Incorporating the parallel processing techniques with the demand-driven model of functional programming languages

This paper presents a hybrid reduction scheme that incorporates parallel processing techniques into the demand-driven model of functional programs. With this scheme, functional language programs can be executed in parallel at three different levels, which are the parallel execution of vectors inside a redex, the parallel reduction of multiple redexes, and the high level pipeline processing. The high level pipeline consists of three stages that are redex detection, redex allocation and reduction execution. Moreover, memory latency and switch network latency that cannot be ignored in usual multiprocessor systems can effectively be eliminated by embedding the memory access and the switch network transmission into the high level pipeline. Experimental results of the software simulator reveal the potential performance of the hybrid reduction scheme. Functional language FL is chosen as the target language in our research work.<>