FastFlow is a multi-cores programming framework implemented as a C++ template
library that offers a set of mechanisms to support low-latency and
high-bandwidth data flows in a network of threads running on
a cache-coherent multi-core architectures. On these architectures, one of the
key performance issues concern memory fences, which are required to keep the
various caches coherent.
FastFlow provides the programmer with two basic mechanisms:
1. efficient point-to-pint communication channels;
2. a memory allocator.
Communication channels, as typical is in streaming applications, are
unidirectional and asynchronous. They are implemented via fence-free FIFO
queues. The memory allocator is built on top of these queues, thus taking
advantage of their efficiency.
On top of these basic machnisms FastFlow provides a library of explicitly
parallel constructs (a.k.a. skeletons) such as pipeline and farm.
The farm skeleton, exploits functional replication and abstracts the
parallel filtering of successive independent items of the stream under the
control of a scheduler.
See the BUILD.ME file for instructions about Cmake builds and testing.
See tests directory for some simple runnable examples.
FastFlow is currently actively supported for:
Linux with gcc >4.8 x86_64
Windows 7, Windows 8 with MSVS 2013 x86_64
Mac OS 10.9 with gcc >4.8 or clang 5.x
Previous Fastflow versions has been tested on:
Linux with gcc (> 4.0.1): i386, x86_64
Mac OS (>= 10.4) with gcc (> 4.0.1): x86_64, PowerPC G4, PowerPC G5
Windows XP 32bit with Visual Studio 9 2008 Win32: i386, x86_64
Windows 7 64bit with Visual Studio 9 2008 Win32: x86_64
Windows 7 64bit with Visual Studio 9 2008 Win64: x86_64
Windows 7 64bit with Visual Studio 10 2010 Win32: x86_64
Windows 7 64bit with Visual Studio 10 2010 Win64: x86_64
Windows with cygwin: not yet tested
Linux/PPC with gcc
Linux/Arm with gcc
Main development platform are Linux and MacOS. Starting from
version 2.0.4 FastFlow eavily uses C++11 features. It can be expected to work on any platform with a C++11 compiler.
Since "core pattern layer" does not stictly requires C++11 features, it can be expected to work also with an older C++ compiler.
Windows platform can be particuarly slow in debug mode due to iterator
debugging and secure crt layers. Release mode appers to be not affected by the
issue. See http://blogs.warwick.ac.uk/ahazelden/entry/visual_studio_stl/
The following preprocessor directives /D_SECURE_SCL=0;
/D_HAS_ITERATOR_DEBUGGING=0 (i.e. ADD_DEFINITIONS(-D_SECURE_SCL=0
-D_HAS_ITERATOR_DEBUGGING=0) in cmake) might ameliorate the problem. Consider
however that they cannot be used if the application links libraries compiled
with different options (e.g. Boost).
Please report tips, improvements or bug to the FastFlow mailing list
(see http://calvados.di.unipi.it/fastflow) or directly to:
Massimo Torquati <email@example.com> or <firstname.lastname@example.org>
Marco Aldinucci <email@example.com>