StreamDrive: a Dynamic Dataflow Framework for Clustered Embedded Architectures

摘要: In this paper, we present StreamDrive, a dynamic dataflow framework for programming clustered embedded multicore architectures. StreamDrive simplifies development of applications starting from sequential reference C code and allows seamless handling heterogeneous application-specific processing elements by applications. We address issues efficient implementation the runtime system in context constrained environments, which have not been sufficiently addressed previous research. conducted detailed performance evaluation on our Application Specific MultiProcessor (ASMP) cluster using Oriented FAST Rotated BRIEF (ORB) algorithm typical image domain. used proposed incremental flow transformation ORB original into an optimized implementation. Our has less than 10% parallelization overhead, near-linear speedup when number processors increases 1 to 8, achieves 15 VGA frames per second with small configuration 4 64KB shared memory, 30 8 128KB memory.