tomoaki/hakmem
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
659 Commits 2 Branches 0 Tags
e51231471be93b210d6e6b94f1fb13a345a18ba5
Commit Graph

25 Commits

Author SHA1 Message Date
Moe Charm (CI)
4f99054fd5 Phase 75-3: C5+C6 Interaction Matrix Test (4-Point A/B) - STRONG GO (+5.41%) 
Comprehensive interaction testing with single binary, ENV-only configuration:

4-Point Matrix Results (Mixed SSOT, WS=400):
- Point A (C5=0, C6=0): 42.36 M ops/s [Baseline]
- Point B (C5=1, C6=0): 43.54 M ops/s (+2.79% vs A)
- Point C (C5=0, C6=1): 44.25 M ops/s (+4.46% vs A)
- Point D (C5=1, C6=1): 44.65 M ops/s (+5.41% vs A) **[COMBINED TARGET]**

Additivity Analysis:
- Expected additive: 45.43 M ops/s (B+C-A)
- Actual: 44.65 M ops/s (D)
- Sub-additivity: 1.72% (near-perfect, minimal negative interaction)

Perf Stat Validation (Point D vs A):
- Instructions: -6.1% (function call elimination confirmed)
- Branches: -6.1% (matches instructions reduction)
- Cache-misses: -31.5% (improved locality, NO code explosion)
- Throughput: +5.41% (net positive)

Decision: ✅ STRONG GO (exceeds +3.0% GO threshold)
- D vs A: +5.41% >> +3.0%
- Sub-additivity: 1.72% << 20% acceptable
- Phase 73 hypothesis validated: -6.1% instructions/branches → +5.41% throughput

Promotion to Defaults:
- core/bench_profile.h: C5+C6 added to bench_apply_mixed_tinyv3_c7_common()
- scripts/run_mixed_10_cleanenv.sh: C5+C6 ENV defaults added
- C5+C6 inline slots now PRESET DEFAULT for MIXED_TINYV3_C7_SAFE

New Baseline: 44.65 M ops/s (36.75% of mimalloc, +5.41% from Phase 75-0)
M2 Target: 55% of mimalloc ≈ 66.8 M ops/s (remaining gap: 22.15 M ops/s)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
 2025-12-18 08:53:01 +09:00
Moe Charm (CI)
043d34ad5a Phase 75-2: C5-only Inline Slots (P2) - GO (+1.10%) 
Extends Phase 75-1 pattern to C5 class (28.5% of C4-C7 ops):
- Created 4 new boxes: env_box, tls_box, fast_path_api, TLS variable
- Integration: 2 minimal boundary points (alloc/free for C5)
- Test strategy: C5-only isolation (baseline C5=OFF+C6=ON, treatment C5=ON+C6=ON)
- Default OFF: zero overhead when disabled

Results (10-run Mixed SSOT, WS=400, C6 already enabled):
- Baseline (C5=OFF, C6=ON): 44.26 M ops/s (σ=0.37)
- Treatment (C5=ON, C6=ON): 44.74 M ops/s (σ=0.54)
- Delta: +0.49 M ops/s (+1.10%)

Status: ✅ GO - C5 individual contribution confirmed
Cumulative since Phase 75-0: +2.87% (C6) + 1.10% (C5) = potential +3.97% combined
Next: Phase 75-3 (test C5+C6 interaction + non-additivity + promote to preset default)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
 2025-12-18 08:39:48 +09:00
Moe Charm (CI)
0009ce13b3 Phase 75-1: C6-only Inline Slots (P2) - GO (+2.87%) 
Modular implementation of hot-class inline slots optimization:
- Created 5 new boxes: env_box, tls_box, fast_path_api, integration_box, test_script
- Single decision point at TLS init (ENV gate: HAKMEM_TINY_C6_INLINE_SLOTS=0/1)
- Integration: 2 minimal boundary points (alloc/free paths for C6 class)
- Default OFF: zero overhead when disabled (full backward compatibility)

Results (10-run Mixed SSOT, WS=400):
- Baseline (C6 inline OFF):  44.24 M ops/s
- Treatment (C6 inline ON):  45.51 M ops/s
- Delta: +1.27 M ops/s (+2.87%)

Status: ✅ GO - Strong improvement via C6 ring buffer fast-path
Mechanism: Branch elimination on unified_cache_push/pop for C6 allocations
Next: Phase 75-2 (add C5 inline slots, target 85% C4-C7 coverage)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
 2025-12-18 08:22:09 +09:00
Moe Charm (CI)
b6212bbe31 Phase 69: Refill tuning completion (Warm Pool Size=16 optimized) 
- Promoted Warm Pool Size=16 as the new baseline (+3.26% gain).
- Updated PERFORMANCE_TARGETS_SCORECARD.md with Phase 69 results.
- Updated scripts/run_mixed_10_cleanenv.sh and core/bench_profile.h to use HAKMEM_WARM_POOL_SIZE=16 by default.
- Clarified that TINY_REFILL_BATCH_SIZE is not currently connected.
 2025-12-18 01:55:27 +09:00
Moe Charm (CI)
b2e861db12 Phase 67a: Layout tax forensics foundation (SSOT + measurement box) 
Changes:
- scripts/box/layout_tax_forensics_box.sh: New measurement harness
  * Baseline vs treatment 10-run throughput comparison
  * Automated perf stat collection (cycles, IPC, branches, misses, TLB)
  * Binary metadata (size, section info)
  * Output to results/layout_tax_forensics/

- docs/analysis/PHASE67A_LAYOUT_TAX_FORENSICS_SSOT.md: Diagnostic reference
  * Decision tree: GO/NEUTRAL/NO-GO classification
  * Symptom→root-cause mapping (IPC/branch-miss/dTLB/cache-miss)
  * Phase 64 case study analysis (IPC 2.05→1.98)
  * Operational guidelines for Phase 67b+ optimizations

