NPAC Technical Report SCCS-380

Language Support for Loosely Synchronous Data Parallelism

Pankaj Kumar, Nancy McCracken

Submitted November 17 1992


Abstract

Exploiting data parallelism in programs that construct and manipulate pointer based data structures of arbitrary shape and varying size require support for expressing multiple data parallel operations on scattered data. This paper extends the 'C' language with a new data type called Troupe and associated for_each statement to express data parallelism in pointer based dynamic data structures and describes the implementation strategy for efficiently executing such programs on message passing parallel computers. A variety of memory allocators are also introduced that automatically distribute the data in a load balanced manner.


PostScript version of the paper