On compile-time evaluation of process partitioning transformations for Kahn process networks

Abstract

Kahn Process Networks is an appealing model of computation for programming and mapping applications onto multi-processor platforms. Autonomous processes communicate through unbounded FIFO channels in absence of a global scheduler. We derive Kahn process networks from sequential applications using the pn compiler, but the derived networks do not necessarily meet the performance requirements. Process partitioning transformations can achieve a more balanced network improving the performance results significantly. There are a number of process partitioning transformations that can be used, but no hints are given to the designer which transformation should be applied to minimize, for example, the execution time. Therefore, we investigate a compile-time approach for selecting the best transformation candidate and show results on a Xilinx Virtex 2 FPGA and the Cell BE processor.

Verification of parallelising transformations of KPN models

“…Too much concurrency, in contrast, may require too large overhead to eventually manage in the architecture, which may increase the overall latency. The transformation techniques like computation migration, process splitting, channel merging, process clustering, and unfolding and skewing may be used to control the concurrency in the KPN behaviour according to the architectural constraints [10][11][12][13]. The verification task, therefore, is to show the equivalence between two KPN behaviours.…”

Throughput modeling to evaluate process merging transformations in polyhedral process networks

“…To meet these requirements, a designer can apply algorithmic transformations to increase parallelism by unfolding processes [3] or to decrease parallelism by merging processes into a single component [4]. For the unfolding transformation, it has been shown in [5] that processes can be unfolded in many different ways which can result in significant differences in performance. Our current paper shows that the same holds for the process merging transformation: many solutions exist to merge different processes in a PPN with great differences in performance results and it's not trivial to select the best solution.…”