AutoPilot: message passing parallel programming for a cache incoherent embedded manycore processor

Abstract

The Renesas Electronics IMAPCAR2 embedded realtime image processor combines a single core with a 128-way SIMD array. At runtime, sections of the SIMD array can be reconfigured as additional CPU cores, interconnected via a message ring. Effective use is made difficult by the low-level message passing API and lack of cache coherency between processors. The AutoPilot library addresses this by providing a high-level message-oriented parallel programming model mirroring that of Pilot, itself a wrapper around the Message Passing Interface (MPI) for cluster computing. AutoPilot shows that Pilot's processes-and-channels architecture is a viable choice for parallel programming on cache-incoherent multicore and manycore architectures. It provides a simpler API for programmers, with built-in safety checks that eliminate some common sources of errors. Since the IMAPCAR2 is targeted chiefly at automotive applications, open source AutoPilot has a large degree of MISRA-C compliance.