// pool_tls_ring.inc.h — Box: L2 Pool TLS ring/shard helpers
#ifndef POOL_TLS_RING_INC_H
#define POOL_TLS_RING_INC_H

// Minimal header write for Mid allocations (fast-path friendly)
static inline void mid_set_header(AllocHeader* hdr, size_t class_sz, uintptr_t site_id) {
    if (g_hdr_light_enabled >= 1) return;
    hdr->magic = HAKMEM_MAGIC;
    hdr->method = ALLOC_METHOD_POOL;
    hdr->size = class_sz;
    if (!g_hdr_light_enabled) {
        hdr->alloc_site = site_id;
        hdr->class_bytes = 0;
        hdr->owner_tid = (uintptr_t)(uintptr_t)pthread_self();
    }
}

// Branchless size→class LUT for 0..52KB（Bridge classes対応）
static inline int hak_pool_get_class_index(size_t size) {
    for (int i = 0; i < POOL_NUM_CLASSES; i++) {
        size_t cs = g_class_sizes[i];
        if (cs != 0 && size == cs) return i;
    }
    uint32_t kb = (uint32_t)((size + 1023) >> 10);
    extern const uint8_t SIZE_TO_CLASS[53];
    return (kb < 53) ? SIZE_TO_CLASS[kb] : -1;
}

// site_id→shard（64 shards）
static inline int hak_pool_get_shard_index(uintptr_t site_id) {
    if (!g_shard_mix_enabled) {
        return (int)((site_id >> 4) & (POOL_NUM_SHARDS - 1));
    }
    uint64_t x = (uint64_t)site_id;
    uint64_t tid = (uint64_t)(uintptr_t)pthread_self();
    x ^= (tid << 1);
    x += 0x9e3779b97f4a7c15ULL;
    x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9ULL;
    x = (x ^ (x >> 27)) * 0x94d049bb133111ebULL;
    x = (x ^ (x >> 31));
    return (int)((uint32_t)x & (POOL_NUM_SHARDS - 1));
}

// Bitmap helpers（O(1)）
static inline void set_nonempty_bit(int class_idx, int shard_idx) {
    atomic_fetch_or(&g_pool.nonempty_mask[class_idx], (uint64_t)(1ULL << shard_idx));
}
static inline void clear_nonempty_bit(int class_idx, int shard_idx) {
    atomic_fetch_and(&g_pool.nonempty_mask[class_idx], ~(uint64_t)(1ULL << shard_idx));
}
static inline int is_shard_nonempty(int class_idx, int shard_idx) {
    uint64_t mask = atomic_load(&g_pool.nonempty_mask[class_idx]);
    return (mask & (1ULL << shard_idx)) != 0;
}

// Remote MPSC → freelist（ロック下）
static inline void drain_remote_locked(int class_idx, int shard_idx) {
    uintptr_t head = atomic_exchange_explicit(&g_pool.remote_head[class_idx][shard_idx], (uintptr_t)0, memory_order_acq_rel);
    unsigned drained = 0;
    while (head) {
        PoolBlock* b = (PoolBlock*)head; head = (uintptr_t)b->next;
        b->next = g_pool.freelist[class_idx][shard_idx];
        g_pool.freelist[class_idx][shard_idx] = b; drained++;
    }
    if (drained) {
        atomic_fetch_sub_explicit(&g_pool.remote_count[class_idx][shard_idx], drained, memory_order_relaxed);
        if (g_pool.freelist[class_idx][shard_idx]) set_nonempty_bit(class_idx, shard_idx);
    }
}

// 近傍のnon-empty shardを選ぶ（無ければpreferred）
static inline int choose_nonempty_shard(int class_idx, int preferred) {
    uint64_t mask = atomic_load_explicit(&g_pool.nonempty_mask[class_idx], memory_order_acquire);
    if (!mask) return preferred;
    int shift = preferred & 63; uint64_t rot = (mask >> shift) | (mask << (64 - shift));
    if (!rot) return preferred; int off = __builtin_ctzll(rot);
    return (preferred + off) & (POOL_NUM_SHARDS - 1);
}

// TLSアクティブページの確保（bump-run）
static inline int alloc_tls_page(int class_idx, PoolTLSPage* ap) {
    size_t user_size = g_class_sizes[class_idx]; size_t block_size = HEADER_SIZE + user_size;
    int blocks_per_page = POOL_PAGE_SIZE / block_size; if (blocks_per_page <= 0) return 0;
    void* page = mmap(NULL, POOL_PAGE_SIZE, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); if (!page) return 0;
    ap->page = page; ap->bump = (char*)page; ap->end = (char*)page + POOL_PAGE_SIZE; ap->count = blocks_per_page;
    mid_desc_register(page, class_idx, (uint64_t)(uintptr_t)pthread_self());
    g_pool.refills[class_idx]++; g_pool.total_pages_allocated++; g_pool.pages_by_class[class_idx]++; g_pool.total_bytes_allocated += POOL_PAGE_SIZE; return 1;
}

// TLS ring/LIFO への補充（リンク無し）
static inline int refill_tls_from_active_page(int class_idx, PoolTLSRing* ring, PoolTLSBin* bin, PoolTLSPage* ap, int need) {
    if (!ap || !ap->page || ap->count <= 0 || ap->bump >= ap->end) return 0; size_t blk = HEADER_SIZE + g_class_sizes[class_idx];
    int moved = 0, to_add = need;
    while (to_add > 0 && ap->bump < ap->end && ap->count > 0) {
        PoolBlock* b = (PoolBlock*)(void*)ap->bump;
        if (g_tls_ring_enabled && ring->top < POOL_L2_RING_CAP) { ring->items[ring->top++] = b; } else { b->next = bin->lo_head; bin->lo_head = b; bin->lo_count++; }
        ap->bump += blk; ap->count--; moved++; to_add--;
    }
    return moved;
}

#endif // POOL_TLS_RING_INC_H