A new WG21 paper is available. A copy is linked below, and the paper will also appear in the next normal WG21 mailing. If you are not a committee member, please use the comments section below or the std-proposals forum for public discussion.

Document number: N3872

Date: 2014-01-15

A Primer on Scheduling Fork-Join Parallelism with Work Stealing

by Arch Robison

Excerpt:

This paper is a primer, not a proposal, on some issues related to implementing fork-join parallelism. It is intended to introduce readers to two key design choices for implementing fork-join parallelism and their impact. Both choices concern how fork-join computations are mapped (scheduled) onto threads. The key point is that the most efficient resolution for these choices is the opposite of what programmers new to parallelism might consider “obvious”. This paper also makes clear that the two choices can be made independently of each other.

The separate subject of how to map fork-join call stacks (“cactus stacks”) onto memory is not covered. Reference [CilkM] surveys various methods for such.