Configurable computers based on Field Programmable Gate Arrays (FPGAs) are capable of accelerating suitable applications by several orders of magnitude when compared to
DynamicComputationStructures:
User-levelHardwareforRecon gurableArchitectures
JonathanBabbandMatthewFrank
jbabb@lcs.mit.edu,mfrank@lcs.mit.edu
MITLaboratoryforComputerScience
Introduction
Con gurablecomputersbasedonFieldProgrammableGateArrays(FPGAs)arecapableofacceleratingsuitableappli-cationsbyseveralordersofmagnitudewhencomparedtotraditionalprocessor-basedarchitectures[3].However,theprogrammingparadigmforsuchmachinesisoftennomorethanasetofhand-craftedcircuits,oneperFPGAinthesys-tem.Programmersmustexplicitlyaccountformachine-leveldetails,suchasFPGAcapacityandinterconnect,prohibit-ingthedevelopmentofarchitecturally-independentapplica-tions.
WeareaddressingtheseissuesasapartoftheRecon- gurableArchitectureWorkstation(RAW)project,worktoimplementarecon gurablecomputerbasedonacomputa-tionalmodelthatdoesnotmandateeithera xedinstructionsetora xedmemorysystemmodel;ratheritbuildsausercon gurableinstructionsetontopof xed,coreinstruc-tionset,andsynthesizesauser-con gurablesystemmodelontopofprimitive,commodity,hardwareelementssuchasDRAMandFPGAs.ThisworkleveragesourpreviousFPGAwork,VirtualWires[1],inconjunctionwithbehav-ioralcompilationtechnology,toviewanarrayofFPGAsasamachine-independentcomputingfabric.
TheRAWmachinecomprisesverysimplecommod-ityhardwareincludingSRAMs,DRAMs,alargeamountofprimitiveprogrammablelogicintheformofField-ProgrammableGateArrays(FPGAs),andfastmicropro-cessors.Unlikecurrentcomputersinwhichthesystemlogicishardwiredforalltime,RAWmachinespermituser-level,compiler,oroperatingsystemassistedcon gurationofthearchitectureofthecomputer.Possiblecon gurationmod-elsfortheRAWmachineinclude:simplechangestothememorymodel,creationofnewinstructions,andcompletecon gurationofapplication-speci c,user-levelhardware.Weintroducesacompilationparadigmforthelastmodel.
Hardware-software Interfaces for Reconfigurable ......computation + OS + VM CPU RH high-ILP ...Dynamic stack f(x) { return x+1; } f() {...
structures to be created from a linear array of... and optimal scheduling of computation kernels for...the reconfigurable hardware for efficient execution....
structures to be created from a linear array of... and optimal scheduling of computation kernels for...the reconfigurable hardware for efficient execution....
热门文档