Demystify Scalable Parallelism with Intel Threading Building Block’s Generic Parallel Algorithms
Multithreaded applications are notoriously difficult to write, test, and debug, but for many client-server developers, this task will soon become necessary for some high-performance applications. If it's in your future, start here to get an introduction to the algorithms behind the scenes in Intel's C++ template library for multithreading and start demystifing the multicore future.
by Michael Voss
October 26, 2006
o exploit the performance potential of multicore processors, applications must be threaded for performance. While performance-oriented threading is new for many developers of desktop and laptop systems, there is a long history of threading in the scientific and high-performance computing domains that can be leveraged on the desktop—at least theoretically. Unfortunately, the languages and libraries developed in those domains do not match the unique demands of mainstream client and server applications, and therefore are difficult to apply directly.
This article introduces the generic parallel algorithms provided by Intel Threading Building Blocks (Intel TBB). This C++ template library includes several common algorithms such as parallel_for, parallel_reduce, and pipeline, which are known through experience to be useful in a wide range of application domains. In Intel TBB, these algorithms have been specifically tuned to provide scalable performance for applications targeting current and future generation multicore and multithreaded systems.
It's quick, easy and you get access to all the articles on DevX.
This registration/login is to allow you to read articles on devx.com. Already a member?
To become a member of DevX.com create your Member Profile by completing the form below. Membership is free!
o exploit the performance potential of multicore processors, applications must be threaded for performance. While performance-oriented threading is new for many developers of desktop and laptop systems, there is a long history of threading in the scientific and high-performance computing domains that can be leveraged on the desktop—at least theoretically. Unfortunately, the languages and libraries developed in those domains do not match the unique demands of mainstream client and server applications, and therefore are difficult to apply directly.
