Automatic selection of dynamic data partitioning schemes for distributed-memory multicomputers

Compiler Algorithms for Optimizing Locality and Parallelism on Shared and Distributed-Memory Machines

“…Previous work on parallelism has concentrated, among other topics, on compilation techniques for multicomputers [5,8,51,24], for multiprocessors [47,7], and for automatic discovery of parallelism [21,48,39,18,36,26]. Since neither data layout transformations nor cache locality was the central issue in any of these papers, we do not discuss them here any further.…”

DDT: A Research Tool for Automatic Data Distribution in High Performance Fortran

“…Some of the proposals presented in the literature about array r~mapping [:'i. [13][14][15][16][17][18][19][20] art> ~umm:uized in the rest of this section.…”

“…and rt>mapping costs into account. Palermo and Barwrjcc [20] also usc a dividc-and-cmi<pu:'r approach in which the program is n·cllrsivt>ly decomposed into a hierarchy of nmdidatP phases. Then.…”

Dynamic Data Partitioning for Distributed-Memory Multicomputers

“…Currently, the compiler can automatically select a static distribution of data (using a constraint-based algorithm [16]) specifying both the configuration of an abstract multi-dimensional mesh topology along with how program data should be distributed on the mesh. We have previously presented a technique [27] which extends the static partitioning algorithm to select dynamic data distributions which can further improve the performance of the resulting parallel program. In this paper we describe the implementation in more detail, introduce the idea of the hierarchical component affinity graph for performing interprocedural data partitioning, as well as present experimental results obtained using the current implementation.…”