https://www.mortati.com/glusker/
http://madrona.ca/e/eec/sims/index.html
http://www.silab.it/frox/p101/
http://www.kompjutry.cz/kalkulackyceskoslovenske.html
all images copyright Mark Glusker 1999 to 2005 – all rights reserved 
https://www.mortati.com/glusker/
http://madrona.ca/e/eec/sims/index.html
http://www.silab.it/frox/p101/
http://www.kompjutry.cz/kalkulackyceskoslovenske.html
all images copyright Mark Glusker 1999 to 2005 – all rights reserved 
http://www.finetune.co.jp/~lyuka/interests/calc/pioneer/
PCB of the HP pioneer series calculators
Membrane switches
Key pad and back plate (with LCD)
Partial schematic of the HP 17BII/27S/42S
SEE ALSO
HP 42S memory upgrade and overclock
Pioneer “Observational Internals”, Paul J. Brogger
Copyright (c) HOSODA Takayuki. All rights reserved.
Powered by
http://www.anitasimulators.org.uk/

Welcome the Anita Calculator simulator site. In the pages of this site you will find a collection of simulators for various calculators associated with the Sumlock Anita company that shone brightly, but all too briefly, in the sixties and early seventies. All the simulators have an association with this company; either they were made at the U.K. factory in Portsmouth, are variants of calculators that were, or were made by companies that were associated with Sumlock Anita during it’s existence. The simulators are meant
to give a flavour of what these devices were like during that ‘golden age’ of calculator development. All the images of calculators here (except one of a comptometer) give access to a simulator so that you can try them out for yourselves (simply click on the image). Where possible a manual is provided on this site, or a link to where one may be found is given, if this is available.
Not all the simulators are represented in my collection, but more details of those calculators that are can be found on their individual pages accessible via the galleries page. These pages have working simulators on them as well. All of the simulators can be downloaded for running offline. The ZIP download packages can be found on the downloads page. For general comments and feedback (it’s always nice to hear from people), please drop me an email
You can jump directly to the relevant simulator pages using the links below, or use the navigation bar on the left to view all my other pages, including pictures of the items in my collection of Sumlock Anita calculators. If you would like just to browse, follow the ‘next’ / ‘prev’ sequential links at the bottom of each page. If you don’t see any navigation bar please click here.
Sumlock Anita Calculators 
Rockwell Calculators 
Associated Company Calculators 
© 20032020 Simon Southwell. All rights reserved.
http://viceemu.sourceforge.net/
VICE is a program that runs on a Unix, MSDOS, Win32, OS/2, BeOS, QNX 4.x, QNX 6.x, Amiga, Syllable or Mac OS X machine and executes programs intended for the old 8bit computers. The current version emulates the C64, the C64DTV, the C128, the VIC20, practically all PET models, the PLUS4 and the CBMII (aka C610/C510). An extra emulator is provided for C64 expanded with the CMD SuperCPU.
(24 December 2019) Version 3.4 released
______________________________________________
In 1972 the great American company HewlettPackard shipped their first pocket calculator. It was called HP35, and it was the first pocket calculator with transcendental functions and the first with the socalled Reverse Polish Notation (RPN). The HP35 was a huge success with scientific/engineering users, so HP soon came out with a similar model for business users. It was called HP80, and could solve problems like present and future values of payment streams, trend line calculations (least squares linear regression), ammortization, date calculation, etc. HewlettPackard continued with the HP45, an improved scientific model (which I happen to own :), and (after the models 65 and 70) with the HP55, which was programmable and supported a quartzcontrolled stopwatch function. These (socalled “classic”) calculators of HP were based upon a proprietary 4bit CPU developed just for that purpose.I wrote emulators for these four calculators for the Atari 800XL and for the Commodore 64, which simulate the HP calculator CPU to interpret the original program ROMs of the four models. This enables you to work with your Atari 800 or C64 exactly like with the great classic HP calculators with their sophisticated scientific and business functions.
The emulators for the Atari 800XL can be downloaded here:
The emulators for the Commodore 64 can be downloaded here:
You can use these files with an Atari or Commodore emulator or put it on floppy disks and try it on the real machines.
Much more information on HP’s classic (and more modern) calculators can be found at the Museum of HP Calculators.
In the following you find screenshots of the emulators running on the Atari 800XL and on the C64.
Atari 800XL:  
https://www.hpmuseum.org/cgisys/cgiwrap/hpmuseum/archv021.cgi?read=242802
Re: HELP WANTED ON ALGORITHMS
Message #19 Posted by Eric Smith on 27 Apr 2013, 12:49 p.m.,
in response to message #18 by Eddie W. Shore
Usually CORDIC isn’t used for logs and exponentials, although it is possible to do so by using hyperbolic CORDIC. CORDIC is in the general class of shiftandadd algorithms, and there is a simpler one for logs and exponential first published in 1624, only ten years after the invention of logarithms by Napier. I’m not sure whether the HP 9100A (1968) used Briggs’ algorithm, but all of HP’s handheld and handheldderived calculators from the HP35 (1972) through the Saturnbased and Saturnemulating calculators have used it. Recent HPdesigned calculators such as the HP 10bII+, 20b, 30b, and 49gII apparently use a math library based on that from Saturn calculators, so they presumably also use Briggs’ algorithm.
There is a good description of Briggs’ algorithm on Jacques Laporte’s “Briggs and the HP35” web page.
Another excellent reference for algorithms for transcendental functions is Elementary Functions: Algorithms and Implementations by JeanMichel Muller, second edition, which covers many classes of algorithms including shiftandadd (Briggs’, CORDIC, and others), and has especially good coverage of accurate argument range reduction algorithms.
It should be noted that the HP algorithms for sine and cosine use CORDIC, but not in the most basic method normally described. Instead, they compute the tangent (or cotangent), and use an identity to compute the sine (or cosine) from it. This avoids an issue with each CORDIC iteration effectively multiplying the resulting sine and cosine by a scale factor; by using tangent (or cotangent), these scale factors cancel out. In binary CORDIC, the number of CORDIC iteration is fixed, so the result can be multiplied by the inverse of the product of the scale factors, which is constant. In decimal CORDIC as implemented by HP, the number of CORDIC iterations is variable, so it would take more work to compensate for the nonconstant product of the scale factors.
For specifics of the algorithms used by HP, see:
Edited: 27 Apr 2013, 1:04 p.m.
(best my Marchant from Robert Mařík)
Pi= 3.141592653589793238462643383279502884197169399375105820974…
22/7 = 3,14…
22/7Pi = 0.0012644892673496186802137595776400046…
355/113 = 3.141592…
355/113Pi = 2,667641e7
103993/33102 = 3.141592653…
103993/33102 Pi = 5,77890… e10
\\\\\\\\\\\\\\\\
http://qin.laya.com/tech_projects_approxpi.html
After reading the American Scientist article, On the Teeth of Wheels, which describes the intricate interplay between pure and applied mathematics, and how clock makers independently developed mathematical methods to approximate gear ratios that were not feasibly made (such as representing gear ratios that were two primes with other gear ratios that were close in value), I decided to write a program to find fractional approximations of Pi. A couple of minutes searching google brought me to an existing fractional approximations of pi page, which attempted to find decimal approximations using an iterative approach, which turned out to be very slow. However since the code was open source, I decided to modify for my own uses. The sternbrocot method pretty much changes the O(n^{2}) to about O(1) for the calculation of successive values.
You can get several versions of the source:
The original modification to use the sternbrocot method
A 64 bit version
pi = 3.14159265358979323846264338327950288419716939937510 Num. Den. = Result (Accuracy )   =  ( ) 7/ 2 = 3.50000000000000000000000000000000000000 (0.35840734641020676153735661672049711581) [ 0] 10/ 3 = 3.33333333333333333333333333333333333333 (0.19174067974354009487068995005383044914) [ 0] 13/ 4 = 3.25000000000000000000000000000000000000 (0.10840734641020676153735661672049711581) [ 0] 16/ 5 = 3.20000000000000000000000000000000000000 (0.05840734641020676153735661672049711581) [ 1] 19/ 6 = 3.16666666666666666666666666666666666666 (0.02507401307687342820402328338716378247) [ 1] 22/ 7 = 3.14285714285714285714285714285714285714 (0.00126448926734961868021375957763997295) [ 2] 179/ 57 = 3.14035087719298245614035087719298245614 ( 0.00124177639681078232229250608652042805) [ 2] 201/ 64 = 3.14062500000000000000000000000000000000 ( 0.00096765358979323846264338327950288419) [ 3] 223/ 71 = 3.14084507042253521126760563380281690140 ( 0.00074758316725802719503774947668598279) [ 3] 245/ 78 = 3.14102564102564102564102564102564102564 ( 0.00056701256415221282161774225386185855) [ 3] 267/ 85 = 3.14117647058823529411764705882352941176 ( 0.00041618300155794434499632445597347243) [ 3] 289/ 92 = 3.14130434782608695652173913043478260869 ( 0.00028830576370628194090425284472027550) [ 3] 311/ 99 = 3.14141414141414141414141414141414141414 ( 0.00017851217565182432122924186536147005) [ 3] 333/ 106 = 3.14150943396226415094339622641509433962 ( 0.00008321962752908751924715686440854457) [ 4] 355/ 113 = 3.14159292035398230088495575221238938053 (0.00000026676418906242231236893288649634) [ 6] 52163/ 16604 = 3.14159238737653577451216574319441098530 ( 0.00000026621325746395047764008509189889) [ 6] 52518/ 16717 = 3.14159239097924268708500329006400669976 ( 0.00000026261055055137764009321549618443) [ 6] 52873/ 16830 = 3.14159239453357100415923945335710041592 ( 0.00000025905622223430340392992240246827) [ 6] 53228/ 16943 = 3.14159239804048869739715516732573924334 ( 0.00000025554930454106548821595376364085) [ 6] 53583/ 17056 = 3.14159240150093808630393996247654784240 ( 0.00000025208885515215870342080295504179) [ 6] 53938/ 17169 = 3.14159240491583668239268448948686586289 ( 0.00000024867395655606995889379263702130) [ 6] 54293/ 17282 = 3.14159240828607800023145469274389538247 ( 0.00000024530371523823118869053560750172) [ 6] 54648/ 17395 = 3.14159241161253233687841333716585225639 ( 0.00000024197726090158423004611365062780) [ 6] 55003/ 17508 = 3.14159241489604752113319625314142106465 ( 0.00000023869374571732944713013808181954) [ 6] 55358/ 17621 = 3.14159241813744963395948016571136711877 ( 0.00000023545234360450316321756813576542) [ 6] 55713/ 17734 = 3.14159242133754370136461035299424833652 ( 0.00000023225224953709803303028525454767) [ 6] 56068/ 17847 = 3.14159242449711436095702358939877850619 ( 0.00000022909267887750561979388072437800) [ 6] 56423/ 17960 = 3.14159242761692650334075723830734966592 ( 0.00000022597286673512188614497215321827) [ 6] 56778/ 18073 = 3.14159243069772588944834836496431140375 ( 0.00000022289206734901429501831519148044) [ 6] 57133/ 18186 = 3.14159243374023974485868250302430440998 ( 0.00000021984955349360396088025519847421) [ 6] 57488/ 18299 = 3.14159243674517733209464998087327176348 ( 0.00000021684461590636799340240623112071) [ 6] 57843/ 18412 = 3.14159243971323050184662176841190527916 ( 0.00000021387656273661602161486759760503) [ 6] 58198/ 18525 = 3.14159244264507422402159244264507422402 ( 0.00000021094471901444105094063442866017) [ 6] 58553/ 18638 = 3.14159244554136709947419250992595772078 ( 0.00000020804842613898845087335354516341) [ 6] 58908/ 18751 = 3.14159244840275185323449416031145005599 ( 0.00000020518704138522814922296805282820) [ 6] 59263/ 18864 = 3.14159245122985581000848176420695504664 ( 0.00000020235993742845416161907254783755) [ 6] 59618/ 18977 = 3.14159245402329135269009854033830426305 ( 0.00000019956650188577254484294119862114) [ 6] 59973/ 19090 = 3.14159245678365636458878994237820848611 ( 0.00000019680613687387385344090129439808) [ 6] 60328/ 19203 = 3.14159245951153465604332656355777743060 ( 0.00000019407825858241931681972172545359) [ 6] 60683/ 19316 = 3.14159246220749637606129633464485400704 ( 0.00000019138229686240134704863464887715) [ 6] 61038/ 19429 = 3.14159246487209840959390601677904163878 ( 0.00000018871769482886873736650046124541) [ 6] 61393/ 19542 = 3.14159246750588476102753044724183809231 ( 0.00000018608390847743511293603766479188) [ 6] 61748/ 19655 = 3.14159247010938692444670567285677944543 ( 0.00000018348040631401593771042272343876) [ 6] 62103/ 19768 = 3.14159247268312424119789558883043302306 ( 0.00000018090666899726474779444906986113) [ 6] 62458/ 19881 = 3.14159247522760424525929279211307278305 ( 0.00000017836218899320335059116643010114) [ 6] 62813/ 19994 = 3.14159247774332299689906972091627488246 ( 0.00000017584647024156357366236322800173) [ 6] 63168/ 20107 = 3.14159248023076540508280698264286069528 ( 0.00000017335902783337983640063664218891) [ 6] 63523/ 20220 = 3.14159248269040553907022749752720079129 ( 0.00000017089938769939241588575230209290) [ 6] 63878/ 20333 = 3.14159248512270692962179707864063345300 ( 0.00000016846708630884084630463886943119) [ 6] 64233/ 20446 = 3.14159248752812286021715739019857184779 ( 0.00000016606167037824548599308093103640) [ 6] 64588/ 20559 = 3.14159248990709664866968237754754608687 ( 0.00000016368269658979296100573195679732) [ 6] 64943/ 20672 = 3.14159249226006191950464396284829721362 ( 0.00000016132973131895799942043120567057) [ 6] 65298/ 20785 = 3.14159249458744286745248977628097185470 ( 0.00000015900235037101015360699853102949) [ 6] 65653/ 20898 = 3.14159249688965451239353048138577854340 ( 0.00000015670013872606911290189372434079) [ 6] 66008/ 21011 = 3.14159249916710294607586502308314692304 ( 0.00000015442269029238677836019635596115) [ 6] 66363/ 21124 = 3.14159250142018557091459950766900208293 ( 0.00000015216960766754804387561050080126) [ 6] 66718/ 21237 = 3.14159250364929133116730234967274097094 ( 0.00000014994050190729534103360676191325) [ 6] 67073/ 21350 = 3.14159250585480093676814988290398126463 ( 0.00000014773499230169449350037552161956) [ 6] 67428/ 21463 = 3.14159250803708708009131994595350137445 ( 0.00000014555270615837132343732600150974) [ 6] 67783/ 21576 = 3.14159251019651464590285502410085279940 ( 0.00000014339327859255978835917865008479) [ 6] 68138/ 21689 = 3.14159251233344091474941214440499792521 ( 0.00000014125635232371323123887450495898) [ 6] 68493/ 21802 = 3.14159251444821576002201632877717640583 ( 0.00000013914157747844062705450232647836) [ 6] 68848/ 21915 = 3.14159251654118183892311202372804015514 ( 0.00000013704861139953953135955146272905) [ 6] 69203/ 22028 = 3.14159251861267477755583802433266751407 ( 0.00000013497711846090680535894683537012) [ 6] 69558/ 22141 = 3.14159252066302335034551284946479382141 ( 0.00000013292676988811713053381470906278) [ 6] 69913/ 22254 = 3.14159252269254965399478745394086456367 ( 0.00000013089724358446785592933863832052) [ 6] 70268/ 22367 = 3.14159252470156927616577994366700943354 ( 0.00000012888822396229686343961249345065) [ 6] 70623/ 22480 = 3.14159252669039145907473309608540925266 ( 0.00000012689940177938791028719409363153) [ 6] 70978/ 22593 = 3.14159252865931925817731155667684681095 ( 0.00000012493047398028533182660265607324) [ 6] 71333/ 22706 = 3.14159253060864969611556416806130538183 ( 0.00000012298114354234707921521819750236) [ 6] 71688/ 22819 = 3.14159253253867391209080152504491870809 ( 0.00000012105111932637184185823458417610) [ 6] 72043/ 22932 = 3.14159253444967730682016396302110587824 ( 0.00000011914011593164247942025839700595) [ 6] 72398/ 23045 = 3.14159253634193968322846604469516164026 ( 0.00000011724785355523417733858434124393) [ 6] 72753/ 23158 = 3.14159253821573538302098626824423525347 ( 0.00000011537405785544165711503526763072) [ 6] 73108/ 23271 = 3.14159254007133341927721198057668342572 ( 0.00000011351845981918543140270281945847) [ 6] 73463/ 23384 = 3.14159254190899760520013684570646595963 ( 0.00000011168079563326250653757303692456) [ 6] 73818/ 23497 = 3.14159254372898667915052985487509043707 ( 0.00000010986080655931211352840441244712) [ 6] 74173/ 23610 = 3.14159254553155442609063955950868276154 ( 0.00000010805823881237200382377082012265) [ 6] 74528/ 23723 = 3.14159254731694979555705433545504362854 ( 0.00000010627284344290558904782445925565) [ 6] 74883/ 23836 = 3.14159254908541701627789897633831179728 ( 0.00000010450437622218474440694119108691) [ 6] 75238/ 23949 = 3.14159255083719570754520021712806380224 ( 0.00000010275259753091744316615143908195) [ 6] 75593/ 24062 = 3.14159255257252098744908985121768764026 ( 0.00000010101727225101355353206181524393) [ 6] 75948/ 24175 = 3.14159255429162357807652533609100310237 ( 0.00000009929816966038611804718849978182) [ 7] 76303/ 24288 = 3.14159255599472990777338603425559947299 ( 0.00000009759506333068925734902390341120) [ 7] 76658/ 24401 = 3.14159255768206221056514077291914265808 ( 0.00000009590773102789750261036036022611) [ 7] 77013/ 24514 = 3.14159255935383862282777188545321041037 ( 0.00000009423595461563487149782629247382) [ 7] 77368/ 24627 = 3.14159256101027327729727534819507045113 ( 0.00000009257951996116536803508443243306) [ 7] 77723/ 24740 = 3.14159256265157639450282942603071948261 ( 0.00000009093821684395981395724878340158) [ 7] 78078/ 24853 = 3.14159256427795437170562909910272401722 ( 0.00000008931183886675701428417677886697) [ 7] 78433/ 24966 = 3.14159256588960986942241448369782904750 ( 0.00000008770018336904022889958167383669) [ 7] 78788/ 25079 = 3.14159256748674189560987280194585111049 ( 0.00000008610305134285277058133365177370) [ 7] 79143/ 25192 = 3.14159256906954588758335979676087646872 ( 0.00000008452024735087928358651862641547) [ 7] 79498/ 25305 = 3.14159257063821379174076269511954159257 ( 0.00000008295157944672188068815996129162) [ 7] 79853/ 25418 = 3.14159257219293414115980801007160280116 ( 0.00000008139685909730283537320790008303) [ 7] 80208/ 25531 = 3.14159257373389213113469899338059613802 ( 0.00000007985590110732794438989890674617) [ 7] 80563/ 25644 = 3.14159257526126969271564498518171892060 ( 0.00000007832852354574699839809778396359) [ 7] 80918/ 25757 = 3.14159257677524556431261404666692549598 ( 0.00000007681454767415002933661257738821) [ 7] 81273/ 25870 = 3.14159257827599536142249710088906068805 ( 0.00000007531379787704014628239044219614) [ 7] 81628/ 25983 = 3.14159257976369164453681253127044606088 ( 0.00000007382610159392583085200905682331) [ 7] 81983/ 26096 = 3.14159258123850398528510116492949110974 ( 0.00000007235128925317754221835001177445) [ 7] 82338/ 26209 = 3.14159258270059903086725933839520775306 ( 0.00000007088919420759538404488429513113) [ 7] 82693/ 26322 = 3.14159258415014056682622900995365093837 ( 0.00000006943965267163641437332585194582) [ 7] 83048/ 26435 = 3.14159258558728957821070550406657839984 ( 0.00000006800250366025193787921292448435) [ 7] 83403/ 26548 = 3.14159258701220430917583245442217869519 ( 0.00000006657758892928681092885732418900) [ 7] 83758/ 26661 = 3.14159258842504032106822699823712538914 ( 0.00000006516475291739441638504237749505) [ 7] 84113/ 26774 = 3.14159258982595054904011354298946739374 ( 0.00000006376384268942252984029003549045) [ 7] 84468/ 26887 = 3.14159259121508535723583888124372373265 ( 0.00000006237470788122680450203577915154) [ 7] 84823/ 27000 = 3.14159259259259259259259259259259259259 ( 0.00000006099720064587005079068691029160) [ 7] 85178/ 27113 = 3.14159259395861763729576218050381735698 ( 0.00000005963117560116688120277568552721) [ 7] 85533/ 27226 = 3.14159259531330345992800999045030485565 ( 0.00000005827648977853463339282919802854) [ 7] 85888/ 27339 = 3.14159259665679066534986649109330992355 ( 0.00000005693300257311277689218619296064) [ 7] 86243/ 27452 = 3.14159259798921754334838991694594200786 ( 0.00000005560057569511425346633356087633) [ 7] 86598/ 27565 = 3.14159259931072011608924360602212951206 ( 0.00000005427907312237339977725737337213) [ 7] 86953/ 27678 = 3.14159260062143218440638774477924705542 ( 0.00000005296836105405625563850025582877) [ 7] 87308/ 27791 = 3.14159260192148537296246986434457198373 ( 0.00000005166830786550017351893493090046) [ 7] 87663/ 27904 = 3.14159260321100917431192660550458715596 ( 0.00000005037878406415071677777491572823) [ 7] 88018/ 28017 = 3.14159260449013099189777635007316986115 ( 0.00000004909966224656486703320633302304) [ 7] 88373/ 28130 = 3.14159260575897618201208674013508709562 ( 0.00000004783081705645055664314441578857) [ 7] 88728/ 28243 = 3.14159260701766809474914138016499663633 ( 0.00000004657212514371350200311450624786) [ 7] 89083/ 28356 = 3.14159260826632811397940471152489772887 ( 0.00000004532346512448323867175460515532) [ 7] 89438/ 28469 = 3.14159260950507569637149179809617478660 ( 0.00000004408471754209115158518332809759) [ 7] 89793/ 28582 = 3.14159261073402840948848925897417955356 ( 0.00000004285576482897415412430532333063) [ 7] 90148/ 28695 = 3.14159261195330196898414357902073531974 ( 0.00000004163649126947849980425876756445) [ 7] 90503/ 28808 = 3.14159261316301027492363232435434601499 ( 0.00000004042678296353901105892515686920) [ 7] 90858/ 28921 = 3.14159261436326544725286124269561910030 ( 0.00000003922652779120978214058388378389) [ 7] 91213/ 29034 = 3.14159261555417786043948474202658951574 ( 0.00000003803561537802315864125291336845) [ 7] 91568/ 29147 = 3.14159261673585617730812776615089031461 ( 0.00000003685393706115451561712861256958) [ 7] 91923/ 29260 = 3.14159261790840738209159261790840738209 ( 0.00000003568138585637105076537109550210) [ 7] 92278/ 29373 = 3.14159261907193681271916385796479760324 ( 0.00000003451785642574347952531470528095) [ 7] 92633/ 29486 = 3.14159262022654819236247710777996337244 ( 0.00000003336324504610016627549953951175) [ 7] 92988/ 29599 = 3.14159262137234366025879252677455319436 ( 0.00000003221744957820385085650494968983) [ 7] 93343/ 29712 = 3.14159262250942380183091007000538502961 ( 0.00000003108036943663173331327411785458) [ 7] 93698/ 29825 = 3.14159262363788767812238055322715842414 ( 0.00000002995190556034026283005234446005) [ 7] 94053/ 29938 = 3.14159262475783285456610328011223194602 ( 0.00000002883196038389654010316727093817) [ 7] 94408/ 30051 = 3.14159262586935542910385677681275165551 ( 0.00000002772043780935878660646675122868) [ 7] 94763/ 30164 = 3.14159262697255005967378331786235247314 ( 0.00000002661724317878886006541715041105) [ 7] 95118/ 30277 = 3.14159262806750999108233972982792218515 ( 0.00000002552228324738030365345158069904) [ 7] 95473/ 30390 = 3.14159262915432708127673576834485027969 ( 0.00000002443546615718590761493465260450) [ 7] 95828/ 30503 = 3.14159263023309182703340655017539258433 ( 0.00000002335670141142923683310411029986) [ 7] 96183/ 30616 = 3.14159263130389338907760648027175333159 ( 0.00000002228589984938503690300774955260) [ 7] 96538/ 30729 = 3.14159263236681961664876826450584138761 ( 0.00000002122297362181387511877366149658) [ 7] 96893/ 30842 = 3.14159263342195707152584138512418131119 ( 0.00000002016783616693680199815532157300) [ 7] 97248/ 30955 = 3.14159263446939105152640930382813761912 ( 0.00000001912040218693623407945136526507) [ 7] 97603/ 31068 = 3.14159263550920561349298313377108278614 ( 0.00000001808058762496966024950842009805) [ 7] 97958/ 31181 = 3.14159263654148359577948109425611750745 ( 0.00000001704830964268316228902338537674) [ 7] 98313/ 31294 = 3.14159263756630664025052725762126925289 ( 0.00000001602348659821211612565823363130) [ 7] 98668/ 31407 = 3.14159263858375521380583946254019804502 ( 0.00000001500603802465680392073930483917) [ 7] 99023/ 31520 = 3.14159263959390862944162436548223350253 ( 0.00000001399588460902101901779726938166) [ 7] 99378/ 31633 = 3.14159264059684506686055701324566117661 ( 0.00000001299294817160208637003384170758) [ 7] 99733/ 31746 = 3.14159264159264159264159264159264159264 ( 0.00000001199715164582105074168686129155) [ 7] 100088/ 31859 = 3.14159264258137417998053925107504943657 ( 0.00000001100841905848210413220445344762) [ 7] 100443/ 31972 = 3.14159264356311772801201050919554610283 ( 0.00000001002667551045063287408395678136) [ 7] 100798/ 32085 = 3.14159264453794608072307932055477637525 ( 0.00000000905184715773956406272472650894) [ 8] 101153/ 32198 = 3.14159264550593204546866264985402820050 ( 0.00000000808386119299398073342547468369) [ 8] 101508/ 32311 = 3.14159264646714741109838754603695335953 ( 0.00000000712264582736425583724254952466) [ 8] 101863/ 32424 = 3.14159264742166296570441648161855415741 ( 0.00000000616813027275822690166094872678) [ 8] 102218/ 32537 = 3.14159264836954851399944678366167747487 ( 0.00000000522024472446319659961782540932) [ 8] 102573/ 32650 = 3.14159264931087289433384379785604900459 ( 0.00000000427892034412879958542345387960) [ 8] 102928/ 32763 = 3.14159265024570399536062021182431401275 ( 0.00000000334408924310202317145518887144) [ 8] 103283/ 32876 = 3.14159265117410877235673439591191142474 ( 0.00000000241568446610590898736759145945) [ 8] 103638/ 32989 = 3.14159265209615326320894843735790718118 ( 0.00000000149363997525369494592159570301) [ 8] 103993/ 33102 = 3.14159265301190260407226149477372968400 ( 0.00000000057789063439038188850577320019) [ 9] 104348/ 33215 = 3.14159265392142104470871594159265392142 (0.00000000033162780624607255831315103723) [ 9] 208341/ 66317 = 3.14159265346743670552045478534915632492 ( 0.00000000012235653294218859793034655927) [ 9] 312689/ 99532 = 3.14159265361893662339750030141060161556 (0.00000000002914338493485691813109873137) [10] 833719/ 265381 = 3.14159265358107777120441930658185778183 ( 0.00000000000871546725822407669764510236) [11] 1146408/ 364913 = 3.14159265359140397848254241421927966391 (0.00000000000161074001989903093977677972) [11] 3126535/ 995207 = 3.14159265358865040137378454934501063597 ( 0.00000000000114283708885883393449224822) [11] 4272943/ 1360120 = 3.14159265358938917154368732170690821398 ( 0.00000000000040406691895606157259467021) [12] 5419351/ 1725033 = 3.14159265358981538324194377730744861112 (0.00000000000002214477930039402794572693) [13] 42208400/ 13435351 = 3.14159265358977223594679439338801048070 ( 0.00000000000002100251584898989149240349) [13] 47627751/ 15160384 = 3.14159265358977714548655231951908342163 ( 0.00000000000001609297609106376041946256) [13] 53047102/ 16885417 = 3.14159265358978105189821489158366654492 ( 0.00000000000001218656442849169583633927) [13] 58466453/ 18610450 = 3.14159265358978423412652568852445803298 ( 0.00000000000000900433611769475504485121) [14] 63885804/ 20335483 = 3.14159265358978687646612573696921779531 ( 0.00000000000000636199651764631028508888) [14] 69305155/ 22060516 = 3.14159265358978910556761228975786423128 ( 0.00000000000000413289503109352163865291) [14] 74724506/ 23785549 = 3.14159265358979101134054126730478241221 ( 0.00000000000000222712210211597472047198) [14] 80143857/ 25510582 = 3.14159265358979265937562694571217544154 ( 0.00000000000000057908701643756732744265) [15] 165707065/ 52746197 = 3.14159265358979340254615892023457160333 (0.00000000000000016408351553695506871914) [15] 245850922/ 78256779 = 3.14159265358979316028327718420406748404 ( 0.00000000000000007817936619907543540015) [16] 411557987/ 131002976 = 3.14159265358979325782644815641440084536 (0.00000000000000001936380477313489796117) [16] 657408909/ 209259755 = 3.14159265358979322134827119529027452029 ( 0.00000000000000001711437218798922836390) [16] 1068966896/ 340262731 = 3.14159265358979323539256492948091926059 ( 0.00000000000000000307007845379858362360) [17] 2549491779/ 811528438 = 3.14159265358979323901400975919959073572 (0.00000000000000000055136637592008785153) [18] 3618458675/ 1151791169 = 3.14159265358979323794416069185889026381 ( 0.00000000000000000051848269142061262038) [18] 6167950454/ 1963319607 = 3.14159265358979323838637750639037956696 ( 0.00000000000000000007626587688912331723) [19] 14885392687/ 4738167652 = 3.14159265358979323849387505801156062596 (0.00000000000000000003123167473205774177) [19] 21053343141/ 6701487259 = 3.14159265358979323846238174277486901359 ( 0.00000000000000000000026164050463387060) [21] 899125804609/ 286200632530 = 3.14159265358979323846290312739302875642 (0.00000000000000000000025974411352587223) [21] 920179147750/ 292902119789 = 3.14159265358979323846289119831454290206 (0.00000000000000000000024781503504001787) [21] 941232490891/ 299603607048 = 3.14159265358979323846287980289163130623 (0.00000000000000000000023641961212842204) [21] 962285834032/ 306305094307 = 3.14159265358979323846286890609758924162 (0.00000000000000000000022552281808635743) [21] 983339177173/ 313006581566 = 3.14159265358979323846285847590540629124 (0.00000000000000000000021509262590340705) [21] 1004392520314/ 319708068825 = 3.14159265358979323846284848297337933163 (0.00000000000000000000020509969387644744) [21] 1025445863455/ 326409556084 = 3.14159265358979323846283890036945343710 (0.00000000000000000000019551708995055291) [21] 1046499206596/ 333111043343 = 3.14159265358979323846282970332883684002 (0.00000000000000000000018632004933395583) [21] 1067552549737/ 339812530602 = 3.14159265358979323846282086904029653302 (0.00000000000000000000017748576079364883) [21] 1088605892878/ 346514017861 = 3.14159265358979323846281237645725178231 (0.00000000000000000000016899317774889812) [21] 1109659236019/ 353215505120 = 3.14159265358979323846280420613037215131 (0.00000000000000000000016082285086926712) [21] 1130712579160/ 359916992379 = 3.14159265358979323846279634005887720665 (0.00000000000000000000015295677937432246) [21] 1151765922301/ 366618479638 = 3.14159265358979323846278876155814494589 (0.00000000000000000000014537827864206170) [21] 1172819265442/ 373319966897 = 3.14159265358979323846278145514157963557 (0.00000000000000000000013807186207675138) [21] 1193872608583/ 380021454156 = 3.14159265358979323846277440641497885695 (0.00000000000000000000013102313547597276) [21] 1214925951724/ 386722941415 = 3.14159265358979323846276760198188357586 (0.00000000000000000000012421870238069167) [21] 1235979294865/ 393424428674 = 3.14159265358979323846276102935860166316 (0.00000000000000000000011764607909877897) [21] 1257032638006/ 400125915933 = 3.14159265358979323846275467689777075912 (0.00000000000000000000011129361826787493) [21] 1278085981147/ 406827403192 = 3.14159265358979323846274853371947582775 (0.00000000000000000000010515043997294356) [21] 1299139324288/ 413528890451 = 3.14159265358979323846274258964906440289 (0.00000000000000000000009920636956151870) [22] 1320192667429/ 420230377710 = 3.14159265358979323846273683516091186105 (0.00000000000000000000009345188140897686) [22] 1341246010570/ 426931864969 = 3.14159265358979323846273126132748294414 (0.00000000000000000000008787804798005995) [22] 1362299353711/ 433633352228 = 3.14159265358979323846272585977311658438 (0.00000000000000000000008247649361370019) [22] 1383352696852/ 440334839487 = 3.14159265358979323846272062263203084137 (0.00000000000000000000007723935252795718) [22] 1404406039993/ 447036326746 = 3.14159265358979323846271554251010510785 (0.00000000000000000000007215923060222366) [22] 1425459383134/ 453737814005 = 3.14159265358979323846271061245004906498 (0.00000000000000000000006722917054618079) [22] 1446512726275/ 460439301264 = 3.14159265358979323846270582589961333896 (0.00000000000000000000006244262011045477) [22] 1467566069416/ 467140788523 = 3.14159265358979323846270117668253641040 (0.00000000000000000000005779340303352621) [22] 1488619412557/ 473842275782 = 3.14159265358979323846269665897195688730 (0.00000000000000000000005327569245400311) [22] 1509672755698/ 480543763041 = 3.14159265358979323846269226726605047424 (0.00000000000000000000004888398654759005) [22] 1530726098839/ 487245250300 = 3.14159265358979323846268799636567745111 (0.00000000000000000000004461308617456692) [22] 1551779441980/ 493946737559 = 3.14159265358979323846268384135384972222 (0.00000000000000000000004045807434683803) [22] 1572832785121/ 500648224818 = 3.14159265358979323846267979757684694309 (0.00000000000000000000003641429734405890) [22] 1593886128262/ 507349712077 = 3.14159265358979323846267586062682924856 (0.00000000000000000000003247734732636437) [22] 1614939471403/ 514051199336 = 3.14159265358979323846267202632581000779 (0.00000000000000000000002864304630712360) [22] 1635992814544/ 520752686595 = 3.14159265358979323846266829071086609244 (0.00000000000000000000002490743136320825) [22] 1657046157685/ 527454173854 = 3.14159265358979323846266465002047559662 (0.00000000000000000000002126674097271243) [22] 1678099500826/ 534155661113 = 3.14159265358979323846266110068188399415 (0.00000000000000000000001771740238110996) [22] 1699152843967/ 540857148372 = 3.14159265358979323846265763929940953314 (0.00000000000000000000001425601990664895) [22] 1720206187108/ 547558635631 = 3.14159265358979323846265426264360740155 (0.00000000000000000000001087936410451736) [22] 1741259530249/ 554260122890 = 3.14159265358979323846265096764121998082 (0.00000000000000000000000758436171709663) [23] 1762312873390/ 560961610149 = 3.14159265358979323846264775136584745085 (0.00000000000000000000000436808634456666) [23] 1783366216531/ 567663097408 = 3.14159265358979323846264461102927921823 (0.00000000000000000000000122774977633404) [23] 3587785776203/ 1142027682075 = 3.14159265358979323846264306850252143875 ( 0.00000000000000000000000031477698144544) [24] 5371151992734/ 1709690779483 = 3.14159265358979323846264358066268961876 (0.00000000000000000000000019738318673457) [24] 8958937768937/ 2851718461558 = 3.14159265358979323846264337555790890412 ( 0.00000000000000000000000000772159398007) [26] 77042654144230/ 24523438471947 = 3.14159265358979323846264338985711710108 (0.00000000000000000000000000657761421689) [26] 86001591913167/ 27375156933505 = 3.14159265358979323846264338836754332073 (0.00000000000000000000000000508804043654) [26] 94960529682104/ 30226875395063 = 3.14159265358979323846264338715903365924 (0.00000000000000000000000000387953077505) [26] 103919467451041/ 33078593856621 = 3.14159265358979323846264338615889617509 (0.00000000000000000000000000287939329090) [26] 112878405219978/ 35930312318179 = 3.14159265358979323846264338531751659443 (0.00000000000000000000000000203801371024) [26] 121837342988915/ 38782030779737 = 3.14159265358979323846264338459987358119 (0.00000000000000000000000000132037069700) [26] 130796280757852/ 41633749241295 = 3.14159265358979323846264338398054099481 (0.00000000000000000000000000070103811062) [27] 139755218526789/ 44485467702853 = 3.14159265358979323846264338344061241435 (0.00000000000000000000000000016110953016) [27] 288469374822515/ 91822653867264 = 3.14159265358979323846264338319580081089 ( 0.00000000000000000000000000008370207330) [28] 428224593349304/ 136308121570117 = 3.14159265358979323846264338327569743446 ( 0.00000000000000000000000000000380544973) [29] 3137327371971917/ 998642318693672 = 3.14159265358979323846264338328304372840 (0.00000000000000000000000000000354084421) [29] 3565551965321221/ 1134950440263789 = 3.14159265358979323846264338328216143479 (0.00000000000000000000000000000265855060) [29] 3993776558670525/ 1271258561833906 = 3.14159265358979323846264338328146834546 (0.00000000000000000000000000000196546127) [29] 4422001152019829/ 1407566683404023 = 3.14159265358979323846264338328090949305 (0.00000000000000000000000000000140660886) [29] 4850225745369133/ 1543874804974140 = 3.14159265358979323846264338328044932237 (0.00000000000000000000000000000094643818) [30] 5278450338718437/ 1680182926544257 = 3.14159265358979323846264338328006381618 (0.00000000000000000000000000000056093199) [30] 5706674932067741/ 1816491048114374 = 3.14159265358979323846264338327973616618 (0.00000000000000000000000000000023328199) [30] 6134899525417045/ 1952799169684491 = 3.14159265358979323846264338327945425704 ( 0.00000000000000000000000000000004862715) [31] 17976473982901831/ 5722089387483356 = 3.14159265358979323846264338327954374978 (0.00000000000000000000000000000004086559) [31] 24111373508318876/ 7674888557167847 = 3.14159265358979323846264338327952097924 (0.00000000000000000000000000000001809505) [31] 30246273033735921/ 9627687726852338 = 3.14159265358979323846264338327950744587 (0.00000000000000000000000000000000456168) [32] 36381172559152966/ 11580486896536829 = 3.14159265358979323846264338327949847672 ( 0.00000000000000000000000000000000440747) [32] 66627445592888887/ 21208174623389167 = 3.14159265358979323846264338327950254837 ( 0.00000000000000000000000000000000033582) [33] 230128609812402582/ 73252211597019839 = 3.14159265358979323846264338327950319206 (0.00000000000000000000000000000000030787) [33] 296756055405291469/ 94460386220409006 = 3.14159265358979323846264338327950304754 (0.00000000000000000000000000000000016335) [33] 363383500998180356/ 115668560843798173 = 3.14159265358979323846264338327950295601 (0.00000000000000000000000000000000007182) [34] 430010946591069243/ 136876735467187340 = 3.14159265358979323846264338327950289285 (0.00000000000000000000000000000000000866) [35] 1356660285366096616/ 431838381024951187 = 3.14159265358979323846264338327950287593 ( 0.00000000000000000000000000000000000826) [35] 1786671231957165859/ 568715116492138527 = 3.14159265358979323846264338327950288000 ( 0.00000000000000000000000000000000000419) [35] 2216682178548235102/ 705591851959325867 = 3.14159265358979323846264338327950288250 ( 0.00000000000000000000000000000000000169) [35] 2646693125139304345/ 842468587426513207 = 3.14159265358979323846264338327950288418 ( 0.00000000000000000000000000000000000001) [37]
https://polska.pl/science/famousscientists/abrahamstern17691842doesanyonehavecalculator/
The Polish Jew Abraham Jakub Stern (see biography of Abraham Stern), a mathematician, inventor, translator, and censor, was born in 1768 in Hrubieszów, in a poor Jewish family. Around 1800, while working at a clockmaker’s shop in his home town, he was lucky to be noticed by Stanisław Wawrzyniec Staszic (17551826), a leading figure in the Polish Enlightenment: a Catholic priest, statesman, philosopher, geologist, writer and translator. Staszic, who studied at the Hrubieszów and Lublin secondary school in early 1770s, bought an estate in Hrubieszów in 1800. Staszic obviously noticed the extraordinary talent of the humble clockmaker and encouraged him to devote himself to the study of mathematics, Latin, and German, later sending him to Warsaw to continue his studies.
His first computing machine Stern designed around 1810, and in 1811 he sent a report to Staszic, outlining the device and asking for financial help. Later on he designed two more calculating devices. His inventions became popular, at the time of their development. In 1816 and 1818, Stern demonstrated his machine to the Tzar of Russia Alexander I, who received him cordially and granted him an annual pension of 350 roubles, promising, in case of his death, to pay half of this sum to his widow.
For his inventions, Stern was admitted to the Warsaw Society of the Friends of Science (Warszawskiego Towarzystwa Przyjaciół Nauk, the predecessor of the Polish Academy of Sciences), first as a corresponding member (1817), then as a qualifying member (1821), and finally as a full member (1830). He presented his inventions multiple times at the Society’s meetings.
Stern was a fatherinlaw and heavily influenced another inventor of calculating machines—Chaim Zelig Slonimski. Stern most probably had strong influence also over another Polish Jew and inventor—Izrael Abraham Staffel.
Stern presented to the Society three calculating machines. First machine for four arithmetic operations was designed around 1810 and was presented on 7 January, 1813, then a different machine for extracting square roots (presented on 13 January, 1817), and finally the combined machine for four operations and square roots (30 April, 1818). A lengthy description of Stern’s machines had been given by himself, and you can see it below. Unfortunately, an original of any of his machines did not survive to the present time, only a later replica of one of the machines, shown below. There is also a low quality reproduction of Stern with one of his calculating machines (see below).
The only detailed description of the Stern’s machines is his treatise, prepared for the presentation to the Warsaw Scientific Society of the combined machine for four operations and square roots on 30 April 1817, which you can see below (translated from Polish language by Phil Boiarski and Janusz Zalewski):
For the third time, in this earnest place of gatherings of the Society, comprising a selection of learned and enlightened men, I reveal the fruits of my thought—first in the month of January 1813, I presented an invention of a Machine for 4 arithmetic operations—secondly, in January of the current year 1817, an invention of the Machine for extracting roots with fractions—and then finally today, the 30th of April of the current year 1817, an invention combining both these Machines into a single one.
This memorable day is the anniversary of establishing the glorious Warsaw Society of the Friends of Science and honoring it with the title of a Royal Society. I consider myself to be extremely fortunate that on this celebrated day, I can report concerning my inventions, regarding both their historical development and the thought that inspired them as well as regarding properties of said Machines.
Remarks that initially led me to this thought are the following:
A man, although he comes into the world without any means to meet his inevitable needs, nevertheless, being above all creatures with his invaluable gift of mind elevated, with his unlimited ingenuity, uncountable for meeting his stressing needs finds means; because every need he feels inspires him to search and devise means corresponding to this need. That Man, feeling his superiority, thinks that all of nature for his benefit and service has been created and to him subdued; as the Psalmist with astonishment about a man says:
For thou hast made him a little lower than the angels, and hast crowned him with glory and honor. Thou modest him to have dominion over the works of thy hands; thou hast put all things under his feet: All sheep and oxen, yea, and the beasts of the field; The fowl of the air, and the fish of the sea, etc.
This feeling of his undetermined power over all of nature, causes him to regard whatever he finds for himself in nature useful for his needs, even though it should be considered more as a luxury. Experience teaches us that many things that initially were luxurious only because of having been used by a small number of people, with time, however, have become so common that they have shifted from the level intrinsic to luxury to the level of essential need.
From all this ensues, that when in the mankind a number of needs increases, then by this very thing, the ingenuity in methods and means to meet these needs has to multiply. Since such means are commonly based on physical acts, that is, works of the body, which often become onerous, or even beyond human power, so in such case the mind, as a primary Leader of the Man, makes every effort to invent intermediary means to replace the work of the body, or at least to ease it. Following this purpose, numerous mechanical tools have been invented to protect human physical power and support it.
From the depths of this convincing truth, another one equally undeniable have I drawn, that while no one spared efforts to bring assistance and relief to the physical condition, it becomes at least equally necessary to launch a search for mechanical means, which would offer help in human mental activities and relieve the intensity of thought; since the intensity of thought, as it is known, not only often impairs the subtlety of organs, deadens the wit, degrades memory, but also, even causes weakening of the body.
I considered arithmetic or calculation science such a mental activity, one that was necessary, but through an intensity of thought, one that could be harmful. In it, the first 4 types of operations, that is, Addition, Subtraction, Multiplication and Division are the main principles of all calculations, insofar that all other calculi are only the result of combinations of the said 4 kinds. And even though all 4 arithmetic operations, in general, require an uninterrupted presence of mind, that if for a moment gets distracted, a calculation cannot be accurate; since Multiplication and Division, for the reason of higher and more continuous intensity of thought, turn out to be the most difficult ones and therefore so often are subjected to errors.
At this point, I think it would not be unusual, regarding calculation errors, to make the following remark:
In regular calculus, we do not have and even cannot have a test convincing us whether any error was made; this is because a test performed by a reverse calculation, for example, Multiplication by Division, or Division by Multiplication, does not yet constitute a sufficient proof, for it means to test a mental activity by another mental activity. Since the repeated mental action, intending to serve as a test, is subjected to an equal error as in the primary calculation, this error in a test could have obstructed an error made in the calculation itself, and made it invisible.
All these remarks became for me the reason to invent an arithmetic Machine based on mechanical and arithmetic principles, with the assistance of which even people knowing only counting and numbers, all 4 kinds of calculations, and therefore all the other calculi, without slightest application of thought to it, easily could accomplish. And because I thought it to be just, in such an important subject, not to rely solely on the principles of the theory of the mechanism, on which my invention has been based, insofar the slightest error in these principles could have disproved the entire construction of the invention, therefore for better conviction, I elaborated for testing a model of such an arithmetic Machine that worked. And even though the Machine was not of a durable construction, and the required accuracy in the first, rough design, could not have been achieved, however, it exactly performed all the arithmetic calculations, so far that it proved the reality of this important invention.
In the month of December 1812, I submitted this invention for the consideration of the respectable Royal Warsaw Society of the Friends of Science. This Eminent Society, having assessed the invention as corresponding fully to its purpose, deigned to deliver to the public a message about it, in its gathering on January 1813.
I have stated then, that I have planned to make another Machine, made of metal, in a way strong and durable. And although such an endeavor in particular at the initial stage, required time and significant funds for covering expenses, which by then a critical war situation of the Polish state, of which I am a compatriot, made it even more difficult for me, however, not saving efforts on my part, this statement of mine I have put into effect, so far that working continuously on this invention, I have finished a Machine for 4 fundamental arithmetic operations, completely of metal made with the finest work, and performing 13 digit operations.
In conjunction with work on this arithmetic Machine, I also worked on another, by far more difficult, invention of a Machine for extracting roots with fractions. The difference existing between only arithmetic operations and extracting roots already shows the level of difficulty; because in the former, there are always at least two known numbers given and the third unknown is searched for, but in extracting roots, there is only one known number given, and the other one unknown, that multiplied by itself equals the given number. Admittedly I learned that I ventured into such an abyss, from which a recovery is subject to numerous difficulties, both regarding the implementation of a thought as well as the huge costs, which a carefully elaborated plan definitely required. But no difficulty could oppose my keen willingness to finish the invention, which from various points of view seems to be important, both for the intention proper, the relief in the intensity of thought and counteracting unintentional errors, and to create a completely new mechanical means, included in this invention, which could apply great benefits to mechanical tools in other objects.
Thanks to the Almighty, I have passed this difficult and dangerous path, too, and the Machine for extracting roots with fractions I have led to the intended goal. This invention just like the first one, I have submitted for consideration of the glorious Royal Society of the Friends of Science, about which the public has been informed at the past January gathering of this Society.
This way, then, these two inventions, two separate Machines have been formed, one for the 4 arithmetic operations, and another for extracting roots.
I began thinking further on the ways, which these two inventions could combine into a single Machine. It seemed to me, initially, to be impossible, indeed. But finally, at this point mechanical ingenuity showed me the means to put my intent into effect. The importance of this thought so overwhelmed me, that to all unpleasant things stemming from shortages I have been insensitive, doubling my efforts, so I could make this combination sooner.
Thanks to the Highest Being, in this subject I did not fail either. I can say this boldly, since I am referring to the convincing proof, that is, before our eyes: a Machine, which accurately performs all 4 arithmetic operations, as well as extracting roots.
If I wanted to venture into the details and explanations of all principles of the internal Mechanism of this Machine, the purpose would be missed. Because the Mechanism, comprising several wheels of various kind, rotations of a new type, springs and levers, by various means connected with each other, requires an extensive description and many figures, which will be the subject of a work planned for a later date, with figures clearly presenting the matter, but in this treatise clarifications would be an excessive boring of the respected public. Therefore I am moving now only to a brief sketch of the Machine and the explanation of the way of using it, in various arithmetic operations and extracting roots, as well as doing a foolproof test.
This Machine has a shape of a parallelepiped, longish and rectangular, in its length by five rows of wheels divided. The first, uppermost row, just like the second one underneath, is composed of 13 wheels based on axles. The wheels of the first uppermost row have discs, on which there are engraved ordinary digits of numbers of which only one number over the aperture is visible. Because the numbers of these wheels correspond to positions of units, tens, hundreds, etc., thus, this row entails trillions. Wheels of the second row, in turn, do not have discs, and serve only as the Mechanism offering movement to the uppermost numerical wheels. Both these rows do not change (their place in the Machine. Behind these rows underneath, there are two rows of wheels, which similarly have numerical discs visible through the apertures, and are placed in a separate base in the shape of a carriage. This carriage with its two rows of wheels is so imbedded in the Machine, that it can easily move on smaller wheels or rollers. The first row in this carriage has 7 numerical wheels on the axles of which there are as many folding cranks. Besides these cranks, on the diameter of a folding crank, there is another main crank which can be inserted and removed. The second row underneath has 8 wheels. Below this carriage, there is a lowermost row, composed of 7 wheels, equipped with numerical discs visible through apertures. This row has a stable and invariable place in the Machine. In addition to these rows of wheels, in top of the Machine there are two more rows of wheels, on which Roman numerals are engraved, visible through apertures. One of these rows has its place above the ordinary numerical apertures of the uppermost row, and the other above the ordinary numerical apertures of the lowermost row.
The way of using the Machine in operations is the following:
When any of the 4 arithmetic operations is to be performed, then one has to move to the left the handle positioned on the right on the carriage at the second row. As a result of this move, on a carriage on the left hand side, the word Species shows through an aperture, all the numerical apertures of the second row are covered, and thus the Machine is ready for 4 arithmetic operations. If the species of an operation is to be addition or multiplication, then with two handles at the ends of the Machine devised at the right and left hand sides, one moves up, while at the same time the words: Addition – Multiplication show up on the Machine through apertures, and by this the Machine is ready for these operations. If the species of the calculation were subtraction or division, this is done by moving from top down, the same way, words Addition and Multiplication disappear, and in their place the words Subtraction – Division are seen and the Machine by this is ready for to the said operations.
In calculations of addition or subtraction, one puts the first number known participating in the problem, in the uppermost row, and the other one in the first crank row on a carriage. The operation is performed by the main crank in the middle of the carriage base, which gives movement to the entire Machine. If only a single circular rotation is performed, then the brake, located on the left hand side of the carriage, stops further movement of the Machine, and at the same time the unknown number searched for, appears in the uppermost row through apertures as a result of the operation.
In addition, there is one more convenience in the Machine, that is, because in this type of operation it happens that more than two rows of numbers in the calculation have to be combined, for example, in Registers and Tables, one sets on the Machine the first two given rows, as mentioned above, and by making a single circular rotation of a crank the Machine brakes, one touches the brake with a finger, and the Machine becomes available for rotation. Furthermore, one sets, in the third crank row on a carriage, the third given row, and rotates the crank once again, and so on, acting so until all the given rows are exhausted; at that time, in the uppermost row there will be a general Sum of all the given numbers. However, to prevent an error from squeezing in, when all these different numbers are being added, which can especially happen when the operator interrupts the work, the Machine shows, through the aperture, the number corresponding to the value of how many given rows have been taken to the operation thus far.
Multiplication is performed in the following way. One of the factors is set on the crank row in a carriage, and the other on the lowermost row, while on the uppermost row, which is designated for the product searched for, zeroes are placed. After that, one moves the carriage from the right to the left side, to the very end of the Machine, by the handle placed on the left hand side of the carriage. After releasing the handle, the carriage returns by itself, and stops in a position resulting from the nature of the problem. In this position one begins rotation of the main crank. During the rotation, the carriage moves by itself from one number to the other towards the right hand side, back through the end of the Machine; over there, the operation lasts until the ring of a bell warns about the operation’s completion, while at the same time the product searched for appears already on the uppermost row. In this species of operation, the Machine has a particular superiority over calculations in an ordinary manner, that from several given multiplications one can obtain a general product without performing an addition operation, that is, without combining individually calculated products together. This is because in an ordinary calculation, in such case, one has to first calculate a separate product from every two factors, then collect all individual products together and, by addition, derive the general product. On the Machine, however, one sets the first task and operates as long as the ring of a bell indicates to stop; not paying attention to the value of a product, one sets the second task, third, and so on, and when after the last operation the ring of a bell indicates to stop the rotations, at that time the general product of all the tasks appears on the uppermost row.
In division, one proceeds in the following way. The dividend is set on the uppermost row, and the divisor on the crank row in the carriage, while on the lowermost row, designated for the quotient, zeroes are placed. The carriage moves towards the left hand side, until the divisor stands straight under the dividend number being greater or at least equal to the divisor. Then a main crank rotation begins and lasts as long as the dividend number becomes smaller than the divisor, at which point one presses with finger a flap situated on the right hand side of the carriage, after which the carriage moves by itself towards the right hand side and stops at the appropriate place, where further operation continues in a similar manner till the end of the job. And when the divisor located on the carriage, standing in its first place, that is, at the end of the Machine on the right hand side, carries the dividend, then the operation is to stop and the quotient appears on the lowermost row. In case there is a fraction, then the numerator appears on the uppermost row and the denominator on a crank row in the carriage. If on the uppermost row there are only zeroes, this means that quotient is a whole number, without a fraction.
I am now going to describe the way of extracting roots.
If one wants to extract a square root from a given number, first, one has to move to the right the handle at the second row on the right hand side of the carriage, so that at the left hand side of the carriage, the word Species disappears and is replaced by the word Radices in the aperture. Then, numerical apertures of the second row of the carriage open, and the Machine is ready for extracting roots. Next, one has to move the handles at the ends of the Machine from top down, where between the inscriptions Subtraction – Division, one can also see on the Machine the word Extraction. The main crank in the middle of the carriage has to be removed, and the smaller folding cranks replace it in the operation. On the uppermost row, one sets up the known number of a given square, and on the first and second rows of the carriage all zeroes, except at the position of units in the second row, where one places the number 1. At the apertures for ordinary numbers of the uppermost row there are various signs dividing this row into sections, in such an order that for every two numerical wheels there is a sign, that is, at units, hundreds, tens of thousands, millions, and so on. On the said cranks there are identical signs, so that each crank corresponds to two wheels of the uppermost row, for example, the first crank from the right corresponds to units and tens, the second one – hundreds and thousands, and so on. The last sign, at the given number of the square, points to the crank from which the operation has to start, for example, if the given square ends on the wheels of the first sign, then the operation has to be undertaken with the first crank on the right hand side. If, however, a given square ended on the wheels of the second sign, the operation then begins with the second crank, having the same sign. Indicated this way the folding crank unfolds, the carriage moves to the left until the unfolding crank stops in front of the last sign of a given square. The rotation is conducted with this unfolding crank and lasts as long as the number on the uppermost row, in front of the rotating crank, becomes smaller than or, at least, equal to the number positioned in front of the same crank on the second row of the carriage. Next, by folding this crank, the crank to the right of it unfolds, and by pressing with finger a flap on the right hand side of the carriage, the carriage moves by itself to the right hand side, until it is stopped by a folded crank, just in front of the previous section. One performs the same operations as above, for each section up to the last one. After completing the operation, if a given number was a full square, then it is replaced by zeroes and the whole root on the crank row in the carriage appears. Otherwise, except of the whole number root, an additional fraction results, namely, the numerator on the uppermost row and the denominator on the second row in the carriage.
