The c9y library implements algorithms and strategies how to write concurrent code. It builds on top of the C++ threading primitives and enhances them.
The thread_pool implements a thread pool with a similar API as jthread, but
starts the given amount of threads.
The task_pool implements a task oriented thread pool. That is it provides the
means to schedule work at any given time after the creation of the task pool.
The queue class implements a thread safe queue with the ability to wait for
elements to be put into the queue.
Since not all compilers on all platforms have all the new threading primitives,
c9y provides drop in replacements for latch, jthread, stop_source,
stop_token and stop_callback. These classes live in the c9y namesapce and
if available, the std counterpart is used.
c9y implements a set of standard algorithms but as parallel versions. That is the calling function will wait until the algorithm finishes to execute on a thread pool.
The parallel function allows you to execute any number of taks in paralell, waiting for all of them to compleate.
The following standard algorithms are implemented:
parallel_all_ofparallel_any_ofparallel_copyparallel_countparallel_count_ifparallel_for_eachparallel_generateparallel_none_ofparallel_reduceparallel_transform
The following is semantically the same, but parallel_for_each will run on a
thread pool and thus complete quicker.
auto values = std::vector<int>(10000);
std::for_each(begin(values), end(values), [] (auto& value) {
value += 1;
});
c9y::parallel_for_each(begin(values), end(values), [] (auto& value) {
value += 1;
});Also a parallel parallel_map_reduce is provided to implement the map/reduce
algorithm.
c9y provides the ability to triggert "fire and forget" functions. In contrast
to std::async, c9y::async has a version that does not return a future, thus
allowing the calling code to return before the async function finishes.
int main()
{
auto app = MyApp{};
c9y::async([&] () {
auto latest_version = http::get("http://myservice.com/latest-version");
app.set_latest_version(latest_version);
});
app.run();
}A version with a future is also provided for completeness.
c9y provides a set of functions to "synchronize" into other threads. If a thread
calls sync_point regularly, it is possible to schedule a task in this thread
via the sync function and will be executed in the sync_point.
auto scene = std::make_shared<Scene>();
auto render_worker = std::jthread([] (std::stop_token st) {
auto gc = create_graphics_constext();
while (!st.stop_requested())
{
c9y::sync_point();
scene.render(gc);
}
});
// at some time later
auto mesh = load_mesh();
c9y::sync(render_worker.get_id(), [scene, mesh] () {
scene->add_mesh(mesh);
});Both a version with an explicit thread to be used and one using the "main thread".
The main thread can be registered with set_main_thread.
A special case is implemented with the delay function. This will schedule a task
on the same thread as the calling thread. This may be usefully to delay an action
to be preformed outside of a given callback. This allows for such things as
unregistration of event handlers from the handler or self delete in a way that
does not corrupt the call stack.
auto key_down_signal = rsgi::singal<Key>{};
// some place in the code
auto esc_conn = rsig::connection{};
esc_conn = key_down_signal.connect([&esc_conn] (Key key) {
if (key == Key::ESC)
{
do_esc_things();
c9y::delay([&esc_conn] () {
// this would crash / deadlook otherwise
key_down_signal.disconnect(esc_conn);
});
}
});
// in the main function or such
while (running)
{
handle_input();
c9y::sync_point();
}All sync function take a once_tag. This allows for multiple sync calls
to to be accumulated, preventing work to be redone multiple times.
// use render worker from above
auto physics_sync_tag = c9y::once_tag{};
void update_pysics()
{
auto new_state = update_dynamic_bodies();
// this happens only once per frame
c9y::sync(physics_sync_tag, render_worker.get_id(), [new_state] () {
for (auto& body : new_state)
{
auto& object = scene->get_object(body.id);
object.set_transform(body.transform);
}
});
}A real world example of using sync can be seen in spdr's Node class.
c9y implements coroutines support. By using the namespace co_async or
co_sync you can subsequently use co_await and co_return.
#include <c9y/coroutine.h>
std::future<int> compute_async()
{
using namespace c9y::co_async;
int a = co_await c9y::async<int>([] { return 6; });
int b = co_await c9y::async<int>([] { return 7; });
co_return a * b;
}
TEST(coroutine, compute_async)
{
EXPECT_EQ(6 * 7, compute_async().get());
}If you are using co_async the coroutines will be executed with the help of
async. This means that different bits of the function will be executed on
different threads. On the upside this is that the code just works. The
downside is, when interacting with other code, care needs to be taken to not
introduce a race condition.
std::future<int> compute_sync()
{
using namespace c9y::co_sync;
int a = co_await c9y::async<int>([] { return 6; });
int b = co_await c9y::async<int>([] { return 7; });
co_return a * b;
}
TEST(coroutine, compute_sync)
{
using namespace std::chrono_literals;
auto f = compute_sync();
while (f.wait_for(0s) == std::future_status::timeout)
{
c9y::sync_point();
}
EXPECT_EQ(6 * 7, f.get());
}If you are using co_sync the coroutines will be exectued with the help of
sync. This means that the thread calling the coroutine needs to call
sync_point at some point. If this is already the case, coroutines embed
naturally into the code and behave in the expected way, that is they execute
interleaved in the same thread.
The c9y library is distributed under the MIT license. See License.txt for details.
I would like to thank G. Sliepen for reviewing c9y's code.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent