bench: compare uniform and per-chunk bitpacking

[mhk197]

Add a benchmark sweep comparing global-width and per-chunk encoding and decoding across uniform, alternating, drifting, and outlier-heavy distributions. Report packed sizes alongside timing to make compression and runtime tradeoffs visible.

Part 9/9 of the bitpacked-v2 stack (260 added + removed lines). Merge in order; each draft targets the preceding branch, starting from develop.

1 → 2 → 3 → 4 → 5 → 6 → 7 → 8 → 9

Validation: After restacking, the complete stack passes 435 FastLanes/BtrBlocks tests (1 skipped), 4 FastLanes doctests, focused Clippy with --all-targets --all-features -- -D warnings, and cargo +nightly fmt --all -- --check. Focused serde/layout tests also pass on PRs 3, 5, and 6 (10, 17, and 28 tests respectively); git diff --check passes throughout the stack.

GPU execution was not tested. CUDA validation and workspace-wide all-feature Clippy were previously blocked by the nvCOMP 5.1 SDK download in this environment.

The benchmark smoke test (--profile dev -- --test) passed before the serde extraction; this follow-up reorganizes the plugins and preserves v1 validation behavior.

[codspeed]

Merging this PR will degrade performance by 0.43%

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 3 improved benchmarks
❌ 2 regressed benchmarks
✅ 2203 untouched benchmarks
🆕 160 new benchmarks
⏩ 287 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	take_fsl_u32_random[256, 10]	124.5 µs	159.6 µs	-21.97%
❌	WallTime	dbp_split_kernel_mixed_null_avx2[(I128, 1024)]	1.1 µs	1.3 µs	-11.5%
⚡	Simulation	take_fsl_u32_random[128, 100]	237.4 µs	208.2 µs	+14%
⚡	Simulation	take_fsl_f16_random[256, 100]	230.8 µs	204.8 µs	+12.67%
⚡	Simulation	take_fsl_u64_random[64, 10]	124.4 µs	112.7 µs	+10.33%
🆕	Simulation	compress_v1[u16, drift]	N/A	295.7 µs	N/A
🆕	Simulation	compress_v1[u16, drift+exc1%]	N/A	832.6 µs	N/A
🆕	Simulation	compress_v1[u16, drift+null10%]	N/A	523.5 µs	N/A
🆕	Simulation	compress_v1[u16, random]	N/A	300.3 µs	N/A
🆕	Simulation	compress_v1[u16, spiky]	N/A	888.4 µs	N/A
🆕	Simulation	compress_v1[u16, spiky+exc1%]	N/A	895.2 µs	N/A
🆕	Simulation	compress_v1[u16, uniform]	N/A	272.4 µs	N/A
🆕	Simulation	compress_v1[u16, uniform+exc1%]	N/A	814.8 µs	N/A
🆕	Simulation	compress_v1[u16, uniform+null10%]	N/A	500.2 µs	N/A
🆕	Simulation	compress_v1[u16, zero_heavy]	N/A	272.5 µs	N/A
🆕	Simulation	compress_v1[u32, drift]	N/A	432.1 µs	N/A
🆕	Simulation	compress_v1[u32, drift+exc1%]	N/A	992.4 µs	N/A
🆕	Simulation	compress_v1[u32, drift+null10%]	N/A	1.5 ms	N/A
🆕	Simulation	compress_v1[u32, random]	N/A	432 µs	N/A
🆕	Simulation	compress_v1[u32, spiky]	N/A	1 ms	N/A
...	...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing mk/bitpacked-stack-09-benchmarks (545cb8a) with mk/bitpacked-stack-08-fused-encoder (f2104c5)}

Footnotes

1. 287 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