Realizations of the Extended Linearization Model

Alexandru Turjan, Bart Kienhuis, and Ed F. DeprettereLeiden Embedded Research Center, LeidenUniversity, Leiden, The NetherlandsAt the Leiden Embedded Research Center, we are working towards aframework called Compaan that automates the transformation of digitalsignal processing (DSP) applications to Kahn Process Networks (KPNs).These applications are written in Matlab as parameterized nested loop pro-grams. This transformation is interesting as KPNs are well suited for map-ping onto parallel architectures. Although the KPN semantic always as-sumes that FIFO buffers can be used between processes, we have foundcases in which the FIFO is not enough as data may arrive in the wrongorder. To solve this order problem, we previously presented the ExtendedLinearization Model (ELM) that describes a mechanism to reorder tokens.The introduction of the ELM does not violate the Kahn Process Networksemantics; we still use a FIFO between a Producer and Consumer. TheELM relays on some additional memory and a controller to perform thereordering. The ELM model can be implemented in different ways. In thischapter, we investigate four different realizations of the ELM. The realiza-tions differ in the computational complexity of performing the reordering,the kind of reordering memory used, and the size of the reordering mem-ory.1