To approximate the root in decimal fractions, one has to set on the uppermost row as many sections to zeroes as decimal digits in a fraction we want to have, for example, if the root is to be extracted from the number 7, and a fraction approximated with two decimal digits, one sets two sections, or four wheels, to zeroes, and the given number 7 is set in the third section, that is, on the 5th wheel of the uppermost row. To distinguish between the number actually given and the zeroes attached to it, a moving hand always slides out under this sign where the actual number has been set, which warns the operator how many digits for a decimal fraction he has to cut from the right hand side on the crank row in the carriage. This way, then, when a given number is under the third sign, one has to unfold the third crank and perform the operations as above. The root will, then, result on the 3 crank wheels in the carriage, as number 264. Cutting off, following the hand’s warning, two digits for a decimal fraction, will mean 2 wholes and 64 hundredths. In addition to that, in the uppermost row there is number 304, as a numerator, and in the second row of the carriage, 529, as a denominator of the ordinary fraction of the tenth units of the first order.
At the beginning of this treatise I explained that in our ordinary calculus, there is no convincing test that in our mental operation there was not any error, and that a way of testing by reverse calculation does not constitute a sufficient proof. The same remark applies to the operation of the Machine. In case the Machine, due to damage, produced a false result, then the test by a reverse operation would not be proving, because the same damage that caused a false result in Multiplication, for example, would have affected a false result in Division, which would be clear even from the composition of the Machine. But to remedy this, I devised for the Machine a completely different kind of test, which is an absolute proof.
Two rows of wheels with Roman numerals, mentioned above, located on top of the Machine, are designed for this purpose.
To obtain such a reliable test, one proceeds as follows.
Regarding Multiplication: Since from all the factors, the numbers of the factor on the uppermost row disappear during the work, being replaced by zeroes, so to make it visible after the work, what factor was a part of the problem, one sets in advance, on a Roman numbered row located above the apertures of the lowermost row, the signs corresponding to the digits of the factor which is to disappear. And because after completing the operation, the product results on the uppermost row and on the lowermost row there are zeroes, so one shifts as many zeroes to the number 9 as the number of digits of the remaining factor in the Roman numbered row, except of the first digit, being meaningful, at the right hand side of the factor, where the appearing zero remains. After this, the operation of testing begins. The carriage moves to the left and will stop by itself at the last number 9, but the rotation lasts as long as the number appears that is equal to the Roman numeral right above it, that is, the same which previously disappeared. At that time, one presses the flap on the right hand side of the carriage, the carriage moves to the right and the rotations proceed further, as before, until a given factor fully appears on its first place, that is, on the lowermost row.
If after this work it turns out that there are as many digits in the factor on the lowermost row, as the number of zeroes on the uppermost row, at the right hand side, and the numbers following them are equal to the numbers on the crank row of the carriage, then it is an absolute proof that the first product was true, otherwise it had been false.
In Division, the test is conducted as follows. When the dividend is set on the uppermost row, at the same time on the Roman numerals row above it, the same signs are set, so that the dividend, which disappears during the work, that way be preserved; and when after completion of the division, the quotient appears on the lowermost row, it is then moved to the Roman numerals row directly above it. Then one moves the carriage towards the left hand side, until the first number of units on the carriage appears in front of the last number of the quotient. The rotation is conducted in this place, until the carriage moves by itself towards the right hand side, and this continues number by number, until the carriage moves to the first number of units, where one has to rotate as long as a zero appears. After completing this operation, one moves the carriage to the left as long as its first number of units passes all the digits of the preserved quotient. When the carriage is moved thus far, one has to watch if the number on the uppermost row, which has now formed, combined with the number on the crank row of the carriage, will match the dividend preserved on the Roman numerals row, which is a positive proof that the first result was true, otherwise it had been false.
The test of extracting roots is conducted the same way as in the division, except that right before making the test, one has to adjust the Machine from the state of extracting roots to the state of arithmetic operations, and then set on the lowermost row the number equal to the root resulted in the crank row of the carriage. The remaining steps and the proving test are the same as in the division testing.
Remark: I am ending this treatise with a remark that since the Mechanics is an Opener to [meeting] our needs, insofar that not only our physical power but even that of mental power can replace, thus we should put our strongest effort to propagate ingenuity in such a broad and useful field, not venturing, however, into a search for perpetuum mobile, that is, an eternal motion, since this is an incurable disease of Mechanics, just as a philosopher’s stone and an inextinguishable fire in Chemistry, and a squaring of a circle in Geometry—all thoughts in this area have an attribute of an ineffective stubbornness. Let us better strive to make progress in mechanical matters showing promise, since such conduct paves the way to the wellbeing and glory of the Nation.
Contents 

Canon “Canola” Model 167P, 1971 
http://www.vcalc.net/hpcode.htm
Last Update: January 22, 2003 — THE HP REFERENCE
Just as DNA can be called the building blocks of the human organism. Microcode can be called the building blocks of the electronic calculator.
But, while the way the human organism works is determined by heredity, the way an electronic calculator works is determined by the highly personal–and even idiosyncratic–creative impulses of a programming specialist.
The principles of programming can be learned, of course. But anyone who has programmed his own calculator quickly discovers that techniques may vary widely from person to person.
Consider the challenge faced by the professional programmer When you press the key labeled SIN, for example, you expect the calculator to display the sine of the value you have keyed into it–and presto, it does. But in that lessthanasecond interval between keystroke and display, the calculator has executed an internal program of about 3500 steps. And it does this according to the highly individualistic microcode that the programmer has created.
The development of microcode in HewlettPackard personal calculators began with the development developments of the microprocessor in the HP35–and not coincidentally, since both were developed by the same HewlettPackard engineer.
It is the microprocessor that determines the “language” of the internal microcode. If you are familiar with computer languages such as BASIC, FORTRAN, and COBOL, you know that these languages structure the way you write your program on the computer. You can only do what the language lets you do.
The microprocessor is similar to the computer. It provides a language that a clever engineer can then build into a function on the keyboard.
The original HP35 microprocessor has remained essentially unchanged through the years and is the heart of the new HP19C and HP29C. Compared with computer processors, the binarycoded decimal microprocessor is very simple it does not handle byte data well, but is, in fact, specially designed for 10digit floating point numbers (See figure 1.). The resulting microcode language most closely resembles machine language, which is programming at as most basic level.
Most microprocessors use 8bit instructions and two or three instructions are usually combined to perform one operation. The beauty of HP’s calculator microcode is that 10bit instructions are used and each usually performs a complete operation by itself.
The language’s strongest point is its robust arithmetic section of 37 instructions combined with eight fieldselect options. The fieldselect options allow the program to apply the instruction to any wordselect portion of the register (See figure 2.).
The language is also designed to use very little storage; only seven registers were used in the HP35 of which five were user registers. This is done to reduce costs and to save valuable space. For the design engineer it means that he must accomplish all of his miracles within the program itself.
Based on warranty card analysis and other market research, parameters regarding the desired function set and price are given to the design engineer. It is his job then to determine the specific functions for the calculator and to attempt to fit them in the allotted memory. Price is an important factor to the engineer because it directly influences the amount of memory he has to work with.
Only after several months of hard work writing and compacting microcode will he know if the function set will fit. If it is not possible, the product may be redefined at a higher price with greater performance to increase the available memory. More likely however, the engineer will be forced to pare functions until his program fits.
To give you an idea of how much memory is required, the HP35 used three pages of 256 instructions each. Each page required a separate ROM readonly memory.
The HP45 originally took six pages of instructions. But about that time the quad ROM was developed, which, as its name implies, was the equivalent of four conventional ROM’S. So for the HP45, two quad ROM’s were used. It was in the leftover two pages that an enterprising designer placed the celebrated HP45 clock. Later calculators, listed below, continued to use quad ROM’s.
Note: 1 quad = 1024 instructions or 4 pages of 256 instructions each. Rick
Writing the microcode is where the designer’s personality is stamped indelibly on the Calculator. While it is true that the fundamental algorithms for commuting the complex mathematical functions found in HP personal calculators have remained essentially the same since the HP35, the individual code is substantially different.
++  NUMBER REGISTER REPRESENTATION     +++++++++++++++   23.4  0  2  3  4  0  0  0  0  0  0  0  0  0  1    +++++++++++++++     a   +++++++++++++++   123.  9  1  2  3  0  0  0  0  0  0  0  0  0  2    +++++++++++++++     b   +++++++++++++++   .002  0  2  0  0  0  0  0  0  0  0  0  9  9  7    +++++++++++++++     a. A "9" in the sign position indicates a negative number.   b. Exponents are kept in 10's complement form.    ++  Figure 1. All numbers in registers are in scientific notation with   the mantissa portion of the number left justified in the mantissa   portion of the register.  ++
++  mantissa sign exponent sign   pointer     positions+++++++++++++++   > 13 12 11 10  9  8  7  6  5  4  3  2  1  0    +++++++++++++++                + mantissa + exponent +        + mantissa and sign +        + word +    ++ ++ Instructions  +++  A = 0    B = 0  Clears wordselect portion.   C = 0   +++  B = A  Copies wordselect portion   C = B  from specified register   A = C  to specified register.  +++  AB EX  Exchanges wordselect portion   AC EX  between specified registers.   CB EX   +++  A = A + B    A = A + C    C = A + C    C = C + C    A = A  B    C = A  C    A = A  C  Performs stated arithmetic   A = A + 1  on wordselect portion.   C = C + 1    A = A  1    C = C  1    C = C    C = C  1   +++  A SR    B SR  Shifts wordselect portion right.   C SR   +++     A SL  Shifts wordselect portion left.     +++   Circular shifts whole A   A SLC  register but does not have    wordselect option.  +++  ? B = 0    ? C = 0    ? A => C  Tests wordselect portion of   ? A => B  given register.   ? A = 0    ? C = 0   +++  FieldSelect Options     1. Mantissa (M)   2. Mantissa and Sign (MS)   3. Exponent (X)   4. Exponent Sign (XS)   5. Sign of Mantissa (S)   6. Pointer (P)   7. Word (W)   8. Word thru Pointer (WP)    ++  Figure 2. Registers are 14 digits long with each digit being four   bits. An additional four bit register is used as a pointer. The   programmer can set the pointer to any digit position, change that   digit or all digits up to that position.  ++
Some of the major routines such as those for sine, cosine, and tangent are the same from calculator to calculator. But in most cases if the code cannot be borrowed exactly as it exists in another calculator, it must be rewritten.
Some designer begin by strictly flowcharting the entire program Others tackle the code directly and leave all but basic flowcharting till the end for documentation purposes only. The code is written originally on paper just as you would write a program on an HP programming pad. It is then either punched on cards or typed into a handy CRT.
The major task in writing the microcode is not only to have the functions produce the correct answers when a key is pressed, but to fit the code into a given amount of memory and make it execute as fast as possible. Compacting the code, that is, rewriting sections of the program to make them more spaceefficient, is easy enough, but sometimes results in a loss of speed. These are tradeoffs to which the designer must be constantly attentive.
The processor executes approximately three instructions per millisecond, with each instruction taking the same amount of time. The designer takes into consideration the type of function and the necessary speed when writing the code. Straightline code with no branches of any I kind executes the fastest. For the label search function on the HP67 and the HP97 the designer duplicated a great deal of code to; make it faster. The print instructions for the new HP19C, on the other hand, did not need to be as fast to keep up with the printer, mechanism. So the designer compacted these codes, making them very complex.
++  ++ ++ ++    A. Answer  B. Multiplicand  C. Multiplier    ++ ++ ++    +++++  LABEL  CODE   Explanation  +++++   A = 0  W  Clearing space for answer.    P =  3  Starting at least significant digit.    GoTo  Mpy90  Starting in middle of loop for speed.   Mpy90  A = A + B  W  Add Multiplicand to partial product      the number of times specified by      digit being worked on.  +++++  Mpy100  C = C  1  P     GoNC  Mpy90  NC stands for no carry, i.e. a GoTo      is executed unless a digit is      carried (or borrowed).    ? P =  12  Have we reached the end of the world?    GoYes  END        +++++   P = P + 1   Move on the next digit.    A SP  W  This does a divide by 10 to line up      for the next decade.    GoTo  Mpy100  Go reenter loop.       +++++  Figure 3. Multiplying two numbers uses the basic routine shown   above and involves three registers The starting point of this   code assumes the sign and exponent of the answer have already   been calculated. END is a common function return which would   perform operations such as display formatting, printing, etc.  ++
In general, programmable calculators go through more gyrations for each function than do preprogrammed calculators because they must generate an intermediate keycode. Where simple functions such as Change Sign, x exchange y, and ENTER require only 100 steps of code on a preprogrammed calculator, the same functions on a programmable calculator might take 150 steps. And a complex function such as rectangulartopolar conversion might take over 4000 steps, depending on the argument keyed in.
The engineer uses a computer simulator to write and debug his code. Special programs written for this simulator furnish him with status bit information, register contents, and intermediate answers as an aid in this process.
Once the microcode is completed on the computer, it is transferred to an EROM (erasable read onlymemory) simulator for further debugging (See photo.). The simulator is ideal for this because it is portable and easily updated as bugs are found and corrected. Simulated calculators are given to application engineers and quality assurance engineers to help locate problems.
After much editing, the microcode is ready to be converted into hardware. This usually takes several weeks. In the meantime, the simulators continue to be used heavily and, in most cases, additional bugs are found.
When the completed integrated circuit chips return, the first working models of the calculator are constructed. Final testing is then initiated. Some problems can only be discovered at this stage because of the peculiarities of simulated operation. For example, this is the first time low battery indicators can be checked since the EROM simulator does not work on batteries.
At long last, the revised code is ready to be sent for final chips. Although no problems are anticipated at this stage, testing continues to assure traditional HewlettPackard reliability.
The tremendous emphasis on testing of the calculators is for practical reasons as well incorrect programs cannot be easily corrected as they can be on large computers. Once the code is set in hardware, changes are costly and inconvenient.
When the final chips are approved for production, the development cycle is complete. The design engineer has spent anywhere from six months to 13 months perfecting his building block design. And whether the end product is a highpowered financial calculator like the HP92 or a versatile keystroke programmable like the HP19C, it is first an expression of his personality and creativity.
The Calculator Reference by Rick Furr (rfurr@vcalc.net)
https://docs.oracle.com/cd/E1995701/8063568/ncg_goldberg.html
Numerical Computation Guide 
What Every Computer Scientist Should Know About FloatingPoint Arithmetic
Note – This appendix is an edited reprint of the paper What Every Computer Scientist Should Know About FloatingPoint Arithmetic, by David Goldberg, published in the March, 1991 issue of Computing Surveys. Copyright 1991, Association for Computing Machinery, Inc., reprinted by permission.
Abstract
Floatingpoint arithmetic is considered an esoteric subject by many people. This is rather surprising because floatingpoint is ubiquitous in computer systems. Almost every language has a floatingpoint datatype; computers from PCs to supercomputers have floatingpoint accelerators; most compilers will be called upon to compile floatingpoint algorithms from time to time; and virtually every operating system must respond to floatingpoint exceptions such as overflow. This paper presents a tutorial on those aspects of floatingpoint that have a direct impact on designers of computer systems. It begins with background on floatingpoint representation and rounding error, continues with a discussion of the IEEE floatingpoint standard, and concludes with numerous examples of how computer builders can better support floatingpoint.
Categories and Subject Descriptors: (Primary) C.0 [Computer Systems Organization]: General — instruction set design; D.3.4 [Programming Languages]: Processors — compilers, optimization; G.1.0 [Numerical Analysis]: General — computer arithmetic, error analysis, numerical algorithms (Secondary)
D.2.1 [Software Engineering]: Requirements/Specifications — languages; D.3.4 Programming Languages]: Formal Definitions and Theory — semantics; D.4.1 Operating Systems]: Process Management — synchronization.
General Terms: Algorithms, Design, Languages
Additional Key Words and Phrases: Denormalized number, exception, floatingpoint, floatingpoint standard, gradual underflow, guard digit, NaN, overflow, relative error, rounding error, rounding mode, ulp, underflow.
Introduction
Builders of computer systems often need information about floatingpoint arithmetic. There are, however, remarkably few sources of detailed information about it. One of the few books on the subject, FloatingPoint Computation by Pat Sterbenz, is long out of print. This paper is a tutorial on those aspects of floatingpoint arithmetic (floatingpoint hereafter) that have a direct connection to systems building. It consists of three loosely connected parts. The first section, Rounding Error, discusses the implications of using different rounding strategies for the basic operations of addition, subtraction, multiplication and division. It also contains background information on the two methods of measuring rounding error, ulps and
relative
error
. The second part discusses the IEEE floatingpoint standard, which is becoming rapidly accepted by commercial hardware manufacturers. Included in the IEEE standard is the rounding method for basic operations. The discussion of the standard draws on the material in the section Rounding Error. The third part discusses the connections between floatingpoint and the design of various aspects of computer systems. Topics include instruction set design, optimizing compilers and exception handling.I have tried to avoid making statements about floatingpoint without also giving reasons why the statements are true, especially since the justifications involve nothing more complicated than elementary calculus. Those explanations that are not central to the main argument have been grouped into a section called “The Details,” so that they can be skipped if desired. In particular, the proofs of many of the theorems appear in this section. The end of each proof is marked with the z symbol. When a proof is not included, the z appears immediately following the statement of the theorem.
Rounding Error
Squeezing infinitely many real numbers into a finite number of bits requires an approximate representation. Although there are infinitely many integers, in most programs the result of integer computations can be stored in 32 bits. In contrast, given any fixed number of bits, most calculations with real numbers will produce quantities that cannot be exactly represented using that many bits. Therefore the result of a floatingpoint calculation must often be rounded in order to fit back into its finite representation. This rounding error is the characteristic feature of floatingpoint computation. The section Relative Error and Ulps describes how it is measured.
Since most floatingpoint calculations have rounding error anyway, does it matter if the basic arithmetic operations introduce a little bit more rounding error than necessary? That question is a main theme throughout this section. The section Guard Digits discusses guard digits, a means of reducing the error when subtracting two nearby numbers. Guard digits were considered sufficiently important by IBM that in 1968 it added a guard digit to the double precision format in the System/360 architecture (single precision already had a guard digit), and retrofitted all existing machines in the field. Two examples are given to illustrate the utility of guard digits.
The IEEE standard goes further than just requiring the use of a guard digit. It gives an algorithm for addition, subtraction, multiplication, division and square root, and requires that implementations produce the same result as that algorithm. Thus, when a program is moved from one machine to another, the results of the basic operations will be the same in every bit if both machines support the IEEE standard. This greatly simplifies the porting of programs. Other uses of this precise specification are given in Exactly Rounded Operations.
Floatingpoint Formats
Several different representations of real numbers have been proposed, but by far the most widely used is the floatingpoint representation.^{1}Floatingpoint representations have a base (which is always assumed to be even) and a precision p. If = 10 and p = 3, then the number 0.1 is represented as 1.00 × 10^{1}. If = 2 and p = 24, then the decimal number 0.1 cannot be represented exactly, but is approximately 1.10011001100110011001101 × 2^{4}.
In general, a floatingpoint number will be represented as ± d.dd… d × ^{e}, where d.dd… d is called the significand^{2} and has p digits. More precisely ± d_{0} . d_{1} d_{2} … d_{p1} × ^{e} represents the number
(1) .
The term floatingpoint number will be used to mean a real number that can be exactly represented in the format under discussion. Two other parameters associated with floatingpoint representations are the largest and smallest allowable exponents, e_{max} and e_{min}. Since there are ^{p}possible significands, and e_{max} – e_{min} + 1 possible exponents, a floatingpoint number can be encoded in
bits, where the final +1 is for the sign bit. The precise encoding is not important for now.There are two reasons why a real number might not be exactly representable as a floatingpoint number. The most common situation is illustrated by the decimal number 0.1. Although it has a finite decimal representation, in binary it has an infinite repeating representation. Thus when = 2, the number 0.1 lies strictly between two floatingpoint numbers and is exactly representable by neither of them. A less common situation is that a real number is out of range, that is, its absolute value is larger than × or smaller than 1.0 × . Most of this paper discusses issues due to the first reason. However, numbers that are out of range will be discussed in the sections Infinity and Denormalized Numbers.
Floatingpoint representations are not necessarily unique. For example, both 0.01 × 10^{1} and 1.00 × 10^{1} represent 0.1. If the leading digit is nonzero (d_{0} 0 in equation (1) above), then the representation is said to be normalized. The floatingpoint number 1.00 × 10^{1} is normalized, while 0.01 × 10^{1} is not. When = 2, p = 3, e_{min} = 1 and e_{max} = 2 there are 16 normalized floatingpoint numbers, as shown in FIGURE D1. The bold hash marks correspond to numbers whose significand is 1.00. Requiring that a floatingpoint representation be normalized makes the representation unique. Unfortunately, this restriction makes it impossible to represent zero! A natural way to represent 0 is with 1.0 × , since^{ }this preserves the fact that the numerical ordering of nonnegative real numbers corresponds to the lexicographic ordering of their floatingpoint representations.^{3} When the exponent is stored in a k bit field, that means that only 2^{k} – 1 values are available for use as exponents, since one must be reserved to represent 0.
Note that the × in a floatingpoint number is part of the notation, and different from a floatingpoint multiply operation. The meaning of the ×symbol should be clear from the context. For example, the expression (2.5 × 10^{3}) × (4.0 × 10^{2}) involves only a single floatingpoint multiplication.
FIGURE D1 Normalized numbers when = 2, p = 3, e_{min} = 1, e_{max} = 2
Relative Error and Ulps
Since rounding error is inherent in floatingpoint computation, it is important to have a way to measure this error. Consider the floatingpoint format with = 10 and p = 3, which will be used throughout this section. If the result of a floatingpoint computation is 3.12 × 10^{2}, and the answer when computed to infinite precision is .0314, it is clear that this is in error by 2 units in the last place. Similarly, if the real number .0314159 is represented as 3.14 × 10^{2}, then it is in error by .159 units in the last place. In general, if the floatingpoint number d.d…d × ^{e} is used to represent z, then it is in error by d.d…d – (z/^{e})^{p1 }units in the last place.^{4}^{,} ^{5} The term ulps will be used as shorthand for “units in the last place.” If the result of a calculation is the floatingpoint number nearest to the correct result, it still might be in error by as much as .5 ulp. Another way to measure the difference between a floatingpoint number and the real number it is approximating is relative error, which is simply the difference between the two numbers divided by the real number. For example the relative error committed when approximating 3.14159 by 3.14 × 10^{0} is .00159/3.14159 .0005.
To compute the relative error that corresponds to .5 ulp, observe that when a real number is approximated by the closest possible floatingpoint number d.dd…dd × ^{e}, the error can be as large as 0.00…00‘ × ^{e}, where ‘ is the digit /2, there are p units in the significand of the floatingpoint number, and p units of 0 in the significand of the error. This error is ((/2)^{p}) × ^{e}. Since numbers of the form d.dd…dd × ^{e} all have the same absolute error, but have values that range between ^{e} and × ^{e}, the relative error ranges between ((/2)^{p}) × ^{e}/^{e} and ((/2)^{p}) × ^{e}/^{e+1}. That is,
(2)
In particular, the relative error corresponding to .5 ulp can vary by a factor of . This factor is called the wobble. Setting = (/2)^{p} to the largest of the bounds in (2) above, we can say that when a real number is rounded to the closest floatingpoint number, the relative error is always bounded by e, which is referred to as machine epsilon.In the example above, the relative error was .00159/3.14159 .0005. In order to avoid such small numbers, the relative error is normally written as a factor times , which in this case is = (/2)^{p} = 5(10)^{3} = .005. Thus the relative error would be expressed as (.00159/3.14159)/.005) 0.1.
To illustrate the difference between ulps and relative error, consider the real number x = 12.35. It is approximated by = 1.24 × 10^{1}. The error is 0.5 ulps, the relative error is 0.8. Next consider the computation 8 . The exact value is 8x = 98.8, while the computed value is 8 = 9.92 ×10^{1}. The error is now 4.0 ulps, but the relative error is still 0.8. The error measured in ulps is 8 times larger, even though the relative error is the same. In general, when the base is , a fixed relative error expressed in ulps can wobble by a factor of up to . And conversely, as equation (2) above shows, a fixed error of .5 ulps results in a relative error that can wobble by .
The most natural way to measure rounding error is in ulps. For example rounding to the nearest floatingpoint number corresponds to an error of less than or equal to .5 ulp. However, when analyzing the rounding error caused by various formulas, relative error is a better measure. A good illustration of this is the analysis in the section Theorem 9. Since can overestimate the effect of rounding to the nearest floatingpoint number by the wobble factor of , error estimates of formulas will be tighter on machines with a small .
When only the order of magnitude of rounding error is of interest, ulps and may be used interchangeably, since they differ by at most a factor of . For example, when a floatingpoint number is in error by n ulps, that means that the number of contaminated digits is log n. If the relative error in a computation is n, then
(3) contaminated digits log_{} n.
Guard Digits
One method of computing the difference between two floatingpoint numbers is to compute the difference exactly and then round it to the nearest floatingpoint number. This is very expensive if the operands differ greatly in size. Assuming p = 3, 2.15 × 10^{12} – 1.25 × 10^{5} would be calculated as
which rounds to 2.15 × 10^{12}. Rather than using all these digits, floatingpoint hardware normally operates on a fixed number of digits. Suppose that the number of digits kept is p, and that when the smaller operand is shifted right, digits are simply discarded (as opposed to rounding). Then 2.15 × 10^{12} – 1.25 × 10^{5} becomes
The answer is exactly the same as if the difference had been computed exactly and then rounded. Take another example: 10.1 – 9.93. This becomes
The correct answer is .17, so the computed difference is off by 30 ulps and is wrong in every digit! How bad can the error be?
Theorem 1
 Using a floatingpoint format with parameters and p, and computing differences using p digits, the relative error of the result can be as large as – 1.
Proof
 A relative error of – 1 in the expression x – y occurs when x = 1.00…0 and y = .…, where = – 1. Here y has p digits (all equal to ). The exact difference is x – y = ^{–}^{p}. However, when computing the answer using only p digits, the rightmost digit of y gets shifted off, and so the computed difference is ^{–}^{p}^{+1}. Thus the error is ^{–}^{p} – ^{–}^{p}^{+1} = ^{–}^{p} ( – 1), and the relative error is ^{–}^{p}( – 1)/^{–}^{p} = – 1. z
When =2, the relative error can be as large as the result, and when =10, it can be 9 times larger. Or to put it another way, when =2, equation (3) shows that the number of contaminated digits is log_{2}(1/) = log_{2}(2^{p}) = p. That is, all of the p digits in the result are wrong! Suppose that one extra digit is added to guard against this situation (a guard digit). That is, the smaller number is truncated to p + 1 digits, and then the result of the subtraction is rounded to p digits. With a guard digit, the previous example becomes
and the answer is exact. With a single guard digit, the relative error of the result may be greater than , as in 110 – 8.59.
This rounds to 102, compared with the correct answer of 101.41, for a relative error of .006, which is greater than = .005. In general, the relative error of the result can be only slightly larger than . More precisely,
Theorem 2
 If x and y are floatingpoint numbers in a format with parameters and p, and if subtraction is done with p + 1 digits (i.e. one guard digit), then the relative rounding error in the result is less than 2.
This theorem will be proven in Rounding Error. Addition is included in the above theorem since x and y can be positive or negative.
Cancellation
The last section can be summarized by saying that without a guard digit, the relative error committed when subtracting two nearby quantities can be very large. In other words, the evaluation of any expression containing a subtraction (or an addition of quantities with opposite signs) could result in a relative error so large that all the digits are meaningless (Theorem 1). When subtracting nearby quantities, the most significant digits in the operands match and cancel each other. There are two kinds of cancellation: catastrophic and benign.
Catastrophic cancellation occurs when the operands are subject to rounding errors. For example in the quadratic formula, the expression b^{2} – 4ac occurs. The quantities b^{2} and 4ac are subject to rounding errors since they are the results of floatingpoint multiplications. Suppose that they are rounded to the nearest floatingpoint number, and so are accurate to within .5 ulp. When they are subtracted, cancellation can cause many of the accurate digits to disappear, leaving behind mainly digits contaminated by rounding error. Hence the difference might have an error of many ulps. For example, consider b = 3.34, a = 1.22, and c = 2.28. The exact value of b^{2} – 4ac is .0292. But b^{2} rounds to 11.2 and 4ac rounds to 11.1, hence the final answer is .1 which is an error by 70 ulps, even though 11.2 – 11.1 is exactly equal to .1^{6}. The subtraction did not introduce any error, but rather exposed the error introduced in the earlier multiplications.
