enkiTS - Enki Task Scheduler - is a permissively licensed task scheduler for creating multithreaded programs with C++ and C++11 implementations and an optional C interface.
enkiTS is used in our voxel editor and game, Avoyd.
Doug Binks - 05 Sep 2015
This is the second in a series of articles detailing the inner workings and evolution of the permissively open source multithreading task scheduler enkiTS for C and C++ (including C++ 11). In the first article of this series I covered the external interfaces and their implementation. This post will cover the task threading function, running tasks, and waiting for tasks.
Figure 1: Screenshot of Avoyd being profiled with microprofile and ImGui integration available in enkiTSExamples. Solid bars above named tasks show when threads are active - the wait functionality allows the core to idle or other threads to run.
Doug Binks - 22 Aug 2015 - edited 28 Jun 2019
This is the first in a series of articles detailing the inner workings and evolution of the permissively open source multithreading task scheduler enkiTS for C and C++ (including C++ 11). The interface has evolved somewhat since this article, and whilst I've made minor changes to update it, keeping it fully up to date is difficult, so please check out the full code on github.
enkiTS - Code and basic examples for the task scheduler described this article.
enkiTSExamples - Further enkiTS examples adding profiling instrumentation.
If you're writing a compute intensive programming task on consumer hardware, and you want to use as much of the systems resources as possible, then you'll need to consider multithreading on the CPU. There are a number of ways to approach this, but the current gold standard approach for developers who want both simplicity and control is a task scheduler which can handle data-parallelism (in the games industry a task is often referred to as a job). Task parallelism allows you to run different types of computation at the same time, whilst data-parallelism enables you to run the same computation over a set of data across different threads at the same time. Note that I won't consider Single Instruction Multiple Data (SIMD) parallelism here, but if you're doing computations you probably should. Additionally, I won't cover multithreading methods for handling large latency, such as waiting on hard disk or socket transactions.
enkiTS in our game, Avoyd, being profiled using the microprofile and Dear ImGui integration available in enkiTSExamples. This is an old image of the game and the current codebase uses newer enkiTS API features such as priorities and pinned tasks making for a much more complex task graph.