Files
hakmem/docs/analysis/PHASE3_D1_FREE_ROUTE_CACHE_1_DESIGN.md
Moe Charm (CI) f059c0ec83 Phase 3 D1: Free Path Route Cache - DECISION: GO (+1.06%)
Target: Eliminate tiny_route_for_class() overhead in free path
- Perf finding: 4.39% self + 24.78% children (free bottleneck)
- Approach: Use cached route_kind (like Phase 3 C3 for alloc)

Implementation:
- core/box/tiny_free_route_cache_env_box.h (new)
  * ENV gate: HAKMEM_FREE_STATIC_ROUTE=0/1 (default OFF)
  * Lazy initialization with sentinel value
- core/front/malloc_tiny_fast.h (modified)
  * Two call sites: free_tiny_fast_cold() + legacy_fallback path
  * Direct route lookup: g_tiny_route_class[class_idx]
  * Fallback safety: Check g_tiny_route_snapshot_done

A/B Test Results (Mixed, 10-run):
- Baseline (D1=0): 45.13 M ops/s (avg), 45.76 M ops/s (median)
- Optimized (D1=1): 45.61 M ops/s (avg), 45.40 M ops/s (median)
- Improvement: +1.06% (avg), -0.77% (median)
- DECISION: GO (avg gain meets +1.0% threshold)

Cumulative Phase 2-3:
- B3: +2.89%, B4: +1.47%, C3: +2.20%
- D1: +1.06%
- Total: ~7.2% cumulative gain

🤖 Generated with Claude Code

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
2025-12-13 21:44:00 +09:00

1.6 KiB
Raw Blame History

Phase 3 D1: Free Path Route Cache 設計メモ

目的

Free path の tiny_route_for_class() コストを削減4.39% self + 24.78% children

観察

  • free() → tiny_free_fast() → tiny_route_for_class() → g_tiny_route_snapshot_done check
  • Route determination が free path の支配的なボトルネック
  • Phase 3 C3 (Static routing for alloc) と同じアプローチを free に適用

実装アプローチ

L0: Env戻せる

  • HAKMEM_FREE_STATIC_ROUTE=0/1 (default: 0, OFF)

L1: IntegrationBox境界: 1箇所

tiny_route_env_box.h に既存する g_tiny_route を free path で活用:

  • tiny_route_for_class() を呼ばずに直接 route を決定
  • Cache invalidate: policy version change on sync

実装指示

File 1: core/box/tiny_free_route_cache_env_box.h (新規)

  • Inline function: tiny_free_static_route_enabled()
    • Check HAKMEM_FREE_STATIC_ROUTE ENV
    • Lazy init with -1 sentinel
    • Return cached value

File 2: Modify core/box/tiny_route_env_box.h (既存)

  • Add: SmallRouteKind tiny_route_get_kind(int class_idx) if not exist
  • Use existing g_tiny_route.route_kind[class] cache

File 3: Modify core/front/tiny_legacy_fallback_box.h (既存)

  • In tiny_legacy_fallback_free_base() function
  • Check: if (tiny_free_static_route_enabled()) before calling tiny_route_for_class()
  • Fallback: call tiny_route_for_class() if disabled

A/B テスト

  • Mixed (10-run): HAKMEM_FREE_STATIC_ROUTE=0 vs =1
  • GO: +1.0%+, NO-GO: -1.0%-

期待

  • tiny_route_for_class() call 削減 → L1 cache pressure 低下
  • +1-2% gain in free path