Method seq_add

subprojects/llama.cpp/src/llama-kv-cache.cpp:401–444

}

void llama_kv_cache::seq_add(llama_seq_id seq_id, llama_pos p0, llama_pos p1, llama_pos shift) {
    GGML_ASSERT(seq_id >= 0 && (size_t) seq_id < seq_to_stream.size());
    GGML_ASSERT(hparams.n_pos_per_embd() == 1 && "seq_add() is only supported for n_pos_per_embd() == 1");

    auto & cells = v_cells[seq_to_stream[seq_id]];
    auto & head  = v_heads[seq_to_stream[seq_id]];

    if (shift == 0) {
        return;
    }

    uint32_t new_head = cells.size();

    if (p0 < 0) {
        p0 = 0;
    }

    if (p1 < 0) {
        p1 = std::numeric_limits<llama_pos>::max();
    }

    // If there is no range then return early to avoid looping over all cells.
    if (p0 == p1) {
        return;
    }

    for (uint32_t i = 0; i < cells.size(); ++i) {
        if (!cells.pos_in(i, p0, p1)) {
            continue;
        }

        if (cells.seq_has(i, seq_id)) {
            if (cells.pos_add(i, shift)) {
                if (new_head == cells.size()) {
                    new_head = i;
                }
            }
        }
    }

    // If we freed up a slot, set head to it so searching can start there.
    // Otherwise we just start the next search from the beginning.
    head = new_head != cells.size() ? new_head : 0;
}

void llama_kv_cache::seq_div(llama_seq_id seq_id, llama_pos p0, llama_pos p1, int d) {
    GGML_ASSERT(seq_id >= 0 && (size_t) seq_id < seq_to_stream.size());