TI A  C Cross-Compiler for Numerically Intensive Computing on
Scalable Parallel Architectures 

LT 91-017 

YR 1991 

AU Decker, Karsten M. 

AU Rehmann, Ren\'{e} M.  

OR BERN 

AB Computational problems in natural science and engineering are
characterized by the application of local functions to large and
homogeneous data structures.  In this paper we give a technical
description of a two stage cross-compiler which supports the two
major computational models of scientific computing, the
master-slave and the farmer-worker model. Starting from a
sequential {\it C\}-type specification of the problem and
application area specific libraries, the cross-compiler
automatically creates complementary application drivers. Further
capabilities of the compiler include automatic data structure
partitioning and scope adjustment, and the automatic addition of
communication steps providing transparent filesystem IO from any
process, data coherence for system-wide global variables, and
support for unique return code handling.  The cross-compiler has
been developed as part of the SPADE project which aims at
developing an integrated environment for program and application
development for numerically tensive computations on parallel
architectures with distributed memory.