<XML><RECORDS><RECORD><REFERENCE_TYPE>3</REFERENCE_TYPE><REFNUM>5327</REFNUM><AUTHORS><AUTHOR>Goodman,J.</AUTHOR><AUTHOR>O'Donnell,J.T.</AUTHOR><AUTHOR>Ranger,G.</AUTHOR></AUTHORS><YEAR>1998</YEAR><TITLE>Refinement transformation using Abstract Parallel Machines</TITLE><PLACE_PUBLISHED>Glasgow Workshop on Functional Programming </PLACE_PUBLISHED><PUBLISHER>Dept of Computing Science, University of Glasgow</PUBLISHER><PAGES>-</PAGES><LABEL>Goodman:1998:5327</LABEL><KEYWORDS><KEYWORD>parallel combinators; design methodology; functional languages</KEYWORD></KEYWORDS<ABSTRACT>Abstract Parallel Machines support a design methodology for the refinement of a high level specification into an executable parallel program. Parallel combinators are defined in a normal form which allows a calcluation of the cost as well as the semantics. The user can define several versions of these combinators, corresponding to different levels of abstraction. A specification is compiled (either manually or automatically) into an implementation through a sequence of refinements, where each version of the algorithm uses a suitable set of parallel combinators defined in an Abstract Parallel Machine. We illustrate the method with a simple case study: the refinement transformation of a concise functional program to sum the columns of a triangular matrix into a parallel C+MPI program that uses load balancing to improve performance. </ABSTRACT></RECORD></RECORDS></XML>