Combining Program and Data Specialization

  • Authors:
  • Sandrine Chirokoff;Charles Consel;Renaud Marlet

  • Affiliations:
  • Compose project, IRISA/INRIA-Université de Rennes 1, Campus Universitaire de Beaulieu, 35042 Rennes Cedex, France. sandrine.chirokoff@irisa.fr http://www.irisa.fr/compose;Compose project, IRISA/INRIA-Université de Rennes 1, Campus Universitaire de Beaulieu, 35042 Rennes Cedex, France. charles.consel@irisa.fr http://www.irisa.fr/compose;Compose project, IRISA/INRIA-Université de Rennes 1, Campus Universitaire de Beaulieu, 35042 Rennes Cedex, France. renaud.marlet@irisa.fr http://www.irisa.fr/compose

  • Venue:
  • Higher-Order and Symbolic Computation
  • Year:
  • 1999

Quantified Score

Hi-index 0.01

Visualization

Abstract

Program and data specialization have always been studiedseparately, although they are both aimed at processing earlycomputations. Program specialization encodes the result of earlycomputations into a new program; while data specialization encodesthe result of early computations into data structures.In this paper, we present an extension of the Tempo specializer, whichperforms both program and data specialization. We show how thesetwo strategies can be integrated in a single specializer. This newkind of specializer provides the programmer with complementarystrategies which widen the scope of specialization. We illustrate the benefitsand limitations of these strategies and their combinationon a variety of programs.