Copyright (c) 2017 Sean Middleditch [email protected]

This is free and unencumbered software released into the public domain.

Anyone is free to copy, modify, publish, use, compile, sell, or distribute this software, either in source code form or as a compiled binary, for any purpose, commercial or non-commercial, and by any means.

In jurisdictions that recognize copyright laws, the author or authors of this software dedicate any and all copyright interest in the software to the public domain. We make this dedication for the benefit of the public at large and to the detriment of our heirs and successors. We intend this dedication to be an overt act of relinquishment in perpetuity of all present and future rights to this software under copyright law.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

For more information, please refer to http://unlicense.org

Simple task and job scheduling library for C++. The library aims for reasonably good performance, low overhead, light memory usage, and for suitability for games and soft-real-time uses.

jobxx aims to be as light-weight as possible while retaining strong performance. To this end, the basic primitives of the library are few and focused on minimalism.

The core concepts of jobxx are jobs, tasks, queues, and threads. Of the four, jobxx only directly represents the first three; threads are provided by std::thread or the application.

The task is the lowest-level primitive of the core concepts. A task represents a unit of work. Tasks have no return values nor error states. If applications wish to communicate results from a task, they must use an external mechanism such as std::future. jobxx tasks are best suited for small discrete chunks of work with no failure state or individual results, though of course having a task mutate some shared state (with the appropriate care for thread-safety) is a common use case. For instance, spawning a number of tasks to mutate a large array - where each task operates on a distinct subrange of the array - is an excellent case.

A job is a collection of tasks. jobxx allows users to create a job that spawns 0 or more tasks. The job maintains a completion state which is unset while the job has 1 or more tasks that have not yet been completed. Tasks spawned by a job may themselves spawn more tasks as part of the same job, allowing a job to represent the completion state of an entire tree of tasks, sub-tasks, and continuations. An application can either poll a job's completion state or block on its completion, working on tasks (if available) while waiting, or putting the thread to sleep until the job is ready.

Scheduling tasks is performed by a queue. A queue is essentially a list of tasks that have been spawned and are ready to execute. Any number of threads may poll a queue for tasks to execute. Additionally, queues contain a mechanism for parking threads; this is a way for a thread to sleep/block until a particular queue has work available. The parking approach allows for task creation to be efficient (no need to signal the OS if there are no sleeping threads) and is extensible for future needs, such as parking a thread on a job until it completes (not yet supported).

The two primary points of the api are jobxx::queue and jobxx::job.

A jobxx::queue is used to spawn tasks, execute spawned tasks, to create job instances, and to wait for jobs to complete.

Create a new task encapsulating the work to be performed. The task is put into a pending task queue and will be executed when a thread calls queue::work_one.

Creates a new job instance and then invokes initializer with a context object. Tasks spawned via this context will be added to the returned job as child tasks.

A jobxx::job represents the completion state of a set of tasks. The job can be queried to see if all tasks spawned for the job have been fully executed.

Returns true if there are no outstanding tasks associated with the job.

Same as job::complete.

A context allows for spawning tasks as part of a job.

As queue::spawn_task, except that the spawned task will be associated with the context's job.

A delegate is very similar to std::function with two primary differences. First, a delegate is guaranteed to never allocate. It is a compile-time error to attempt to store a function object (or lambda, or other invokable) into a delegate if it is too large or overly aligned.

Second, a delegate can wrap an invokable with one of two different potential signatures: () -> void or (context&) -> void. This allows for convenience when needing to construct a task which has no need for a context while still allowing for tasks which do need a context.

Note: the current incarnation of delegate only works for function objects which are trivially move-constructible and trivially destructible. This limitation is planned to be lifted in future releases.

Constructs a delegate wrapping function.

Constructs a delegate wrapping function.

Returns true if the delegate has been created with a function, or false for a default-constructed (empty) delegate.

Executes the stored function, passing ctx to it if the stored function takes a context& parameter. It is undefined behavior to call this operator on a delegate with no stored function.

A predicate is simple wrapper for a function reference of signature () -> bool. Note that it is only a reference type, meaning that it does not take ownership of a function object used to construct it.

Creates a predicate wrapping the given pred invokable.

Returns true if the predicate was constructed with a function, or false if the predicate has no function reference.

Invokes the reference function and returns its result. It is undefined behavior to call this operator on a predicate with no stored function reference.