hakmem/core/hakmem_tiny_legacy_slow_box.inc

static __attribute__((cold, noinline, unused)) void* tiny_slow_alloc_fast(int class_idx) {
    int tls_enabled = g_tls_list_enable;
    TinyTLSList* tls = &g_tls_lists[class_idx];
    pthread_mutex_t* lock = &g_tiny_class_locks[class_idx].m;
    pthread_mutex_lock(lock);

    TinySlab* slab = g_tiny_pool.free_slabs[class_idx];
    if (slab) {
        g_tiny_pool.free_slabs[class_idx] = slab->next;
    } else {
        slab = allocate_new_slab(class_idx);
        if (!slab) {
            pthread_mutex_unlock(lock);
            return NULL;
        }
    }
    slab->next = NULL;

    if (atomic_load_explicit(&slab->remote_head, memory_order_acquire)) {
        tiny_remote_drain_locked(slab);
    }

    int block_idx = hak_tiny_find_free_block(slab);
    if (block_idx < 0) {
        slab->next = g_tiny_pool.free_slabs[class_idx];
        g_tiny_pool.free_slabs[class_idx] = slab;
        pthread_mutex_unlock(lock);
        return NULL;
    }

    hak_tiny_set_used(slab, block_idx);
    slab->free_count--;
    size_t block_size = g_tiny_class_sizes[class_idx];
    uint8_t* base = (uint8_t*)slab->base;
    void* ret = (void*)(base + ((size_t)block_idx * block_size));
    g_tiny_pool.alloc_count[class_idx]++;

    uint16_t cap = g_fast_cap_defaults[class_idx];
    uint16_t count = g_fast_count[class_idx];
    uint16_t fast_need = (cap > count) ? (uint16_t)(cap - count) : 0;
    if (fast_need > slab->free_count) fast_need = (uint16_t)slab->free_count;

    uint32_t tls_need = 0;
    if (tls_enabled && tls_list_needs_refill(tls)) {
        uint32_t target = tls_list_refill_threshold(tls);
        if (tls->count < target) {
            tls_need = target - tls->count;
        }
    }
    uint32_t remaining = slab->free_count;
    if (fast_need > remaining) fast_need = (uint16_t)remaining;
    remaining -= fast_need;
    if (tls_need > remaining) tls_need = remaining;

    while (fast_need > 0) {
        int extra_idx = hak_tiny_find_free_block(slab);
        if (extra_idx < 0) break;
        hak_tiny_set_used(slab, extra_idx);
        slab->free_count--;
        void* extra = (void*)(base + ((size_t)extra_idx * block_size));
        int pushed = 0;
        if (__builtin_expect(g_fastcache_enable && class_idx <= 3, 1)) {
            pushed = fastcache_push(class_idx, extra);
        } else {
            pushed = tiny_fast_push(class_idx, extra);
        }
        if (!pushed) {
            if (tls_enabled) {
                tiny_tls_list_guard_push(class_idx, tls, extra);
                tls_list_push(tls, extra, class_idx);
            }
        }
        fast_need--;
    }

    while (tls_enabled && tls_need > 0) {
        int extra_idx = hak_tiny_find_free_block(slab);
        if (extra_idx < 0) break;
        hak_tiny_set_used(slab, extra_idx);
        slab->free_count--;
        void* extra = (void*)(base + ((size_t)extra_idx * block_size));
        tiny_tls_list_guard_push(class_idx, tls, extra);
        tls_list_push(tls, extra, class_idx);
        tls_need--;
    }

    if (slab->free_count == 0) {
        move_to_full_list(class_idx, slab);
    } else {
        slab->next = g_tiny_pool.free_slabs[class_idx];
        g_tiny_pool.free_slabs[class_idx] = slab;
    }

    pthread_mutex_unlock(lock);
    return ret;
}