- CURRENT_TASK.md: Phase 67a marked complete, operational

Outcome:
- Layout tax diagnosis now reproducible in single measurement pass
- Enables fast GO/NO-GO decision for future code removal/reordering attempts
- Foundation for M2 (55% target) structural exploration without regression risk

🤖 Generated with Claude Code

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
 2025-12-17 21:09:42 +09:00
Moe Charm (CI)
84f5034e45 Phase 68: PGO training set diversification (seed/WS expansion) 
Changes:
- scripts/box/pgo_fast_profile_config.sh: Expanded WS patterns (3→5) and seeds (1→3)
  for reduced overfitting and better production workload representativeness
- PERFORMANCE_TARGETS_SCORECARD.md: Phase 68 baseline promoted (61.614M = 50.93%)
- CURRENT_TASK.md: Phase 68 marked complete, Phase 67a (layout tax forensics) set Active

Results:
- 10-run verification: +1.19% vs Phase 66 baseline (GO, >+1.0% threshold)
- M1 milestone: 50.93% of mimalloc (target 50%, exceeded by +0.93pp)
- Stability: 10-run mean/median with <2.1% CV

🤖 Generated with Claude Code

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
 2025-12-17 21:08:17 +09:00
Moe Charm (CI)
ef8e2ab9b5 Phase 59b & 61: Speed-first Rebase + C7 ULTRA Header-Light Optimization 
Phase 59b: Speed-first Mode Baseline Rebase
- Rebase on MIXED_TINYV3_C7_SAFE profile (Speed-first, no prewarm suppression)
- hakmem: 58.478 M ops/s (CV 2.52%)
- mimalloc: 120.979 M ops/s (CV 0.90%)
- Ratio: 48.34% of mimalloc (down from 49.13% Balanced mode in Phase 59)
- Reason for difference: Profile selection (Speed-first vs Balanced) and mimalloc environment variance
- Status: COMPLETE (measurement-only, zero code changes)

Phase 61: C7 ULTRA Header-Light Optimization Attempt
- Objective: Skip header write on C7 ULTRA alloc hit (write only on refill)
- Implementation: ENV gate HAKMEM_TINY_C7_ULTRA_HEADER_LIGHT (default OFF)
- Result: +0.31% (NEUTRAL, below +1.0% GO threshold)
  - Baseline: 59.543 M ops/s (CV 1.53%)
  - Treatment: 59.729 M ops/s (CV 2.66%)
- Root cause analysis:
  - tiny_region_id_write_header only 2.32% of time (lower than Phase 42 estimate 4.56%)
  - Header-light mode adds branch to hot path, negating write savings
  - Mixed workload dilutes C7-specific optimization effectiveness
  - Variance increased due to branch prediction variability
- Decision: Kept as research box with ENV gate (default OFF)
- Lesson: Workload-specific optimizations need careful verification with full workloads

Updated Documentation:
- PHASE59B_SPEED_FIRST_REBASE_RESULTS.md: Full measurement results and analysis
- PHASE61_C7_ULTRA_HEADER_LIGHT_RESULTS.md: A/B test results and root cause analysis
- PHASE61_C7_ULTRA_HEADER_LIGHT_IMPLEMENTATION.md: Implementation details and design
- CURRENT_TASK.md: Updated status and next phase planning (Phase 62)
- PERFORMANCE_TARGETS_SCORECARD.md: Updated baseline and M1 milestone status

M1 (50%) Milestone Status:
- Current: 48.34% (Speed-first profile)
- Gap: -1.66% (within measurement noise)
- Profile recommendation: Speed-first as canonical default for throughput focus

🤖 Generated with Claude Code

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
 2025-12-17 16:25:26 +09:00
Moe Charm (CI)
7adbcdfcb6 Phase 54-60: Memory-Lean mode, Balanced mode stabilization, M1 (50%) achievement 
## Summary

Completed Phase 54-60 optimization work:

**Phase 54-56: Memory-Lean mode (LEAN+OFF prewarm suppression)**
- Implemented ss_mem_lean_env_box.h with ENV gates
- Balanced mode (LEAN+OFF) promoted as production default
- Result: +1.2% throughput, better stability, zero syscall overhead
- Added to bench_profile.h: MIXED_TINYV3_C7_BALANCED preset

**Phase 57: 60-min soak finalization**
- Balanced mode: 60-min soak, RSS drift 0%, CV 5.38%
- Speed-first mode: 60-min soak, RSS drift 0%, CV 1.58%
- Syscall budget: 1.25e-7/op (800× under target)
- Status: PRODUCTION-READY

**Phase 59: 50% recovery baseline rebase**
- hakmem FAST (Balanced): 59.184M ops/s, CV 1.31%
- mimalloc: 120.466M ops/s, CV 3.50%
- Ratio: 49.13% (M1 ACHIEVED within statistical noise)
- Superior stability: 2.68× better CV than mimalloc

**Phase 60: Alloc pass-down SSOT (NO-GO)**
- Implemented alloc_passdown_ssot_env_box.h
- Modified malloc_tiny_fast.h for SSOT pattern
- Result: -0.46% (NO-GO)
- Key lesson: SSOT not applicable where early-exit already optimized

## Key Metrics

- Performance: 49.13% of mimalloc (M1 effectively achieved)
- Stability: CV 1.31% (superior to mimalloc 3.50%)
- Syscall budget: 1.25e-7/op (excellent)
- RSS: 33MB stable, 0% drift over 60 minutes

## Files Added/Modified

New boxes:
- core/box/ss_mem_lean_env_box.h
- core/box/ss_release_policy_box.{h,c}
- core/box/alloc_passdown_ssot_env_box.h

Scripts:
- scripts/soak_mixed_single_process.sh
- scripts/analyze_epoch_tail_csv.py
- scripts/soak_mixed_rss.sh
- scripts/calculate_percentiles.py
- scripts/analyze_soak.py

