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: N3556

Date: 2013-03-18

Thread-Local Storage in X-Parallel Computations

by Pablo Halpern, Charles E. Leiserson

Excerpt:

... [W]e designate the model under discussion as being an X-parallel model, where X might represent a variety of individual parallelization technologies, including vector units, GPU’s, task-based multithreading, attached processing, and the like. ...

The purpose of this paper is to develop terminology so that the impact of any X-parallel model on TLS can be described clearly and evaluated effectively. We compare the semantics of accessing a TLS variable within an X-parallel region to the semantics of accessing the same variable in serial execution. We propose a 5-level hierarchical taxonomy of TLS concordance, where each level in the hierarchy is more faithful to the way the existing threading model treats TLS than the level below. Placing an X-parallel model lower within the taxonomy does not necessarily mean that it is a bad model, but lower in the hierarchy does mean that an X-parallel model is a less faithful extension to the existing standard than a higher-level model, at least as far as its adherence to legacy TLS behavior is concerned.

We conclude with a brief discussion in which we advocate that any X-parallel model should provide Level-5 (full) concordance — the highest level — unless there is a compelling reason to the contrary.