//
        // TODO(allenl): There may be smarter ways to do this copy in some
        // cases, i.e. with a collective broadcast. We'll need to be careful
        // about things that are taken as inputs on the host or on their
        // existing device (for multi-device functions).
        std::unique_ptr<ParallelTensor> parallel_tensor(
            parallel_device.CopyToParallelDevice(

    // LRU iterator for the key and block.
    std::shared_ptr<Block> block = Lookup(key);
    if (!block) {
      std::cerr << "No block for key " << key.first << "@" << key.second;
      abort();
    }
    MaybeFetch(key, block, status);
    if (TF_GetCode(status) != TF_OK) return -1;
    UpdateLRU(key, block, status);
    if (TF_GetCode(status) != TF_OK) return -1;