Documentation: Phase 40-60 analysis documents

## Design Decisions

1. Profile separation (core/bench_profile.h):
   - MIXED_TINYV3_C7_SAFE: Speed-first (no LEAN)
   - MIXED_TINYV3_C7_BALANCED: Balanced mode (LEAN+OFF)

2. Box Theory compliance:
   - All ENV gates reversible (HAKMEM_SS_MEM_LEAN, HAKMEM_ALLOC_PASSDOWN_SSOT)
   - Single conversion points maintained
   - No physical deletions (compile-out only)

3. Lessons learned:
   - SSOT effective only where redundancy exists (Phase 60 showed limits)
   - Branch prediction extremely effective (~0 cycles for well-predicted branches)
   - Early-exit pattern valuable even when seemingly redundant

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
 2025-12-17 06:24:01 +09:00
Moe Charm (CI)
b7085c47e1 Phase 35-39: FAST build optimization complete (+7.13% cumulative) 
Phase 35-A: BENCH_MINIMAL gate function elimination (GO +4.39%)
- tiny_front_v3_enabled() → constant true
- tiny_metadata_cache_enabled() → constant 0
- learner_v7_enabled() → constant false
- small_learner_v2_enabled() → constant false

Phase 36: Policy snapshot init-once (GO +0.71%)
- small_policy_v7_snapshot() version check skip in BENCH_MINIMAL
- TLS cache for policy snapshot

Phase 37: Standard TLS cache (NO-GO -0.07%)
- TLS cache for Standard build attempted
- Runtime gate overhead negates benefit

Phase 38: FAST/OBSERVE/Standard workflow established
- make perf_fast, make perf_observe targets
- Scorecard and documentation updates

Phase 39: Hot path gate constantization (GO +1.98%)
- front_gate_unified_enabled() → constant 1
- alloc_dualhot_enabled() → constant 0
- g_bench_fast_front, g_v3_enabled blocks → compile-out
- free_dispatch_stats_enabled() → constant false

Results:
- FAST v3: 56.04M ops/s (47.4% of mimalloc)
- Standard: 53.50M ops/s (45.3% of mimalloc)
- M1 target (50%): 5.5% remaining

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
 2025-12-16 15:01:56 +09:00
Moe Charm (CI)
8052e8b320 Phase 24-26: Hot path atomic telemetry prune (+2.00% cumulative) 
Summary:
- Phase 24 (alloc stats): +0.93% GO
- Phase 25 (free stats): +1.07% GO
- Phase 26 (diagnostics): -0.33% NEUTRAL (code cleanliness)
- Total: 11 atomics compiled-out, +2.00% improvement

Phase 24: OBSERVE tax prune (tiny_class_stats_box.h)
- Added HAKMEM_TINY_CLASS_STATS_COMPILED (default: 0)
- Wrapped 5 stats functions: uc_miss, warm_hit, shared_lock, tls_carve_*
- Result: +0.93% (baseline 56.675M vs compiled-in 56.151M ops/s)

Phase 25: Tiny free stats prune (tiny_superslab_free.inc.h)
- Added HAKMEM_TINY_FREE_STATS_COMPILED (default: 0)
- Wrapped g_free_ss_enter atomic in free hot path
- Result: +1.07% (baseline 57.017M vs compiled-in 56.415M ops/s)

Phase 26: Hot path diagnostic atomics prune
- Added 5 compile gates for low-frequency error counters:
  - HAKMEM_TINY_C7_FREE_COUNT_COMPILED
  - HAKMEM_TINY_HDR_MISMATCH_LOG_COMPILED
  - HAKMEM_TINY_HDR_META_MISMATCH_COMPILED
  - HAKMEM_TINY_METRIC_BAD_CLASS_COMPILED
  - HAKMEM_TINY_HDR_META_FAST_COMPILED
- Result: -0.33% NEUTRAL (within noise, kept for cleanliness)

Alignment with mimalloc principles:
- "No atomics on hot path" - telemetry moved to compile-time opt-in
- Fixed per-op tax elimination
- Production builds: maximum performance (atomics compiled-out)
- Research builds: full diagnostics (COMPILED=1)

Generated with Claude Code
https://claude.com/claude-code

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
 2025-12-16 05:35:11 +09:00
Moe Charm (CI)
ec87025da6 Phase 17 v2 (FORCE_LIBC fix) + Phase 19-1b (FastLane Direct) — GO (+5.88%) 
## Phase 17 v2: FORCE_LIBC Gap Validation Fix

**Critical bug fix**: Phase 17 v1 の測定が壊れていた

**Problem**: HAKMEM_FORCE_LIBC_ALLOC=1 が FastLane より後でしか見えず、
same-binary A/B が実質 "hakmem vs hakmem" になっていた（+0.39% 誤測定）

**Fix**: core/box/hak_wrappers.inc.h:171 と :645 に g_force_libc_alloc==1 の
early bypass を追加、__libc_malloc/__libc_free に最初に直行

**Result**: 正しい同一バイナリ A/B 測定
- hakmem (FORCE_LIBC=0): 48.99M ops/s
- libc (FORCE_LIBC=1): 79.72M ops/s (+62.7%)
- system binary: 88.06M ops/s (+10.5% vs libc)

**Gap 分解**:
- Allocator 差: +62.7% (主戦場)
- Layout penalty: +10.5% (副次的)

**Conclusion**: Case A 確定 (allocator dominant, NOT layout)
Phase 17 v1 の Case B 判定は誤り。

Files:
- docs/analysis/PHASE17_FORCE_LIBC_GAP_VALIDATION_1_AB_TEST_RESULTS.md (v2)
- docs/analysis/PHASE17_FORCE_LIBC_GAP_VALIDATION_1_NEXT_INSTRUCTIONS.md (updated)

---

## Phase 19: FastLane Instruction Reduction Analysis

**Goal**: libc との instruction gap (-35% instructions, -56% branches) を削減

**perf stat 分析** (FORCE_LIBC=0 vs 1, 200M ops):
- hakmem: 209.09 instructions/op, 52.33 branches/op
- libc: 135.92 instructions/op, 22.93 branches/op
- Delta: +73.17 instructions/op (+53.8%), +29.40 branches/op (+128.2%)

