开了几个线程跑，没发现问题

template <typename T> struct LocklessCache {
  std::atomic<T *> caches[4];

  T *get() {
    for (auto &o : caches) {
      if (T *t = o.exchange(nullptr, std::memory_order_acq_rel))
        return t;
    }

    return new T();
  }

  void release(T *t) {
    for (auto &o : caches) {
      t = o.exchange(t, std::memory_order_acq_rel);
      if (t == nullptr)
        return;
    }

    delete t;
  }
};

LocklessCache<int> cache;

int main(int argc, char **argv) {
  std::thread *threads[8];
  for (size_t k = 0; k < _countof(threads); ++k) {
    threads[k] = new std::thread([]() {
      for (size_t k = 0; k < 1000000000; k++) {
        auto p = cache.get();
        //Sleep(1);
        cache.release(p);
      }
    });
  }

  for (size_t k = 0; k < _countof(threads); ++k) {
    threads[k]->join();
    delete threads[k];
  }
--
FROM 221.218.167.*