A parallel variant of
ML using
skeletons being developed
(April 1994) as part of Tore Bratvold's PhD in the Department
of Computing and Electronic Engineering, Heriot-Watt
University, Edinburgh, UK. Programs are written in a subset
of
Standard ML, and parallelism is extracted from the use of
certain
higher-order functions. The SkelML compiler uses
profiling information together with skeleton performance
models to distinguish useful from non-useful parallelism. An
important feature is the ability to perform transformations
between skeletons to improve performance. Skeletons currently
supported are map, filter, fold, pipe (implicitly extracted
from function application) and various combinations of these.
See also
paraML.
E-mail: Tore A Bratvold <
tore@cee.hw.ac.uk>.