**Hot path** (perf report):
- front_fastlane_try_free: 23.97% cycles
- malloc wrapper: 23.84% cycles
- free wrapper: 6.82% cycles
- **Wrapper overhead: ~55% of all cycles**

**Reduction candidates**:
- A: Wrapper layer 削除 (-17.5 inst/op, +10-15% 期待)
- B: ENV snapshot 統合 (-10.0 inst/op, +5-8%)
- C: Stats 削除 (-5.0 inst/op, +3-5%)
- D: Header inline (-4.0 inst/op, +2-3%)
- E: Route fast path (-3.5 inst/op, +2-3%)

Files:
- docs/analysis/PHASE19_FASTLANE_INSTRUCTION_REDUCTION_1_DESIGN.md
- docs/analysis/PHASE19_FASTLANE_INSTRUCTION_REDUCTION_2_NEXT_INSTRUCTIONS.md

---

## Phase 19-1b: FastLane Direct — GO (+5.88%)

**Strategy**: Wrapper layer を bypass し、core allocator を直接呼ぶ
- free() → free_tiny_fast() (not free_tiny_fast_hot)
- malloc() → malloc_tiny_fast()

**Phase 19-1 が NO-GO (-3.81%) だった原因**:
1. __builtin_expect(fastlane_direct_enabled(), 0) が逆効果（A/B 不公平）
2. free_tiny_fast_hot() が誤選択（free_tiny_fast() が勝ち筋）

**Phase 19-1b の修正**:
1. __builtin_expect() 削除
2. free_tiny_fast() を直接呼び出し

**Result** (Mixed, 10-run, 20M iters, ws=400):
- Baseline (FASTLANE_DIRECT=0): 49.17M ops/s
- Optimized (FASTLANE_DIRECT=1): 52.06M ops/s
- **Delta: +5.88%** (GO 基準 +5% クリア)

**perf stat** (200M iters):
- Instructions/op: 199.90 → 169.45 (-30.45, -15.23%)
- Branches/op: 51.49 → 41.52 (-9.97, -19.36%)
- Cycles/op: 88.88 → 84.37 (-4.51, -5.07%)
- I-cache miss: 111K → 98K (-11.79%)

**Trade-offs** (acceptable):
- iTLB miss: +41.46% (front-end cost)
- dTLB miss: +29.15% (backend cost)
- Overall gain (+5.88%) outweighs costs

**Implementation**:
1. **ENV gate**: core/box/fastlane_direct_env_box.{h,c}
   - HAKMEM_FASTLANE_DIRECT=0/1 (default: 0, opt-in)
   - Single _Atomic global (wrapper キャッシュ問題を解決)

2. **Wrapper 修正**: core/box/hak_wrappers.inc.h
   - malloc: direct call to malloc_tiny_fast() when FASTLANE_DIRECT=1
   - free: direct call to free_tiny_fast() when FASTLANE_DIRECT=1
   - Safety: !g_initialized では direct 使わない、fallback 維持

3. **Preset 昇格**: core/bench_profile.h:88
   - bench_setenv_default("HAKMEM_FASTLANE_DIRECT", "1")
   - Comment: +5.88% proven on Mixed, 10-run

4. **cleanenv 更新**: scripts/run_mixed_10_cleanenv.sh:22
   - HAKMEM_FASTLANE_DIRECT=${HAKMEM_FASTLANE_DIRECT:-1}
   - Phase 9/10 と同様に昇格

**Verdict**: GO — 本線採用、プリセット昇格完了

**Rollback**: HAKMEM_FASTLANE_DIRECT=0 で既存 FastLane path に戻る

Files:
- core/box/fastlane_direct_env_box.{h,c} (new)
- core/box/hak_wrappers.inc.h (modified)
- core/bench_profile.h (preset promotion)
- scripts/run_mixed_10_cleanenv.sh (ENV default aligned)
- Makefile (new obj)
- docs/analysis/PHASE19_1B_FASTLANE_DIRECT_REVISED_AB_TEST_RESULTS.md

---

## Cumulative Performance

- Baseline (all optimizations OFF): ~40M ops/s (estimated)
- Current (Phase 19-1b): 52.06M ops/s
- **Cumulative gain: ~+30% from baseline**

Remaining gap to libc (79.72M):
- Current: 52.06M ops/s
- Target: 79.72M ops/s
- **Gap: +53.2%** (was +62.7% before Phase 19-1b)

Next: Phase 19-2 (ENV snapshot consolidation, +5-8% expected)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
 2025-12-15 11:28:40 +09:00
Moe Charm (CI)
f8e7cf05b4 Phase 16 v1 NEUTRAL, Phase 17 Case B confirmed, Phase 18 design added 
## Phase 16 v1: Front FastLane Alloc LEGACY Direct — NEUTRAL (+0.62%)

Target: Reduce alloc-side fixed costs by adding LEGACY direct path to
FastLane entry, mirroring Phase 9/10 free-side winning pattern.

Result: +0.62% on Mixed (below +1.0% GO threshold) → NEUTRAL, freeze as
research box (default OFF).

Critical issue: Initial impl crashed (segfault) for C4-C7. Root cause:
unified_cache_refill() incompatibility. Safety fix: Limited to C0-C3
only (matching existing dualhot pattern).

Files:
- core/box/front_fastlane_alloc_legacy_direct_env_box.{h,c} (new)
- core/box/front_fastlane_box.h (LEGACY direct path, lines 93-119)
- core/bench_profile.h (env refresh sync)
- Makefile (new obj)
- docs/analysis/PHASE16_*.md (design/results/instructions)

ENV: HAKMEM_FRONT_FASTLANE_ALLOC_LEGACY_DIRECT=0 (default OFF, opt-in)

Verdict: Research box frozen. Phase 14-16 plateau confirms dispatch/
routing optimization ROI is exhausted post-Phase-6 FastLane collapse.

