diff --git a/core/tiny_c7_ultra.c b/core/tiny_c7_ultra.c
index 789ba366..31fa6de1 100644
--- a/core/tiny_c7_ultra.c
+++ b/core/tiny_c7_ultra.c
@@ -12,6 +12,11 @@
 #include "box/tiny_front_v3_env_box.h"
 #include "box/free_path_stats_box.h"
 
+// Phase PERF-ULTRA-REFILL-OPT-1a: Import page size shift macro
+// (defined in tiny_c7_ultra_segment.c for consistency)
+// We'll define it locally here as well for convenience
+#define TINY_C7_ULTRA_PAGE_SHIFT 16  // 64KiB = 2^16
+
 #ifndef likely
 #define likely(x)   __builtin_expect(!!(x), 1)
 #define unlikely(x) __builtin_expect(!!(x), 0)
@@ -78,7 +83,8 @@ bool tiny_c7_ultra_refill(tiny_c7_ultra_tls_t* tls) {
         if (!seg) return false;
         tls->seg = seg;
         tls->seg_base = (uintptr_t)seg->base;
-        tls->seg_end = tls->seg_base + ((size_t)seg->num_pages * seg->page_size);
+        // Phase PERF-ULTRA-REFILL-OPT-1a: Use bit shift instead of multiplication
+        tls->seg_end = tls->seg_base + ((size_t)seg->num_pages << TINY_C7_ULTRA_PAGE_SHIFT);
     }
 
     size_t block_sz = tls->block_size;
@@ -107,7 +113,8 @@ bool tiny_c7_ultra_refill(tiny_c7_ultra_tls_t* tls) {
     }
 
     tiny_c7_ultra_page_meta_t* page = &seg->pages[chosen];
-    uint8_t* base = (uint8_t*)seg->base + ((size_t)chosen * seg->page_size);
+    // Phase PERF-ULTRA-REFILL-OPT-1a: Use bit shift instead of multiplication
+    uint8_t* base = (uint8_t*)seg->base + ((size_t)chosen << TINY_C7_ULTRA_PAGE_SHIFT);
 
     // If page is uninitialized, carve it
     if (page->capacity == 0) {
@@ -170,17 +177,17 @@ void tiny_c7_ultra_free(void* ptr) {
     tiny_c7_ultra_tls_t* tls = &g_tiny_c7_ultra_tls;
     void* base = (uint8_t*)ptr - 1;  // Convert USER -> BASE pointer
 
-    // 1) Initial segment learning (cold path, once per thread)
-    if (unlikely(tls->seg_base == 0)) {
-        tiny_c7_ultra_segment_t* seg = tiny_c7_ultra_segment_from_ptr(ptr);
-        if (seg != NULL) {
-            tls->seg = seg;
-            tls->seg_base = (uintptr_t)seg->base;
-            tls->seg_end = tls->seg_base + ((size_t)seg->num_pages * seg->page_size);
-        }
-    }
+    // Phase PERF-ULTRA-REFILL-OPT-1b: Segment learning moved to refill (alloc cold path)
+    // In normal allocation patterns, alloc is always called before free on each thread.
+    // Therefore, seg_base/seg_end are guaranteed to be initialized by refill's
+    // tiny_c7_ultra_segment_acquire() call (line 82-87).
+    //
+    // This optimization removes the per-free segment learning overhead.
+    // Risk: If a thread does free() before any alloc(), it will fallback to so_free().
+    // This is acceptable because it's an unusual pattern.
 
-    // 2) Fast path: range check + TLS push
+    // Fast path: assume segment already learned by refill
+    // No unlikely() guard needed because refill always runs first in normal patterns
     uintptr_t addr = (uintptr_t)base;
     if (likely(tls->seg_base != 0 &&
                addr >= tls->seg_base &&
diff --git a/core/tiny_c7_ultra_segment.c b/core/tiny_c7_ultra_segment.c
index 23cdfec3..a38247b0 100644
--- a/core/tiny_c7_ultra_segment.c
+++ b/core/tiny_c7_ultra_segment.c
@@ -11,6 +11,7 @@
 // 2MiB セグメントを 64KiB ページに分割（C7 専用、pow2 で mask しやすく）
 #define TINY_C7_ULTRA_SEG_SIZE  ((size_t)(2 * 1024 * 1024))
 #define TINY_C7_ULTRA_PAGE_SIZE ((size_t)(64 * 1024))
+#define TINY_C7_ULTRA_PAGE_SHIFT 16  // 64KiB = 2^16 (for O(1) bit shift instead of division)
 
 static __thread tiny_c7_ultra_segment_t* g_ultra_seg;
 
@@ -92,7 +93,8 @@ tiny_c7_ultra_page_meta_t* tiny_c7_ultra_page_of(void* p,
     uintptr_t base = (uintptr_t)seg->base;
     uintptr_t addr = (uintptr_t)p;
     size_t offset = (size_t)(addr - base);
-    uint32_t idx = (uint32_t)(offset / seg->page_size);
+    // Phase PERF-ULTRA-REFILL-OPT-1a: Replace division with bit shift for O(1) lookup
+    uint32_t idx = (uint32_t)(offset >> TINY_C7_ULTRA_PAGE_SHIFT);
     if (idx >= seg->num_pages) {
         return NULL;
     }