The Monadic Processor
Resumo
In a straightforward pipelined implementation of the Explicit Token Store dataflow model there is an imbalance concerning the execution of dyadic instructions (instructions with two input operands). This is due to the fact that two tokens need to be processed for a dyadic instruction, before valid operands are available to be processed by the Arithmetical-logical Unit, which at each pipeline cycle can consume two tokens and produce as a result also two tokens. In this work we investigate an approach to increase the Arithmetical-logical Unit utilization rate in dataflow processors based on the Explicit Token Store model. We propose an abstract realization of this model where there is two tokens queues. One token queue keeps tokens headed for monadic instructions (instractions with a single input operand). These tokens are utilized whenever the main path of the processor pipeline (feeded directly by the two token queues) is unable to produce valid operands for the Arithmetical-logical Unit to process. Therefore, increasing the utilization rate of the Arithmetical-logical Unit. The other token queue holds tokens headed for dyadic instructions (instructions with two input operands).
Referências
Arvind, R.S.Nikhil, and K.K. Pingale. I-structures: Data Structures for Parallel Computing. Technical Report - CSG Memo 269, Laboratory For Computer Science, Computer Structure Group, MIT, February 1987.
S.A Brobst. Instruction Scheduling and Token Storage Requirements in a Dataflow Supercomputer. Master's Thesis, Department of Electrical Engineering and Computer Science, MIT 1986.
F.A.Almeida. Parallel Software Emulation of Multiprocessor Dataftow Machines on Transputer Networks. PhD Thesis. Computing Laboratory. University of Kent at Canterbury, 1992.
G.M-Papadopoulos. Implementation of a General Purpose Dataflow Multiprocessor. PhD thesis, Dept of Electrical Engineering and Computer Scicence, MIT 1988.
G.M. Papadopoulos and D.E.Culler. The Explicit Token Store. Journal of Parallel and Distribuled Computing, pages 189-308, December 1990.
G.M-Papadopoulos. Implementation of a General-Purpose Dataflow Multiprocessor. Research Monographs in Parallel and Distributed Computing, The MIT Press, Cambridge, Massachusetts 1991.
Inmos Limited. OCCAM 2 Reference Manual Inmos Limited, 1088.
R.S.Nikbil and Arvind. A Dataflow Approach to General Purpose Parallel Computing. Technical Report - CSG Memo 302, Laboratory for Computer Science, Computer Structure Group, MIT, June 1989.
