Intel Threading Building Blocks

Working on the Assembly Line: Pipeline

Pipelining is a common parallel pattern that mimics a traditional manufacturing assembly line. Data flows through a series of pipeline stages and each stage processes the data in some way. Given an incoming stream of data, some of these stages can operate in parallel, and others cannot. For example, in video processing, some operations on frames do not depend on other frames and so can be done on multiple frames at the same time. On the other hand, some operations on frames require processing prior frames first.

Pipelined processing is common in multimedia and signal processing applications. A classical thread-per-stage implementation suffers two problems:

• The speedup is limited to the number of stages.

• When a thread finishes a stage, it must pass its data to another thread.

Eliminating these problems is desirable. To do so, you specify whether a stage is serial or parallel. A serial stage processes items one at a time, in order. A parallel stage may process items out of order or concurrently. By allowing some stages to be run concurrently, you make available more opportunities for load balancing and concurrency. Given sufficient processor cores and concurrent opportunities, the throughput of the pipeline is limited to the throughput of the slowest serial filter.

The pipeline and filter classes implement the pipeline pattern. Here we’ll look at a simple text-processing problem to demonstrate the usage of pipeline and filter. The problem is to read ...