---

## Phase 17: FORCE_LIBC Gap Validation — Case B Confirmed

Purpose: Validate "system malloc faster" observation using same-binary
A/B testing to isolate allocator logic差 vs binary layout penalty.

Method:
- Same-binary toggle: HAKMEM_FORCE_LIBC_ALLOC=0/1 (bench_random_mixed_hakmem)
- System binary: bench_random_mixed_system (21K separate binary)
- Perf stat: Hardware counter analysis (I-cache, cycles, instructions)

Result: **Case B confirmed** — Allocator差 negligible, layout penalty dominates.

Gap breakdown (Mixed, 20M iters, ws=400):
- hakmem (FORCE_LIBC=0): 48.12M ops/s
- libc (FORCE_LIBC=1, same binary): 48.31M ops/s → +0.39% (noise level)
- system binary (21K): 83.85M ops/s → +73.57% vs libc, +74.26% vs hakmem

Perf stat (200M iters):
- I-cache misses: 153K (hakmem) → 68K (system) = -55% (smoking gun)
- Cycles: 17.9B → 10.2B = -43%
- Instructions: 41.3B → 21.5B = -48%
- Binary size: 653K → 21K (30x difference)

Root cause: Binary size (30x) causes I-cache thrashing. Code bloat >>
algorithmic efficiency.

Conclusion: Phase 12's "system malloc 1.6x faster" was real, but
misattributed. Gap is layout/I-cache, NOT allocator algorithm.

Files:
- docs/analysis/PHASE17_*.md (results/instructions)
- scripts/run_mixed_10_cleanenv.sh (Phase 9/10 defaults aligned)

Next: Phase 18 Hot Text Isolation (layout optimization, not algorithm opt)

---

## Phase 18: Hot Text Isolation — Design Added

Purpose: Reduce I-cache misses + instruction footprint via layout control
(binary optimization, not allocator algorithm changes).

Strategy (v1 → v2 progression):

v1 (TU split + hot/cold attrs + optional gc-sections):
- Target: +2% throughput (GO threshold, realistic for layout tweaks)
- Secondary: I-cache -10%, instructions -5% (direction confirmation)
- Risk: Low (reversible via build knob)
- Expected: +0-2% (NEUTRAL likely, but validates approach)

v2 (BENCH_MINIMAL compile-out):
- Target: +10-20% throughput (本命)
- Method: Conditional compilation removes stats/ENV/debug from hot path
- Expected: Instruction count -30-40% → significant I-cache improvement

Files:
- docs/analysis/PHASE18_*.md (design/instructions)
- CURRENT_TASK.md (Phase 17 complete, Phase 18 v1/v2 plan)

Build gate: HOT_TEXT_ISOLATION=0/1 (Makefile knob)

Next: Implement Phase 18 v1 (TU split first, BENCH_MINIMAL if v1 NEUTRAL)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
 2025-12-15 05:25:47 +09:00
Moe Charm (CI)
cbb35ee27f Phase 13 v1 + E5-2 retest: Both NEUTRAL, freeze as research boxes 
Phase 13 v1: Header Write Elimination (C7 preserve header)
- Verdict: NEUTRAL (+0.78%)
- Implementation: HAKMEM_TINY_C7_PRESERVE_HEADER ENV gate (default OFF)
- Makes C7 nextptr offset conditional (0→1 when enabled)
- 4-point matrix A/B test results:
  * Case A (baseline): 51.49M ops/s
  * Case B (WRITE_ONCE=1): 52.07M ops/s (+1.13%)
  * Case C (C7_PRESERVE=1): 51.36M ops/s (-0.26%)
  * Case D (both): 51.89M ops/s (+0.78% NEUTRAL)
- Action: Freeze as research box (default OFF, manual opt-in)

Phase 5 E5-2: Header Write-Once retest (promotion test)
- Verdict: NEUTRAL (+0.54%)
- Motivation: Phase 13 Case B showed +1.13%, re-tested with dedicated 20-run
- Results (20-run):
  * Case A (baseline): 51.10M ops/s
  * Case B (WRITE_ONCE=1): 51.37M ops/s (+0.54%)
- Previous test: +0.45% (consistent with NEUTRAL)
- Action: Keep as research box (default OFF, manual opt-in)

Key findings:
- Header write tax optimization shows consistent NEUTRAL results
- Neither Phase 13 v1 nor E5-2 reaches GO threshold (+1.0%)
- Both implemented as reversible ENV gates for future research

Files changed:
- New: core/box/tiny_c7_preserve_header_env_box.{c,h}
- Modified: core/box/tiny_layout_box.h (C7 offset conditional)
- Modified: core/tiny_nextptr.h, core/box/tiny_header_box.h (comments)
- Modified: core/bench_profile.h (refresh sync)
- Modified: Makefile (add new .o files)
- Modified: scripts/run_mixed_10_cleanenv.sh (add C7_PRESERVE ENV)
- Docs: PHASE13_*, PHASE5_E5_2_HEADER_WRITE_ONCE_* (design/results)

Next: Phase 14 (Pointer-chase reduction, tcache-style intrusive LIFO)

🤖 Generated with Claude Code

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
 2025-12-15 00:32:25 +09:00
Moe Charm (CI)
71b1354d32 Phase 10: FREE-TINY-FAST MONO LEGACY DIRECT (GO +1.89%) 
Results:
- A/B test: +1.89% on Mixed (10-run, clean env)
- Baseline: 51.96M ops/s
- Optimized: 52.94M ops/s
- Improvement: +984K ops/s (+1.89%)
- C6-heavy verification: +7.86% (nonlegacy_mask works correctly, no misfires)