Benign cancellation occurs when subtracting exactly known quantities. If x and y have no rounding error, then by Theorem 2 if the subtraction is done with a guard digit, the difference xy has a very small relative error (less than 2).
A formula that exhibits catastrophic cancellation can sometimes be rearranged to eliminate the problem. Again consider the quadratic formula
(4)
When , then does not involve a cancellation and.
But the other addition (subtraction) in one of the formulas will have a catastrophic cancellation. To avoid this, multiply the numerator and denominator of r_{1} by
(and similarly for r_{2}) to obtain(5)
If and , then computing r_{1} using formula (4) will involve a cancellation. Therefore, use formula (5) for computing r_{1} and (4) for r_{2}. On the other hand, if b < 0, use (4) for computing r_{1} and (5) for r_{2}.The expression x^{2} – y^{2} is another formula that exhibits catastrophic cancellation. It is more accurate to evaluate it as (x – y)(x + y).^{7} Unlike the quadratic formula, this improved form still has a subtraction, but it is a benign cancellation of quantities without rounding error, not a catastrophic one. By Theorem 2, the relative error in x – y is at most 2. The same is true of x + y. Multiplying two quantities with a small relative error results in a product with a small relative error (see the section Rounding Error).
In order to avoid confusion between exact and computed values, the following notation is used. Whereas x – y denotes the exact difference of xand y, x y denotes the computed difference (i.e., with rounding error). Similarly , , and denote computed addition, multiplication, and division, respectively. All caps indicate the computed value of a function, as in
LN(x)
orSQRT(x)
. Lowercase functions and traditional mathematical notation denote their exact values as in ln(x) and .Although (x y) (x y) is an excellent approximation to x^{2} – y^{2}, the floatingpoint numbers x and y might themselves be approximations to some true quantities and . For example, and might be exactly known decimal numbers that cannot be expressed exactly in binary. In this case, even though x y is a good approximation to x – y, it can have a huge relative error compared to the true expression , and so the advantage of (x + y)(x – y) over x^{2} – y^{2} is not as dramatic. Since computing (x + y)(x – y) is about the same amount of work as computing x^{2} – y^{2}, it is clearly the preferred form in this case. In general, however, replacing a catastrophic cancellation by a benign one is not worthwhile if the expense is large, because the input is often (but not always) an approximation. But eliminating a cancellation entirely (as in the quadratic formula) is worthwhile even if the data are not exact. Throughout this paper, it will be assumed that the floatingpoint inputs to an algorithm are exact and that the results are computed as accurately as possible.
The expression x^{2} – y^{2} is more accurate when rewritten as (x – y)(x + y) because a catastrophic cancellation is replaced with a benign one. We next present more interesting examples of formulas exhibiting catastrophic cancellation that can be rewritten to exhibit only benign cancellation.
The area of a triangle can be expressed directly in terms of the lengths of its sides a, b, and c as
(6)
(Suppose the triangle is very flat; that is, a b + c. Then s a, and the term (s – a) in formula (6) subtracts two nearby numbers, one of which may have rounding error. For example, if a = 9.0, b = c = 4.53, the correct value of s is 9.03 and A is 2.342…. Even though the computed value of s (9.05) is in error by only 2 ulps, the computed value of A is 3.04, an error of 70 ulps.There is a way to rewrite formula (6) so that it will return accurate results even for flat triangles [Kahan 1986]. It is
(7)
If a, b, and c do not satisfy a b c, rename them before applying (7). It is straightforward to check that the righthand sides of (6) and (7)are algebraically identical. Using the values of a, b, and c above gives a computed area of 2.35, which is 1 ulp in error and much more accurate than the first formula.Although formula (7) is much more accurate than (6) for this example, it would be nice to know how well (7) performs in general.
Theorem 3
 The rounding error incurred when using (7) to compute the area of a triangle is at most 11, provided that subtraction is performed with a guard digit, e .005, and that square roots are computed to within 1/2 ulp.
The condition that e < .005 is met in virtually every actual floatingpoint system. For example when = 2, p 8 ensures that e < .005, and when = 10, p 3 is enough.
In statements like Theorem 3 that discuss the relative error of an expression, it is understood that the expression is computed using floatingpoint arithmetic. In particular, the relative error is actually of the expression
(8)
SQRT
((a (b c)) (c (a b)) (c (a b)) (a (b c))) 4
Because of the cumbersome nature of (8), in the statement of theorems we will usually say the computed value of E rather than writing out Ewith circle notation.Error bounds are usually too pessimistic. In the numerical example given above, the computed value of (7) is 2.35, compared with a true value of 2.34216 for a relative error of 0.7, which is much less than 11. The main reason for computing error bounds is not to get precise bounds but rather to verify that the formula does not contain numerical problems.
A final example of an expression that can be rewritten to use benign cancellation is (1 + x)^{n}, where . This expression arises in financial calculations. Consider depositing $100 every day into a bank account that earns an annual interest rate of 6%, compounded daily. If n = 365 and i = .06, the amount of money accumulated at the end of one year is
100
dollars. If this is computed using = 2 and p = 24, the result is $37615.45 compared to the exact answer of $37614.05, a discrepancy of $1.40. The reason for the problem is easy to see. The expression 1 + i/n involves adding 1 to .0001643836, so the low order bits of i/n are lost. This rounding error is amplified when 1 + i/n is raised to the nth power.The troublesome expression (1 + i/n)^{n} can be rewritten as e^{nln(1 + }^{i}^{/}^{n}^{)}, where now the problem is to compute ln(1 + x) for small x. One approach is to use the approximation ln(1 + x) x, in which case the payment becomes $37617.26, which is off by $3.21 and even less accurate than the obvious formula. But there is a way to compute ln(1 + x) very accurately, as Theorem 4 shows [HewlettPackard 1982]. This formula yields $37614.07, accurate to within two cents!
Theorem 4 assumes that
LN(x)
approximates ln(x) to within 1/2 ulp. The problem it solves is that when x is small,LN
(1 x) is not close to ln(1 + x) because 1 x has lost the information in the low order bits of x. That is, the computed value of ln(1 + x) is not close to its actual value when .Theorem 4
 the relative error is at most 5 when 0 x < 3/4, provided subtraction is performed with a guard digit, e < 0.1, and ln is computed to within 1/2 ulp.
This formula will work for any value of x but is only interesting for , which is where catastrophic cancellation occurs in the naive formula ln(1 + x). Although the formula may seem mysterious, there is a simple explanation for why it works. Write ln(1 + x) as
.
The left hand factor can be computed exactly, but the right hand factor µ(x) = ln(1 + x)/x will suffer a large rounding error when adding 1 to x. However, µ is almost constant, since ln(1 + x) x. So changing x slightly will not introduce much error. In other words, if , computing will be a good approximation to xµ(x) = ln(1 + x). Is there a value for for which and can be computed accurately? There is; namely = (1 x) 1, because then 1 + is exactly equal to 1 x.The results of this section can be summarized by saying that a guard digit guarantees accuracy when nearby precisely known quantities are subtracted (benign cancellation). Sometimes a formula that gives inaccurate results can be rewritten to have much higher numerical accuracy by using benign cancellation; however, the procedure only works if subtraction is performed using a guard digit. The price of a guard digit is not high, because it merely requires making the adder one bit wider. For a 54 bit double precision adder, the additional cost is less than 2%. For this price, you gain the ability to run many algorithms such as formula (6) for computing the area of a triangle and the expression ln(1 + x). Although most modern computers have a guard digit, there are a few (such as Cray systems) that do not.
Exactly Rounded Operations
When floatingpoint operations are done with a guard digit, they are not as accurate as if they were computed exactly then rounded to the nearest floatingpoint number. Operations performed in this manner will be called exactly rounded.^{8} The example immediately preceding Theorem 2 shows that a single guard digit will not always give exactly rounded results. The previous section gave several examples of algorithms that require a guard digit in order to work properly. This section gives examples of algorithms that require exact rounding.
So far, the definition of rounding has not been given. Rounding is straightforward, with the exception of how to round halfway cases; for example, should 12.5 round to 12 or 13? One school of thought divides the 10 digits in half, letting {0, 1, 2, 3, 4} round down, and {5, 6, 7, 8, 9} round up; thus 12.5 would round to 13. This is how rounding works on Digital Equipment Corporation’s VAX computers. Another school of thought says that since numbers ending in 5 are halfway between two possible roundings, they should round down half the time and round up the other half. One way of obtaining this 50% behavior to require that the rounded result have its least significant digit be even. Thus 12.5 rounds to 12 rather than 13 because 2 is even. Which of these methods is best, round up or round to even? Reiser and Knuth [1975] offer the following reason for preferring round to even.
Theorem 5
 Let x and y be floatingpoint numbers, and define x_{0} = x, x_{1} = (x_{0} y) y, …, x_{n} = (x_{n1} y) y. If and are exactly rounded using round to even, then either x_{n} = x for all n or x_{n} = x_{1} for all n 1. z
To clarify this result, consider = 10, p = 3 and let x = 1.00, y = .555. When rounding up, the sequence becomes
x_{0} y = 1.56, x_{1} = 1.56 .555 = 1.01, x_{1} y = 1.01 .555 = 1.57,
and each successive value of x_{n} increases by .01, until x_{n} = 9.45 (n 845)^{9}. Under round to even, x_{n} is always 1.00. This example suggests that when using the round up rule, computations can gradually drift upward, whereas when using round to even the theorem says this cannot happen. Throughout the rest of this paper, round to even will be used.One application of exact rounding occurs in multiple precision arithmetic. There are two basic approaches to higher precision. One approach represents floatingpoint numbers using a very large significand, which is stored in an array of words, and codes the routines for manipulating these numbers in assembly language. The second approach represents higher precision floatingpoint numbers as an array of ordinary floatingpoint numbers, where adding the elements of the array in infinite precision recovers the high precision floatingpoint number. It is this second approach that will be discussed here. The advantage of using an array of floatingpoint numbers is that it can be coded portably in a high level language, but it requires exactly rounded arithmetic.
The key to multiplication in this system is representing a product xy as a sum, where each summand has the same precision as x and y. This can be done by splitting x and y. Writing x = x_{h} + x_{l} and y = y_{h} + y_{l}, the exact product is
xy = x_{h} y_{h} + x_{h} y_{l} + x_{l} y_{h} + x_{l} y_{l}.
If x and y have p bit significands, the summands will also have p bit significands provided that x_{l}, x_{h}, y_{h}, y_{l} can be represented using [p/2] bits. When p is even, it is easy to find a splitting. The number x_{0}.x_{1} … x_{p – 1} can be written as the sum of x_{0}.x_{1} … x_{p/2 – 1} and 0.0 … 0x_{p/2} … x_{p – 1}. When p is odd, this simple splitting method will not work. An extra bit can, however, be gained by using negative numbers. For example, if = 2, p = 5, and x = .10111, x can be split as x_{h} = .11 and x_{l} = .00001. There is more than one way to split a number. A splitting method that is easy to compute is due to Dekker [1971], but it requires more than a single guard digit.Theorem 6
 Let p be the floatingpoint precision, with the restriction that p is even when > 2, and assume that floatingpoint operations are exactly rounded. Then if k = [p/2] is half the precision (rounded up) and m = ^{k} + 1, x can be split as x = x_{h} + x_{l}, where
x_{h} = (m x) (m x x), x_{l} = x x_{h},
and each x_{i} is representable using [p/2] bits of precision.To see how this theorem works in an example, let = 10, p = 4, b = 3.476, a = 3.463, and c = 3.479. Then b^{2} – ac rounded to the nearest floatingpoint number is .03480, while b b = 12.08, a c = 12.05, and so the computed value of b^{2} – ac is .03. This is an error of 480 ulps. Using Theorem 6 to write b = 3.5 – .024, a = 3.5 – .037, and c = 3.5 – .021, b^{2} becomes 3.5^{2} – 2 × 3.5 × .024 + .024^{2}. Each summand is exact, so b^{2} = 12.25 – .168 + .000576, where the sum is left unevaluated at this point. Similarly, ac = 3.5^{2} – (3.5 × .037 + 3.5 × .021) + .037 × .021 = 12.25 – .2030 +.000777. Finally, subtracting these two series term by term gives an estimate for b^{2} – ac of 0 .0350 .000201 = .03480, which is identical to the exactly rounded result. To show that Theorem 6 really requires exact rounding, consider p = 3, = 2, and x = 7. Then m = 5, mx = 35, and m x = 32. If subtraction is performed with a single guard digit, then (m x) x = 28. Therefore, x_{h} = 4 and x_{l}= 3, hence x_{l} is not representable with [p/2] = 1 bit.
As a final example of exact rounding, consider dividing m by 10. The result is a floatingpoint number that will in general not be equal to m/10. When = 2, multiplying m/10 by 10 will restore m, provided exact rounding is being used. Actually, a more general fact (due to Kahan) is true. The proof is ingenious, but readers not interested in such details can skip ahead to section The IEEE Standard.
Theorem 7
 When = 2, if m and n are integers with m < 2^{p – 1} and n has the special form n = 2^{i} + 2^{j}, then (m n) n = m, provided floatingpoint operations are exactly rounded.
Proof
 Scaling by a power of two is harmless, since it changes only the exponent, not the significand. If q = m/n, then scale n so that 2^{p}^{ – 1} n < 2^{p}and scale m so that 1/2 < q < 1. Thus, 2^{p}^{ – 2} < m < 2^{p}. Since m has p significant bits, it has at most one bit to the right of the binary point. Changing the sign of m is harmless, so assume that q > 0.
 If = m n, to prove the theorem requires showing that
 That is because m has at most 1 bit right of the binary point, so n will round to m. To deal with the halfway case when n – m = 1/4, note that since the initial unscaled m had m < 2^{p}^{ – 1}, its loworder bit was 0, so the loworder bit of the scaled m is also 0. Thus, halfway cases will round to m.
 Suppose that q = .q_{1}q_{2} …, and let = .q_{1}q_{2} … q_{p}1. To estimate n – m, first compute
 – q = N/2^{p}^{ + 1} – m/n,
 where N is an odd integer. Since n = 2^{i} + 2^{j} and 2^{p}^{ – 1} n < 2^{p}, it must be that n = 2^{p}^{ – 1} + 2^{k} for some k p – 2, and thus
 – q 1/(n2^{p}^{ + 1 – }^{k}).
 Assume q < (the case q > is similar).^{10} Then n < m, and
mn = mn = n(q ) = n(q( 2^{p1}))
=(2^{p}^{1}+2^{k})2^{p1}2^{p1+k} =
The theorem holds true for any base , as long as 2^{i} + 2^{j} is replaced by ^{i} + ^{j}. As gets larger, however, denominators of the form ^{i} + ^{j} are farther and farther apart.
We are now in a position to answer the question, Does it matter if the basic arithmetic operations introduce a little more rounding error than necessary? The answer is that it does matter, because accurate basic operations enable us to prove that formulas are “correct” in the sense they have a small relative error. The section Cancellation discussed several algorithms that require guard digits to produce correct results in this sense. If the input to those formulas are numbers representing imprecise measurements, however, the bounds of Theorems 3 and 4 become less interesting. The reason is that the benign cancellation x – y can become catastrophic if x and y are only approximations to some measured quantity. But accurate operations are useful even in the face of inexact data, because they enable us to establish exact relationships like those discussed in Theorems 6 and 7. These are useful even if every floatingpoint variable is only an approximation to some actual value.
The IEEE Standard
There are two different IEEE standards for floatingpoint computation. IEEE 754 is a binary standard that requires = 2, p = 24 for single precision and p = 53 for double precision [IEEE 1987]. It also specifies the precise layout of bits in a single and double precision. IEEE 854 allows either = 2 or = 10 and unlike 754, does not specify how floatingpoint numbers are encoded into bits [Cody et al. 1984]. It does not require a particular value for p, but instead it specifies constraints on the allowable values of p for single and double precision. The term IEEE Standard will be used when discussing properties common to both standards.
This section provides a tour of the IEEE standard. Each subsection discusses one aspect of the standard and why it was included. It is not the purpose of this paper to argue that the IEEE standard is the best possible floatingpoint standard but rather to accept the standard as given and provide an introduction to its use. For full details consult the standards themselves [IEEE 1987; Cody et al. 1984].
Formats and Operations
Base
It is clear why IEEE 854 allows = 10. Base ten is how humans exchange and think about numbers. Using = 10 is especially appropriate for calculators, where the result of each operation is displayed by the calculator in decimal.
There are several reasons why IEEE 854 requires that if the base is not 10, it must be 2. The section Relative Error and Ulps mentioned one reason: the results of error analyses are much tighter when is 2 because a rounding error of .5 ulp wobbles by a factor of when computed as a relative error, and error analyses are almost always simpler when based on relative error. A related reason has to do with the effective precision for large bases. Consider = 16, p = 1 compared to = 2, p = 4. Both systems have 4 bits of significand. Consider the computation of 15/8. When = 2, 15 is represented as 1.111 × 2^{3}, and 15/8 as 1.111 × 2^{0}. So 15/8 is exact. However, when = 16, 15 is represented as F × 16^{0}, where F is the hexadecimal digit for 15. But 15/8 is represented as 1 × 16^{0}, which has only one bit correct. In general, base 16 can lose up to 3 bits, so that a precision of p hexadecimal digits can have an effective precision as low as 4p – 3 rather than 4p binary bits. Since large values of have these problems, why did IBM choose = 16 for its system/370? Only IBM knows for sure, but there are two possible reasons. The first is increased exponent range. Single precision on the system/370 has = 16, p = 6. Hence the significand requires 24 bits. Since this must fit into 32 bits, this leaves 7 bits for the exponent and one for the sign bit. Thus the magnitude of representable numbers ranges from about to about = . To get a similar exponent range when = 2 would require 9 bits of exponent, leaving only 22 bits for the significand. However, it was just pointed out that when = 16, the effective precision can be as low as 4p – 3 = 21 bits. Even worse, when = 2 it is possible to gain an extra bit of precision (as explained later in this section), so the = 2 machine has 23 bits of precision to compare with a range of 21 – 24 bits for the = 16 machine.
Another possible explanation for choosing = 16 has to do with shifting. When adding two floatingpoint numbers, if their exponents are different, one of the significands will have to be shifted to make the radix points line up, slowing down the operation. In the = 16, p = 1 system, all the numbers between 1 and 15 have the same exponent, and so no shifting is required when adding any of the ( ) = 105 possible pairs of distinct numbers from this set. However, in the = 2, p = 4 system, these numbers have exponents ranging from 0 to 3, and shifting is required for 70 of the 105 pairs.
In most modern hardware, the performance gained by avoiding a shift for a subset of operands is negligible, and so the small wobble of = 2 makes it the preferable base. Another advantage of using = 2 is that there is a way to gain an extra bit of significance.^{12} Since floatingpoint numbers are always normalized, the most significant bit of the significand is always 1, and there is no reason to waste a bit of storage representing it. Formats that use this trick are said to have a hidden bit. It was already pointed out in Floatingpoint Formats that this requires a special convention for 0. The method given there was that an exponent of e_{min} – 1 and a significand of all zeros represents not , but rather 0.
IEEE 754 single precision is encoded in 32 bits using 1 bit for the sign, 8 bits for the exponent, and 23 bits for the significand. However, it uses a hidden bit, so the significand is 24 bits (p = 24), even though it is encoded using only 23 bits.
Precision
The IEEE standard defines four different precisions: single, double, singleextended, and doubleextended. In IEEE 754, single and double precision correspond roughly to what most floatingpoint hardware provides. Single precision occupies a single 32 bit word, double precision two consecutive 32 bit words. Extended precision is a format that offers at least a little extra precision and exponent range (TABLE D1).
The IEEE standard only specifies a lower bound on how many extra bits extended precision provides. The minimum allowable doubleextended format is sometimes referred to as 80bit format, even though the table shows it using 79 bits. The reason is that hardware implementations of extended precision normally do not use a hidden bit, and so would use 80 rather than 79 bits.^{13}The standard puts the most emphasis on extended precision, making no recommendation concerning double precision, but strongly recommending that Implementations should support the extended format corresponding to the widest basic format supported, …
One motivation for extended precision comes from calculators, which will often display 10 digits, but use 13 digits internally. By displaying only 10 of the 13 digits, the calculator appears to the user as a “black box” that computes exponentials, cosines, etc. to 10 digits of accuracy. For the calculator to compute functions like exp, log and cos to within 10 digits with reasonable efficiency, it needs a few extra digits to work with. It is not hard to find a simple rational expression that approximates log with an error of 500 units in the last place. Thus computing with 13 digits gives an answer correct to 10 digits. By keeping these extra 3 digits hidden, the calculator presents a simple model to the operator.
Extended precision in the IEEE standard serves a similar function. It enables libraries to efficiently compute quantities to within about .5 ulp in single (or double) precision, giving the user of those libraries a simple model, namely that each primitive operation, be it a simple multiply or an invocation of log, returns a value accurate to within about .5 ulp. However, when using extended precision, it is important to make sure that its use is transparent to the user. For example, on a calculator, if the internal representation of a displayed value is not rounded to the same precision as the display, then the result of further operations will depend on the hidden digits and appear unpredictable to the user.
To illustrate extended precision further, consider the problem of converting between IEEE 754 single precision and decimal. Ideally, single precision numbers will be printed with enough digits so that when the decimal number is read back in, the single precision number can be recovered. It turns out that 9 decimal digits are enough to recover a single precision binary number (see the section Binary to Decimal Conversion). When converting a decimal number back to its unique binary representation, a rounding error as small as 1 ulp is fatal, because it will give the wrong answer. Here is a situation where extended precision is vital for an efficient algorithm. When singleextended is available, a very straightforward method exists for converting a decimal number to a single precision binary one. First read in the 9 decimal digits as an integer N, ignoring the decimal point. From TABLE D1, p 32, and since 10^{9} < 2^{32} 4.3 × 10^{9}, N can be represented exactly in singleextended. Next find the appropriate power 10^{P} necessary to scale N. This will be a combination of the exponent of the decimal number, together with the position of the (up until now) ignored decimal point. Compute 10^{}^{P}^{}. If P 13, then this is also represented exactly, because 10^{13} = 2^{13}5^{13}, and 5^{13} < 2^{32}. Finally multiply (or divide if p < 0) N and 10^{}^{P}^{}. If this last operation is done exactly, then the closest binary number is recovered. The section Binary to Decimal Conversion shows how to do the last multiply (or divide) exactly. Thus for P 13, the use of the singleextended format enables 9digit decimal numbers to be converted to the closest binary number (i.e. exactly rounded). If P > 13, then singleextended is not enough for the above algorithm to always compute the exactly rounded binary equivalent, but Coonen [1984] shows that it is enough to guarantee that the conversion of binary to decimal and back will recover the original binary number.
If double precision is supported, then the algorithm above would be run in double precision rather than singleextended, but to convert double precision to a 17digit decimal number and back would require the doubleextended format.
Exponent
Since the exponent can be positive or negative, some method must be chosen to represent its sign. Two common methods of representing signed numbers are sign/magnitude and two’s complement. Sign/magnitude is the system used for the sign of the significand in the IEEE formats: one bit is used to hold the sign, the rest of the bits represent the magnitude of the number. The two’s complement representation is often used in integer arithmetic. In this scheme, a number in the range [2^{p1}, 2^{p1} – 1] is represented by the smallest nonnegative number that is congruent to it modulo 2^{p}.
The IEEE binary standard does not use either of these methods to represent the exponent, but instead uses a biased representation. In the case of single precision, where the exponent is stored in 8 bits, the bias is 127 (for double precision it is 1023). What this means is that if is the value of the exponent bits interpreted as an unsigned integer, then the exponent of the floatingpoint number is – 127. This is often called the unbiased exponent to distinguish from the biased exponent .
Referring to TABLE D1, single precision has e_{max} = 127 and e_{min} = 126. The reason for having e_{min} < e_{max} is so that the reciprocal of the smallest number will not overflow. Although it is true that the reciprocal of the largest number will underflow, underflow is usually less serious than overflow. The section Base explained that e_{min} – 1 is used for representing 0, and Special Quantities will introduce a use for e_{max} + 1. In IEEE single precision, this means that the biased exponents range between e_{min} – 1 = 127 and e_{max} + 1 = 128, whereas the unbiased exponents range between 0 and 255, which are exactly the nonnegative numbers that can be represented using 8 bits.
Operations
The IEEE standard requires that the result of addition, subtraction, multiplication and division be exactly rounded. That is, the result must be computed exactly and then rounded to the nearest floatingpoint number (using round to even). The section Guard Digits pointed out that computing the exact difference or sum of two floatingpoint numbers can be very expensive when their exponents are substantially different. That section introduced guard digits, which provide a practical way of computing differences while guaranteeing that the relative error is small. However, computing with a single guard digit will not always give the same answer as computing the exact result and then rounding. By introducing a second guard digit and a third sticky bit, differences can be computed at only a little more cost than with a single guard digit, but the result is the same as if the difference were computed exactly and then rounded [Goldberg 1990]. Thus the standard can be implemented efficiently.
One reason for completely specifying the results of arithmetic operations is to improve the portability of software. When a program is moved between two machines and both support IEEE arithmetic, then if any intermediate result differs, it must be because of software bugs, not from differences in arithmetic. Another advantage of precise specification is that it makes it easier to reason about floatingpoint. Proofs about floatingpoint are hard enough, without having to deal with multiple cases arising from multiple kinds of arithmetic. Just as integer programs can be proven to be correct, so can floatingpoint programs, although what is proven in that case is that the rounding error of the result satisfies certain bounds. Theorem 4 is an example of such a proof. These proofs are made much easier when the operations being reasoned about are precisely specified. Once an algorithm is proven to be correct for IEEE arithmetic, it will work correctly on any machine supporting the IEEE standard.
Brown [1981] has proposed axioms for floatingpoint that include most of the existing floatingpoint hardware. However, proofs in this system cannot verify the algorithms of sections Cancellation and Exactly Rounded Operations, which require features not present on all hardware. Furthermore, Brown’s axioms are more complex than simply defining operations to be performed exactly and then rounded. Thus proving theorems from Brown’s axioms is usually more difficult than proving them assuming operations are exactly rounded.
There is not complete agreement on what operations a floatingpoint standard should cover. In addition to the basic operations +, , × and /, the IEEE standard also specifies that square root, remainder, and conversion between integer and floatingpoint be correctly rounded. It also requires that conversion between internal formats and decimal be correctly rounded (except for very large numbers). Kulisch and Miranker [1986] have proposed adding inner product to the list of operations that are precisely specified. They note that when inner products are computed in IEEE arithmetic, the final answer can be quite wrong. For example sums are a special case of inner products, and the sum ((2 ×10^{30} + 10^{30}) – 10^{30}) – 10^{30} is exactly equal to 10^{30}, but on a machine with IEEE arithmetic the computed result will be 10^{30}. It is possible to compute inner products to within 1 ulp with less hardware than it takes to implement a fast multiplier [Kirchner and Kulish 1987].^{14} ^{15}
All the operations mentioned in the standard are required to be exactly rounded except conversion between decimal and binary. The reason is that efficient algorithms for exactly rounding all the operations are known, except conversion. For conversion, the best known efficient algorithms produce results that are slightly worse than exactly rounded ones [Coonen 1984].
The IEEE standard does not require transcendental functions to be exactly rounded because of the table maker’s dilemma. To illustrate, suppose you are making a table of the exponential function to 4 places. Then exp(1.626) = 5.0835. Should this be rounded to 5.083 or 5.084? If exp(1.626) is computed more carefully, it becomes 5.08350. And then 5.083500. And then 5.0835000. Since exp is transcendental, this could go on arbitrarily long before distinguishing whether exp(1.626) is 5.083500…0ddd or 5.0834999…9ddd. Thus it is not practical to specify that the precision of transcendental functions be the same as if they were computed to infinite precision and then rounded. Another approach would be to specify transcendental functions algorithmically. But there does not appear to be a single algorithm that works well across all hardware architectures. Rational approximation, CORDIC,^{16} and large tables are three different techniques that are used for computing transcendentals on contemporary machines. Each is appropriate for a different class of hardware, and at present no single algorithm works acceptably over the wide range of current hardware.
