Lecture4Data-FlowScheduling.ppt

Lecture4Data-works(KPN)(work)worksspecialcaseofKPNactors,tokensandfiringsStaticDataFlow(ClockedAutomata)specialcaseofDNstaticschedulingcodegenerationbuffersizing(resources!!)OtherClockedDataFlowmodelsBooleanDataFlowDynamicDataFlowSequenceGraphs,DependencyGraphs,DataFlowGraphsControlDataFlowDataFlowModelsPowerfulformalismfordata-dominatedsystemspecificationPartially-orderedmodel(over-specification)DeterministicexecutionindependentofschedulingUsedforsimulationschedulingmemoryallocationcodegenerationforDigitalSignalProcessors(HWandSW)municateoverunboundedFIFOqueuesActorsfiringfollowsfiringrulesFiringrule:numberofrequiredtokensoninputsFunction:(scheduling)Tokenscarryvaluesinteger,float,audiosamples,workstate:numberoftokensinFIFOsIntuitivesemanticsAteachtime,oneactorisfiredCanfiremore–butoneisalwayssafe(atomicfiring)Whenfiring,actorsconsumeinputtokensandproduceoutputtokensActorscanbefiredonlyifthereareenoughtokensintheinputqueuesFilterexampleExample:FIRfiltersingleinputsequencei(n)singleoutputsequenceo(n)o(n)=c1i(n)+c2i(n-1)*c1+oi*c2i(-1)FilterexampleExample:FIRfiltersingleinputsequencei(n)singleoutputsequenceo(n)o(n)=c1i(n)+c2i(n-1)*c1+oi*c2i(-1)FilterexampleExample:FIRfiltersingleinputsequencei(n)singleoutputsequenceo(n)o(n)=c1i(n)+c2i(n-1)*c1+oi*c2i(-1)