bluuewhale/hash-smith v0.2.0

on GitHub

What's Changed

• feat: benchmark infrastructure — per-structure tasks, fast mode, JSON persistence, regression compare by @bluuewhale in #14
• perf(SwissMap): exp-001 optimization — 13–33% improvement across all benchmarks by @bluuewhale in #15

Detailed Change Logs

Added

• Benchmark infrastructure (#14): Added per-structure Gradle tasks (jmhSwissMap, jmhConcurrent, jmhSimd) and jmhFast mode (reduced fork/warmup/iteration counts for quick iteration). Results are now auto-written to timestamped JSON files under benchmark-results/. Added jmhCompare task backed by scripts/jmh_compare.py to detect performance regressions between two result files (exits non-zero on regression above threshold).

Improved

• SwissMap hot-path performance (#15): exp-001 optimization experiment applied to findIndexHashed and putValHashed — 13–33% improvement across all key benchmarks with no regressions.

  Metric	Baseline	Optimized	Δ
  GetHit @ 12K	5.59 ns	4.53 ns	−18.9%
  GetHit @ 784K	17.98 ns	14.95 ns	−16.8%
  GetMiss @ 12K	5.84 ns	4.80 ns	−17.8%
  GetMiss @ 784K	16.51 ns	14.28 ns	−13.5%
  PutHit @ 12K	8.09 ns	6.48 ns	−19.9%
  PutHit @ 784K	30.59 ns	23.34 ns	−23.7%
  PutMiss @ 12K	23.70 ns	15.98 ns	−32.6%
  PutMiss @ 784K	109.84 ns	83.79 ns	−23.7%

  Techniques applied: ILP hoisting of SWAR eqMask/emptyMask operations, SWAR match shortcut (skip emptyMask on hit path), lazy emptyMask in putValHashed tombstone update path.

Changed

• README example code simplified (removed SwissSimdMap references).

Full Changelog: v0.1.8...v0.2.0