Automatic Parallelization of Models using Pipeline Extraction from Combined RHS and Inlined Solvers
2009 (English)In: the 14th Workshop on Compilers for Parallel Computing, 2009Conference paper (Other scientific)
In this work we report preliminary results from a new integrated method of automatically generating parallel code from equationbased object-oriented mathematical models by combining parallelization at two levels of abstraction. Performing inline expansion of a Runge-Kutta solver combined with fine-grained automatic parallelization of the resulting RHS opens up new possibilities for generating high performance code, which is becoming increasingly relevant when multi-core computers are becoming commonplace. We have introduced a new way of scheduling the task graph generated from the simulation problem which utilizes knowledge about locality of the simulation problem and generates a computation pipeline such that processors early in the pipeline can carry on with subsequent time steps while the end of the pipeline still computes the current step. A prototype implementation has been developed as part of the OpenModelica compiler, which has been used for preliminary measurements.
Place, publisher, year, edition, pages
Modelica, automatic parallelization, equation-based modeling
IdentifiersURN: urn:nbn:se:liu:diva-12438OAI: oai:DiVA.org:liu-12438DiVA: diva2:85
Submitted, but not yet accepted.2008-09-042008-09-042013-06-12