Frequently Asked Questions

Q1: What's the goal of Taskflow?

Taskflow aims to help C++ developers quickly implement efficient parallel decomposition strategies using task-based approaches.

Q2: How do I use Taskflow in my projects?

Taskflow is a header-only library with zero dependencies. The only thing you need is a C++17 compiler. To use Taskflow, simply drop the folder taskflow/ to your project and include taskflow.hpp.

Q3: What is the difference between static tasking and dynamic tasking?

Static tasking refers to those tasks created before execution, while dynamic tasking refers to those tasks created during the execution of static tasks or dynamic tasks (nested). Dynamic tasks created by the same task node are grouped together to a subflow.

Q4: How many tasks can Taskflow handle?

Benchmarks showed Taskflow can efficiently handle millions or billions of tasks (both large and small tasks) on a machine with up to 64 CPUs.

Q5: What is the weird hex value, like 0x7fc39d402ab0, in the dumped graph?

The hex value represents the memory address of the task. Each task has a method tf::Task::name(const std::string&) for user to assign a human readable string to ease the debugging process. If a task is not assigned a name or is an internal node, its address value in the memory is used instead.

Q6: Does Taskflow have backward compatibility with C++03/98/11/14?

Unfortunately, Taskflow is heavily relying on modern C++17's features/idoms/STL and it is very difficult to provide a version that compiles under older C++ versions.

Q7: How does Taskflow schedule tasks?

Taskflow implemented a very efficient work-stealing scheduler to execute task dependency graphs. The source code is available at taskflow/core/executor.hpp.

Q8: What is the overhead of taskflow?

Creating a taskflow has certain overhead. For example, creating a task and a dependency takes about 61 and 14 nanoseconds in our system (Intel 4-core CPU at 2.00GHz). The time is amortized over 1M operations, since we have implemented an object pool to recycle tasks for minimal overhead.

Q9: How does it compare to existing task programming systems?

There is a large amount of work on programming systems (e.g., StarPU, Intel TBB, OpenMP, PaRSEC, Kokkos, HPX) in the interest of simplifying the programming complexity of parallel and heterogeneous computing. Each of these systems has its own pros and cons and deserves a reason to exist. However, they do have some problems, particularly from the standpoint of ease of use, static control flow, and scheduling efficiency. Taskflow addresses these limitations through a simple, expressive, and transparent graph programming model.

Q10: Do you try to simplify the GPU kernel programming?

No, we do not develop new programming models to simplify the kernel programming. The rationale is simple: Writing efficient kernels requires domain-specific knowledge and developers often require direct access to the native GPU programming interface. High-level kernel programming models or abstractions all come with restricted applicability. Despite non-trivial kernel programming, we believe what makes heterogeneous computing difficult are surrounding tasks. A mistake made by task scheduling can outweigh all speed-up benefits from a highly optimized kernel. Therefore, Taskflow focuses on heterogeneous tasking that affects the overall system performance to a large extent.

Q11: Do you have any real use cases?

We have applied Taskflow to solve many realistic workloads and demonstrated promising performance scalability and programming productivity. Please refer to Real Use Cases and References.

Q12: Who is the Principal Investigator of Taskflow I can talk to?

Please visit this page or email the investigator Dr. Tsung-Wei Huang.

Q13: Who are developing and maintaining Taskflow?

Taskflow is in active development with core functionalities contributed by an academic group at the University of Wisconsin at Madison, led by Dr. Tsung-Wei Huang. While coming out of an academic lab, Taskflow aims to be industrial-strength and is committed to long-term support.

Q14: Is Taskflow just an another API or model?

OK, let me ask this first: Is your new car just another vehicle? Or, is your new home just another place to live?

The answer to this question is the question itself. As technology advances, we can always find new ways to solve computational problems and achieve new performance milestones that were previously out-of-reach.

Q15: How can I contribute?

New contributors are always welcome! Please visit Contributing.

Q16: Does Taskflow support pipeline parallelism?

Yes, Taskflow has a specialized programming model to create a pipeline scheduling framework. Please visit Task-parallel Pipeline and Data-parallel Pipeline.

Q1: What is the difference between Taskflow threads and workers?

The master thread owns the thread pool and can spawn workers to run tasks or shutdown the pool. Giving taskflow N threads means using N threads to do the works, and there is a total of N+1 threads (including the master thread) in the program. Please refer to Create an Executor for more details.

Q2: What is the Lifetime of a Task and a Graph?

The lifetime of a task sticks with its parent graph. A task is not destroyed until its parent graph is destroyed. Please refer to Understand the Lifetime of a Task for more details.

Q3: Is taskflow thread-safe?

No, the taskflow object is not thread-safe. Multiple threads cannot create tasks from the same taskflow at the same time.

Q4: Is executor thread-safe?

Yes, the executor object is thread-safe. You can have multiple threads submit different taskflows to the same executor.

Q5: My program hangs and never returns after dispatching a taskflow graph. What's wrong?

When the program hangs forever it is very likely your taskflow graph has a cycle or not properly conditioned (see Conditional Tasking). Try the tf::Taskflow::dump method to debug the graph before dispatching your taskflow graph.

Q6: In the following example where B spawns a joined subflow of three tasks B1, B2, and B3, do they run concurrently with task A?

No. The subflow is spawned during the execution of B, and at this point A must have finished because A precedes B. This gives rise to the fact B1 and B2 must run after A.

Q7: What is the purpose of a condition task?

A condition task lets you perform in-task decision making so you can integrate control flow into a task graph with end-to-end parallelism without synchronizing or partitioning your parallelism across conditionals.

Q8: Is the program master thread involved in running tasks?

No, the program master thread is not involved in running taskflows. The executor keeps a set of private worker threads spawned upon construction time to run tasks.

Q9: Are there any limits on the branches of conditional tasking?

No, as long as the return value points to a valid successors, your conditional tasking is valid.

Q10: Why does Taskflow program GPU in a task graph?

We ask users to describe a GPU workload in a task graph and execute it in a second moment. This organization minimizes kernels launch overhead and allows the GPU runtime (e.g., CUDA) to optimize the whole workflow.

Q11: Can I limit the concurrency in certain sections of tasks?

Yes, Taskflow provides a lightweight mechanism, tf::Semaphore, for you to limit the maximum concurrency (i.e., the number of workers) in a section of tasks. Please refer to Limit the Maximum Concurrency.

Q12: How can I attach custom data to a task and access it?

Each node in a taskflow is associated with a C-styled data pointer (i.e., void*) you can use to point to user data and access it in the body of a task callable. Please refer to Attach User Data to a Task.