What is llama_batch,and why we need this in llama.android?

[FranzKafkaYu]

I am reading the code of llama.android,and I found that when this App load a model,it will call new_batch method,here is the code:

suspend fun load(pathToModel: String) {
        withContext(runLoop) {
            when (threadLocalState.get()) {
                is State.Idle -> {
                    val model = load_model(pathToModel)
                    if (model == 0L)  throw IllegalStateException("load_model() failed")

                    val context = new_context(model)
                    if (context == 0L) throw IllegalStateException("new_context() failed")

                    val batch = new_batch(512, 0, 1)
                    if (batch == 0L) throw IllegalStateException("new_batch() failed")

                    Log.i(tag, "Loaded model $pathToModel")
                    threadLocalState.set(State.Loaded(model, context, batch))
                }
                else -> throw IllegalStateException("Model already loaded")
            }
        }
    }

I noticed that the implemention is copy of llama_batch_init,details:

ava_android_llama_cpp_LLamaAndroid_new_1batch(JNIEnv *, jobject, jint n_tokens, jint embd, jint n_seq_max) {

    // Source: Copy of llama.cpp:llama_batch_init but heap-allocated.

    llama_batch *batch = new llama_batch {
        0,
        nullptr,
        nullptr,
        nullptr,
        nullptr,
        nullptr,
        nullptr,
        0,
       0,
        0,
    };

    if (embd) {
        batch->embd = (float *) malloc(sizeof(float) * n_tokens * embd);
    } else {
        batch->token = (llama_token *) malloc(sizeof(llama_token) * n_tokens);
    }

    batch->pos      = (llama_pos *)     malloc(sizeof(llama_pos)      * n_tokens);
    batch->n_seq_id = (int32_t *)       malloc(sizeof(int32_t)        * n_tokens);
    batch->seq_id   = (llama_seq_id **) malloc(sizeof(llama_seq_id *) * n_tokens);
    for (int i = 0; i < n_tokens; ++i) {
        batch->seq_id[i] = (llama_seq_id *) malloc(sizeof(llama_seq_id) * n_seq_max);
    }
    batch->logits   = (int8_t *)        malloc(sizeof(int8_t)         * n_tokens);

    return reinterpret_cast<jlong>(batch);
}

somebody can tell me what is the purpose of this API,does it affect inference performance?