boost中的lower_bound长什么样儿

就是下面这段 lower_bound 的核心代码，把我引向了 OpenCL 的不归路。
她长这样。

///
/// \brief Binary find algorithm
///
/// Finds the end of true values in the partitioned range [first, last).
/// \return Iterator pointing to end of true values
///
/// \param first Iterator pointing to start of range
/// \param last Iterator pointing to end of range
/// \param predicate Predicate according to which the range is partitioned
/// \param queue Queue on which to execute
///
template<class InputIterator, class UnaryPredicate>
inline InputIterator binary_find(InputIterator first,
                                 InputIterator last,
                                 UnaryPredicate predicate,
                                 command_queue &queue = system::default_queue())
{
    const device &device = queue.get_device();

    boost::shared_ptr<parameter_cache> parameters =
        detail::parameter_cache::get_global_cache(device);

    const std::string cache_key = "__boost_binary_find";

    size_t find_if_limit = 128;
    size_t threads = parameters->get(cache_key, "tpb", 128);
    size_t count = iterator_range_size(first, last);

    InputIterator search_first = first;
    InputIterator search_last = last;

    scalar<uint_> index(queue.get_context());

    // construct and compile binary_find kernel
    binary_find_kernel<InputIterator, UnaryPredicate>
        binary_find_kernel(search_first, search_last, predicate);
    ::boost::compute::kernel kernel = binary_find_kernel.compile(queue.get_context());

    // set buffer for index
    kernel.set_arg(binary_find_kernel.m_index_arg, index.get_buffer());

    while(count > find_if_limit) {
        index.write(static_cast<uint_>(count), queue);

        // set block and run binary_find kernel
        uint_ block = static_cast<uint_>((count - 1)/(threads - 1));
        kernel.set_arg(binary_find_kernel.m_block_arg, block);
        queue.enqueue_1d_range_kernel(kernel, 0, threads, 0);

        size_t i = index.read(queue);

        if(i == count) {
            search_first = search_last - ((count - 1)%(threads - 1));
            break;
        } else {
            search_last = search_first + i;
            search_first = search_last - ((count - 1)/(threads - 1));
        }

        // Make sure that first and last stay within the input range
        search_last = (std::min)(search_last, last);
        search_last = (std::max)(search_last, first);

        search_first = (std::max)(search_first, first);
        search_first = (std::min)(search_first, last);

        count = iterator_range_size(search_first, search_last);
    }

    return find_if(search_first, search_last, predicate, queue);
}

心中浮现多个问号，“我是谁，我在哪”。
command_queue 是什么， device 和 kenerl 又在干什么的？ 这个方法是实现什么功能来着？
这就是第一印象。

配合着文档开始 helloworld 吧。 ¹

---

1. boost.compute. ↩