Files
VoxelForge/Standalone/Research/WorldAssemblerM3/Linux105/REPORT.md
T
Fr0zkaandClaude Opus 5.5 f228d6319f startup: parallel owner-shape compile and far-parent build (L105, exact); R45 simplification audit
Owner shapes compile per stratum on the Core parallel scheduler and merge in owner-layout order;
far-parent ellipsoids/connectors generate in parallel and merge canonically; 512-unit far bins.
Base compile 239 -> 203 ms, far-parent build 128 -> 33 ms. Near-ready @W12: SW 0.84 -> 0.75,
FI 1.51 -> 1.42, TN 1.38 -> 1.30, VS 0.92 -> 0.85 s. All hashes equal L104 (Clang/GCC W1/W8/W12).
Fast tier PASS; Linux UE builds; 6/6 Unreal arrival traces pass.
R45 (read-only, owner's "are we overcomplicating?"): next simplifications = tunnel floor/mouth support
inside the bounded passage field; boundary bands once per XY column; certified early SW tiles.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
2026-10-28 21:17:15 +01:00

50 lines
3.5 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# Linux 105 — serial startup phases
## Changes and phase timings
Owner-shape compilation now runs by stratum on the core parallel scheduler. Results are merged in owner-layout order. Far-parent ellipsoid and connector generation also uses parallel workers, then merges in canonical order. The far-parent spatial index uses larger 512-unit bins; the exact primitive predicate and insertion order remain unchanged. LOD2 still waits for far parents.
| Phase | L104 median | L105 median | Change |
|---|---:|---:|---:|
| Base-shape compile | 239.1 ms | 202.8 ms | −15.2% |
| Far-parent build | 128.5 ms | 33.4 ms | −74.0% |
L105 base-shape substage timers across FI/TN at W8/W12:
| Substage | Range |
|---|---:|
| Owner shapes (parallel wall; aggregate worker time) | 22.1–23.8 ms; 47.8–51.7 worker-ms |
| Owner bins / visual detail prep | 1.18–1.40 ms / below 0.001 ms |
| Packed intervals/signs | 23.7–25.9 ms |
| Passage prep | 10.1–11.7 ms |
| Candidate sweeps | 23.3–27.6 ms |
| Surface/detail band | 54.0–93.2 ms |
| Sign-edge repair | 9.6–10.7 ms |
| Route validation | 26.5–29.3 ms |
| Far-parent geometry (parallel wall; aggregate worker time) | 2.4–2.7 ms; 16.2–18.2 worker-ms |
| Far shared fields / bin rebuild | 4.0–4.3 ms / 16.7–18.3 ms |
The timeline tracks streaming tile workers. In L104 they were about 99.99% idle during the 232–251 ms base compile, 99.4% busy during LOD0–1, and idle throughout the 110–144 ms far-parent phase. In L105 the tile pool remains 99.96–99.99% idle during base compile while the separate core parallel scheduler performs owner-shape work (22–24 ms wall from 48–52 worker-ms). LOD0–1 is 99.2–99.6% busy; far-parent takes 33–34 ms with the tile pool idle; LOD2 is 92.8–95.9% busy. Full event data and the utilization breakdown are in `Saved/Linux/l105/timeline/` (`analysis.json`, `utilization.csv`, and four logs).
## L80 readiness pairs
Sixteen warmups total, then three interleaved measured runs per build and cell. Each cell has complete near and whole-view tile sets; values are medians in seconds, L104 → L105.
| View | W8 near | W8 whole | W12 near | W12 whole |
|---|---:|---:|---:|---:|
| SW | 0.840 → 0.744 (−11.4%) | 1.957 → 1.835 (−6.2%) | 0.839 → 0.746 (−11.2%) | 1.878 → 1.762 (−6.2%) |
| FI | 1.597 → 1.556 (−2.6%) | 2.594 → 2.516 (−3.0%) | 1.514 → 1.420 (−6.2%) | 2.438 → 2.338 (−4.1%) |
| TN | 1.516 → 1.455 (−4.0%) | 2.475 → 2.341 (−5.4%) | 1.384 → 1.301 (−6.0%) | 2.228 → 2.140 (−3.9%) |
| VS | 1.085 → 1.024 (−5.6%) | 2.016 → 1.920 (−4.8%) | 0.917 → 0.849 (−7.4%) | 1.774 → 1.703 (−4.0%) |
## Exactness and verification
- All 24 hash runs passed: SW/FI/TN/VS × Clang/GCC × W1/W8/W12. Every tile hash and whole-view hash matched L104 and the Clang W1 reference.
- Candidate source stamp: `c6b2740c67bc96252ba57eb303b40d97d3c3451fcb4f1daed7bae773351681ab`.
- Fast tier: **PASS**, 278.4 s.
- Linux Unreal build: **Succeeded** with `-NoUBA -MaxParallelActions=4`; it compiled the modified core shim. Build log: `Saved/Linux/l105/unreal-build.log`.
- Three FI and three SW Unreal arrival traces: **6/6 passed**, exit 0. FI settled 1,216 near tiles in 1.545–1.562 s from first admission; SW settled 1,430 in 1.285–1.317 s. Unreal ran with `taskset -c 0-11`, without `prlimit`.
- Parallel full-bin assembly variants were retired after regressions. The retained serial exact bin rebuild measured 16.7–18.3 ms with the 512-unit far bins.
Hash, timing, readiness, and arrival logs are under `Saved/Linux/l105/`. No commit was made.