Special Quantities
On some floatingpoint hardware every bit pattern represents a valid floatingpoint number. The IBM System/370 is an example of this. On the other hand, the VAX^{TM} reserves some bit patterns to represent special numbers called reserved operands. This idea goes back to the CDC 6600, which had bit patterns for the special quantities
INDEFINITE
andINFINITY
.The IEEE standard continues in this tradition and has NaNs (Not a Number) and infinities. Without any special quantities, there is no good way to handle exceptional situations like taking the square root of a negative number, other than aborting computation. Under IBM System/370 FORTRAN, the default action in response to computing the square root of a negative number like 4 results in the printing of an error message. Since every bit pattern represents a valid number, the return value of square root must be some floatingpoint number. In the case of System/370 FORTRAN, is returned. In IEEE arithmetic, a NaN is returned in this situation.
The IEEE standard specifies the following special values (see TABLE D2): ± 0, denormalized numbers, ± and NaNs (there is more than one NaN, as explained in the next section). These special values are all encoded with exponents of either e_{max} + 1 or e_{min} – 1 (it was already pointed out that 0 has an exponent of e_{min} – 1).
TABLE D2 IEEE 754 Special Values Exponent Fraction Represents e = e_{min} – 1 f = 0 ±0 e = e_{min} – 1 f 0 e_{min} e e_{max} — 1.f × 2^{e} e = e_{max} + 1 f = 0 ± e = e_{max} + 1 f 0 NaN
NaNs
Traditionally, the computation of 0/0 or has been treated as an unrecoverable error which causes a computation to halt. However, there are examples where it makes sense for a computation to continue in such a situation. Consider a subroutine that finds the zeros of a function f, say
zero(f)
. Traditionally, zero finders require the user to input an interval [a, b] on which the function is defined and over which the zero finder will search. That is, the subroutine is called aszero(f
,a
,b)
. A more useful zero finder would not require the user to input this extra information. This more general zero finder is especially appropriate for calculators, where it is natural to simply key in a function, and awkward to then have to specify the domain. However, it is easy to see why most zero finders require a domain. The zero finder does its work by probing the functionf
at various values. If it probed for a value outside the domain off
, the code forf
might well compute 0/0 or , and the computation would halt, unnecessarily aborting the zero finding process.This problem can be avoided by introducing a special value called NaN, and specifying that the computation of expressions like 0/0 and produce NaN, rather than halting. A list of some of the situations that can cause a NaN are given in TABLE D3. Then when
zero(f)
probes outside the domain off
, the code forf
will return NaN, and the zero finder can continue. That is,zero(f)
is not “punished” for making an incorrect guess. With this example in mind, it is easy to see what the result of combining a NaN with an ordinary floatingpoint number should be. Suppose that the final statement off
isreturn(b +
sqrt(d))/(2*a)
. If d < 0, thenf
should return a NaN. Since d < 0,sqrt(d)
is a NaN, andb + sqrt(d)
will be a NaN, if the sum of a NaN and any other number is a NaN. Similarly if one operand of a division operation is a NaN, the quotient should be a NaN. In general, whenever a NaN participates in a floatingpoint operation, the result is another NaN.
TABLE D3 Operations That Produce a NaN Operation NaN
Produced By+ + ( ) × 0 × / 0/0, / REM
x REM
0,REM
y(when x < 0)
Another approach to writing a zero solver that doesn’t require the user to input a domain is to use signals. The zerofinder could install a signal handler for floatingpoint exceptions. Then iff
was evaluated outside its domain and raised an exception, control would be returned to the zero solver. The problem with this approach is that every language has a different method of handling signals (if it has a method at all), and so it has no hope of portability.In IEEE 754, NaNs are often represented as floatingpoint numbers with the exponent e_{max} + 1 and nonzero significands. Implementations are free to put systemdependent information into the significand. Thus there is not a unique NaN, but rather a whole family of NaNs. When a NaN and an ordinary floatingpoint number are combined, the result should be the same as the NaN operand. Thus if the result of a long computation is a NaN, the systemdependent information in the significand will be the information that was generated when the first NaN in the computation was generated. Actually, there is a caveat to the last statement. If both operands are NaNs, then the result will be one of those NaNs, but it might not be the NaN that was generated first.
Infinity
Just as NaNs provide a way to continue a computation when expressions like 0/0 or are encountered, infinities provide a way to continue when an overflow occurs. This is much safer than simply returning the largest representable number. As an example, consider computing , when = 10, p = 3, and e_{max} = 98. If x = 3 × 10^{70} and y = 4 × 10^{70}, then x^{2} will overflow, and be replaced by 9.99 × 10^{98}. Similarly y^{2}, and x^{2} + y^{2} will each overflow in turn, and be replaced by 9.99 × 10^{98}. So the final result will be , which is drastically wrong: the correct answer is 5 × 10^{70}. In IEEE arithmetic, the result of x^{2} is , as is y^{2}, x^{2} + y^{2} and . So the final result is , which is safer than returning an ordinary floatingpoint number that is nowhere near the correct answer.^{17}
The division of 0 by 0 results in a NaN. A nonzero number divided by 0, however, returns infinity: 1/0 = , 1/0 = –. The reason for the distinction is this: if f(x) 0 and g(x) 0 as x approaches some limit, then f(x)/g(x) could have any value. For example, when f(x) = sin x and g(x) = x, then f(x)/g(x) 1 as x 0. But when f(x) = 1 – cos x, f(x)/g(x) 0. When thinking of 0/0 as the limiting situation of a quotient of two very small numbers, 0/0 could represent anything. Thus in the IEEE standard, 0/0 results in a NaN. But when c > 0, f(x) c, and g(x)0, then f(x)/g(x) ±, for any analytic functions f and g. If g(x) < 0 for small x, then f(x)/g(x) –, otherwise the limit is +. So the IEEE standard defines c/0 = ±, as long as c 0. The sign of depends on the signs of c and 0 in the usual way, so that 10/0 = –, and 10/0 = +. You can distinguish between getting because of overflow and getting because of division by zero by checking the status flags (which will be discussed in detail in section Flags). The overflow flag will be set in the first case, the division by zero flag in the second.
The rule for determining the result of an operation that has infinity as an operand is simple: replace infinity with a finite number x and take the limit as x . Thus 3/ = 0, because
.
Similarly, 4 – = –, and = . When the limit doesn’t exist, the result is a NaN, so / will be a NaN (TABLE D3 has additional examples). This agrees with the reasoning used to conclude that 0/0 should be a NaN.When a subexpression evaluates to a NaN, the value of the entire expression is also a NaN. In the case of ± however, the value of the expression might be an ordinary floatingpoint number because of rules like 1/ = 0. Here is a practical example that makes use of the rules for infinity arithmetic. Consider computing the function x/(x^{2} + 1). This is a bad formula, because not only will it overflow when x is larger than , but infinity arithmetic will give the wrong answer because it will yield 0, rather than a number near 1/x. However, x/(x^{2} + 1) can be rewritten as 1/(x + x^{1}). This improved expression will not overflow prematurely and because of infinity arithmetic will have the correct value when x = 0: 1/(0 + 0^{1}) = 1/(0 + ) = 1/ = 0. Without infinity arithmetic, the expression 1/(x + x^{1}) requires a test for x = 0, which not only adds extra instructions, but may also disrupt a pipeline. This example illustrates a general fact, namely that infinity arithmetic often avoids the need for special case checking; however, formulas need to be carefully inspected to make sure they do not have spurious behavior at infinity (as x/(x^{2} + 1) did).
Signed Zero
Zero is represented by the exponent e_{min} – 1 and a zero significand. Since the sign bit can take on two different values, there are two zeros, +0 and 0. If a distinction were made when comparing +0 and 0, simple tests like
if
(x
=
0)
would have very unpredictable behavior, depending on the sign ofx
. Thus the IEEE standard defines comparison so that +0 = 0, rather than 0 < +0. Although it would be possible always to ignore the sign of zero, the IEEE standard does not do so. When a multiplication or division involves a signed zero, the usual sign rules apply in computing the sign of the answer. Thus 3·(+0) = +0, and +0/3 = 0. If zero did not have a sign, then the relation 1/(1/x) = x would fail to hold when x = ±. The reason is that 1/ and 1/+ both result in 0, and 1/0 results in +, the sign information having been lost. One way to restore the identity 1/(1/x) = x is to only have one kind of infinity, however that would result in the disastrous consequence of losing the sign of an overflowed quantity.Another example of the use of signed zero concerns underflow and functions that have a discontinuity at 0, such as log. In IEEE arithmetic, it is natural to define log 0 = – and log x to be a NaN when x < 0. Suppose that x represents a small negative number that has underflowed to zero. Thanks to signed zero, x will be negative, so log can return a NaN. However, if there were no signed zero, the log function could not distinguish an underflowed negative number from 0, and would therefore have to return –. Another example of a function with a discontinuity at zero is the signum function, which returns the sign of a number.
Probably the most interesting use of signed zero occurs in complex arithmetic. To take a simple example, consider the equation . This is certainly true when z 0. If z = 1, the obvious computation gives and . Thus, ! The problem can be traced to the fact that square root is multivalued, and there is no way to select the values so that it is continuous in the entire complex plane. However, square root is continuous if a branch cut consisting of all negative real numbers is excluded from consideration. This leaves the problem of what to do for the negative real numbers, which are of the form –x + i0, where x > 0. Signed zero provides a perfect way to resolve this problem. Numbers of the form x + i(+0) have one sign and numbers of the form x + i(0) on the other side of the branch cut have the other sign . In fact, the natural formulas for computing will give these results.
Back to . If z =1 = 1 + i0, then
1/z = 1/(1 + i0) = [(1 i0)]/[(1 + i0)(1 – i0)] = (1 — i0)/((1)^{2} – 0^{2}) = 1 + i(0),
and so , while . Thus IEEE arithmetic preserves this identity for all z. Some more sophisticated examples are given by Kahan [1987]. Although distinguishing between +0 and 0 has advantages, it can occasionally be confusing. For example, signed zero destroys the relation x = y 1/x = 1/y, which is false when x = +0 and y = 0. However, the IEEE committee decided that the advantages of utilizing the sign of zero outweighed the disadvantages.Denormalized Numbers
Consider normalized floatingpoint numbers with = 10, p = 3, and e_{min} = 98. The numbers x = 6.87 × 10^{97} and y = 6.81 × 10^{97} appear to be perfectly ordinary floatingpoint numbers, which are more than a factor of 10 larger than the smallest floatingpoint number 1.00 × 10^{98}. They have a strange property, however: x y = 0 even though x y! The reason is that x – y = .06 × 10^{ 97} = 6.0 × 10^{99} is too small to be represented as a normalized number, and so must be flushed to zero. How important is it to preserve the property
(10) x = y x – y = 0 ?
It’s very easy to imagine writing the code fragment,if
(x
y)
then
z
=
1/(xy)
, and much later having a program fail due to a spurious division by zero. Tracking down bugs like this is frustrating and time consuming. On a more philosophical level, computer science textbooks often point out that even though it is currently impractical to prove large programs correct, designing programs with the idea of proving them often results in better code. For example, introducing invariants is quite useful, even if they aren’t going to be used as part of a proof. Floatingpoint code is just like any other code: it helps to have provable facts on which to depend. For example, when analyzing formula (6), it was very helpful to know that x/2 < y < 2x x y = x – y. Similarly, knowing that (10) is true makes writing reliable floatingpoint code easier. If it is only true for most numbers, it cannot be used to prove anything.The IEEE standard uses denormalized^{18} numbers, which guarantee (10), as well as other useful relations. They are the most controversial part of the standard and probably accounted for the long delay in getting 754 approved. Most high performance hardware that claims to be IEEE compatible does not support denormalized numbers directly, but rather traps when consuming or producing denormals, and leaves it to software to simulate the IEEE standard.^{19} The idea behind denormalized numbers goes back to Goldberg [1967] and is very simple. When the exponent is e_{min}, the significand does not have to be normalized, so that when = 10, p = 3 and e_{min} = 98, 1.00 × 10^{98} is no longer the smallest floatingpoint number, because 0.98 × 10^{98} is also a floatingpoint number.
There is a small snag when = 2 and a hidden bit is being used, since a number with an exponent of e_{min} will always have a significand greater than or equal to 1.0 because of the implicit leading bit. The solution is similar to that used to represent 0, and is summarized in TABLE D2. The exponent e_{min} is used to represent denormals. More formally, if the bits in the significand field are b_{1}, b_{2}, …, b_{p 1}, and the value of the exponent is e, then when e > e_{min} – 1, the number being represented is 1.b_{1}b_{2}…b_{p – 1} × 2^{e} whereas when e = e_{min} – 1, the number being represented is 0.b_{1}b_{2}…b_{p – 1} × 2^{e}^{ + 1}. The +1 in the exponent is needed because denormals have an exponent of e_{min}, not e_{min} – 1.
Recall the example of = 10, p = 3, e_{min} = 98, x = 6.87 × 10^{97} and y = 6.81 × 10^{97} presented at the beginning of this section. With denormals, x – y does not flush to zero but is instead represented by the denormalized number .6 × 10^{98}. This behavior is called gradual underflow. It is easy to verify that (10) always holds when using gradual underflow.
FIGURE D2 Flush To Zero Compared With Gradual Underflow
FIGURE D2 illustrates denormalized numbers. The top number line in the figure shows normalized floatingpoint numbers. Notice the gap between 0 and the smallest normalized number . If the result of a floatingpoint calculation falls into this gulf, it is flushed to zero. The bottom number line shows what happens when denormals are added to the set of floatingpoint numbers. The “gulf” is filled in, and when the result of a calculation is less than , it is represented by the nearest denormal. When denormalized numbers are added to the number line, the spacing between adjacent floatingpoint numbers varies in a regular way: adjacent spacings are either the same length or differ by a factor of . Without denormals, the
spacing abruptly changes from to , which is a factor of , rather than the orderly change by a factor of . Because of this, many algorithms that can have large relative error for normalized numbers close to the underflow threshold are wellbehaved in this range when gradual underflow is used.Without gradual underflow, the simple expression x – y can have a very large relative error for normalized inputs, as was seen above for x = 6.87 × 10^{97} and y = 6.81 × 10^{97}. Large relative errors can happen even without cancellation, as the following example shows [Demmel 1984]. Consider dividing two complex numbers, a + ib and c + id. The obvious formula
· i
suffers from the problem that if either component of the denominator c + id is larger than , the formula will overflow, even though the final result may be well within range. A better method of computing the quotients is to use Smith’s formula:Applying Smith’s formula to (2 · 10^{98} + i10^{98})/(4 · 10^{98} + i(2 · 10^{98})) gives the correct answer of 0.5 with gradual underflow. It yields 0.4 with flush to zero, an error of 100 ulps. It is typical for denormalized numbers to guarantee error bounds for arguments all the way down to 1.0 x .
Exceptions, Flags and Trap Handlers
When an exceptional condition like division by zero or overflow occurs in IEEE arithmetic, the default is to deliver a result and continue. Typical of the default results are NaN for 0/0 and , and for 1/0 and overflow. The preceding sections gave examples where proceeding from an exception with these default values was the reasonable thing to do. When any exception occurs, a status flag is also set. Implementations of the IEEE standard are required to provide users with a way to read and write the status flags. The flags are “sticky” in that once set, they remain set until explicitly cleared. Testing the flags is the only way to distinguish 1/0, which is a genuine infinity from an overflow.
Sometimes continuing execution in the face of exception conditions is not appropriate. The section Infinity gave the example of x/(x^{2} + 1). When x > , the denominator is infinite, resulting in a final answer of 0, which is totally wrong. Although for this formula the problem can be solved by rewriting it as 1/(x + x^{1}), rewriting may not always solve the problem. The IEEE standard strongly recommends that implementations allow trap handlers to be installed. Then when an exception occurs, the trap handler is called instead of setting the flag. The value returned by the trap handler will be used as the result of the operation. It is the responsibility of the trap handler to either clear or set the status flag; otherwise, the value of the flag is allowed to be undefined.
The IEEE standard divides exceptions into 5 classes: overflow, underflow, division by zero, invalid operation and inexact. There is a separate status flag for each class of exception. The meaning of the first three exceptions is selfevident. Invalid operation covers the situations listed in TABLE D3, and any comparison that involves a NaN. The default result of an operation that causes an invalid exception is to return a NaN, but the converse is not true. When one of the operands to an operation is a NaN, the result is a NaN but no invalid exception is raised unless the operation also satisfies one of the conditions in TABLE D3.^{20}
*x is the exact result of the operation, = 192 for single precision, 1536 for double, and x_{max} = 1.11 …11 × .The inexact exception is raised when the result of a floatingpoint operation is not exact. In the = 10, p = 3 system, 3.5 4.2 = 14.7 is exact, but 3.5 4.3 = 15.0 is not exact (since 3.5 · 4.3 = 15.05), and raises an inexact exception. Binary to Decimal Conversion discusses an algorithm that uses the inexact exception. A summary of the behavior of all five exceptions is given in TABLE D4.
There is an implementation issue connected with the fact that the inexact exception is raised so often. If floatingpoint hardware does not have flags of its own, but instead interrupts the operating system to signal a floatingpoint exception, the cost of inexact exceptions could be prohibitive. This cost can be avoided by having the status flags maintained by software. The first time an exception is raised, set the software flag for the appropriate class, and tell the floatingpoint hardware to mask off that class of exceptions. Then all further exceptions will run without interrupting the operating system. When a user resets that status flag, the hardware mask is reenabled.
Trap Handlers
One obvious use for trap handlers is for backward compatibility. Old codes that expect to be aborted when exceptions occur can install a trap handler that aborts the process. This is especially useful for codes with a loop like
do
S
until
(x
>=
100)
. Since comparing a NaN to a number with <, , >, , or = (but not ) always returns false, this code will go into an infinite loop ifx
ever becomes a NaN.There is a more interesting use for trap handlers that comes up when computing products such as that could potentially overflow. One solution is to use logarithms, and compute exp instead. The problem with this approach is that it is less accurate, and that it costs more than the simple expression , even if there is no overflow. There is another solution using trap handlers called over/underflow counting that avoids both of these problems [Sterbenz 1974].
The idea is as follows. There is a global counter initialized to zero. Whenever the partial product overflows for some k, the trap handler increments the counter by one and returns the overflowed quantity with the exponent wrapped around. In IEEE 754 single precision, e_{max} = 127, so if p_{k} = 1.45 × 2^{130}, it will overflow and cause the trap handler to be called, which will wrap the exponent back into range, changing p_{k}to 1.45 × 2^{62} (see below). Similarly, if p_{k} underflows, the counter would be decremented, and negative exponent would get wrapped around into a positive one. When all the multiplications are done, if the counter is zero then the final product is p_{n}. If the counter is positive, the product overflowed, if the counter is negative, it underflowed. If none of the partial products are out of range, the trap handler is never called and the computation incurs no extra cost. Even if there are over/underflows, the calculation is more accurate than if it had been computed with logarithms, because each p_{k} was computed from p_{k – 1} using a full precision multiply. Barnett [1987] discusses a formula where the full accuracy of over/underflow counting turned up an error in earlier tables of that formula.
IEEE 754 specifies that when an overflow or underflow trap handler is called, it is passed the wrappedaround result as an argument. The definition of wrappedaround for overflow is that the result is computed as if to infinite precision, then divided by 2, and then rounded to the relevant precision. For underflow, the result is multiplied by 2. The exponent is 192 for single precision and 1536 for double precision. This is why 1.45 x 2^{130} was transformed into 1.45 × 2^{62} in the example above.
Rounding Modes
In the IEEE standard, rounding occurs whenever an operation has a result that is not exact, since (with the exception of binary decimal conversion) each operation is computed exactly and then rounded. By default, rounding means round toward nearest. The standard requires that three other rounding modes be provided, namely round toward 0, round toward +, and round toward –. When used with the convert to integer operation, round toward – causes the convert to become the floor function, while round toward + is ceiling. The rounding mode affects overflow, because when round toward 0 or round toward – is in effect, an overflow of positive magnitude causes the default result to be the largest representable number, not +. Similarly, overflows of negative magnitude will produce the largest negative number when round toward + or round toward 0 is in effect.
One application of rounding modes occurs in interval arithmetic (another is mentioned in Binary to Decimal Conversion). When using interval arithmetic, the sum of two numbers x and y is an interval , where is x y rounded toward –, and is x y rounded toward +. The exact result of the addition is contained within the interval . Without rounding modes, interval arithmetic is usually implemented by computing and , where is machine epsilon.^{21} This results in overestimates for the size of the intervals. Since the result of an operation in interval arithmetic is an interval, in general the input to an operation will also be an interval. If two intervals , and , are added, the result is , where is with the rounding mode set to round toward –, and is with the rounding mode set to round toward +.
When a floatingpoint calculation is performed using interval arithmetic, the final answer is an interval that contains the exact result of the calculation. This is not very helpful if the interval turns out to be large (as it often does), since the correct answer could be anywhere in that interval. Interval arithmetic makes more sense when used in conjunction with a multiple precision floatingpoint package. The calculation is first performed with some precision p. If interval arithmetic suggests that the final answer may be inaccurate, the computation is redone with higher and higher precisions until the final interval is a reasonable size.
Flags
The IEEE standard has a number of flags and modes. As discussed above, there is one status flag for each of the five exceptions: underflow, overflow, division by zero, invalid operation and inexact. There are four rounding modes: round toward nearest, round toward +, round toward 0, and round toward –. It is strongly recommended that there be an enable mode bit for each of the five exceptions. This section gives some simple examples of how these modes and flags can be put to good use. A more sophisticated example is discussed in the section Binary to Decimal Conversion.
Consider writing a subroutine to compute x^{n}, where n is an integer. When n > 0, a simple routine like
PositivePower(x,n) {while (n is even) {x = x*xn = n/2}u = xwhile (true) {n = n/2if (n==0) return ux = x*xif (n is odd) u = u*x}
If n < 0, then a more accurate way to compute x^{n} is not to callPositivePower(1/x,
n)
but rather1/PositivePower(x,
n)
, because the first expression multiplies n quantities each of which have a rounding error from the division (i.e., 1/x). In the second expression these are exact (i.e., x), and the final division commits just one additional rounding error. Unfortunately, these is a slight snag in this strategy. IfPositivePower(x,
n)
underflows, then either the underflow trap handler will be called, or else the underflow status flag will be set. This is incorrect, because if x^{–}^{n}underflows, then x^{n} will either overflow or be in range.^{22} But since the IEEE standard gives the user access to all the flags, the subroutine can easily correct for this. It simply turns off the overflow and underflow trap enable bits and saves the overflow and underflow status bits. It then computes1/PositivePower(x,
n)
. If neither the overflow nor underflow status bit is set, it restores them together with the trap enable bits. If one of the status bits is set, it restores the flags and redoes the calculation usingPositivePower(1/x,
n)
, which causes the correct exceptions to occur.Another example of the use of flags occurs when computing arccos via the formula
arccos x = 2 arctan .
If arctan() evaluates to /2, then arccos(1) will correctly evaluate to 2·arctan() =, because of infinity arithmetic. However, there is a small snag, because the computation of (1 – x)/(1 + x) will cause the divide by zero exception flag to be set, even though arccos(1) is not exceptional. The solution to this problem is straightforward. Simply save the value of the divide by zero flag before computing arccos, and then restore its old value after the computation.Systems Aspects
The design of almost every aspect of a computer system requires knowledge about floatingpoint. Computer architectures usually have floatingpoint instructions, compilers must generate those floatingpoint instructions, and the operating system must decide what to do when exception conditions are raised for those floatingpoint instructions. Computer system designers rarely get guidance from numerical analysis texts, which are typically aimed at users and writers of software, not at computer designers. As an example of how plausible design decisions can lead to unexpected behavior, consider the following BASIC program.
q = 3.0/7.0if q = 3.0/7.0 then print "Equal":else print "Not Equal"
When compiled and run using Borland’s Turbo Basic on an IBM PC, the program printsNot
Equal
! This example will be analyzed in the next sectionIncidentally, some people think that the solution to such anomalies is never to compare floatingpoint numbers for equality, but instead to consider them equal if they are within some error bound E. This is hardly a cureall because it raises as many questions as it answers. What should the value of E be? If x < 0 and y > 0 are within E, should they really be considered to be equal, even though they have different signs? Furthermore, the relation defined by this rule, a ~ b a – b < E, is not an equivalence relation because a ~ b and b ~ c does not imply that a~ c.
Instruction Sets
It is quite common for an algorithm to require a short burst of higher precision in order to produce accurate results. One example occurs in the quadratic formula ( )/2a. As discussed in the section Proof of Theorem 4, when b^{2} 4ac, rounding error can contaminate up to half the digits in the roots computed with the quadratic formula. By performing the subcalculation of b^{2} – 4ac in double precision, half the double precision bits of the root are lost, which means that all the single precision bits are preserved.
The computation of b^{2} – 4ac in double precision when each of the quantities a, b, and c are in single precision is easy if there is a multiplication instruction that takes two single precision numbers and produces a double precision result. In order to produce the exactly rounded product of two pdigit numbers, a multiplier needs to generate the entire 2p bits of product, although it may throw bits away as it proceeds. Thus, hardware to compute a double precision product from single precision operands will normally be only a little more expensive than a single precision multiplier, and much cheaper than a double precision multiplier. Despite this, modern instruction sets tend to provide only instructions that produce a result of the same precision as the operands.^{23}
If an instruction that combines two single precision operands to produce a double precision product was only useful for the quadratic formula, it wouldn’t be worth adding to an instruction set. However, this instruction has many other uses. Consider the problem of solving a system of linear equations,
a_{11}x_{1} + a_{12}x_{2} +_{ }· · · + a_{1n}x_{n}= b_{1}
a_{21}x_{1} + a_{22}x_{2} +_{ }· · · + a_{2n}x_{n}= b_{2}
· · ·
a_{n1}x_{1} + a_{n2}x_{2} +_{ }· · ·+ a_{nn}x_{n}= b_{n}
which can be written in matrix form as Ax = b, where
Suppose that a solution x^{(1)} is computed by some method, perhaps Gaussian elimination. There is a simple way to improve the accuracy of the result called iterative improvement. First compute(12) = Ax^{(1)} – b
and then solve the system(13) Ay =
Note that if x^{(1)} is an exact solution, then is the zero vector, as is y. In general, the computation of and y will incur rounding error, so Ay Ax^{(1)} – b = A(x^{(1)} – x), where x is the (unknown) true solution. Then y x^{(1)} – x, so an improved estimate for the solution is(14) x^{(2)} = x^{(1)} – y
The three steps (12), (13), and (14) can be repeated, replacing x^{(1)} with x^{(2)}, and x^{(2)} with x^{(3)}. This argument that x^{(}^{i}^{ + 1)} is more accurate than x^{(}^{i}^{)} is only informal. For more information, see [Golub and Van Loan 1989].When performing iterative improvement, is a vector whose elements are the difference of nearby inexact floatingpoint numbers, and so can suffer from catastrophic cancellation. Thus iterative improvement is not very useful unless = Ax^{(1)} – b is computed in double precision. Once again, this is a case of computing the product of two single precision numbers (A and x^{(1)}), where the full double precision result is needed.
To summarize, instructions that multiply two floatingpoint numbers and return a product with twice the precision of the operands make a useful addition to a floatingpoint instruction set. Some of the implications of this for compilers are discussed in the next section.
Languages and Compilers
The interaction of compilers and floatingpoint is discussed in Farnum [1988], and much of the discussion in this section is taken from that paper.
Ambiguity
Ideally, a language definition should define the semantics of the language precisely enough to prove statements about programs. While this is usually true for the integer part of a language, language definitions often have a large grey area when it comes to floatingpoint. Perhaps this is due to the fact that many language designers believe that nothing can be proven about floatingpoint, since it entails rounding error. If so, the previous sections have demonstrated the fallacy in this reasoning. This section discusses some common grey areas in language definitions, including suggestions about how to deal with them.
Remarkably enough, some languages don’t clearly specify that if
x
is a floatingpoint variable (with say a value of3.0/10.0
), then every occurrence of (say)10.0*x
must have the same value. For example Ada, which is based on Brown’s model, seems to imply that floatingpoint arithmetic only has to satisfy Brown’s axioms, and thus expressions can have one of many possible values. Thinking about floatingpoint in this fuzzy way stands in sharp contrast to the IEEE model, where the result of each floatingpoint operation is precisely defined. In the IEEE model, we can prove that(3.0/10.0)*10.0
evaluates to3
(Theorem 7). In Brown’s model, we cannot.Another ambiguity in most language definitions concerns what happens on overflow, underflow and other exceptions. The IEEE standard precisely specifies the behavior of exceptions, and so languages that use the standard as a model can avoid any ambiguity on this point.