Strategy:
- Extend Phase 9 (C0-C3 DUALHOT) to C4-C7 LEGACY DIRECT
- Fail-Fast principle: Never misclassify MID/ULTRA/V7 as LEGACY
- nonlegacy_mask: Cached at init, hot path uses single bit operation

Success factors:
1. Performance improvement: +1.89% (1.9x GO threshold)
2. Safety verified: nonlegacy_mask prevents MID v3 misfire in C6-heavy
3. Phase 9 coexistence: C0-C3 (Phase 9) + C4-C7 (Phase 10) = full LEGACY coverage
4. Minimal overhead: Single bit operation in hot path (mask & (1u<<class))

Implementation:
- Patch 1: ENV gate box (free_tiny_fast_mono_legacy_direct_env_box.h)
  - ENV: HAKMEM_FREE_TINY_FAST_MONO_LEGACY_DIRECT=0/1 (default 0)
  - nonlegacy_mask cached (reuses free_policy_fast_v2_nonlegacy_mask())
  - Probe window: 64 (avoid bench_profile putenv race)
- Patch 2: Early-exit in free_tiny_fast() (malloc_tiny_fast.h)
  - Conditions: !nonlegacy_mask, route==LEGACY, !LARSON_FIX, done==1
  - Direct call: tiny_legacy_fallback_free_base()
- Patch 3: Visibility (free_path_stats_box.h)
  - mono_legacy_direct_hit counter (compile-out in release)
- Patch 4: cleanenv extension (run_mixed_10_cleanenv.sh)
  - ENV leak protection

Safety verification (C6-heavy):
- OFF: 19.75M ops/s
- ON: 21.30M ops/s (+7.86%)
- nonlegacy_mask correctly excludes C6 (MID v3 active)
- Improvement from C0-C5, C7 direct path acceleration

Files modified:
- core/bench_profile.h: add to MIXED_TINYV3_C7_SAFE preset
- core/front/malloc_tiny_fast.h: early-exit insertion
- core/box/free_path_stats_box.h: counter
- core/box/free_tiny_fast_mono_legacy_direct_env_box.h: NEW (ENV gate + nonlegacy_mask)
- scripts/run_mixed_10_cleanenv.sh: ENV leak protection

Health check: PASSED (all profiles)

Promotion: Added to MIXED_TINYV3_C7_SAFE preset (default ON, opt-out)

Rollback: HAKMEM_FREE_TINY_FAST_MONO_LEGACY_DIRECT=0

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
 2025-12-14 20:09:40 +09:00
Moe Charm (CI)
871034da1f Phase 9: FREE-TINY-FAST MONO DUALHOT (GO +2.72%) 
Results:
- A/B test: +2.72% on Mixed (10-run, clean env)
- Baseline: 48.89M ops/s
- Optimized: 50.22M ops/s
- Improvement: +1.33M ops/s (+2.72%)
- Stability: Standard deviation reduced by 60.8% (2.44M → 955K ops/s)

Strategy:
- Transplant C0-C3 "second hot" path to monolithic free_tiny_fast()
- Early-exit within monolithic (no hot/cold split)
- FastLane free now benefits from C0-C3 direct path

Success factors:
1. Performance improvement: +2.72% (2.7x GO threshold)
2. Stability improvement: 2.6x more stable (stdev 60.8% reduction)
3. Learned from Phase 7 failure:
   - Phase 7: Function split (hot/cold) → NO-GO
   - Phase 9: Early-exit within monolithic → GO
4. FastLane free compatibility: C0-C3 direct path now works with FastLane
5. Policy snapshot overhead reduction: C0-C3 (48% of Mixed) skip route lookup

Implementation:
- Patch 1: ENV gate box (free_tiny_fast_mono_dualhot_env_box.h)
  - ENV: HAKMEM_FREE_TINY_FAST_MONO_DUALHOT=0/1 (default 0)
  - Probe window: 64 (avoid bench_profile putenv race)
- Patch 2: Early-exit in free_tiny_fast() (malloc_tiny_fast.h)
  - Conditions: class_idx <= 3, !LARSON_FIX, route==LEGACY
  - Direct call: tiny_legacy_fallback_free_base()
- Patch 3: Visibility (free_path_stats_box.h)
  - mono_dualhot_hit counter (compile-out in release)
- Patch 4: cleanenv extension (run_mixed_10_cleanenv.sh)
  - ENV leak protection

Files modified:
- core/bench_profile.h: add to MIXED_TINYV3_C7_SAFE preset
- core/front/malloc_tiny_fast.h: early-exit insertion
- core/box/free_path_stats_box.h: counter
- core/box/free_tiny_fast_mono_dualhot_env_box.h: NEW (ENV gate)
- scripts/run_mixed_10_cleanenv.sh: ENV leak protection

Health check: PASSED (all profiles)

Promotion: Added to MIXED_TINYV3_C7_SAFE preset (default ON, opt-out)

Rollback: HAKMEM_FREE_TINY_FAST_MONO_DUALHOT=0

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
 2025-12-14 19:16:49 +09:00
Moe Charm (CI)
6d38511318 Phase 5: E7 prune no-go (keep frozen boxes); add clean-env runner 2025-12-14 08:11:20 +09:00
Moe Charm (CI)
21e2e4ac2b Phase 4 E3-4: ENV Constructor Init (+4.75% GO) 
Target: Eliminate E1 lazy init check overhead (3.22% self%)
- E1 consolidated ENV gates but lazy check remained in hot path
- Strategy: __attribute__((constructor(101))) for pre-main init

Implementation:
- ENV gate: HAKMEM_ENV_SNAPSHOT_CTOR=0/1 (default 0, research box)
- core/box/hakmem_env_snapshot_box.c: Constructor function added
  - Reads ENV before main() when CTOR=1
  - Refresh also syncs gate state for bench_profile putenv
- core/box/hakmem_env_snapshot_box.h: Dual-mode enabled check
  - CTOR=1 fast path: direct global read (no lazy branch)
  - CTOR=0 fallback: legacy lazy init (rollback safe)
  - Branch hints adjusted for default OFF baseline

