Optimal Compilation of Hpf

Applications with varying array access patterns in a parallel distributed memory machine require to dynamically change array mappings. Hpf (High Performance Fortran) provides such remappings, possibly on partially replicated data, explicitly through the realign and redistribute directives and implicitly at procedure calls and returns. However such features are left out of the hpf subset or of the currently discussed hpf kernel for eeciency reason. This paper presents a new compilation technique to handle hpf remappings for message-passing parallel architectures. The rst phase is global and removes all useless remappings that naturally appear in procedures. The second phase generates the actual remap-ping code and takes advantage of replications to shorten the remapping time. A minimal number of messages, containing only the required data, is sent over the network.