Another grey area concerns the interpretation of parentheses. Due to roundoff errors, the associative laws of algebra do not necessarily hold for floatingpoint numbers. For example, the expression
(x+y)+z
has a totally different answer thanx+(y+z)
when x = 10^{30}, y = 10^{30} and z = 1 (it is 1 in the former case, 0 in the latter). The importance of preserving parentheses cannot be overemphasized. The algorithms presented in theorems 3, 4 and 6 all depend on it. For example, in Theorem 6, the formula x_{h} = mx – (mx – x) would reduce to x_{h} = x if it weren’t for parentheses, thereby destroying the entire algorithm. A language definition that does not require parentheses to be honored is useless for floatingpoint calculations.Subexpression evaluation is imprecisely defined in many languages. Suppose that
ds
is double precision, butx
andy
are single precision. Then in the expressionds
+
x*y
is the product performed in single or double precision? Another example: inx
+
m/n
wherem
andn
are integers, is the division an integer operation or a floatingpoint one? There are two ways to deal with this problem, neither of which is completely satisfactory. The first is to require that all variables in an expression have the same type. This is the simplest solution, but has some drawbacks. First of all, languages like Pascal that have subrange types allow mixing subrange variables with integer variables, so it is somewhat bizarre to prohibit mixing single and double precision variables. Another problem concerns constants. In the expression0.1*x
, most languages interpret 0.1 to be a single precision constant. Now suppose the programmer decides to change the declaration of all the floatingpoint variables from single to double precision. If 0.1 is still treated as a single precision constant, then there will be a compile time error. The programmer will have to hunt down and change every floatingpoint constant.The second approach is to allow mixed expressions, in which case rules for subexpression evaluation must be provided. There are a number of guiding examples. The original definition of C required that every floatingpoint expression be computed in double precision [Kernighan and Ritchie 1978]. This leads to anomalies like the example at the beginning of this section. The expression
3.0/7.0
is computed in double precision, but ifq
is a singleprecision variable, the quotient is rounded to single precision for storage. Since 3/7 is a repeating binary fraction, its computed value in double precision is different from its stored value in single precision. Thus the comparison q = 3/7 fails. This suggests that computing every expression in the highest precision available is not a good rule.Another guiding example is inner products. If the inner product has thousands of terms, the rounding error in the sum can become substantial. One way to reduce this rounding error is to accumulate the sums in double precision (this will be discussed in more detail in the section Optimizers). If
d
is a double precision variable, andx[]
andy[]
are single precision arrays, then the inner product loop will look liked
=
d
+
x[i]*y[i]
. If the multiplication is done in single precision, than much of the advantage of double precision accumulation is lost, because the product is truncated to single precision just before being added to a double precision variable.A rule that covers both of the previous two examples is to compute an expression in the highest precision of any variable that occurs in that expression. Then
q
=
3.0/7.0
will be computed entirely in single precision^{24} and will have the boolean value true, whereasd
=
d
+
x[i]*y[i]
will be computed in double precision, gaining the full advantage of double precision accumulation. However, this rule is too simplistic to cover all cases cleanly. Ifdx
anddy
are double precision variables, the expressiony
=
x
+
single(dxdy)
contains a double precision variable, but performing the sum in double precision would be pointless, because both operands are single precision, as is the result.A more sophisticated subexpression evaluation rule is as follows. First assign each operation a tentative precision, which is the maximum of the precisions of its operands. This assignment has to be carried out from the leaves to the root of the expression tree. Then perform a second pass from the root to the leaves. In this pass, assign to each operation the maximum of the tentative precision and the precision expected by the parent. In the case of
q
=
3.0/7.0
, every leaf is single precision, so all the operations are done in single precision. In the case ofd
=
d
+
x[i]*y[i]
, the tentative precision of the multiply operation is single precision, but in the second pass it gets promoted to double precision, because its parent operation expects a double precision operand. And iny
=
x
+
single(dxdy)
, the addition is done in single precision. Farnum [1988] presents evidence that this algorithm in not difficult to implement.The disadvantage of this rule is that the evaluation of a subexpression depends on the expression in which it is embedded. This can have some annoying consequences. For example, suppose you are debugging a program and want to know the value of a subexpression. You cannot simply type the subexpression to the debugger and ask it to be evaluated, because the value of the subexpression in the program depends on the expression it is embedded in. A final comment on subexpressions: since converting decimal constants to binary is an operation, the evaluation rule also affects the interpretation of decimal constants. This is especially important for constants like
0.1
which are not exactly representable in binary.Another potential grey area occurs when a language includes exponentiation as one of its builtin operations. Unlike the basic arithmetic operations, the value of exponentiation is not always obvious [Kahan and Coonen 1982]. If
**
is the exponentiation operator, then(3)**3
certainly has the value 27. However,(3.0)**3.0
is problematical. If the**
operator checks for integer powers, it would compute(3.0)**3.0
as 3.0^{3} = 27. On the other hand, if the formula x^{y} = e^{ylog}^{x} is used to define**
for real arguments, then depending on the log function, the result could be a NaN (using the natural definition of log(x) =NaN
when x < 0). If the FORTRANCLOG
function is used however, then the answer will be 27, because the ANSI FORTRAN standard definesCLOG(3.0)
to be i + log 3 [ANSI 1978]. The programming language Ada avoids this problem by only defining exponentiation for integer powers, while ANSI FORTRAN prohibits raising a negative number to a real power.In fact, the FORTRAN standard says that
Unfortunately, with the introduction of ± by the IEEE standard, the meaning of not mathematically defined is no longer totally clear cut. One definition might be to use the method shown in section Infinity. For example, to determine the value of a^{b}, consider nonconstant analytic functions f and g with the property that f(x) a and g(x) b as x 0. If f(x)^{g}^{(}^{x}^{)} always approaches the same limit, then this should be the value of a^{b}. This definition would set 2^{} = which seems quite reasonable. In the case of 1.0^{}, when f(x) = 1 and g(x) = 1/x the limit approaches 1, but when f(x) = 1 – x and g(x) = 1/x the limit is e^{1}. So 1.0^{}, should be a NaN. In the case of 0^{0}, f(x)^{g}^{(}^{x}^{)} = e^{g(}^{x}^{)log }^{f}^{(}^{x}^{)}. Since fand g are analytic and take on the value 0 at 0, f(x) = a_{1}x^{1} + a_{2}x^{2} + … and g(x) = b_{1}x^{1} + b_{2}x^{2} + …. Thus lim_{x}_{ 0}g(x) log f(x) = lim_{x}_{ 0}x log(x(a_{1} + a_{2}x + …)) = lim_{x}_{ 0}x log(a_{1}x) = 0. So f(x)^{g}^{(}^{x}^{)} e^{0} = 1 for all f and g, which means that 0^{0} = 1.^{25} ^{26} Using this definition would unambiguously define the exponential function for all arguments, and in particular would define
(3.0)**3.0
to be 27.The IEEE Standard
The section The IEEE Standard,” discussed many of the features of the IEEE standard. However, the IEEE standard says nothing about how these features are to be accessed from a programming language. Thus, there is usually a mismatch between floatingpoint hardware that supports the standard and programming languages like C, Pascal or FORTRAN. Some of the IEEE capabilities can be accessed through a library of subroutine calls. For example the IEEE standard requires that square root be exactly rounded, and the square root function is often implemented directly in hardware. This functionality is easily accessed via a library square root routine. However, other aspects of the standard are not so easily implemented as subroutines. For example, most computer languages specify at most two floatingpoint types, while the IEEE standard has four different precisions (although the recommended configurations are single plus singleextended or single, double, and doubleextended). Infinity provides another example. Constants to represent ± could be supplied by a subroutine. But that might make them unusable in places that require constant expressions, such as the initializer of a constant variable.
A more subtle situation is manipulating the state associated with a computation, where the state consists of the rounding modes, trap enable bits, trap handlers and exception flags. One approach is to provide subroutines for reading and writing the state. In addition, a single call that can atomically set a new value and return the old value is often useful. As the examples in the section Flags show, a very common pattern of modifying IEEE state is to change it only within the scope of a block or subroutine. Thus the burden is on the programmer to find each exit from the block, and make sure the state is restored. Language support for setting the state precisely in the scope of a block would be very useful here. Modula3 is one language that implements this idea for trap handlers [Nelson 1991].
There are a number of minor points that need to be considered when implementing the IEEE standard in a language. Since x – x = +0 for all x,^{27} (+0) – (+0) = +0. However, (+0) = 0, thus –x should not be defined as 0 – x. The introduction of NaNs can be confusing, because a NaN is never equal to any other number (including another NaN), so x = x is no longer always true. In fact, the expression x x is the simplest way to test for a NaN if the IEEE recommended function
Isnan
is not provided. Furthermore, NaNs are unordered with respect to all other numbers, so x y cannot be defined as not x > y. Since the introduction of NaNs causes floatingpoint numbers to become partially ordered, acompare
function that returns one of <, =, >, or unordered can make it easier for the programmer to deal with comparisons.Although the IEEE standard defines the basic floatingpoint operations to return a NaN if any operand is a NaN, this might not always be the best definition for compound operations. For example when computing the appropriate scale factor to use in plotting a graph, the maximum of a set of values must be computed. In this case it makes sense for the max operation to simply ignore NaNs.
Finally, rounding can be a problem. The IEEE standard defines rounding very precisely, and it depends on the current value of the rounding modes. This sometimes conflicts with the definition of implicit rounding in type conversions or the explicit
round
function in languages. This means that programs which wish to use IEEE rounding can’t use the natural language primitives, and conversely the language primitives will be inefficient to implement on the ever increasing number of IEEE machines.Optimizers
Compiler texts tend to ignore the subject of floatingpoint. For example Aho et al. [1986] mentions replacing
x/2.0
withx*0.5
, leading the reader to assume thatx/10.0
should be replaced by0.1*x
. However, these two expressions do not have the same semantics on a binary machine, because 0.1 cannot be represented exactly in binary. This textbook also suggests replacingx*yx*z
byx*(yz)
, even though we have seen that these two expressions can have quite different values when y z. Although it does qualify the statement that any algebraic identity can be used when optimizing code by noting that optimizers should not violate the language definition, it leaves the impression that floatingpoint semantics are not very important. Whether or not the language standard specifies that parenthesis must be honored,(x+y)+z
can have a totally different answer thanx+(y+z)
, as discussed above. There is a problem closely related to preserving parentheses that is illustrated by the following code
eps = 1;do eps = 0.5*eps; while (eps + 1 > 1);
:This is designed to give an estimate for machine epsilon. If an optimizing compiler notices that eps + 1 > 1 eps > 0, the program will be changed completely. Instead of computing the smallest number x such that 1 x is still greater than x (x e ), it will compute the largest number x for which x/2 is rounded to 0 (x ). Avoiding this kind of “optimization” is so important that it is worth presenting one more very useful algorithm that is totally ruined by it.Many problems, such as numerical integration and the numerical solution of differential equations involve computing sums with many terms. Because each addition can potentially introduce an error as large as .5 ulp, a sum involving thousands of terms can have quite a bit of rounding error. A simple way to correct for this is to store the partial summand in a double precision variable and to perform each addition using double precision. If the calculation is being done in single precision, performing the sum in double precision is easy on most computer systems. However, if the calculation is already being done in double precision, doubling the precision is not so simple. One method that is sometimes advocated is to sort the numbers and add them from smallest to largest. However, there is a much more efficient method which dramatically improves the accuracy of sums, namely
Theorem 8 (Kahan Summation Formula)
 Suppose that is computed using the following algorithm
S = X[1];C = 0;for j = 2 to N {Y = X[j]  C;T = S + Y;C = (T  S)  Y;S = T;} Then the computed sum S is equal to where .
Using the naive formula , the computed sum is equal to where _{j} < (n – j)e. Comparing this with the error in the Kahan summation formula shows a dramatic improvement. Each summand is perturbed by only 2e, instead of perturbations as large as ne in the simple formula. Details are in, Errors In Summation.
An optimizer that believed floatingpoint arithmetic obeyed the laws of algebra would conclude that C = [T–S] – Y = [(S+Y)S] – Y = 0, rendering the algorithm completely useless. These examples can be summarized by saying that optimizers should be extremely cautious when applying algebraic identities that hold for the mathematical real numbers to expressions involving floatingpoint variables.
Another way that optimizers can change the semantics of floatingpoint code involves constants. In the expression
1.0E40*x
, there is an implicit decimal to binary conversion operation that converts the decimal number to a binary constant. Because this constant cannot be represented exactly in binary, the inexact exception should be raised. In addition, the underflow flag should to be set if the expression is evaluated in single precision. Since the constant is inexact, its exact conversion to binary depends on the current value of the IEEE rounding modes. Thus an optimizer that converts1.0E40
to binary at compile time would be changing the semantics of the program. However, constants like 27.5 which are exactly representable in the smallest available precision can be safely converted at compile time, since they are always exact, cannot raise any exception, and are unaffected by the rounding modes. Constants that are intended to be converted at compile time should be done with a constant declaration, such asconst
pi
=
3.14159265
.Common subexpression elimination is another example of an optimization that can change floatingpoint semantics, as illustrated by the following code
C = A*B;RndMode = UpD = A*B;
AlthoughA*B
can appear to be a common subexpression, it is not because the rounding mode is different at the two evaluation sites. Three final examples: x = x cannot be replaced by the boolean constanttrue
, because it fails when x is a NaN; –x = 0 – x fails for x = +0; and x < y is not the opposite of x y, because NaNs are neither greater than nor less than ordinary floatingpoint numbers.Despite these examples, there are useful optimizations that can be done on floatingpoint code. First of all, there are algebraic identities that are valid for floatingpoint numbers. Some examples in IEEE arithmetic are x + y = y + x, 2 × x = x + x, 1 × x = x, and 0.5× x = x/2. However, even these simple identities can fail on a few machines such as CDC and Cray supercomputers. Instruction scheduling and inline procedure substitution are two other potentially useful optimizations.^{28}
As a final example, consider the expression
dx
=
x*y
, wherex
andy
are single precision variables, anddx
is double precision. On machines that have an instruction that multiplies two single precision numbers to produce a double precision number,dx
=
x*y
can get mapped to that instruction, rather than compiled to a series of instructions that convert the operands to double and then perform a double to double precision multiply.Some compiler writers view restrictions which prohibit converting (x + y) + z to x + (y + z) as irrelevant, of interest only to programmers who use unportable tricks. Perhaps they have in mind that floatingpoint numbers model real numbers and should obey the same laws that real numbers do. The problem with real number semantics is that they are extremely expensive to implement. Every time two n bit numbers are multiplied, the product will have 2n bits. Every time two n bit numbers with widely spaced exponents are added, the number of bits in the sum is n + the space between the exponents. The sum could have up to (e^{max} – e^{min}) + n bits, or roughly 2·e^{max} + n bits. An algorithm that involves thousands of operations (such as solving a linear system) will soon be operating on numbers with many significant bits, and be hopelessly slow. The implementation of library functions such as sin and cos is even more difficult, because the value of these transcendental functions aren’t rational numbers. Exact integer arithmetic is often provided by lisp systems and is handy for some problems. However, exact floatingpoint arithmetic is rarely useful.
The fact is that there are useful algorithms (like the Kahan summation formula) that exploit the fact that (x + y) + z x + (y + z), and work whenever the bound
a b = (a + b)(1 + )
holds (as well as similar bounds for , × and /). Since these bounds hold for almost all commercial hardware, it would be foolish for numerical programmers to ignore such algorithms, and it would be irresponsible for compiler writers to destroy these algorithms by pretending that floatingpoint variables have real number semantics.Exception Handling
The topics discussed up to now have primarily concerned systems implications of accuracy and precision. Trap handlers also raise some interesting systems issues. The IEEE standard strongly recommends that users be able to specify a trap handler for each of the five classes of exceptions, and the section Trap Handlers, gave some applications of user defined trap handlers. In the case of invalid operation and division by zero exceptions, the handler should be provided with the operands, otherwise, with the exactly rounded result. Depending on the programming language being used, the trap handler might be able to access other variables in the program as well. For all exceptions, the trap handler must be able to identify what operation was being performed and the precision of its destination.
The IEEE standard assumes that operations are conceptually serial and that when an interrupt occurs, it is possible to identify the operation and its operands. On machines which have pipelining or multiple arithmetic units, when an exception occurs, it may not be enough to simply have the trap handler examine the program counter. Hardware support for identifying exactly which operation trapped may be necessary.
Another problem is illustrated by the following program fragment.
x = y*z;z = x*w;a = b + c;d = a/x;
Suppose the second multiply raises an exception, and the trap handler wants to use the value ofa
. On hardware that can do an add and multiply in parallel, an optimizer would probably move the addition operation ahead of the second multiply, so that the add can proceed in parallel with the first multiply. Thus when the second multiply traps,a
=
b
+
c
has already been executed, potentially changing the result ofa
. It would not be reasonable for a compiler to avoid this kind of optimization, because every floatingpoint operation can potentially trap, and thus virtually all instruction scheduling optimizations would be eliminated. This problem can be avoided by prohibiting trap handlers from accessing any variables of the program directly. Instead, the handler can be given the operands or result as an argument.But there are still problems. In the fragment
x = y*z;z = a + b;
the two instructions might well be executed in parallel. If the multiply traps, its argumentz
could already have been overwritten by the addition, especially since addition is usually faster than multiply. Computer systems that support the IEEE standard must provide some way to save the value ofz
, either in hardware or by having the compiler avoid such a situation in the first place.W. Kahan has proposed using presubstitution instead of trap handlers to avoid these problems. In this method, the user specifies an exception and the value he wants to be used as the result when the exception occurs. As an example, suppose that in code for computing (sin x)/x, the user decides that x = 0 is so rare that it would improve performance to avoid a test for x = 0, and instead handle this case when a 0/0 trap occurs. Using IEEE trap handlers, the user would write a handler that returns a value of 1 and install it before computing sin x/x. Using presubstitution, the user would specify that when an invalid operation occurs, the value 1 should be used. Kahan calls this presubstitution, because the value to be used must be specified before the exception occurs. When using trap handlers, the value to be returned can be computed when the trap occurs.
The advantage of presubstitution is that it has a straightforward hardware implementation.^{29} As soon as the type of exception has been determined, it can be used to index a table which contains the desired result of the operation. Although presubstitution has some attractive attributes, the widespread acceptance of the IEEE standard makes it unlikely to be widely implemented by hardware manufacturers.
The Details
A number of claims have been made in this paper concerning properties of floatingpoint arithmetic. We now proceed to show that floatingpoint is not black magic, but rather is a straightforward subject whose claims can be verified mathematically. This section is divided into three parts. The first part presents an introduction to error analysis, and provides the details for the section Rounding Error. The second part explores binary to decimal conversion, filling in some gaps from the section The IEEE Standard. The third part discusses the Kahan summation formula, which was used as an example in the section Systems Aspects.
Rounding Error
In the discussion of rounding error, it was stated that a single guard digit is enough to guarantee that addition and subtraction will always be accurate (Theorem 2). We now proceed to verify this fact. Theorem 2 has two parts, one for subtraction and one for addition. The part for subtraction is
Theorem 9
If x and y are positive floatingpoint numbers in a format with parameters and p, and if subtraction is done with p + 1 digits (i.e. one guard digit), then the relative rounding error in the result is less than
Proof
 Interchange x and y if necessary so that x > y. It is also harmless to scale x and y so that x is represented by x_{0}.x_{1} … x_{p – 1} × ^{0}. If y is represented as y_{0}.y_{1} … y_{p1}, then the difference is exact. If y is represented as 0.y_{1} … y_{p}, then the guard digit ensures that the computed difference will be the exact difference rounded to a floatingpoint number, so the rounding error is at most e. In general, let y = 0.0 … 0y_{k + 1}… y_{k + }_{p} and be y truncated to p + 1 digits. Then
(15) y – < ( – 1)(^{–}^{p}^{ – 1} + ^{–}^{p}^{ – 2} + …^{ }+ ^{–}^{p}^{ – }^{k}).
 From the definition of guard digit, the computed value of x – y is x – rounded to be a floatingpoint number, that is, (x – ) + , where the rounding error satisfies
 The exact difference is x – y, so the error is (x – y) – (x – + ) = – y + . There are three cases. If x – y 1 then the relative error is bounded by
(17) ^{–}^{p} [( – 1)(^{1} + … + ^{–}^{k}) + /2] < ^{–}^{p}(1 + /2) .
> (– 1)(^{1} + … + ^{–}^{k}), where = – 1,
 The final case is when x – y < 1 but x – 1. The only way this could happen is if x – = 1, in which case = 0. But if = 0, then (18)applies, so that again the relative error is bounded by ^{–}^{p} < ^{–}^{p}(1 + /2). z
When = 2, the bound is exactly 2e, and this bound is achieved for x= 1 + 2^{2 – }^{p} and y = 2^{1 – }^{p} – 2^{1 – 2}^{p} in the limit as p . When adding numbers of the same sign, a guard digit is not necessary to achieve good accuracy, as the following result shows.
Theorem 10
If x 0 and y 0, then the relative error in computing x + y is at most 2, even if no guard digits are used.
Proof
 The algorithm for addition with k guard digits is similar to that for subtraction. If x y, shift y right until the radix points of x and y are aligned. Discard any digits shifted past the p + k position. Compute the sum of these two p + k digit numbers exactly. Then round to p digits.
 We will verify the theorem when no guard digits are used; the general case is similar. There is no loss of generality in assuming that x y 0 and that x is scaled to be of the form d.dd…d × ^{0}. First, assume there is no carry out. Then the digits shifted off the end of y have a value less than ^{–}^{p}^{ + 1}, and the sum is at least 1, so the relative error is less than ^{–}^{p}^{+1}/1 = 2e. If there is a carry out, then the error from shifting must be added to the rounding error of
.
The sum is at least , so the relative error is less than2. z
It is obvious that combining these two theorems gives Theorem 2. Theorem 2 gives the relative error for performing one operation. Comparing the rounding error of x^{2} – y^{2} and (x + y) (x – y) requires knowing the relative error of multiple operations. The relative error of x y is _{1} = [(x y) – (x – y)] / (x – y), which satisfies _{1} 2e. Or to write it another way(19) x y = (x – y) (1 + _{1}), _{1} 2e
Similarly(20) x y = (x + y) (1 + _{2}), _{2} 2e
Assuming that multiplication is performed by computing the exact product and then rounding, the relative error is at most .5 ulp, so(21) u v = uv (1 + _{3}), _{3} e
for any floatingpoint numbers u and v. Putting these three equations together (letting u = x y and v = x y) gives(22) (x y) (x y) = (x – y) (1 + _{1}) (x + y) (1 + _{2}) (1 + _{3})
So the relative error incurred when computing (x – y) (x + y) is(23)
This relative error is equal to _{1} + _{2} + _{3} + _{1}_{2} + _{1}_{3} + _{2}_{3} + _{1}_{2}_{3}, which is bounded by 5 + 8^{2}. In other words, the maximum relative error is about 5 rounding errors (since e is a small number, e^{2} is almost negligible).A similar analysis of (x x) (y y) cannot result in a small value for the relative error, because when two nearby values of x and y are plugged into x^{2} – y^{2}, the relative error will usually be quite large. Another way to see this is to try and duplicate the analysis that worked on (x y) (x y), yielding
(x x) (y y) = [x^{2}(1 + _{1}) – y^{2}(1 + _{2})] (1 + _{3})
= ((x^{2} – y^{2}) (1 + _{1}) + (_{1} – _{2})y^{2}) (1 + _{3})
When x and y are nearby, the error term (_{1} – _{2})y^{2} can be as large as the result x^{2} – y^{2}. These computations formally justify our claim that (x– y) (x + y) is more accurate than x^{2} – y^{2}.We next turn to an analysis of the formula for the area of a triangle. In order to estimate the maximum error that can occur when computing with (7), the following fact will be needed.
Theorem 11
Proof
 Note that if x and y have the same exponent, then certainly x y is exact. Otherwise, from the condition of the theorem, the exponents can differ by at most 1. Scale and interchange x and y if necessary so that 0 y x, and x is represented as x_{0}.x_{1} … x_{p – 1} and y as 0.y_{1} … y_{p}. Then the algorithm for computing x y will compute x – y exactly and round to a floatingpoint number. If the difference is of the form 0.d_{1} …d_{p}, the difference will already be p digits long, and no rounding is necessary. Since x 2y, x – y y, and since y is of the form 0.d_{1} … d_{p}, so is x – y. z
When > 2, the hypothesis of Theorem 11 cannot be replaced by y/x y; the stronger condition y/2 x 2y is still necessary. The analysis of the error in (x – y) (x + y), immediately following the proof of Theorem 10, used the fact that the relative error in the basic operations of addition and subtraction is small (namely equations (19) and (20)). This is the most common kind of error analysis. However, analyzing formula (7) requires something more, namely Theorem 11, as the following proof will show.
Theorem 12
 If subtraction uses a guard digit, and if a,b and c are the sides of a triangle (a b c), then the relative error in computing (a + (b + c))(c – (a – b))(c + (a – b))(a +(b – c)) is at most 16, provided e < .005.
Proof
 Let’s examine the factors one by one. From Theorem 10, b c = (b + c) (1 + _{1}), where _{1} is the relative error, and _{1} 2. Then the value of the first factor is
(a (b c)) = (a + (b c)) (1 + _{2}) = (a + (b + c) (1 + _{1}))(1 + _{2}),
(a + b + c) (1 – 2)^{2} [a + (b + c) (1 – 2)] · (12)
a (b c)
[a + (b + c) (1 + 2)] (1 + 2)
(a + b + c) (1 + 2)^{2}
(24) (a (b c)) = (a + b + c) (1 + _{1})^{2}, _{1} 2.
 The next term involves the potentially catastrophic subtraction of c and a
b
, because a b may have rounding error. Because a, b and c are the sides of a triangle, a b+ c, and combining this with the ordering c b a gives a b + c 2b 2a. So a – b satisfies the conditions of Theorem 11. This means that a – b = a b is exact, hence c (a – b) is a harmless subtraction which can be estimated from Theorem 9 to be(25) (c (a b)) = (c – (a – b)) (1 + _{2}), _{2} 2
(26) (c (a b)) = (c + (a – b)) (1 + _{3}), _{3} 2
(27) (a (b c)) = (a + (b – c)) (1 + _{4})^{2}, _{4} 2,
 using both Theorem 9 and Theorem 10. If multiplication is assumed to be exactly rounded, so that x y = xy(1 + ) with  , then combining (24), (25), (26) and (27) gives
(a (b c)) (c (a b)) (c (a b)) (a (b c))
(a + (b + c)) (c – (a – b)) (c + (a – b)) (a + (b – c)) E
E = (1 + _{1})^{2} (1 + _{2}) (1 + _{3}) (1 +_{4})^{2 }(1 + _{1})(1 + _{2}) (1 + _{3})
 An upper bound for E is (1 + 2)^{6}(1 + )^{3}, which expands out to 1 + 15 + O(^{2}). Some writers simply ignore the O(e^{2}) term, but it is easy to account for it. Writing (1 + 2)^{6}(1 + )^{3} = 1 + 15 + R(), R() is a polynomial in e with positive coefficients, so it is an increasing function of . Since R(.005) = .505, R() < 1 for all < .005, and hence E (1 + 2)^{6}(1 + )^{3} < 1 + 16. To get a lower bound on E, note that 1 – 15 – R() < E, and so when < .005, 1 – 16 < (1 – 2)^{6}(1 – )^{3}. Combining these two bounds yields 1 – 16 < E < 1 + 16. Thus the relative error is at most 16. z
Theorem 12 certainly shows that there is no catastrophic cancellation in formula (7). So although it is not necessary to show formula (7) is numerically stable, it is satisfying to have a bound for the entire formula, which is what Theorem 3 of Cancellation gives.
Proof of Theorem 3
q = (a + (b + c)) (c – (a – b)) (c + (a – b)) (a + (b – c))
Q = (a (b c)) (c (a b)) (c (a b)) (a (b c)).
 with _{1} .52 8.5. If square roots are computed to within .5 ulp, then the error when computing is (1 + _{1})(1 + _{2}), with _{2} . If = 2, then there is no further error committed when dividing by 4. Otherwise, one more factor 1 + _{3} with _{3} is necessary for the division, and using the method in the proof of Theorem 12, the final error bound of (1 +_{1}) (1 + _{2}) (1 + _{3}) is dominated by 1 + _{4}, with _{4} 11. z
To make the heuristic explanation immediately following the statement of Theorem 4 precise, the next theorem describes just how closely µ(x) approximates a constant.
