Fr0zkaandClaude Opus 5.5
89504ecc97
Fine pairs 48.9M -> 38.8M (-21%) with L115-identical hashes, but no readiness change (still +15..+30% near-ready vs L109). Pair evaluation is not where the regression lives. Kept as patch. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
1.5 KiB
Linux L116
L116 preserves L115 output: all 18 Clang/GCC hash-matrix cases matched, including all four owner views at W1/W8/W12. The tightened chord bound and z-column cache reduced exact fine-segment/sample evaluations from 48,947,915 in L115 to 38,783,101 (-20.8%). L109 was 19,812,455; L116 still performs 1.96x as many pairs.
Timing used five interleaved L80 runs per version, stratum, and worker count. The pre-run one-minute load was 0.94; top reported 97.1% idle.
| Stratum | Workers | LOD0-2 ready L109 / L115 / L116 (s) | L116 vs L109 | Whole view L109 / L115 / L116 (s) |
|---|---|---|---|---|
| 0 | 8 | 0.770 / 0.908 / 0.947 | +23.1% | 1.791 / 1.940 / 1.907 |
| 0 | 12 | 0.788 / 0.917 / 0.930 | +18.1% | 1.732 / 1.871 / 1.886 |
| 1 | 8 | 0.951 / 1.236 / 1.235 | +29.9% | 1.833 / 2.163 / 2.194 |
| 1 | 12 | 0.906 / 1.174 / 1.174 | +29.5% | 1.756 / 2.040 / 2.038 |
| 2 | 8 | 1.015 / 1.264 / 1.248 | +22.9% | 1.905 / 2.166 / 2.102 |
| 2 | 12 | 0.940 / 1.177 / 1.173 | +24.8% | 1.757 / 1.999 / 1.995 |
| 3 | 8 | 0.882 / 1.011 / 1.016 | +15.2% | 1.850 / 1.958 / 2.009 |
| 3 | 12 | 0.796 / 0.933 / 0.950 | +19.4% | 1.663 / 1.809 / 1.826 |
The requested near-ready target of about 5% over L109 was not met. Fast tier passed (344.3 s). The Linux VFHostEditor build passed with -NoUBA -MaxParallelActions=4 (71.11 s; no prlimit).
Logs and summaries are under Saved/Linux/l116/; timing summary is timing/summary.json, pair counts are in pairs/summary.json, and the hash result is hashes/summary.json.