Compiling dynamic mappings with array copies

Abstract

Array remappings are useful to many applications on distributed memory parallel machines. They are available in High Performance Fortran, a Fortran-based data-parallel language. This paper describes techniques to handle dynamic mappings through simple array copies: array r e m a ppings are translated into copies between statically mapped distinct versions of the array. It discusses the language restrictions required to do so. The remapping graph which captures all remapping and liveness information is presented, as well as additional data-ow optimizations that can be performed on this graph, so as to avoid useless remappings at run time. Such useless remappings appear for arrays that are not used after a remapping. Live a r r a y copies are also ke p t t o a void other ow-dependent useless remappings. Finally the code generation and runtime required by our scheme are discussed. These techniques are implemented in our prototype HPF compiler.

Hardware synthesis for systems of recurrence equations with multidimensional schedule

“…The RAMs used in each cell are not included in the area complexity because these RAM blocks are already present on the chip: if not used, they would be lost anyway. It is also linked to research concerning efficient code generation for data parallel programs on SIMD or SPMD architectures, as studied for High-Performance Fortran (HPF) compilers [4], [17] for example. However, results obtained for HPF cannot be applied directly, because they rely upon the strong assumption that communications between processors are much more (up to one thousand times more) costly than computations.…”