Filter
Top Products
3.5The16-BIS/32-BISConcept
3.5.116-BIS/32-BISAdvantages
www.ti.com
The16-BIS/32-BISConcept
Thekeyideabehind16-BISisthatofasuper-reducedinstructionset.Essentially,theARM926EJ
processorhastwoinstructionsets:
•ARMmodeor32-BIS:thestandard32-bitinstructionset
•Thumbmodeor16-BIS:a16-bitinstructionset
The16-bitinstructionlength(16-BIS)allowsthe16-BIStoapproachtwicethedensityofstandard32-BIS
codewhileretainingmostofthe32-BIS’sperformanceadvantageoveratraditional16-bitprocessorusing
16-bitregisters.Thisispossiblebecause16-BIScodeoperatesonthesame32-bitregistersetas32-BIS
code.16-bitcodecanprovideupto65%ofthecodesizeofthe32-bitcodeand160%oftheperformance
ofanequivalent32-BISprocessorconnectedtoa16-bitmemorysystem.
16-bitinstructionsoperatewiththestandard32-bitregisterconfiguration,allowingexcellent
inter-operabilitybetween32-BISand16-BISstates.Each16-bitinstructionhasacorresponding32-bit
instructionwiththesameeffectontheprocessormodel.Themajoradvantageofa32-bitarchitectureover
a16-bitarchitectureisitsabilitytomanipulate32-bitintegerswithsingleinstructions,andtoaddressa
largeaddressspaceefficiently.Whenprocessing32-bitdata,a16-bitarchitecturetakesatleasttwo
instructionstoperformthesametaskasasingle32-bitinstruction.However,notallofthecodeina
programprocesses32-bitdata(forexample,codethatperformscharacterstringhandling),andsome
instructions(likebranches)donotprocessanydataatall.Ifa16-bitarchitectureonlyhas16-bit
instructions,anda32-bitarchitectureonlyhas32-bitinstructions,thenthe16-bitarchitecturehasbetter
codedensityoverall,andhasbetterthanonehalfoftheperformanceofthe32-bitarchitecture.Clearly,
32-bitperformancecomesatthecostofcodedensity.The16-bitinstructionbreaksthisconstraintby
implementinga16-bitinstructionlengthona32-bitarchitecture,makingtheprocessingof32-bitdata
efficientwithcompactinstructioncoding.Thisprovidesfarbetterperformancethana16-bitarchitecture,
withbettercodedensitythana32-bitarchitecture.The16-BISalsohasamajoradvantageoverother
32-bitarchitectureswith16-bitinstructions.Theadvantageistheabilitytoswitchbacktofull32-bitcode
andexecuteatfullspeed.Thus,criticalloopsforapplicationssuchasfastinterruptsandDSPalgorithms
canbecodedusingthefull32-BISandlinkedwith16-BIScode.Theoverheadofswitchingfrom16-bit
codeto32-bitcodeisfoldedintosub-routineentrytime.Variousportionsofasystemcanbeoptimizedfor
speedorforcodedensitybyswitchingbetween16-BISand32-BISexecution,asappropriate.
SPRUEP9A–May2008ARMCore23
SubmitDocumentationFeedback