Release Notes » Release 2.1.0 (2019/02/15)

Taskflow 2.1.0 is the second release in the 2.x line! This release includes several new changes such as tf::Framework, tf::WorkStealingThreadpool, tf::SpeculativeThreadpool, allocators, benchmarks, and so forth. In addition, this release improved usability, stability, and performance.

Download

Taskflow 2.1.0 can be downloaded from here.

New Features

• A new reuseable task dependency graph tf::Framework.
• New API (tf::Taskflow::run, tf::Taskflow::run_n, tf::Taskflow::run_until) to execute a framework

Improvements and Enhancements

• Improved the performance of tf::WorkStealingThreadpool (non-blocking notifier, work stealing strategy, etc.)
• Changed the data structure to store nodes and topologies to std::list.
• Added memory pool and allocator to manage the memory allocation of nodes and topologies.
• Added a customized vector (PassiveVector) to hold POD data.
• Added a dependents vector to keep track of the predecessors, in addition to successors, of a node.
• Added more examples to demonstrate the usage of Taskflow.
• Added a benchmarking environment with comparison to OpenMP and Intel TBB libraries (micro-benchmarks and machine learning)
• Removed the target node from a topology.

Breaks and Deprecated Features

• tf::Taskflow::emplace is now merged with tf::Taskflow::silent_emplace (both are the same) and no longer returns std::future in order to support tf::Framework

Bug Fixes

There are no major bug fixes in this release.