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>

core/hakmem_debug.h

@@ -0,0 +1,197 @@
+// hakmem_debug.h - Debug Timing Infrastructure (Box理論: 計測箱)
+// Purpose: Lightweight timing measurement for performance analysis
+//
+// License: MIT
+// Date: 2025-10-21
+
+#pragma once
+#include <stdint.h>
+#include <time.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// ============================================================================
+// Configuration (環境変数で制御可能)
+// ============================================================================
+
+// Build-time guard: HAKMEM_DEBUG_TIMING=1 enables all timing code
+#ifndef HAKMEM_DEBUG_TIMING
+#define HAKMEM_DEBUG_TIMING 0  // Default: OFF (zero overhead)
+#endif
+
+// Runtime guard: HAKMEM_TIMING=1 enables timing at runtime
+// Sampling rate: HAKMEM_TIMING_SAMPLE=N (default: 64, means 1/64 sampling)
+
+// ============================================================================
+// Timing Categories (計測カテゴリID)
+// ============================================================================
+
+typedef enum {
+    // Syscalls and fallbacks
+    HKM_CAT_SYSCALL_MMAP = 0,
+    HKM_CAT_SYSCALL_MUNMAP,
+    HKM_CAT_SYSCALL_MADVISE,
+    HKM_CAT_FALLBACK_MALLOC,
+
+    // Whale Fast-Path
+    HKM_CAT_WHALE_GET,
+    HKM_CAT_WHALE_PUT,
+
+    // L2.5 Pool (Phase 6.11.3: Added for profiling)
+    HKM_CAT_L25_GET,
+    HKM_CAT_L25_PUT,
+    HKM_CAT_L25_REFILL,
+
+    // Tiny Pool (Phase 6.11.3: Added for profiling)
+    HKM_CAT_TINY_ALLOC,
+    HKM_CAT_TINY_FREE,
+    HKM_CAT_TINY_SLAB_SEARCH,
+
+    // BigCache (Phase 6.11.3: Added for profiling)
+    HKM_CAT_BIGCACHE_GET,
+    HKM_CAT_BIGCACHE_PUT,
+    HKM_CAT_BIGCACHE_EVICT_SCAN,
+
+    // Site Rules (Phase 6.11.3: Added for profiling)
+    HKM_CAT_SITE_RULES_LOOKUP,
+    HKM_CAT_SITE_RULES_ADOPT,
+
+    // ELO Learning (Phase 6.11.3: Added for profiling)
+    HKM_CAT_ELO_SELECT,
+    HKM_CAT_ELO_UPDATE,
+
+    // Top-level API (Phase 6.11.3: Added for profiling)
+    HKM_CAT_HAK_ALLOC,
+    HKM_CAT_HAK_FREE,
+
+    // Legacy (keep for compatibility)
+    HKM_CAT_REGION_GET,
+    HKM_CAT_REGION_PUT,
+    HKM_CAT_HASH_FNV1A,
+
+    // Mid (L2 Pool) fine-grained
+    HKM_CAT_POOL_GET,
+    HKM_CAT_POOL_LOCK,
+    HKM_CAT_POOL_REFILL,
+    HKM_CAT_TC_DRAIN,
+    HKM_CAT_TLS_FAST,
+    HKM_CAT_SHARD_STEAL,
+    // Additional Mid (L2) fine-grained (NEW)
+    HKM_CAT_POOL_TLS_RING_POP,
+    HKM_CAT_POOL_TLS_LIFO_POP,
+    HKM_CAT_POOL_REMOTE_PUSH,
+    HKM_CAT_POOL_ALLOC_TLS_PAGE,
+
+    // L2.5 (LargePool) fine-grained (NEW)
+    HKM_CAT_L25_LOCK,
+    HKM_CAT_L25_TLS_RING_POP,
+    HKM_CAT_L25_TLS_LIFO_POP,
+    HKM_CAT_L25_TC_DRAIN,
+    HKM_CAT_L25_REMOTE_PUSH,
+    HKM_CAT_L25_ALLOC_TLS_PAGE,
+    HKM_CAT_L25_SHARD_STEAL,
+
+    HKM_CAT_MAX  // Total number of categories
+} HkmTimingCategory;
+
+// ============================================================================
+// Timing Statistics (TLS per-thread)
+// ============================================================================
+
+typedef struct {
+    uint64_t count;     // Number of calls
+    uint64_t cycles;    // Total cycles (TSC or nanoseconds)
+} HkmTimingStat;
+
+// ============================================================================
+// Public API
+// ============================================================================
+
+// Initialize timing subsystem (called from hak_init)
+void hkm_timing_init(void);
+
+// Shutdown timing subsystem (called from hak_shutdown, dumps stats)
+void hkm_timing_shutdown(void);
+
+// Get current timestamp (TSC or clock_gettime)
+static inline uint64_t hkm_tsc_now(void);
+
+// Increment counter for category
+void hkm_count_inc(HkmTimingCategory cat);
+
+// Add cycles to category
+void hkm_cycles_add(HkmTimingCategory cat, uint64_t cycles);
+
+// Dump all statistics (called at exit or on demand)
+void hkm_timing_dump(void);
+
+// ============================================================================
+// Timing Macros (inline化、ビルド時に消える)
+// ============================================================================
+
+#if HAKMEM_DEBUG_TIMING
+
+// HKM_TIME_START(var) - Start timing, stores start time in var
+#define HKM_TIME_START(var) \
+    uint64_t var = hkm_tsc_now()
+
+// HKM_TIME_END(cat, var) - End timing, add elapsed time to category
+#define HKM_TIME_END(cat, var) \
+    do { \
+        uint64_t _end = hkm_tsc_now(); \
+        hkm_cycles_add(cat, _end - var); \
+        hkm_count_inc(cat); \
+    } while (0)
+
+// HKM_CNT(cat) - Increment count only (no timing)
+#define HKM_CNT(cat) \
+    hkm_count_inc(cat)
+
+#else
+
+// Build-time disabled: declare variables but no timing code (zero overhead)
+#define HKM_TIME_START(var) \
+    uint64_t var = 0
+
+#define HKM_TIME_END(cat, var) \
+    do { (void)(cat); (void)(var); } while (0)
+
+#define HKM_CNT(cat) \
+    do { (void)(cat); } while (0)
+
+#endif
+
+// ============================================================================
+// Internal: TSC/Clock Implementation (inline for zero overhead)
+// ============================================================================
+
+#if HAKMEM_DEBUG_TIMING
+
+static inline uint64_t hkm_tsc_now(void) {
+#if defined(__x86_64__) || defined(__i386__)
+    // x86/x64: Use RDTSC (fast, ~10 cycles)
+    uint32_t lo, hi;
+    __asm__ __volatile__ ("rdtsc" : "=a"(lo), "=d"(hi));
+    return ((uint64_t)hi << 32) | lo;
+#else
+    // Non-x86: Use clock_gettime (slower, but portable)
+    struct timespec ts;
+    clock_gettime(CLOCK_MONOTONIC_RAW, &ts);
+    return (uint64_t)ts.tv_sec * 1000000000ULL + (uint64_t)ts.tv_nsec;
+#endif
+}
+
+#else
+
+// Build-time disabled: empty inline function
+static inline uint64_t hkm_tsc_now(void) {
+    return 0;
+}
+
+#endif
+
+#ifdef __cplusplus
+}
+#endif