Release Notes » Release 3.3.0 (2022/01/03)

Taskflow 3.3.0 is the 4th release in the 3.x line! This release includes several new changes, such as sanitized data race, pipeline parallelism, documentation, and unit tests.

Download

Taskflow 3.3.0 can be downloaded from here.

System Requirements

To use Taskflow v3.3.0, you need a compiler that supports C++17:

GNU C++ Compiler at least v8.4 with -std=c++17
Clang C++ Compiler at least v6.0 with -std=c++17
Microsoft Visual Studio at least v19.27 with /std:c++17
AppleClang Xcode Version at least v12.0 with -std=c++17
Nvidia CUDA Toolkit and Compiler (nvcc) at least v11.1 with -std=c++17
Intel C++ Compiler at least v19.0.1 with -std=c++17
Intel DPC++ Clang Compiler at least v13.0.0 with -std=c++17 and SYCL20

Taskflow works on Linux, Windows, and Mac OS X.

Release Summary

This release has resolved data race issues reported by tsan and has incorporated essential sanitizers into the continuous integration workflows for detecting data race, illegal memory access, and memory leak of the Taskflow codebase.
This release has introduced a new pipeline interface (tf::Pipeline) that allow users to create a pipeline scheduling framework for implementing pipeline algorithms.
This release has introduced a new thread-id mapping algorithm to resolve unexpected thread-local storage (TLS) errors when building Taskflow projects in a shared library environment.

New Features

Taskflow Core

Changed all lambda operators in parallel algorithms to copy by default
Cleaned up data race errors in tsan caused by incorrect memory order
Enhanced scheduling performance by caching tasks in the invoke loop
Added tf::Task::data to allow associating a task with user-level data
Added tf::Executor::named_async to allow associating an asynchronous task a name
Added tf::Executor::named_silent_async to allow associating a silent asynchronous task a name
Added tf::Subflow::named_async to allow associating an asynchronous task a name
Added tf::Subflow::named_silent_async to allow associating a silent asynchronous task a name
Added multi-conditional tasking to allow a task to jump to multiple successors
Added tf::Runtime tasking interface to enable in-task scheduling control
Added tf::Taskflow::transform to perform parallel-transform algorithms
Added tf::Graph interface to allow users to create custom module tasks
Added tf::FlowBuilder::erase to remove a task from the associated graph

cudaFlow

Starting from v3.3, using tf::cudaFlow needs to include the header, taskflow/cuda/cudaflow.hpp. See Breaking Changes.

syclFlow

This release does not have any update on syclFlow.

Utilities

Added tf::SmallVector to the documentation
Added relax_cpu call to optimize the work-stealing loop

Taskflow Profiler (TFProf)

This release does not have any update on the profiler.

Bug Fixes

Fixed incorrect static TLS access when building Taskflow in a shared lib
Fixed memory leak in updating tf::cudaFlowCapturer of undestroyed graph
Fixed data race in the object-pool when accessing the heap pointer
Fixed invalid lambda capture by reference in tf::Taskflow::sort
Fixed invalid lambda capture by reference in tf::Taskflow::reduce
Fixed invalid lambda capture by reference in tf::Taskflow::transform_reduce
Fixed invalid lambda capture by reference in tf::Taskflow::for_each
Fixed invalid lambda capture by reference in tf::Taskflow::for_each_index

If you encounter any potential bugs, please submit an issue at issue tracker.

Breaking Changes

For the purpose of compilation speed, you will need to separately include the follwoing files for using specific features and algorithms:

taskflow/algorithm/reduce.hpp for creating a parallel-reduction task
taskflow/algorithm/sort.hpp for creating a parallel-sort task
taskflow/algorithm/transform.hpp for creating a parallel-transform task
taskflow/algorithm/pipeline.hpp for creating a parallel-pipeline task
taskflow/cuda/cudaflow.hpp for creating a tf::cudaFlow and a tf::cudaFlowCapturer tasks
taskflow/cuda/algorithm/for_each.hpp for creating a single-threaded task on a CUDA GPU
taskflow/cuda/algorithm/for_each.hpp for creating a parallel-iteration task on a CUDA GPU
taskflow/cuda/algorithm/transform.hpp for creating a parallel-transform task on a CUDA GPU
taskflow/cuda/algorithm/reduce.hpp for creating a parallel-reduce task on a CUDA GPU
taskflow/cuda/algorithm/scan.hpp for creating a parallel-scan task on a CUDA GPU
taskflow/cuda/algorithm/merge.hpp for creating a parallel-merge task on a CUDA GPU
taskflow/cuda/algorithm/sort.hpp for creating a parallel-sort task on a CUDA GPU
taskflow/cuda/algorithm/find.hpp for creating a parallel-find task on a CUDA GPU

Deprecated and Removed Items

This release does not have any deprecated and removed items.

Documentation

Revised Building and Installing
- Build Sanitizers
Revised Static Tasking
- Attach User Data to a Task
Revised Composable Tasking
- Create a Custom Composable Graph
Revised Conditional Tasking
- Create a Multi-condition Task
Revised GPU Tasking (cudaFlow)
Revised GPU Tasking (cudaFlowCapturer)
Revised Limit the Maximum Concurrency
- Define a Conflict Graph
Revised Parallel Sort to add header-include information
Revised Parallel Reduction to add header-include information
Revised cudaFlow Algorithms to add header-include information
Revised CUDA Standard Algorithms to add header-include information
Added Interact with the Runtime
Added Parallel Transforms
Added Task-parallel Pipeline

Miscellaneous Items

We have published Taskflow in the following venues:

Tsung-Wei Huang, Dian-Lun Lin, Chun-Xun Lin, and Yibo Lin, "Taskflow: A Lightweight Parallel and Heterogeneous Task Graph Computing System," IEEE Transactions on Parallel and Distributed Systems (TPDS), vol. 33, no. 6, pp. 1303-1320, June 2022
Tsung-Wei Huang, "TFProf: Profiling Large Taskflow Programs with Modern D3 and C++," IEEE International Workshop on Programming and Performance Visualization Tools (ProTools), St. Louis, Missouri, 2021

Please do not hesitate to contact Dr. Tsung-Wei Huang if you intend to collaborate with us on using Taskflow in your scientific computing projects.