Files
hakmem/core/box/front_gate_classifier.h
Moe Charm (CI) 03ba62df4d Phase 23 Unified Cache + PageFaultTelemetry generalization: Mid/VM page-fault bottleneck identified
Summary:
- Phase 23 Unified Cache: +30% improvement (Random Mixed 256B: 18.18M → 23.68M ops/s)
- PageFaultTelemetry: Extended to generic buckets (C0-C7, MID, L25, SSM)
- Measurement-driven decision: Mid/VM page-faults (80-100K) >> Tiny (6K) → prioritize Mid/VM optimization

Phase 23 Changes:
1. Unified Cache implementation (core/front/tiny_unified_cache.{c,h})
   - Direct SuperSlab carve (TLS SLL bypass)
   - Self-contained pop-or-refill pattern
   - ENV: HAKMEM_TINY_UNIFIED_CACHE=1, HAKMEM_TINY_UNIFIED_C{0-7}=128

2. Fast path pruning (tiny_alloc_fast.inc.h, tiny_free_fast_v2.inc.h)
   - Unified ON → direct cache access (skip all intermediate layers)
   - Alloc: unified_cache_pop_or_refill() → immediate fail to slow
   - Free: unified_cache_push() → fallback to SLL only if full

PageFaultTelemetry Changes:
3. Generic bucket architecture (core/box/pagefault_telemetry_box.{c,h})
   - PF_BUCKET_{C0-C7, MID, L25, SSM} for domain-specific measurement
   - Integration: hak_pool_try_alloc(), l25_alloc_new_run(), shared_pool_allocate_superslab_unlocked()

4. Measurement results (Random Mixed 500K / 256B):
   - Tiny C2-C7: 2-33 pages, high reuse (64-3.8 touches/page)
   - SSM: 512 pages (initialization footprint)
   - MID/L25: 0 (unused in this workload)
   - Mid/Large VM benchmarks: 80-100K page-faults (13-16x higher than Tiny)

Ring Cache Enhancements:
5. Hot Ring Cache (core/front/tiny_ring_cache.{c,h})
   - ENV: HAKMEM_TINY_HOT_RING_ENABLE=1, HAKMEM_TINY_HOT_RING_C{0-7}=size
   - Conditional compilation cleanup

Documentation:
6. Analysis reports
   - RANDOM_MIXED_BOTTLENECK_ANALYSIS.md: Page-fault breakdown
   - RANDOM_MIXED_SUMMARY.md: Phase 23 summary
   - RING_CACHE_ACTIVATION_GUIDE.md: Ring cache usage
   - CURRENT_TASK.md: Updated with Phase 23 results and Phase 24 plan

Next Steps (Phase 24):
- Target: Mid/VM PageArena/HotSpanBox (page-fault reduction 80-100K → 30-40K)
- Tiny SSM optimization deferred (low ROI, ~6K page-faults already optimal)
- Expected improvement: +30-50% for Mid/Large workloads

Generated with Claude Code

Co-Authored-By: Claude <noreply@anthropic.com>
2025-11-17 02:47:58 +09:00

80 lines
2.7 KiB
C

// front_gate_classifier.h - Box FG: Pointer Classification Front Gate
//
// Purpose: Single point of truth for classifying pointers (Tiny/Pool/Mid/Large)
// Design: Heuristic-free, safe header probe + Registry lookup fallback
//
// Key Rules:
// 1. ptr-1 is read ONLY here (never elsewhere)
// 2. Header probe only when safe (same page + readable)
// 3. C7 (headerless) always identified via Registry
// 4. No 1KB alignment heuristics (eliminated false positives)
//
// Architecture:
// - Box FG (this): Classification authority
// - Box REG: SuperSlab registry (O(1) reverse lookup)
// - Box TLS: next pointer clearing for C7
//
// Performance:
// - Fast path (C0-C6 header): 5-10 cycles (unchanged)
// - Slow path (C7 REG): 50-100 cycles (rare)
// - Safety: SEGV eliminated, false positive = 0%
#ifndef FRONT_GATE_CLASSIFIER_H
#define FRONT_GATE_CLASSIFIER_H
#include <stdint.h>
#include <stddef.h>
// Forward declaration
struct SuperSlab;
// Pointer classification kinds
typedef enum {
PTR_KIND_TINY_HEADER, // C0-C6: Has 1-byte header (fast path)
PTR_KIND_TINY_HEADERLESS, // C7: Headerless 1KB blocks (REG path)
PTR_KIND_POOL_TLS, // Pool TLS 8KB-52KB
PTR_KIND_MID_LARGE, // Mid/Large allocations
PTR_KIND_UNKNOWN // Unknown/external allocation
} tiny_ptr_kind_t;
// Classification result
typedef struct {
tiny_ptr_kind_t kind; // Classification result
int class_idx; // Tiny class (0-7), or -1 if not Tiny
struct SuperSlab* ss; // SuperSlab pointer (from Registry, or NULL)
int slab_idx; // Slab index within SuperSlab (or -1)
} ptr_classification_t;
// ========== Front Gate API ==========
// Classify pointer (single point of truth)
// Returns: Classification result with kind, class_idx, SuperSlab
//
// Strategy:
// 1. Try safe header probe (C0-C6 fast path: 5-10 cycles)
// 2. Fallback to Registry lookup (C7 or header failed)
// 3. Check Pool TLS magic
// 4. Check AllocHeader (16-byte malloc/mmap)
// 5. Return UNKNOWN if all fail
//
// Safety:
// - Header probe only if: (ptr & 0xFFF) >= 1 (same page)
// - No 1KB alignment heuristics
// - Registry provides ground truth for headerless
ptr_classification_t classify_ptr(void* ptr);
// ========== Debug/Stats (optional) ==========
#if !HAKMEM_BUILD_RELEASE
// Track classification hit rates
extern __thread uint64_t g_classify_header_hit;
extern __thread uint64_t g_classify_headerless_hit;
extern __thread uint64_t g_classify_pool_hit;
extern __thread uint64_t g_classify_mid_large_hit;
extern __thread uint64_t g_classify_unknown_hit;
void front_gate_print_stats(void);
#endif
#endif // FRONT_GATE_CLASSIFIER_H