Supported Compilers

To use Taskflow, you only 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 v15.7 (MSVC++ 19.14)
AppleClang Xode 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 (nvcc) 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.

Integrate Taskflow to Your Project

Taskflow is header-only and there is no need for installation. Simply download the source and copy the headers under the directory taskflow/ to your project.

~$ git clone https://github.com/taskflow/taskflow.git
~$ cd taskflow/
~$ cp -r taskflow myproject/include/

Taskflow is written in C++17 and is built on top of C++ standardized threading libraries to improve portability. To compile a Taskflow program, say simple.cpp, you need to tell the compiler where to find the Taskflow header files and link it through the system thread library (usually POSIX threads in Linux-like systems). Take gcc for an example:

~$ g++ simple.cpp -std=c++17 -I myproject/include/ -O2 -pthread -o simple

Build Examples and Unit Tests

Taskflow uses CMake to build examples and unit tests. We recommend using out-of-source build.

~$ cd path/to/taskflow
~$ mkdir build
~$ cd build
~$ cmake ../
~$ make                         # compile all examples and unittests
~$ make test
 
Running tests...
/usr/bin/ctest --force-new-ctest-process
Test project /home/tsung-wei/Code/taskflow/build
        Start   1: passive_vector
  1/254 Test   #1: passive_vector ...................   Passed    0.04 sec
        Start   2: function_traits
  2/254 Test   #2: function_traits ..................   Passed    0.00 sec
        Start   3: object_pool.sequential
  3/254 Test   #3: object_pool.sequential ...........   Passed    0.10 sec
...
 
100% tests passed, 0 tests failed out of 254
 
Total Test time (real) =  29.67 sec

When the building completes, you can find the executables for examples and tests under the two folders, examples/ and unittests/. You can list a set of available options in the cmake.

~$ cmake -LA
...
TF_BUILD_EXAMPLES:BOOL=ON       # by default, we compile examples
TF_BUILD_TESTS:BOOL=ON          # by default, we compile tests
TF_BUILD_BENCHMARKS:BOOL=OFF    # by default, we don't compile benchmarks 
TF_BUILD_CUDA:BOOL=OFF          # by default, we don't compile CUDA code
...
... more options

Currently, our CMake script supports the following options:

CMake Option	Default	Usage
TF_BUILD_EXAMPLES	ON	enable/disable building examples
TF_BUILD_TESTS	ON	enable/disable building unit tests
TF_BUILD_BENCHMARKS	OFF	enable/disable building benchmarks
TF_BUILD_CUDA	OFF	enable/disable building CUDA code
TF_BUILD_SYCL	OFF	enable/disable building SYCL code

To enable or disable a specific option, use -D in the CMake build. For example:

~$ cmake ../ -DTF_BUILD_EXAMPLES=OFF

The above command turns off building Taskflow examples.

Build CUDA Examples and Unit Tests

To build CUDA code, including unit tests and examples, enable the CMake option TF_BUILD_CUDA to ON. Cmake will automatically detect the existence of nvcc and use it to compile and link .cu code.

~$ cmake ../ -DTF_BUILD_CUDA=ON

~$ make

Please visit the page Compile Taskflow with CUDA for details.

Build SYCL Code Examples and Unit Tests

To build SYCL code, including unit tests and examples, enable the CMake option TF_BUILD_SYCL to ON and direct the CMake compiler CMAKE_CXX_COMPILER to the location of DPC++ clang compiler. You can acquire the DPC++ clang compiler at the official page of Getting Started with oneAPI DPC++.

# after succesfully build Intel DPCPP clang: https://intel.github.io/llvm-docs/
~$ export DPCPP_HOME=/path/to/your/dpcpp/workspace
~$ export PATH=$DPCPP_HOME/llvm/build/bin:$PATH
~$ export LD_LIBRARY_PATH=$DPCPP_HOME/llvm/build/lib:$LD_LIBRARY_PATH
 
# build Taskflow SYCL code using Intel oneAPI dpc++ clang compiler
~$ cmake ../ -DTF_BUILD_SYCL=ON -DCMAKE_CXX_COMPILER=clang++
~$ make

Please visit the page Compile Taskflow with SYCL for details.

Build Sanitizers

You can build Taskflow with sanitizers to detect a variety of errors, such as data race, memory leak, undefined behavior, and others. To enable a sanitizer, add the sanitizer flag to the CMake variable CMAKE_CXX_FLAGS. The following example enables thread sanitizer in building Taskflow code to detect data race:

# build Taskflow code with thread sanitizer to detect data race
~$ cmake ../ -DCMAKE_CXX_FLAGS="-fsanitize=thread -g"
 
# build Taskflow code with address sanitizer to detect illegal memory access
~$ cmake ../ -DCMAKE_CXX_FLAGS="-fsanitize=address -g"
 
# build Taskflow code with ub sanitizer to detect undefined behavior
~$ cmake ../ -DCMAKE_CXX_FLAGS="-fsanitize=undefined -g"

Our continuous integration workflows incorporates thread sanitizer (-fsanitize=thread), address sanitizer (-fsanitize=address), and leak sanitizer (-fsanitize=leak) to detect data race, illegal memory address, and memory leak. To our best knowledge, Taskflow is one of the very few parallel programming libraries that are free from data race.

Note: Some sanitizers are supported by certain computing architectures. You can find the information about architecture support of each sanitizer at Clang Documentation and GCC Instrumentation Options.

Build Benchmarks

The Taskflow project contains a set of benchmarks to evaluate and compare the performance of Taskflow with existing parallel programming libraries. To build the benchmark code, enable the CMake option TF_BUILD_BENCHMARKS to ON as follows:

~$ cmake ../ -DTF_BUILD_BENCHMARKS=ON

~$ make

Please visit the page Benchmark Taskflow for details.

Build Documentation

Taskflow uses Doxygen and m.css to generate this documentation. The source of documentation is located in the folder taskflow/doxygen and the generated html is output to the folder taskflow/docs. To generate the documentation, clone the m.css project and enter the m.css/documentation directory:

~$ git clone git://github.com/mosra/m.css

~$ cd m.css/documentation

The script doxygen.py requires Python 3.6, depends on Jinja2 for templating and Pygments for code block highlighting. You can install the dependencies via pip or your distribution package manager:

# You may need sudo here
# More details are available at https://mcss.mosra.cz/documentation/doxygen/
~$ pip3 install jinja2 Pygments

Next, invoke doxygen.py and point it to the taskflow/doxygen/conf.py:

~$ ./doxygen.py path/to/taskflow/doxygen/conf.py

You can find the documentation output in taskflow/docs.

Table of Contents