Release Notes » Release 4.0.0 (2026/01/01)

Release Summary

Happy new year! Starting in v4, Taskflow migrates the codebase to the C++20 standard for improved runtime performance and engineering efficiency. This release enhances scheduling efficiency through bulk scheduling, reduced atomic overhead, and a streamlined executor design.It also introduces new features such as tf::TaskGroup to better support recursive task parallelism, along with Taskflow Academy, a new video suite for learning Taskflow.

Download

To download the newest version of Taskflow, please clone the master branch from Taskflow's GitHub.

System Requirements

To use Taskflow v4.0.0, you need a compiler that supports C++20:

• GNU C++ Compiler at least v11.0 with -std=c++20
• Clang C++ Compiler at least v12.0 with -std=c++20
• Microsoft Visual Studio at least v19.29 (VS 2019) with /std:c++20
• Apple Clang (Xcode) at least v13.0 with -std=c++20
• NVIDIA CUDA Toolkit and Compiler (nvcc) at least v12.0 with host compiler supporting C++20
• Intel oneAPI DPC++/C++ Compiler at least v2022.0 with -std=c++20

New Features

Taskflow Core

• replaced std::function in topology with virtual dispatch for better performance

• replaced TF_LIKELY/_UNLIKELY with C++20 attributes

• extended the work-stealing queue to support the following new features:

  • non-pointer type with std::optional
  • bulk insertion to reduce the overhead of atomic operations

• enhanced the executor performance with the following changes:

  • bulk scheduling to reduce the overhead of atomic operations
  • removed executor pointer from worker
  • removed redundant executor pointer validation from the hot scheduling path

• adopted tf::NonblockingNotifier as the default notifier due to its robustness

  • split notify_one and notify_all to two separate, optimized implementation
  • added notify_n to improve the performance

• split essential members of tf::Node into tf::NodeBase to improve modularity

• split anchor nodes to implicit and explicit types to improve exception handling

• added unit test for tf::NonblockingNotifier

• added embarrassing parallelism benchmark

• added merge sort benchmark

• added tf::TaskGroup to enable efficient implementation of recursive parallelism

• revised the following benchmarks with tf::TaskGroup for recursive parallelism

  • integrate
  • skynet
  • fibonacci
  • nqueens

• enhanced the scheduling performance of dependent-async tasks with state merging

Utilities

• removed ctz as it is duplicate with the C++20 bit manipulation library

Bug Fixes

• fixed the compiler warning in Windows MSVC W4 (#737)
• fixed the compiler warning in NVCC due to auto function argument (#746)
• fixed the regression problem caused by notifier (#746)
• fixed the bug of uncaught exception from silent_async within corun (#748)

Breaking Changes

Starting in v4, Taskflow will adopt the C++20 standard for better runtime performance and engineering efficiency. If your project does not support C++20, please use Taskflow v3.11.0, which can be downloaded here.

Documentation

• introduced Taskflow Academy - a new video suite for learning Taskflow
• introduced tf::NonblockingNotifier
• introduced Task Group
• revised Executor
• revised tf::BoundedWSQ
• revised tf::UnboundedWSQ
• revised Exception Handling
• revised Fibonacci Number

Miscellaneous Items

If you are interested in collaborating with us on applying Taskflow to your projects, please feel free to reach out to Dr. Tsung-Wei Huang!