Phase V6-HDR-4: Headerless 最適化 (P0 + P1)

## P0: Double validation 排除
- region_id_lookup_v6() で TLS segment 登録済み + 範囲内なら
  small_page_meta_v6_of() を呼ばずに直接 page_meta を計算
- 削除された重複チェック:
  - slot->in_use (TLS登録で保証)
  - small_ptr_in_segment_v6() (addr範囲で既にチェック済み)
  - 関数呼び出しオーバーヘッド
- 推定効果: +1-2% (6-8 instructions 削減)

## P1: TLS cache に page_meta キャッシュ追加
- RegionIdTlsCache に追加:
  - last_page_base / last_page_end (ページ範囲)
  - last_page (SmallPageMetaV6* 直接ポインタ)
- region_id_lookup_cached_v6() で same-page hit 時は
  page_meta lookup を完全スキップ
- 推定効果: +1.5-2.5% (10-12 instructions 削減)

## ベンチマーク結果 (揺れあり)
- V6-HDR-3 (P0/P1 前): -3.5% ~ -8.3% 回帰
- V6-HDR-4 (P0+P1 後): +2.7% ~ +12% 改善 (一部の run で)

設計原則:
- RegionIdBox は薄く保つ (分類のみ)
- キャッシュは TLS 側に寄せる
- same-page 判定で last_page_base/end を使用

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

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

core/box/region_id_v6_box.h

@@ -122,9 +122,13 @@ void region_id_observe_unregister(uint32_t id);
 // ============================================================================
 
 /// TLS cache for fast lookup (single entry)
+/// Phase V6-HDR-4 P1: page_meta キャッシュ追加
 typedef struct RegionIdTlsCache {
-    uintptr_t last_base;        // Cached region base
-    uintptr_t last_end;         // Cached region end
+    uintptr_t last_base;        // Cached segment base
+    uintptr_t last_end;         // Cached segment end
+    uintptr_t last_page_base;   // Cached page base (for same-page optimization)
+    uintptr_t last_page_end;    // Cached page end
+    SmallPageMetaV6* last_page; // Cached page_meta (direct pointer)
     RegionLookupV6 last_result; // Cached result
 } RegionIdTlsCache;
 

core/region_id_v6.c

@@ -103,21 +103,38 @@ RegionLookupV6 region_id_lookup_v6(void* ptr) {
 
     uintptr_t addr = (uintptr_t)ptr;
 
-    // Phase V6-HDR-3: Check TLS segment registration first (fast path)
+    // Phase V6-HDR-4 P0: Direct page_meta calculation (Double validation 排除)
+    // TLS segment が登録済みかつ範囲内なら、small_page_meta_v6_of() を呼ばずに
+    // 直接 page_meta を計算する。これにより:
+    // - slot->in_use チェックの重複を排除
+    // - small_ptr_in_segment_v6() の重複を排除
+    // - 関数呼び出しオーバーヘッドを削減
     if (g_v6_segment_registered && addr >= g_v6_segment_base && addr < g_v6_segment_end) {
-        // ptr is in this thread's v6 segment - get page_meta
-        SmallPageMetaV6* page = small_page_meta_v6_of(ptr);
-        if (page != NULL) {
+        // Calculate page_idx directly using TLS cached base
+        size_t page_idx = (addr - g_v6_segment_base) >> SMALL_PAGE_V6_SHIFT;
+
+        // Bounds check (should always pass if segment is valid)
+        if (likely(page_idx < SMALL_PAGES_PER_SEGMENT)) {
+            SmallPageMetaV6* page = &g_v6_segment->page_meta[page_idx];
+
+            // Minimal validation: only check capacity (page is in use)
+            if (likely(page->capacity > 0)) {
                 result.kind = REGION_KIND_SMALL_V6;
-            result.region_id = 1;  // Single TLS segment per thread
+                result.region_id = 1;
                 result.page_meta = page;
 
-            // Update TLS cache
+                // Phase V6-HDR-4 P1: Update TLS cache with page-level info
                 RegionIdTlsCache* cache = region_id_tls_cache_get();
                 cache->last_base = g_v6_segment_base;
                 cache->last_end = g_v6_segment_end;
+                // Page-level cache: same-page ptr は page_meta lookup をスキップ可能
+                uintptr_t page_base = g_v6_segment_base + (page_idx << SMALL_PAGE_V6_SHIFT);
+                cache->last_page_base = page_base;
+                cache->last_page_end = page_base + SMALL_PAGE_V6_SIZE;
+                cache->last_page = page;
                 cache->last_result = result;
             }
+        }
         return result;
     }
 
@@ -144,18 +161,35 @@ RegionLookupV6 region_id_lookup_cached_v6(void* ptr) {
     RegionIdTlsCache* cache = region_id_tls_cache_get();
     uintptr_t addr = (uintptr_t)ptr;
 
-    // TLS cache hit?
+    // Phase V6-HDR-4 P1: Same-page hit optimization
+    // 同一ページ内なら page_meta lookup を完全にスキップ
+    if (addr >= cache->last_page_base && addr < cache->last_page_end &&
+        cache->last_page != NULL && cache->last_page->capacity > 0) {
+        // Same page hit - return cached page_meta directly
+        RegionLookupV6 result = cache->last_result;
+        result.page_meta = cache->last_page;
+        return result;
+    }
+
+    // Segment hit but different page - need to recalculate page_meta
     if (addr >= cache->last_base && addr < cache->last_end &&
         cache->last_result.kind != REGION_KIND_UNKNOWN) {
-        // Cache hit - but still need to get the correct page_meta
-        // for this specific ptr (page_idx may differ)
-        SmallPageMetaV6* page = small_page_meta_v6_of(ptr);
-        if (page) {
+        // Calculate new page_idx
+        size_t page_idx = (addr - cache->last_base) >> SMALL_PAGE_V6_SHIFT;
+        if (likely(page_idx < SMALL_PAGES_PER_SEGMENT && g_v6_segment != NULL)) {
+            SmallPageMetaV6* page = &g_v6_segment->page_meta[page_idx];
+            if (likely(page->capacity > 0)) {
                 RegionLookupV6 result = cache->last_result;
                 result.page_meta = page;
+                // Update page-level cache
+                uintptr_t page_base = cache->last_base + (page_idx << SMALL_PAGE_V6_SHIFT);
+                cache->last_page_base = page_base;
+                cache->last_page_end = page_base + SMALL_PAGE_V6_SIZE;
+                cache->last_page = page;
                 return result;
             }
         }
+    }
 
     // Cache miss -> slow path
     return region_id_lookup_v6(ptr);