Theorem 13
Proof
 Note that µ(x) = 1 – x/2 + x^{2}/3 – … is an alternating series with decreasing terms, so for x 1, µ(x) 1 – x/2 1/2. It is even easier to see that because the series for µ is alternating, µ(x) 1. The Taylor series of µ'(x) is also alternating, and if x has decreasing terms, so – µ'(x) – + 2x/3, or – µ'(x) 0, thus µ'(x) . z
Proof of Theorem 4
 is an alternating series, 0 < x – ln(1 + x) < x^{2}/2, the relative error incurred when approximating ln(1 + x) by x is bounded by x/2. If 1 x = 1, then x < , so the relative error is bounded by /2.
 When 1 x 1, define via 1 x = 1 + . Then since 0 x < 1, (1 x) 1 = . If division and logarithms are computed to within ulp, then the computed value of the expression ln(1 + x)/((1 + x) – 1) is
(29) (1 + _{1}) (1 + _{2}) = (1 + _{1}) (1 + _{2}) = µ( ) (1 + _{1}) (1 + _{2})
where _{1} and _{2} . To estimate µ( ), use the mean value theorem, which says that
 for some between x and . From the definition of , it follows that  – x , and combining this with Theorem 13 gives µ( ) – µ(x) /2, or µ( )/µ(x) – 1 /(2µ(x)) which means that µ( ) = µ(x) (1 + _{3}), with _{3} . Finally, multiplying by x introduces a final _{4}, so the computed value of
(1 + _{1}) (1 + _{2}) (1 + _{3}) (1 + _{4}) = 1 + ,
with  5. zAn interesting example of error analysis using formulas (19), (20), and (21) occurs in the quadratic formula . The section Cancellation, explained how rewriting the equation will eliminate the potential cancellation caused by the ± operation. But there is another potential cancellation that can occur when computing d = b_{2} – 4ac. This one cannot be eliminated by a simple rearrangement of the formula. Roughly speaking, when b^{2} 4ac, rounding error can contaminate up to half the digits in the roots computed with the quadratic formula. Here is an informal proof (another approach to estimating the error in the quadratic formula appears in Kahan [1972]).
If b^{2} 4ac, rounding error can contaminate up to half the digits in the roots computed with the quadratic formula .
Proof: Write (b b) (4a c) = (b^{2}(1 + _{1}) – 4ac(1 + _{2})) (1 + _{3}), where _{i} . ^{30} Using d = b^{2} – 4ac, this can be rewritten as (d(1 + _{1}) – 4ac(_{2} – _{1})) (1 + _{3}). To get an estimate for the size of this error, ignore second order terms in _{i}, in which case the absolute error is d(_{1} + _{3}) – 4ac_{4}, where _{4} = _{1} – _{2} 2. Since , the first term d(_{1} + _{3}) can be ignored. To estimate the second term, use the fact that ax^{2} + bx + c = a(x – r_{1}) (x – r_{2}), so ar_{1}r_{2} = c. Since b^{2} 4ac, then r_{1} _{r}_{2}, so the second error term is . Thus the computed value of is
,
so the absolute error in a is about . Since _{4} ^{–}^{p}, , and thus the absolute error of destroys the bottom half of the bits of the roots r_{1} r_{2}. In other words, since the calculation of the roots involves computing with , and this expression does not have meaningful bits in the position corresponding to the lower order half of r_{i}, then the lower order bits of r_{i} cannot be meaningful. zFinally, we turn to the proof of Theorem 6. It is based on the following fact, which is proven in the section Theorem 14 and Theorem 8.
Theorem 14
 Let 0 < k < p, and set m = ^{k} + 1, and assume that floatingpoint operations are exactly rounded. Then (m x) (m x x) is exactly equal to x rounded to p – k significant digits. More precisely, x is rounded by taking the significand of x, imagining a radix point just left of the k least significant digits and rounding to an integer.
Proof of Theorem 6
 By Theorem 14, x_{h} is x rounded to p – k = places. If there is no carry out, then certainly x_{h} can be represented with significant digits. Suppose there is a carryout. If x = x_{0}.x_{1} … x_{p – 1} × ^{e}, then rounding adds 1 to x_{p – }_{k}_{ – 1}, and the only way there can be a carryout is if x_{p – }_{k}_{ – 1} = – 1, but then the low order digit of x_{h} is 1 + x_{p – }_{k}_{– 1} = 0, and so again x_{h} is representable in digits.
 To deal with x_{l}, scale x to be an integer satisfying ^{p}^{ – 1} x ^{p} – 1. Let where is the p – k high order digits of x, and is the k low order digits. There are three cases to consider. If , then rounding x to p – k places is the same as chopping and , and . Since has at most k digits, if p is even, then has at most k = = digits. Otherwise, = 2 and is representable with k – 1 significant bits. The second case is when , and then computing x_{h} involves rounding up, so x_{h} = + ^{k}, and x_{l} = x – x_{h} = x – –^{k} = – ^{k}. Once again, has at most k digits, so is representable with p/2 digits. Finally, if = (/2)^{k}^{ – 1}, then x_{h} = or + ^{k}depending on whether there is a round up. So x_{l} is either (/2)^{k}^{ – 1} or (/2)^{k}^{ – 1} – ^{k} = –^{k}/2, both of which are represented with 1 digit. z
Theorem 6 gives a way to express the product of two working precision numbers exactly as a sum. There is a companion formula for expressing a sum exactly. If x y then x + y = (x y) + (x (x y)) y [Dekker 1971; Knuth 1981, Theorem C in section 4.2.2]. However, when using exactly rounded operations, this formula is only true for = 2, and not for = 10 as the example x = .99998, y = .99997 shows.
Binary to Decimal Conversion
Since single precision has p = 24, and 2^{24} < 10^{8}, you might expect that converting a binary number to 8 decimal digits would be sufficient to recover the original binary number. However, this is not the case.
Theorem 15
 When a binary IEEE single precision number is converted to the closest eight digit decimal number, it is not always possible to uniquely recover the binary number from the decimal one. However, if nine decimal digits are used, then converting the decimal number to the closest binary number will recover the original floatingpoint number.
Proof
 Binary single precision numbers lying in the half open interval [10^{3}, 2^{10}) = [1000, 1024) have 10 bits to the left of the binary point, and 14 bits to the right of the binary point. Thus there are (2^{10} – 10^{3})2^{14} = 393,216 different binary numbers in that interval. If decimal numbers are represented with 8 digits, then there are (2^{10} – 10^{3})10^{4} = 240,000 decimal numbers in the same interval. There is no way that 240,000 decimal numbers could represent 393,216 different binary numbers. So 8 decimal digits are not enough to uniquely represent each single precision binary number.
 To show that 9 digits are sufficient, it is enough to show that the spacing between binary numbers is always greater than the spacing between decimal numbers. This will ensure that for each decimal number N, the interval
 contains at most one binary number. Thus each binary number rounds to a unique decimal number which in turn rounds to a unique binary number.
 To show that the spacing between binary numbers is always greater than the spacing between decimal numbers, consider an interval [10^{n}, 10^{n}^{+ 1}]. On this interval, the spacing between consecutive decimal numbers is 10^{(}^{n}^{ + 1) – 9}. On [10^{n}, 2^{m}], where m is the smallest integer so that 10^{n} < 2^{m}, the spacing of binary numbers is 2^{m}^{ – 24}, and the spacing gets larger further on in the interval. Thus it is enough to check that 10^{(}^{n}^{ + 1) – 9} < 2^{m}^{ – 24}. But in fact, since 10^{n} < 2^{m}, then 10^{(}^{n}^{ + 1) – 9} = 10^{n}10^{8} < 2^{m}10^{8} < 2^{m}2^{24}. z
The same argument applied to double precision shows that 17 decimal digits are required to recover a double precision number.
Binarydecimal conversion also provides another example of the use of flags. Recall from the section Precision, that to recover a binary number from its decimal expansion, the decimal to binary conversion must be computed exactly. That conversion is performed by multiplying the quantities N and 10^{}^{P}^{} (which are both exact if p < 13) in singleextended precision and then rounding this to single precision (or dividing if p < 0; both cases are similar). Of course the computation of N · 10^{}^{P}^{} cannot be exact; it is the combined operation round(N · 10^{}^{P}^{}) that must be exact, where the rounding is from singleextended to single precision. To see why it might fail to be exact, take the simple case of = 10, p = 2 for single, and p = 3 for singleextended. If the product is to be 12.51, then this would be rounded to 12.5 as part of the singleextended multiply operation. Rounding to single precision would give 12. But that answer is not correct, because rounding the product to single precision should give 13. The error is due to double rounding.
By using the IEEE flags, double rounding can be avoided as follows. Save the current value of the inexact flag, and then reset it. Set the rounding mode to roundtozero. Then perform the multiplication N · 10^{}^{P}^{}. Store the new value of the inexact flag in
ixflag
, and restore the rounding mode and inexact flag. Ifixflag
is 0, then N · 10^{}^{P}^{} is exact, so round(N · 10^{}^{P}^{}) will be correct down to the last bit. Ifixflag
is 1, then some digits were truncated, since roundtozero always truncates. The significand of the product will look like 1.b_{1}…b_{22}b_{23}…b_{31}. A double rounding error may occur if b_{23} …b_{31} = 10…0. A simple way to account for both cases is to perform a logicalOR
ofixflag
with b_{31}. Then round(N· 10^{}^{P}^{}) will be computed correctly in all cases.Errors In Summation
The section Optimizers, mentioned the problem of accurately computing very long sums. The simplest approach to improving accuracy is to double the precision. To get a rough estimate of how much doubling the precision improves the accuracy of a sum, let s_{1} = x_{1}, s_{2} = s_{1} x_{2}…, s_{i}= s_{i} – 1 x_{i}. Then s_{i} = (1 + _{i}) (s_{i – 1} + x_{i}), where _{i} , and ignoring second order terms in _{i} gives
(31)
The first equality of (31) shows that the computed value of is the same as if an exact summation was performed on perturbed values of x_{j}. The first term x_{1} is perturbed by n, the last term x_{n} by only . The second equality in (31) shows that error term is bounded by . Doubling the precision has the effect of squaring . If the sum is being done in an IEEE double precision format, 1/ 10^{16}, so that for any reasonable value of n. Thus, doubling the precision takes the maximum perturbation of n and changes it to . Thus the 2 error bound for the Kahan summation formula (Theorem 8) is not as good as using double precision, even though it is much better than single precision.For an intuitive explanation of why the Kahan summation formula works, consider the following diagram of the procedure.
Each time a summand is added, there is a correction factor C which will be applied on the next loop. So first subtract the correction C computed in the previous loop from X_{j}, giving the corrected summand Y. Then add this summand to the running sum S. The low order bits of Y (namely Y_{l}) are lost in the sum. Next compute the high order bits of Y by computing T – S. When Y is subtracted from this, the low order bits of Y will be recovered. These are the bits that were lost in the first sum in the diagram. They become the correction factor for the next loop. A formal proof of Theorem 8, taken from Knuth [1981] page 572, appears in the section Theorem 14 and Theorem 8.”Summary
It is not uncommon for computer system designers to neglect the parts of a system related to floatingpoint. This is probably due to the fact that floatingpoint is given very little (if any) attention in the computer science curriculum. This in turn has caused the apparently widespread belief that floatingpoint is not a quantifiable subject, and so there is little point in fussing over the details of hardware and software that deal with it.
This paper has demonstrated that it is possible to reason rigorously about floatingpoint. For example, floatingpoint algorithms involving cancellation can be proven to have small relative errors if the underlying hardware has a guard digit, and there is an efficient algorithm for binarydecimal conversion that can be proven to be invertible, provided that extended precision is supported. The task of constructing reliable floatingpoint software is made much easier when the underlying computer system is supportive of floatingpoint. In addition to the two examples just mentioned (guard digits and extended precision), the section Systems Aspects of this paper has examples ranging from instruction set design to compiler optimization illustrating how to better support floatingpoint.
The increasing acceptance of the IEEE floatingpoint standard means that codes that utilize features of the standard are becoming ever more portable. The section The IEEE Standard, gave numerous examples illustrating how the features of the IEEE standard can be used in writing practical floatingpoint codes.
Acknowledgments
This article was inspired by a course given by W. Kahan at Sun Microsystems from May through July of 1988, which was very ably organized by David Hough of Sun. My hope is to enable others to learn about the interaction of floatingpoint and computer systems without having to get up in time to attend 8:00 a.m. lectures. Thanks are due to Kahan and many of my colleagues at Xerox PARC (especially John Gilbert) for reading drafts of this paper and providing many useful comments. Reviews from Paul Hilfinger and an anonymous referee also helped improve the presentation.
References
Aho, Alfred V., Sethi, R., and Ullman J. D. 1986. Compilers: Principles, Techniques and Tools, AddisonWesley, Reading, MA.
ANSI 1978. American National Standard Programming Language FORTRAN, ANSI Standard X3.91978, American National Standards Institute, New York, NY.
Barnett, David 1987. A Portable FloatingPoint Environment, unpublished manuscript.
Brown, W. S. 1981. A Simple but Realistic Model of FloatingPoint Computation, ACM Trans. on Math. Software 7(4), pp. 445480.
Cody, W. J et. al. 1984. A Proposed Radix and Wordlengthindependent Standard for Floatingpoint Arithmetic, IEEE Micro 4(4), pp. 86100.
Cody, W. J. 1988. FloatingPoint Standards — Theory and Practice, in “Reliability in Computing: the role of interval methods in scientific computing”, ed. by Ramon E. Moore, pp. 99107, Academic Press, Boston, MA.
Coonen, Jerome 1984. Contributions to a Proposed Standard for Binary FloatingPoint Arithmetic, PhD Thesis, Univ. of California, Berkeley.
Dekker, T. J. 1971. A FloatingPoint Technique for Extending the Available Precision, Numer. Math. 18(3), pp. 224242.
Demmel, James 1984. Underflow and the Reliability of Numerical Software, SIAM J. Sci. Stat. Comput. 5(4), pp. 887919.
Farnum, Charles 1988. Compiler Support for Floatingpoint Computation, SoftwarePractice and Experience, 18(7), pp. 701709.
Forsythe, G. E. and Moler, C. B. 1967. Computer Solution of Linear Algebraic Systems, PrenticeHall, Englewood Cliffs, NJ.
Goldberg, I. Bennett 1967. 27 Bits Are Not Enough for 8Digit Accuracy, Comm. of the ACM. 10(2), pp 105106.
Goldberg, David 1990. Computer Arithmetic, in “Computer Architecture: A Quantitative Approach”, by David Patterson and John L. Hennessy, Appendix A, Morgan Kaufmann, Los Altos, CA.
Golub, Gene H. and Van Loan, Charles F. 1989. Matrix Computations, 2nd edition,The Johns Hopkins University Press, Baltimore Maryland.
Graham, Ronald L. , Knuth, Donald E. and Patashnik, Oren. 1989. Concrete Mathematics, AddisonWesley, Reading, MA, p.162.
Hewlett Packard 1982. HP15C Advanced Functions Handbook.
IEEE 1987. IEEE Standard 7541985 for Binary Floatingpoint Arithmetic, IEEE, (1985). Reprinted in SIGPLAN 22(2) pp. 925.
Kahan, W. 1972. A Survey Of Error Analysis, in Information Processing 71, Vol 2, pp. 1214 – 1239 (Ljubljana, Yugoslavia), North Holland, Amsterdam.
Kahan, W. 1986. Calculating Area and Angle of a Needlelike Triangle, unpublished manuscript.
Kahan, W. 1987. Branch Cuts for Complex Elementary Functions, in “The State of the Art in Numerical Analysis”, ed. by M.J.D. Powell and A. Iserles (Univ of Birmingham, England), Chapter 7, Oxford University Press, New York.
Kahan, W. 1988. Unpublished lectures given at Sun Microsystems, Mountain View, CA.
Kahan, W. and Coonen, Jerome T. 1982. The Near Orthogonality of Syntax, Semantics, and Diagnostics in Numerical Programming Environments, in “The Relationship Between Numerical Computation And Programming Languages”, ed. by J. K. Reid, pp. 103115, NorthHolland, Amsterdam.
Kahan, W. and LeBlanc, E. 1985. Anomalies in the IBM Acrith Package, Proc. 7th IEEE Symposium on Computer Arithmetic (Urbana, Illinois), pp. 322331.
Kernighan, Brian W. and Ritchie, Dennis M. 1978. The C Programming Language, PrenticeHall, Englewood Cliffs, NJ.
Kirchner, R. and Kulisch, U. 1987. Arithmetic for Vector Processors, Proc. 8th IEEE Symposium on Computer Arithmetic (Como, Italy), pp. 256269.
Knuth, Donald E., 1981. The Art of Computer Programming, Volume II, Second Edition, AddisonWesley, Reading, MA.
Kulisch, U. W., and Miranker, W. L. 1986. The Arithmetic of the Digital Computer: A New Approach, SIAM Review 28(1), pp 136.
Matula, D. W. and Kornerup, P. 1985. Finite Precision Rational Arithmetic: Slash Number Systems, IEEE Trans. on Comput. C34(1), pp 318.
Nelson, G. 1991. Systems Programming With Modula3, PrenticeHall, Englewood Cliffs, NJ.
Reiser, John F. and Knuth, Donald E. 1975. Evading the Drift in Floatingpoint Addition, Information Processing Letters 3(3), pp 8487.
Sterbenz, Pat H. 1974. FloatingPoint Computation, PrenticeHall, Englewood Cliffs, NJ.
Swartzlander, Earl E. and Alexopoulos, Aristides G. 1975. The Sign/Logarithm Number System, IEEE Trans. Comput. C24(12), pp. 12381242.
Walther, J. S., 1971. A unified algorithm for elementary functions, Proceedings of the AFIP Spring Joint Computer Conf. 38, pp. 379385.
Theorem 14 and Theorem 8
This section contains two of the more technical proofs that were omitted from the text.
Theorem 14
 Let 0 < k < p, and set m = ^{k} + 1, and assume that floatingpoint operations are exactly rounded. Then (m x) (m x x) is exactly equal to x rounded to p – k significant digits. More precisely, x is rounded by taking the significand of x, imagining a radix point just left of the k least significant digits, and rounding to an integer.
Proof
 The proof breaks up into two cases, depending on whether or not the computation of mx = ^{k}x + x has a carryout or not.
 Assume there is no carry out. It is harmless to scale x so that it is an integer. Then the computation of mx = x + ^{k}x looks like this:
 where x has been partitioned into two parts. The low order k digits are marked
b
and the high order p – k digits are markeda
. To compute m xfrom mx involves rounding off the low order k digits (the ones marked withb
) so(32) m x = mx – x mod(^{k}) + r^{k}
(33) r = 1 if
a.bb...b
rounds to a + 1, r = 0 otherwise.
 Next compute m x – x = mx – x mod(^{k}) + r^{k} – x = ^{k}(x + r) – x mod(^{k}). The picture below shows the computation of m x – x rounded, that is, (m x) x. The top line is ^{k}(x + r), where
B
is the digit that results from addingr
to the lowest order digitb
. If
.bb...b
< then r = 0, subtracting causes a borrow from the digit markedB
, but the difference is rounded up, and so the net effect is that the rounded difference equals the top line, which is ^{k}x. If.bb...b
> then r = 1, and 1 is subtracted fromB
because of the borrow, so the result is ^{k}x. Finally consider the case.bb...b
= . If r = 0 thenB
is even,Z
is odd, and the difference is rounded up, giving ^{k}x. Similarly when r = 1,B
is odd,Z
is even, the difference is rounded down, so again the difference is ^{k}x. To summarize
 Combining equations (32) and (34) gives (m x) – (m x x) = x – x mod(^{k}) + ·^{k}. The result of performing this computation is
 The rule for computing r, equation (33), is the same as the rule for rounding
a...
ab...b
to p – k places. Thus computing mx – (mx – x) in floatingpoint arithmetic precision is exactly equal to rounding x to p – k places, in the case when x + ^{k}x does not carry out. When x + ^{k}x does carry out, then mx = ^{k}x + x looks like this:
 Thus, m x = mx – x mod(^{k}) + w^{k}, where w = –Z if Z < /2, but the exact value of w is unimportant. Next, m x – x = ^{k}x – x mod(^{k}) + w^{k}. In a picture
aa...aabb...bb00...00
w
 bb... bb
+
zz ... zZbb ...bb^{31}
 Rounding gives (m x) x = ^{k}x + w^{k} – r^{k}, where r = 1 if
.bb...b
> or if.bb...b
= and b_{0} = 1.^{32} Finally,(m x) – (m x x) = mx – x mod(^{k}) + w^{k} – (^{k}x + w^{k} – r^{k})
= x – x mod(^{k}) + r^{k}.
 And once again, r = 1 exactly when rounding
a...ab...b
to p – k places involves rounding up. Thus Theorem 14 is proven in all cases. zTheorem 8 (Kahan Summation Formula)
 Suppose that is computed using the following algorithm
S = X [1]; C = 0; for j = 2 to N { Y = X [j]  C; T = S + Y; C = (T  S)  Y; S = T; } Then the computed sum S is equal to S = x_{j} (1 + _{j}) + O(N^{2}) x_{j}, where _{j} 2.
Proof
 First recall how the error estimate for the simple formula x_{i} went. Introduce s_{1} = x_{1}, s_{i} = (1 + _{i}) (s_{i – 1} + x_{i}). Then the computed sum is s_{n}, which is a sum of terms, each of which is an x_{i} multiplied by an expression involving _{j}‘s. The exact coefficient of x_{1} is (1 + _{2})(1 + _{3}) … (1 + _{n}), and so by renumbering, the coefficient of x_{2} must be (1 + _{3})(1 + _{4}) … (1 + _{n}), and so on. The proof of Theorem 8 runs along exactly the same lines, only the coefficient of x_{1} is more complicated. In detail s_{0} = c_{0} = 0 and
 y_{k }= x_{k} c_{k – 1} = (x_{k} – c_{k – 1}) (1 + _{k})
 s_{k }= s_{k – 1} y_{k }= (s_{k1} + y_{k}) (1 + _{k})
 c_{k }= (s_{k} s_{k – 1}) y_{k}= [(s_{k} – s_{k – 1}) (1 + _{k}) – y_{k}] (1 + _{k})
 where all the Greek letters are bounded by . Although the coefficient of x_{1} in s_{k} is the ultimate expression of interest, in turns out to be easier to compute the coefficient of x_{1} in s_{k} – c_{k} and c_{k}.
 When k = 1,
 c_{1} = (s_{1}(1 + _{1}) – y_{1}) (1 + d_{1})
 = y_{1}((1 + s_{1}) (1 + _{1}) – 1) (1 + d_{1})
 = x_{1}(s_{1} +_{1} + s_{1}g_{1}) (1 + d_{1}) (1 + h_{1})
 s_{1} – c_{1} = x_{1}[(1 + s_{1}) – (s_{1} + g_{1} + s_{1}g_{1}) (1 + d_{1})](1 + h_{1})
 = x_{1}[1 – g_{1} – s_{1}d_{1} – s_{1}g_{1} – d_{1}g_{1} – s_{1}g_{1}d_{1}](1 + h_{1})
 Calling the coefficients of x_{1} in these expressions C_{k} and S_{k} respectively, then
C_{1 }= 2 + O(^{2})
S_{1 }= + _{1} – _{1} + 4^{2} + O(^{3})
 To get the general formula for S_{k} and C_{k}, expand the definitions of s_{k} and c_{k}, ignoring all terms involving x_{i} with i > 1 to get
 s_{k }= (s_{k – 1} + y_{k})(1 + _{k})
 = [s_{k – 1} + (x_{k} – c_{k – 1}) (1 + _{k})](1 + _{k})
 = [(s_{k – 1} – c_{k – 1}) – _{k}c_{k – 1}](1+_{k})
 c_{k }= [{s_{k} – s_{k – 1}}(1 + _{k}) – y_{k}](1 + _{k})
 = [{((s_{k – 1} – c_{k – 1}) – _{k}c_{k – 1})(1 + _{k}) – s_{k – 1}}(1 + _{k}) + c_{k – 1}(1 + _{k})](1 + _{k})
 = [{(s_{k – 1} – c_{k – 1})_{k} – _{k}c_{k}_{1}(1 + _{k}) – c_{k – 1}}(1 + _{k}) + c_{k}_{ – 1}(1 + _{k})](1 + _{k})
 = [(s_{k – 1} – c_{k – 1})_{k}(1 + _{k}) – c_{k – 1}(_{k} + _{k}(_{k} + _{k} + _{k}_{k}))](1 + _{k}),
 s_{k} – c_{k }= ((s_{k – 1} – c_{k – 1}) – _{k}c_{k – 1}) (1 + _{k})
 – [(s_{k – 1} – c_{k – 1})_{k}(1 + _{k}) – c_{k – 1}(_{k} + _{k}(_{k} + _{k} + _{k}_{k})](1 + _{k})
 = (s_{k 1} – c_{k – 1})((1 + _{k}) – _{k}(1 + _{k})(1 + _{k}))
 + c_{k – 1}(_{k}(1 + _{k}) + (_{k} + _{k}(_{k} + _{k} + _{k}_{k})) (1 + _{k}))
 = (s_{– 1} – c_{k – 1}) (1 – _{k}(_{k} + _{k} + _{k}_{k}))
 + c_{k – 1} – [_{k} + _{k} + _{k}(_{k} + _{k}_{k}) + (_{k} + _{k}(_{k} + _{k} + _{k}_{k}))_{k}]
 Since S_{k} and C_{k} are only being computed up to order ^{2}, these formulas can be simplified to
 C_{k}= (_{k} + O(^{2}))S_{k – 1} + (_{k} + O(^{2}))C_{k – 1}
 S_{k}= ((1 + 2^{2} + O(^{3}))S_{k – 1} + (2 + (^{2}))C_{k – 1}
 Using these formulas gives
C_{2} = _{2} + O(^{2})
S_{2 }= 1 + _{1} – _{1} + 10^{2} + O(^{3})
C_{k }= _{k} + O(^{2})
S_{k }= 1 + _{1} – _{1} + (4_{k}+2)^{2} + O(^{3})
Finally, what is wanted is the coefficient of x_{1} in s_{k}. To get this value, let x_{n + 1} = 0, let all the Greek letters with subscripts of n + 1 equal 0, and compute s_{n + 1}. Then s_{n + 1} = s_{n} – c_{n}, and the coefficient of x_{1} in s_{n} is less than the coefficient in s_{n + 1}, which is S_{n} = 1 + _{1} – _{1} + (4n + 2)^{2} = (1 + 2 + (n^{2})). zDifferences Among IEEE 754 Implementations
Note – This section is not part of the published paper. It has been added to clarify certain points and correct possible misconceptions about the IEEE standard that the reader might infer from the paper. This material was not written by David Goldberg, but it appears here with his permission.
The preceding paper has shown that floatingpoint arithmetic must be implemented carefully, since programmers may depend on its properties for the correctness and accuracy of their programs. In particular, the IEEE standard requires a careful implementation, and it is possible to write useful programs that work correctly and deliver accurate results only on systems that conform to the standard. The reader might be tempted to conclude that such programs should be portable to all IEEE systems. Indeed, portable software would be easier to write if the remark “When a program is moved between two machines and both support IEEE arithmetic, then if any intermediate result differs, it must be because of software bugs, not from differences in arithmetic,” were true.
Unfortunately, the IEEE standard does not guarantee that the same program will deliver identical results on all conforming systems. Most programs will actually produce different results on different systems for a variety of reasons. For one, most programs involve the conversion of numbers between decimal and binary formats, and the IEEE standard does not completely specify the accuracy with which such conversions must be performed. For another, many programs use elementary functions supplied by a system library, and the standard doesn’t specify these functions at all. Of course, most programmers know that these features lie beyond the scope of the IEEE standard.
Many programmers may not realize that even a program that uses only the numeric formats and operations prescribed by the IEEE standard can compute different results on different systems. In fact, the authors of the standard intended to allow different implementations to obtain different results. Their intent is evident in the definition of the term destination in the IEEE 754 standard: “A destination may be either explicitly designated by the user or implicitly supplied by the system (for example, intermediate results in subexpressions or arguments for procedures). Some languages place the results of intermediate calculations in destinations beyond the user’s control. Nonetheless, this standard defines the result of an operation in terms of that destination’s format and the operands’ values.” (IEEE 7541985, p. 7) In other words, the IEEE standard requires that each result be rounded correctly to the precision of the destination into which it will be placed, but the standard does not require that the precision of that destination be determined by a user’s program. Thus, different systems may deliver their results to destinations with different precisions, causing the same program to produce different results (sometimes dramatically so), even though those systems all conform to the standard.