A/B Test Results (Mixed, 10-run, 20M iters, E1=1):
- Baseline (CTOR=0): 44.28M ops/s (mean), 44.60M ops/s (median)
- Optimized (CTOR=1): 46.38M ops/s (mean), 46.53M ops/s (median)
- Improvement: +4.75% mean, +4.35% median

Decision: GO (+4.75% >> +0.5% threshold)
- Expected +0.5-1.5%, achieved +4.75%
- Lazy init branch overhead was larger than expected
- Action: Keep as research box (default OFF), evaluate promotion

Phase 4 Cumulative:
- E1 (ENV Snapshot): +3.92%
- E2 (Alloc Per-Class): -0.21% (NEUTRAL, frozen)
- E3-4 (Constructor Init): +4.75%
- Total Phase 4: ~+8.5%

Deliverables:
- docs/analysis/PHASE4_E3_ENV_CONSTRUCTOR_INIT_DESIGN.md
- docs/analysis/PHASE4_E3_ENV_CONSTRUCTOR_INIT_NEXT_INSTRUCTIONS.md
- docs/analysis/PHASE4_COMPREHENSIVE_STATUS_ANALYSIS.md
- docs/analysis/PHASE4_EXECUTIVE_SUMMARY.md
- scripts/verify_health_profiles.sh (sanity check script)
- CURRENT_TASK.md (E3-4 complete, next instructions)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
 2025-12-14 02:57:35 +09:00
Moe Charm (CI)
b51b600e8d Phase 4-Step1: Add PGO workflow automation (+6.25% performance) 
Implemented automated Profile-Guided Optimization workflow using Box pattern:

Performance Improvement:
- Baseline:      57.0 M ops/s
- PGO-optimized: 60.6 M ops/s
- Gain: +6.25% (within expected +5-10% range)

Implementation:
1. scripts/box/pgo_tiny_profile_config.sh - 5 representative workloads
2. scripts/box/pgo_tiny_profile_box.sh - Automated profile collection
3. Makefile PGO targets:
   - pgo-tiny-profile: Build instrumented binaries
   - pgo-tiny-collect: Collect .gcda profile data
   - pgo-tiny-build:   Build optimized binaries
   - pgo-tiny-full:    Complete workflow (profile → collect → build → test)
4. Makefile help target: Added PGO instructions for discoverability

Design:
- Box化: Single responsibility, clear contracts
- Deterministic: Fixed seeds (42) for reproducibility
- Safe: Validation, error detection, timeout protection (30s/workload)
- Observable: Progress reporting, .gcda verification (33 files generated)

Workload Coverage:
- Random mixed: 3 working set sizes (128/256/512 slots)
- Tiny hot: 2 size classes (16B/64B)
- Total: 5 workloads covering hot/cold paths

Documentation:
- PHASE4_STEP1_COMPLETE.md - Completion report
- CURRENT_TASK.md - Phase 4 roadmap (Step 1 complete ✓)
- docs/design/PHASE4_TINY_FRONT_BOX_DESIGN.md - Complete Phase 4 design

Next: Phase 4-Step2 (Hot/Cold Path Box, target +10-15%)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
 2025-11-29 11:28:38 +09:00
Moe Charm (CI)
131cdb7b88 Doc: Add benchmark reports, atomic freelist docs, and .gitignore update 
Phase 1 Commit: Comprehensive documentation and build system cleanup

Added Documentation:
- BENCHMARK_SUMMARY_20251122.md: Current performance baseline
- COMPREHENSIVE_BENCHMARK_REPORT_20251122.md: Detailed analysis
- LARSON_SLOWDOWN_INVESTIGATION_REPORT.md: Larson benchmark deep dive
- ATOMIC_FREELIST_*.md (5 files): Complete atomic freelist documentation
  - Implementation strategy, quick start, site-by-site guide
  - Index and summary for easy navigation

Added Scripts:
- run_comprehensive_benchmark.sh: Automated benchmark runner
- scripts/analyze_freelist_sites.sh: Freelist analysis tool
- scripts/verify_atomic_freelist_conversion.sh: Conversion verification

Build System:
- Updated .gitignore: Added *.d (build dependency files)
- Cleaned up tracked .d files (will be ignored going forward)

Performance Status (2025-11-22):
- Random Mixed 256B: 59.6M ops/s (VERIFIED WORKING)
- Benchmark command: ./out/release/bench_random_mixed_hakmem 10000000 256 42
- Known issue: workset=8192 causes SEGV (to be fixed separately)

Notes:
- bench_random_mixed.c already tracked, working state confirmed
- Ultra SLIM implementation backed up to /tmp/ (Phase 2 restore pending)
- Documentation covers atomic freelist conversion and benchmarking methodology

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
 2025-11-22 06:11:55 +09:00
Moe Charm (CI)
7975e243ee Phase 7 Task 3: Pre-warm TLS cache (+180-280% improvement!) 
MAJOR SUCCESS: HAKMEM now achieves 85-92% of System malloc on tiny
allocations (128-512B) and BEATS System at 146% on 1024B allocations!

Performance Results:
- Random Mixed 128B: 21M → 59M ops/s (+181%) 🚀
- Random Mixed 256B: 19M → 70M ops/s (+268%) 🚀
- Random Mixed 512B: 21M → 68M ops/s (+224%) 🚀
- Random Mixed 1024B: 21M → 65M ops/s (+210%, 146% of System!) 🏆
- Larson 1T: 2.68M ops/s (stable, no regression)

Implementation:
1. Task 3a: Remove profiling overhead in release builds
   - Wrapped RDTSC calls in #if !HAKMEM_BUILD_RELEASE
   - Compiler can eliminate profiling code completely
   - Effect: +2% (2.68M → 2.73M Larson)

2. Task 3b: Simplify refill logic
   - Use constants from hakmem_build_flags.h
   - TLS cache already optimal
   - Effect: No regression

3. Task 3c: Pre-warm TLS cache (GAME CHANGER!)
   - Pre-allocate 16 blocks per class at init
   - Eliminates cold-start penalty
   - Effect: +180-280% improvement 🚀

Root Cause:
The bottleneck was cold-start, not the hot path! First allocation in
each class triggered a SuperSlab refill (100+ cycles). Pre-warming
eliminated this penalty, revealing Phase 7's true potential.

