Files

Moe Charm (CI) 52386401b3 Debug Counters Implementation - Clean History

Major Features:
- Debug counter infrastructure for Refill Stage tracking
- Free Pipeline counters (ss_local, ss_remote, tls_sll)
- Diagnostic counters for early return analysis
- Unified larson.sh benchmark runner with profiles
- Phase 6-3 regression analysis documentation

Bug Fixes:
- Fix SuperSlab disabled by default (HAKMEM_TINY_USE_SUPERSLAB)
- Fix profile variable naming consistency
- Add .gitignore patterns for large files

Performance:
- Phase 6-3: 4.79 M ops/s (has OOM risk)
- With SuperSlab: 3.13 M ops/s (+19% improvement)

This is a clean repository without large log files.

🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>

2025-11-05 12:31:14 +09:00

3.3 KiB

Raw Blame History

📊 包括的ベンチマーク結果: HAKMEM vs System

Test 1-6 詳細比較 (6パターン × 4サイズ = 24テスト)

16B サイズクラス

Test Pattern	HAKMEM	System	vs System
1. Sequential LIFO	66.00 M/s	161.15 M/s	-59.0% ❌
2. Sequential FIFO	66.38 M/s	159.71 M/s	-58.4% ❌
3. Random Free	54.03 M/s	102.09 M/s	-47.1% ❌
4. Interleaved	66.28 M/s	141.72 M/s	-53.2% ❌
6. Long/Short-lived	65.30 M/s	160.62 M/s	-59.3% ❌

平均: 63.60 M/s vs 145.06 M/s → -56.2% 💀

32B サイズクラス

Test Pattern	HAKMEM	System	vs System
1. Sequential LIFO	60.22 M/s	163.51 M/s	-63.2% ❌
2. Sequential FIFO	61.13 M/s	155.76 M/s	-60.8% ❌
3. Random Free	50.30 M/s	106.87 M/s	-52.9% ❌
4. Interleaved	61.00 M/s	166.53 M/s	-63.4% ❌
6. Long/Short-lived	59.40 M/s	174.06 M/s	-65.9% ❌

平均: 58.41 M/s vs 153.35 M/s → -61.9% 💀

64B サイズクラス

Test Pattern	HAKMEM	System	vs System
1. Sequential LIFO	51.83 M/s	168.55 M/s	-69.2% ❌
2. Sequential FIFO	51.76 M/s	169.14 M/s	-69.4% ❌
3. Random Free	43.96 M/s	107.04 M/s	-58.9% ❌
4. Interleaved	51.94 M/s	158.50 M/s	-67.2% ❌
6. Long/Short-lived	51.14 M/s	162.62 M/s	-68.6% ❌

平均: 50.13 M/s vs 153.17 M/s → -67.3% 💀💀

128B サイズクラス

Test Pattern	HAKMEM	System	vs System
1. Sequential LIFO	39.54 M/s	75.34 M/s	-47.5% ❌
2. Sequential FIFO	39.54 M/s	80.91 M/s	-51.1% ❌
3. Random Free	35.49 M/s	48.50 M/s	-26.8% 🤔
4. Interleaved	39.96 M/s	81.28 M/s	-50.8% ❌
6. Long/Short-lived	40.20 M/s	86.92 M/s	-53.7% ❌

平均: 38.95 M/s vs 74.59 M/s → -47.8% ❌

Mixed Sizes (8B, 16B, 32B, 64B)

Test Pattern	HAKMEM	System	vs System
5. Mixed Sizes	62.37 M/s	161.77 M/s	-61.4% ❌

📉 全体サマリー (21テスト)

Metric	HAKMEM	System	Ratio
全体平均	52.59 M/s	135.94 M/s	-61.3% 💀💀
最高値	66.38 M/s	174.06 M/s	-61.9%
最低値	35.49 M/s	48.50 M/s	-26.8%

🔥 フラグメンテーションストレス

Allocator	Throughput	vs System
HAKMEM	4.68 M/s	-74.6% 💥💥💥
System	18.43 M/s (推定)	100%

重大な問題: フラグメンテーション耐性が極めて弱い！

🎯 結論

HAKMEM の弱点が明確化:

Tiny (≤1KB): System の 1/3 以下 (-60~-70%) 💀
フラグメンテーション: System の 1/4 (-75%) 💥
パターン感度: LIFO/FIFO/Random すべてで劣る

HAKMEM が強い領域:

Mid-Large (8-32KB): System の 2.08倍 (+108%) ✅
専用最適化 (HAKX 8-32KB): System の 2.71倍 (+171%) 🏆

戦略的判断:

Tiny を諦め、Mid-Large に集中すべき

3.3 KiB Raw Blame History Unescape Escape