ICS '95 Proceedings of the 9th international conference on Supercomputing
| Hi-index | 0.00 |
We have been experimenting with combining three powerful language tools for large, scientific, parallel Fortran codes. One tool is ParaScope, a programming envirnment; another tool is the Lambda Toolkit, a collection of routines for performing loop transformations using invertible matrices; the third is FORGE 90, a collection of tools for parallelizing Fortran programs. Initial sucess with incorporating the Lambda Toolkit into ParaScope led us to undertake the work leading to a new program preparation strategy, in which one first uses a modified ParaScope to perform Data Access Normalization, then uses FORGE 90 to produce a parallel program for a distributed memory platform. We describe the details of this strategy and present some performance results for the IBM SP1. We conclude that the combination of ParaScope and the Lambda Toolkit (called "ped-Lambda") is a useful transformation tool. Keywords: tools, tool combinations, data locality, APR FORGE 90 DMP, HPF, data access normaliztion, ParaScope, Lambda Toolkit, SP1, Fortran, parallel