Files Modified:
- core/hakmem_tiny.c: Pre-warm function implementation
- core/box/hak_core_init.inc.h: Pre-warm initialization call
- core/tiny_alloc_fast.inc.h: Profiling overhead removal
- core/hakmem_phase7_config.h: Task 3 constants (NEW)
- core/hakmem_build_flags.h: Phase 7 feature flags
- Makefile: PREWARM_TLS flag, phase7 targets
- CLAUDE.md: Phase 7 success summary
- PHASE7_TASK3_RESULTS.md: Comprehensive results report (NEW)

Build:
make HEADER_CLASSIDX=1 AGGRESSIVE_INLINE=1 PREWARM_TLS=1 phase7-bench

🎉 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
 2025-11-08 12:54:52 +09:00
Moe Charm (CI)
382980d450 Phase 6-2.4: Fix SuperSlab free SEGV: remove guess loop and add memory readability check; add registry atomic consistency (base as _Atomic uintptr_t with acq/rel); add debug toggles (SUPER_REG_DEBUG/REQTRACE); update CURRENT_TASK with results and next steps; capture suite results. 2025-11-07 18:07:48 +09:00
Moe Charm (CI)
8f3095fb85 CI-safe debug runners: add ASan LD_PRELOAD + UBSan mailbox targets; add asan_preload script; document sanitizer-safe workflows and results in CURRENT_TASK.md (debug complete). 2025-11-07 12:09:28 +09:00
Moe Charm (CI)
1da8754d45 CRITICAL FIX: TLS 未初期化による 4T SEGV を完全解消 
**問題:**
- Larson 4T で 100% SEGV (1T は 2.09M ops/s で完走)
- System/mimalloc は 4T で 33.52M ops/s 正常動作
- SS OFF + Remote OFF でも 4T で SEGV

**根本原因: (Task agent ultrathink 調査結果)**
```
CRASH: mov (%r15),%r13
R15 = 0x6261  ← ASCII "ba" (ゴミ値、未初期化TLS)
```

Worker スレッドの TLS 変数が未初期化:
- `__thread void* g_tls_sll_head[TINY_NUM_CLASSES];`  ← 初期化なし
- pthread_create() で生成されたスレッドでゼロ初期化されない
- NULL チェックが通過 (0x6261 != NULL) → dereference → SEGV

**修正内容:**
全 TLS 配列に明示的初期化子 `= {0}` を追加:

1. **core/hakmem_tiny.c:**
   - `g_tls_sll_head[TINY_NUM_CLASSES] = {0}`
   - `g_tls_sll_count[TINY_NUM_CLASSES] = {0}`
   - `g_tls_live_ss[TINY_NUM_CLASSES] = {0}`
   - `g_tls_bcur[TINY_NUM_CLASSES] = {0}`
   - `g_tls_bend[TINY_NUM_CLASSES] = {0}`

2. **core/tiny_fastcache.c:**
   - `g_tiny_fast_cache[TINY_FAST_CLASS_COUNT] = {0}`
   - `g_tiny_fast_count[TINY_FAST_CLASS_COUNT] = {0}`
   - `g_tiny_fast_free_head[TINY_FAST_CLASS_COUNT] = {0}`
   - `g_tiny_fast_free_count[TINY_FAST_CLASS_COUNT] = {0}`

3. **core/hakmem_tiny_magazine.c:**
   - `g_tls_mags[TINY_NUM_CLASSES] = {0}`

4. **core/tiny_sticky.c:**
   - `g_tls_sticky_ss[TINY_NUM_CLASSES][TINY_STICKY_RING] = {0}`
   - `g_tls_sticky_idx[TINY_NUM_CLASSES][TINY_STICKY_RING] = {0}`
   - `g_tls_sticky_pos[TINY_NUM_CLASSES] = {0}`

**効果:**
```
Before: 1T: 2.09M ✅  |  4T: SEGV 💀
After:  1T: 2.41M ✅  |  4T: 4.19M ✅  (+15% 1T, SEGV解消)
```

**テスト:**
```bash
# 1 thread: 完走
./larson_hakmem 2 8 128 1024 1 12345 1
→ Throughput = 2,407,597 ops/s ✅

# 4 threads: 完走（以前は SEGV）
./larson_hakmem 2 8 128 1024 1 12345 4
→ Throughput = 4,192,155 ops/s ✅
```

**調査協力:** Task agent (ultrathink mode) による完璧な根本原因特定

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
 2025-11-07 01:27:04 +09:00
Claude
f0c87d0cac Add Larson performance analysis and optimized profile 
Ultrathink analysis reveals root cause of 4x performance gap:

Key Findings:
- Single-thread: HAKMEM 0.46M ops/s vs system 4.29M ops/s (10.7%)
- Multi-thread: HAKMEM 1.81M ops/s vs system 7.23M ops/s (25.0%)
- Root cause: malloc() entry point has 8+ branch checks
- Bottleneck: Fast Path is structurally complex vs system tcache

Files Added:
- LARSON_PERFORMANCE_ANALYSIS_2025_11_05.md: Detailed analysis with 3 optimization strategies
- scripts/profiles/tinyhot_optimized.env: CLAUDE.md-based optimized config

Proposed Solutions:
- Option A: Optimize malloc() guard checks (+200-400% expected)
- Option B: Improve refill efficiency (+30-50% expected)
- Option C: Complete Fast Path simplification (+400-800% expected)

Target: Achieve 60-80% of system malloc performance
 2025-11-05 04:03:10 +00:00
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