Architectures based on the data flow computing model provide an alternative to the conventional Von-Neumann architecture that are widelyused for general purpose computing.Processors based on the data flow architecture...Architectures based on the data flow computing model provide an alternative to the conventional Von-Neumann architecture that are widelyused for general purpose computing.Processors based on the data flow architecture employ fine-grain data-driven parallelism.These architectures have thepotential to exploit the inherent parallelism in compute intensive applicationslike signal processing,image and video processing and so on and can thusachieve faster throughputs and higher power efficiency.In this paper,severaldata flow computing architectures are explored,and their main architecturalfeatures are studied.Furthermore,a classification of the processors is presented based on whether they employ either the data flow execution modelexclusively or in combination with the control flow model and are accordinglygrouped as exclusive data flow or hybrid architectures.The hybrid categoryis further subdivided as conjoint or accelerator-style architectures dependingon how they deploy and separate the data flow and control flow executionmodel within their execution blocks.Lastly,a brief comparison and discussionof their advantages and drawbacks is also considered.From this study weconclude that although the data flow architectures are seen to have maturedsignificantly,issues like data-structure handling and lack of efficient placementand scheduling algorithms have prevented these from becoming commerciallyviable.展开更多
文摘Architectures based on the data flow computing model provide an alternative to the conventional Von-Neumann architecture that are widelyused for general purpose computing.Processors based on the data flow architecture employ fine-grain data-driven parallelism.These architectures have thepotential to exploit the inherent parallelism in compute intensive applicationslike signal processing,image and video processing and so on and can thusachieve faster throughputs and higher power efficiency.In this paper,severaldata flow computing architectures are explored,and their main architecturalfeatures are studied.Furthermore,a classification of the processors is presented based on whether they employ either the data flow execution modelexclusively or in combination with the control flow model and are accordinglygrouped as exclusive data flow or hybrid architectures.The hybrid categoryis further subdivided as conjoint or accelerator-style architectures dependingon how they deploy and separate the data flow and control flow executionmodel within their execution blocks.Lastly,a brief comparison and discussionof their advantages and drawbacks is also considered.From this study weconclude that although the data flow architectures are seen to have maturedsignificantly,issues like data-structure handling and lack of efficient placementand scheduling algorithms have prevented these from becoming commerciallyviable.
