/benchmark
joseph-isaacs
changed the title
Contributor
🚨🚨🚨❌❌❌ SQL BENCHMARK FAILED ❌❌❌🚨🚨🚨Benchmark |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Removes the CUDA oriented patches transpose. From the PatchedArray.
Open question should be have a PatchesChildArray to enable easy reduce pushdown to the inner and patches child?