Several of the examples in the preceding paper depend on some knowledge of the way floatingpoint arithmetic is rounded. In order to rely on examples such as these, a programmer must be able to predict how a program will be interpreted, and in particular, on an IEEE system, what the precision of the destination of each arithmetic operation may be. Alas, the loophole in the IEEE standard’s definition of destinationundermines the programmer’s ability to know how a program will be interpreted. Consequently, several of the examples given above, when implemented as apparently portable programs in a highlevel language, may not work correctly on IEEE systems that normally deliver results to destinations with a different precision than the programmer expects. Other examples may work, but proving that they work may lie beyond the average programmer’s ability.
In this section, we classify existing implementations of IEEE 754 arithmetic based on the precisions of the destination formats they normally use. We then review some examples from the paper to show that delivering results in a wider precision than a program expects can cause it to compute wrong results even though it is provably correct when the expected precision is used. We also revisit one of the proofs in the paper to illustrate the intellectual effort required to cope with unexpected precision even when it doesn’t invalidate our programs. These examples show that despite all that the IEEE standard prescribes, the differences it allows among different implementations can prevent us from writing portable, efficient numerical software whose behavior we can accurately predict. To develop such software, then, we must first create programming languages and environments that limit the variability the IEEE standard permits and allow programmers to express the floatingpoint semantics upon which their programs depend.
Current IEEE 754 Implementations
Current implementations of IEEE 754 arithmetic can be divided into two groups distinguished by the degree to which they support different floatingpoint formats in hardware. Extendedbased systems, exemplified by the Intel x86 family of processors, provide full support for an extended double precision format but only partial support for single and double precision: they provide instructions to load or store data in single and double precision, converting it onthefly to or from the extended double format, and they provide special modes (not the default) in which the results of arithmetic operations are rounded to single or double precision even though they are kept in registers in extended double format. (Motorola 68000 series processors round results to both the precision and range of the single or double formats in these modes. Intel x86 and compatible processors round results to the precision of the single or double formats but retain the same range as the extended double format.) Single/double systems, including most RISC processors, provide full support for single and double precision formats but no support for an IEEEcompliant extended double precision format. (The IBM POWER architecture provides only partial support for single precision, but for the purpose of this section, we classify it as a single/double system.)
To see how a computation might behave differently on an extendedbased system than on a single/double system, consider a C version of the example from the section Systems Aspects:
int main() {double q;q = 3.0/7.0;if (q == 3.0/7.0) printf("Equal\n");else printf("Not Equal\n");return 0;}
Here the constants 3.0 and 7.0 are interpreted as double precision floatingpoint numbers, and the expression 3.0/7.0 inherits thedouble
data type. On a single/double system, the expression will be evaluated in double precision since that is the most efficient format to use. Thus,q
will be assigned the value 3.0/7.0 rounded correctly to double precision. In the next line, the expression 3.0/7.0 will again be evaluated in double precision, and of course the result will be equal to the value just assigned toq
, so the program will print “Equal” as expected.On an extendedbased system, even though the expression 3.0/7.0 has type
double
, the quotient will be computed in a register in extended double format, and thus in the default mode, it will be rounded to extended double precision. When the resulting value is assigned to the variableq
, however, it may then be stored in memory, and sinceq
is declareddouble
, the value will be rounded to double precision. In the next line, the expression 3.0/7.0 may again be evaluated in extended precision yielding a result that differs from the double precision value stored inq
, causing the program to print “Not equal”. Of course, other outcomes are possible, too: the compiler could decide to store and thus round the value of the expression 3.0/7.0 in the second line before comparing it withq
, or it could keepq
in a register in extended precision without storing it. An optimizing compiler might evaluate the expression 3.0/7.0 at compile time, perhaps in double precision or perhaps in extended double precision. (With one x86 compiler, the program prints “Equal” when compiled with optimization and “Not Equal” when compiled for debugging.) Finally, some compilers for extendedbased systems automatically change the rounding precision mode to cause operations producing results in registers to round those results to single or double precision, albeit possibly with a wider range. Thus, on these systems, we can’t predict the behavior of the program simply by reading its source code and applying a basic understanding of IEEE 754 arithmetic. Neither can we accuse the hardware or the compiler of failing to provide an IEEE 754 compliant environment; the hardware has delivered a correctly rounded result to each destination, as it is required to do, and the compiler has assigned some intermediate results to destinations that are beyond the user’s control, as it is allowed to do.Pitfalls in Computations on ExtendedBased Systems
Conventional wisdom maintains that extendedbased systems must produce results that are at least as accurate, if not more accurate than those delivered on single/double systems, since the former always provide at least as much precision and often more than the latter. Trivial examples such as the C program above as well as more subtle programs based on the examples discussed below show that this wisdom is naive at best: some apparently portable programs, which are indeed portable across single/double systems, deliver incorrect results on extendedbased systems precisely because the compiler and hardware conspire to occasionally provide more precision than the program expects.
Current programming languages make it difficult for a program to specify the precision it expects. As the section Languages and Compilersmentions, many programming languages don’t specify that each occurrence of an expression like
10.0*x
in the same context should evaluate to the same value. Some languages, such as Ada, were influenced in this respect by variations among different arithmetics prior to the IEEE standard. More recently, languages like ANSI C have been influenced by standardconforming extendedbased systems. In fact, the ANSI C standard explicitly allows a compiler to evaluate a floatingpoint expression to a precision wider than that normally associated with its type. As a result, the value of the expression10.0*x
may vary in ways that depend on a variety of factors: whether the expression is immediately assigned to a variable or appears as a subexpression in a larger expression; whether the expression participates in a comparison; whether the expression is passed as an argument to a function, and if so, whether the argument is passed by value or by reference; the current precision mode; the level of optimization at which the program was compiled; the precision mode and expression evaluation method used by the compiler when the program was compiled; and so on.Language standards are not entirely to blame for the vagaries of expression evaluation. Extendedbased systems run most efficiently when expressions are evaluated in extended precision registers whenever possible, yet values that must be stored are stored in the narrowest precision required. Constraining a language to require that
10.0*x
evaluate to the same value everywhere would impose a performance penalty on those systems. Unfortunately, allowing those systems to evaluate10.0*x
differently in syntactically equivalent contexts imposes a penalty of its own on programmers of accurate numerical software by preventing them from relying on the syntax of their programs to express their intended semantics.Do real programs depend on the assumption that a given expression always evaluates to the same value? Recall the algorithm presented in Theorem 4 for computing ln(1 + x), written here in Fortran:
real function log1p(x)real xif (1.0 + x .eq. 1.0) thenlog1p = xelselog1p = log(1.0 + x) * x / ((1.0 + x)  1.0)endifreturnOn an extendedbased system, a compiler may evaluate the expression
1.0
+
x
in the third line in extended precision and compare the result with1.0
. When the same expression is passed to the log function in the sixth line, however, the compiler may store its value in memory, rounding it to single precision. Thus, ifx
is not so small that1.0
+
x
rounds to1.0
in extended precision but small enough that1.0
+
x
rounds to1.0
in single precision, then the value returned bylog1p(x)
will be zero instead ofx
, and the relative error will be one–rather larger than 5. Similarly, suppose the rest of the expression in the sixth line, including the reoccurrence of the subexpression1.0
+
x
, is evaluated in extended precision. In that case, ifx
is small but not quite small enough that1.0
+
x
rounds to1.0
in single precision, then the value returned bylog1p(x)
can exceed the correct value by nearly as much asx
, and again the relative error can approach one. For a concrete example, takex
to be 2^{24} + 2^{47}, sox
is the smallest single precision number such that1.0
+
x
rounds up to the next larger number, 1 + 2^{23}. Thenlog(1.0
+
x)
is approximately 2^{23}. Because the denominator in the expression in the sixth line is evaluated in extended precision, it is computed exactly and deliversx
, solog1p(x)
returns approximately 2^{23}, which is nearly twice as large as the exact value. (This actually happens with at least one compiler. When the preceding code is compiled by the Sun WorkShop Compilers 4.2.1 Fortran 77 compiler for x86 systems using theO
optimization flag, the generated code computes1.0
+
x
exactly as described. As a result, the function delivers zero forlog1p(1.0e10)
and1.19209E07
forlog1p(5.97e8)
.)For the algorithm of Theorem 4 to work correctly, the expression
1.0
+
x
must be evaluated the same way each time it appears; the algorithm can fail on extendedbased systems only when1.0
+
x
is evaluated to extended double precision in one instance and to single or double precision in another. Of course, sincelog
is a generic intrinsic function in Fortran, a compiler could evaluate the expression1.0
+
x
in extended precision throughout, computing its logarithm in the same precision, but evidently we cannot assume that the compiler will do so. (One can also imagine a similar example involving a userdefined function. In that case, a compiler could still keep the argument in extended precision even though the function returns a single precision result, but few if any existing Fortran compilers do this, either.) We might therefore attempt to ensure that1.0
+
x
is evaluated consistently by assigning it to a variable. Unfortunately, if we declare that variablereal
, we may still be foiled by a compiler that substitutes a value kept in a register in extended precision for one appearance of the variable and a value stored in memory in single precision for another. Instead, we would need to declare the variable with a type that corresponds to the extended precision format. Standard FORTRAN 77 does not provide a way to do this, and while Fortran 95 offers theSELECTED_REAL_KIND
mechanism for describing various formats, it does not explicitly require implementations that evaluate expressions in extended precision to allow variables to be declared with that precision. In short, there is no portable way to write this program in standard Fortran that is guaranteed to prevent the expression1.0
+
x
from being evaluated in a way that invalidates our proof.There are other examples that can malfunction on extendedbased systems even when each subexpression is stored and thus rounded to the same precision. The cause is doublerounding. In the default precision mode, an extendedbased system will initially round each result to extended double precision. If that result is then stored to double precision, it is rounded again. The combination of these two roundings can yield a value that is different than what would have been obtained by rounding the first result correctly to double precision. This can happen when the result as rounded to extended double precision is a “halfway case”, i.e., it lies exactly halfway between two double precision numbers, so the second rounding is determined by the roundtiestoeven rule. If this second rounding rounds in the same direction as the first, the net rounding error will exceed half a unit in the last place. (Note, though, that doublerounding only affects double precision computations. One can prove that the sum, difference, product, or quotient of two pbit numbers, or the square root of a pbit number, rounded first to q bits and then to p bits gives the same value as if the result were rounded just once to p bits provided q 2p + 2. Thus, extended double precision is wide enough that single precision computations don’t suffer doublerounding.)
Some algorithms that depend on correct rounding can fail with doublerounding. In fact, even some algorithms that don’t require correct rounding and work correctly on a variety of machines that don’t conform to IEEE 754 can fail with doublerounding. The most useful of these are the portable algorithms for performing simulated multiple precision arithmetic mentioned in the section Exactly Rounded Operations. For example, the procedure described in Theorem 6 for splitting a floatingpoint number into high and low parts doesn’t work correctly in doublerounding arithmetic: try to split the double precision number 2^{52} + 3 × 2^{26} – 1 into two parts each with at most 26 bits. When each operation is rounded correctly to double precision, the high order part is 2^{52} + 2^{27} and the low order part is 2^{26} – 1, but when each operation is rounded first to extended double precision and then to double precision, the procedure produces a high order part of 2^{52} + 2^{28} and a low order part of 2^{26} – 1. The latter number occupies 27 bits, so its square can’t be computed exactly in double precision. Of course, it would still be possible to compute the square of this number in extended double precision, but the resulting algorithm would no longer be portable to single/double systems. Also, later steps in the multiple precision multiplication algorithm assume that all partial products have been computed in double precision. Handling a mixture of double and extended double variables correctly would make the implementation significantly more expensive.
Likewise, portable algorithms for adding multiple precision numbers represented as arrays of double precision numbers can fail in doublerounding arithmetic. These algorithms typically rely on a technique similar to Kahan’s summation formula. As the informal explanation of the summation formula given on Errors In Summation suggests, if
s
andy
are floatingpoint variables with s
 y
 and we compute:
t = s + y;e = (s  t) + y;then in most arithmetics,
e
recovers exactly the roundoff error that occurred in computingt
. This technique doesn’t work in doublerounded arithmetic, however: ifs
= 2^{52} + 1 andy
= 1/2 – 2^{54}, thens
+
y
rounds first to 2^{52} + 3/2 in extended double precision, and this value rounds to 2^{52} + 2 in double precision by the roundtiestoeven rule; thus the net rounding error in computingt
is 1/2 + 2^{54}, which is not representable exactly in double precision and so can’t be computed exactly by the expression shown above. Here again, it would be possible to recover the roundoff error by computing the sum in extended double precision, but then a program would have to do extra work to reduce the final outputs back to double precision, and doublerounding could afflict this process, too. For this reason, although portable programs for simulating multiple precision arithmetic by these methods work correctly and efficiently on a wide variety of machines, they do not work as advertised on extendedbased systems.Finally, some algorithms that at first sight appear to depend on correct rounding may in fact work correctly with doublerounding. In these cases, the cost of coping with doublerounding lies not in the implementation but in the verification that the algorithm works as advertised. To illustrate, we prove the following variant of Theorem 7:
Theorem 7′
 If m and n are integers representable in IEEE 754 double precision with m < 2^{52} and n has the special form n = 2^{i} + 2^{j}, then (m n) n = m, provided both floatingpoint operations are either rounded correctly to double precision or rounded first to extended double precision and then to double precision.
Proof
Assume without loss that m > 0. Let q = m n. Scaling by powers of two, we can consider an equivalent setting in which 2^{52} m < 2^{53} and likewise for q, so that both m and q are integers whose least significant bits occupy the units place (i.e., ulp(m) = ulp(q) = 1). Before scaling, we assumed m < 2^{52}, so after scaling, m is an even integer. Also, because the scaled values of m and q satisfy m/2 < q < 2m, the corresponding value of n must have one of two forms depending on which of m or q is larger: if q < m, then evidently 1 < n < 2, and since n is a sum of two powers of two, n = 1 + 2^{–}^{k} for some k; similarly, if q > m, then 1/2 < n < 1, so n = 1/2 + 2^{(}^{k}^{ + 1)}. (As n is the sum of two powers of two, the closest possible value of n to one is n = 1 + 2^{52}. Because m/(1 + 2^{52}) is no larger than the next smaller double precision number less than m, we can’t have q = m.)
Let e denote the rounding error in computing q, so that q = m/n + e, and the computed value q n will be the (once or twice) rounded value of m + ne. Consider first the case in which each floatingpoint operation is rounded correctly to double precision. In this case, e < 1/2. If nhas the form 1/2 + 2^{(}^{k}^{ + 1)}, then ne = nq – m is an integer multiple of 2^{(}^{k}^{ + 1)} and ne < 1/4 + 2^{(}^{k}^{ + 2)}. This implies that ne 1/4. Recall that the difference between m and the next larger representable number is 1 and the difference between m and the next smaller representable number is either 1 if m > 2^{52} or 1/2 if m = 2^{52}. Thus, as ne 1/4, m + ne will round to m. (Even if m = 2^{52} and ne = 1/4, the product will round to m by the roundtiestoeven rule.) Similarly, if n has the form 1 + 2^{–}^{k}, then ne is an integer multiple of 2^{–}^{k} and ne < 1/2 + 2^{(}^{k}^{ + 1)}; this implies ne 1/2. We can’t have m = 2^{52} in this case because m is strictly greater than q, so m differs from its nearest representable neighbors by ±1. Thus, as ne 1/2, again m + ne will round to m. (Even if ne = 1/2, the product will round to m by the roundtiestoeven rule because m is even.) This completes the proof for correctly rounded arithmetic.
In doublerounding arithmetic, it may still happen that q is the correctly rounded quotient (even though it was actually rounded twice), so e < 1/2 as above. In this case, we can appeal to the arguments of the previous paragraph provided we consider the fact that q n will be rounded twice. To account for this, note that the IEEE standard requires that an extended double format carry at least 64 significant bits, so that the numbers m ± 1/2 and m ± 1/4 are exactly representable in extended double precision. Thus, if n has the form 1/2 + 2^{(}^{k}^{ + 1)}, so that ne 1/4, then rounding m + ne to extended double precision must produce a result that differs from m by at most 1/4, and as noted above, this value will round to m in double precision. Similarly, if n has the form 1 + 2^{–}^{k}, so that ne 1/2, then rounding m + ne to extended double precision must produce a result that differs from m by at most 1/2, and this value will round to m in double precision. (Recall that m > 2^{52} in this case.)
Finally, we are left to consider cases in which q is not the correctly rounded quotient due to doublerounding. In these cases, we have e < 1/2 + 2^{(}^{d}^{ + 1)} in the worst case, where d is the number of extra bits in the extended double format. (All existing extendedbased systems support an extended double format with exactly 64 significant bits; for this format, d = 64 – 53 = 11.) Because doublerounding only produces an incorrectly rounded result when the second rounding is determined by the roundtiestoeven rule, q must be an even integer. Thus if n has the form 1/2 + 2^{(}^{k}^{ + 1)}, then ne = nq – m is an integer multiple of 2^{–}^{k}, and
ne < (1/2 + 2^{(}^{k}^{ + 1)})(1/2 + 2^{(}^{d}^{ + 1)}) = 1/4 + 2^{(}^{k}^{ + 2)} + 2^{(}^{d}^{ + 2)} + 2^{(}^{k}^{ + }^{d}^{ + 2)}.
If k d, this implies ne 1/4. If k > d, we have ne 1/4 + 2^{(}^{d}^{ + 2)}. In either case, the first rounding of the product will deliver a result that differs from m by at most 1/4, and by previous arguments, the second rounding will round to m. Similarly, if n has the form 1 + 2^{–}^{k}, then ne is an integer multiple of 2^{(}^{k}^{ – 1)}, andne < 1/2 + 2^{(}^{k}^{ + 1)} + 2^{(}^{d}^{ + 1)} + 2^{(}^{k}^{ + }^{d}^{ + 1)}.
If k d, this implies ne 1/2. If k > d, we have ne 1/2 + 2^{(}^{d}^{ + 1)}. In either case, the first rounding of the product will deliver a result that differs from m by at most 1/2, and again by previous arguments, the second rounding will round to m. zThe preceding proof shows that the product can incur doublerounding only if the quotient does, and even then, it rounds to the correct result. The proof also shows that extending our reasoning to include the possibility of doublerounding can be challenging even for a program with only two floatingpoint operations. For a more complicated program, it may be impossible to systematically account for the effects of doublerounding, not to mention more general combinations of double and extended double precision computations.
Programming Language Support for Extended Precision
The preceding examples should not be taken to suggest that extended precision per se is harmful. Many programs can benefit from extended precision when the programmer is able to use it selectively. Unfortunately, current programming languages do not provide sufficient means for a programmer to specify when and how extended precision should be used. To indicate what support is needed, we consider the ways in which we might want to manage the use of extended precision.
In a portable program that uses double precision as its nominal working precision, there are five ways we might want to control the use of a wider precision:
 Compile to produce the fastest code, using extended precision where possible on extendedbased systems. Clearly most numerical software does not require more of the arithmetic than that the relative error in each operation is bounded by the “machine epsilon”. When data in memory are stored in double precision, the machine epsilon is usually taken to be the largest relative roundoff error in that precision, since the input data are (rightly or wrongly) assumed to have been rounded when they were entered and the results will likewise be rounded when they are stored. Thus, while computing some of the intermediate results in extended precision may yield a more accurate result, extended precision is not essential. In this case, we might prefer that the compiler use extended precision only when it will not appreciably slow the program and use double precision otherwise.
 Use a format wider than double if it is reasonably fast and wide enough, otherwise resort to something else. Some computations can be performed more easily when extended precision is available, but they can also be carried out in double precision with only somewhat greater effort. Consider computing the Euclidean norm of a vector of double precision numbers. By computing the squares of the elements and accumulating their sum in an IEEE 754 extended double format with its wider exponent range, we can trivially avoid premature underflow or overflow for vectors of practical lengths. On extendedbased systems, this is the fastest way to compute the norm. On single/double systems, an extended double format would have to be emulated in software (if one were supported at all), and such emulation would be much slower than simply using double precision, testing the exception flags to determine whether underflow or overflow occurred, and if so, repeating the computation with explicit scaling. Note that to support this use of extended precision, a language must provide both an indication of the widest available format that is reasonably fast, so that a program can choose which method to use, and environmental parameters that indicate the precision and range of each format, so that the program can verify that the widest fast format is wide enough (e.g., that it has wider range than double).
 Use a format wider than double even if it has to be emulated in software. For more complicated programs than the Euclidean norm example, the programmer may simply wish to avoid the need to write two versions of the program and instead rely on extended precision even if it is slow. Again, the language must provide environmental parameters so that the program can determine the range and precision of the widest available format.
 Don’t use a wider precision; round results correctly to the precision of the double format, albeit possibly with extended range. For programs that are most easily written to depend on correctly rounded double precision arithmetic, including some of the examples mentioned above, a language must provide a way for the programmer to indicate that extended precision must not be used, even though intermediate results may be computed in registers with a wider exponent range than double. (Intermediate results computed in this way can still incur doublerounding if they underflow when stored to memory: if the result of an arithmetic operation is rounded first to 53 significant bits, then rounded again to fewer significant bits when it must be denormalized, the final result may differ from what would have been obtained by rounding just once to a denormalized number. Of course, this form of doublerounding is highly unlikely to affect any practical program adversely.)
 Round results correctly to both the precision and range of the double format. This strict enforcement of double precision would be most useful for programs that test either numerical software or the arithmetic itself near the limits of both the range and precision of the double format. Such careful test programs tend to be difficult to write in a portable way; they become even more difficult (and error prone) when they must employ dummy subroutines and other tricks to force results to be rounded to a particular format. Thus, a programmer using an extendedbased system to develop robust software that must be portable to all IEEE 754 implementations would quickly come to appreciate being able to emulate the arithmetic of single/double systems without extraordinary effort.
No current language supports all five of these options. In fact, few languages have attempted to give the programmer the ability to control the use of extended precision at all. One notable exception is the ISO/IEC 9899:1999 Programming Languages – C standard, the latest revision to the C language, which is now in the final stages of standardization.
The C99 standard allows an implementation to evaluate expressions in a format wider than that normally associated with their type, but the C99 standard recommends using one of only three expression evaluation methods. The three recommended methods are characterized by the extent to which expressions are “promoted” to wider formats, and the implementation is encouraged to identify which method it uses by defining the preprocessor macro
FLT_EVAL_METHOD
: ifFLT_EVAL_METHOD
is 0, each expression is evaluated in a format that corresponds to its type; ifFLT_EVAL_METHOD
is 1,float
expressions are promoted to the format that corresponds todouble
; and ifFLT_EVAL_METHOD
is 2,float
anddouble
expressions are promoted to the format that corresponds tolong double
. (An implementation is allowed to setFLT_EVAL_METHOD
to 1 to indicate that the expression evaluation method is indeterminable.) The C99 standard also requires that the<math.h>
header file define the typesfloat_t
anddouble_t
, which are at least as wide asfloat
anddouble
, respectively, and are intended to match the types used to evaluatefloat
anddouble
expressions. For example, ifFLT_EVAL_METHOD
is 2, bothfloat_t
anddouble_t
arelong double
. Finally, the C99 standard requires that the<float.h>
header file define preprocessor macros that specify the range and precision of the formats corresponding to each floatingpoint type.The combination of features required or recommended by the C99 standard supports some of the five options listed above but not all. For example, if an implementation maps the
long double
type to an extended double format and definesFLT_EVAL_METHOD
to be 2, the programmer can reasonably assume that extended precision is relatively fast, so programs like the Euclidean norm example can simply use intermediate variables of typelong double
(ordouble_t
). On the other hand, the same implementation must keep anonymous expressions in extended precision even when they are stored in memory (e.g., when the compiler must spill floatingpoint registers), and it must store the results of expressions assigned to variables declareddouble
to convert them to double precision even if they could have been kept in registers. Thus, neither thedouble
nor thedouble_t
type can be compiled to produce the fastest code on current extendedbased hardware.Likewise, the C99 standard provides solutions to some of the problems illustrated by the examples in this section but not all. A C99 standard version of the
log1p
function is guaranteed to work correctly if the expression1.0
+
x
is assigned to a variable (of any type) and that variable used throughout. A portable, efficient C99 standard program for splitting a double precision number into high and low parts, however, is more difficult: how can we split at the correct position and avoid doublerounding if we cannot guarantee thatdouble
expressions are rounded correctly to double precision? One solution is to use thedouble_t
type to perform the splitting in double precision on single/double systems and in extended precision on extendedbased systems, so that in either case the arithmetic will be correctly rounded. Theorem 14 says that we can split at any bit position provided we know the precision of the underlying arithmetic, and theFLT_EVAL_METHOD
and environmental parameter macros should give us this information.The following fragment shows one possible implementation:
Of course, to find this solution, the programmer must know that
double
expressions may be evaluated in extended precision, that the ensuing doublerounding problem can cause the algorithm to malfunction, and that extended precision may be used instead according to Theorem 14. A more obvious solution is simply to specify that each expression be rounded correctly to double precision. On extendedbased systems, this merely requires changing the rounding precision mode, but unfortunately, the C99 standard does not provide a portable way to do this. (Early drafts of the FloatingPoint C Edits, the working document that specified the changes to be made to the C90 standard to support floatingpoint, recommended that implementations on systems with rounding precision modes providefegetprec
andfesetprec
functions to get and set the rounding precision, analogous to thefegetround
andfesetround
functions that get and set the rounding direction. This recommendation was removed before the changes were made to the C99 standard.)Coincidentally, the C99 standard’s approach to supporting portability among systems with different integer arithmetic capabilities suggests a better way to support different floatingpoint architectures. Each C99 standard implementation supplies an
<stdint.h>
header file that defines those integer types the implementation supports, named according to their sizes and efficiency: for example,int32_t
is an integer type exactly 32 bits wide,int_fast16_t
is the implementation’s fastest integer type at least 16 bits wide, andintmax_t
is the widest integer type supported. One can imagine a similar scheme for floatingpoint types: for example,float53_t
could name a floatingpoint type with exactly 53 bit precision but possibly wider range,float_fast24_t
could name the implementation’s fastest type with at least 24 bit precision, andfloatmax_t
could name the widest reasonably fast type supported. The fast types could allow compilers on extendedbased systems to generate the fastest possible code subject only to the constraint that the values of named variables must not appear to change as a result of register spilling. The exact width types would cause compilers on extendedbased systems to set the rounding precision mode to round to the specified precision, allowing wider range subject to the same constraint. Finally,double_t
could name a type with both the precision and range of the IEEE 754 double format, providing strict double evaluation. Together with environmental parameter macros named accordingly, such a scheme would readily support all five options described above and allow programmers to indicate easily and unambiguously the floatingpoint semantics their programs require.Must language support for extended precision be so complicated? On single/double systems, four of the five options listed above coincide, and there is no need to differentiate fast and exact width types. Extendedbased systems, however, pose difficult choices: they support neither pure double precision nor pure extended precision computation as efficiently as a mixture of the two, and different programs call for different mixtures. Moreover, the choice of when to use extended precision should not be left to compiler writers, who are often tempted by benchmarks (and sometimes told outright by numerical analysts) to regard floatingpoint arithmetic as “inherently inexact” and therefore neither deserving nor capable of the predictability of integer arithmetic. Instead, the choice must be presented to programmers, and they will require languages capable of expressing their selection.
Conclusion
The foregoing remarks are not intended to disparage extendedbased systems but to expose several fallacies, the first being that all IEEE 754 systems must deliver identical results for the same program. We have focused on differences between extendedbased systems and single/double systems, but there are further differences among systems within each of these families. For example, some single/double systems provide a single instruction to multiply two numbers and add a third with just one final rounding. This operation, called a fused multiplyadd, can cause the same program to produce different results across different single/double systems, and, like extended precision, it can even cause the same program to produce different results on the same system depending on whether and when it is used. (A fused multiplyadd can also foil the splitting process of Theorem 6, although it can be used in a nonportable way to perform multiple precision multiplication without the need for splitting.) Even though the IEEE standard didn’t anticipate such an operation, it nevertheless conforms: the intermediate product is delivered to a “destination” beyond the user’s control that is wide enough to hold it exactly, and the final sum is rounded correctly to fit its single or double precision destination.
The idea that IEEE 754 prescribes precisely the result a given program must deliver is nonetheless appealing. Many programmers like to believe that they can understand the behavior of a program and prove that it will work correctly without reference to the compiler that compiles it or the computer that runs it. In many ways, supporting this belief is a worthwhile goal for the designers of computer systems and programming languages. Unfortunately, when it comes to floatingpoint arithmetic, the goal is virtually impossible to achieve. The authors of the IEEE standards knew that, and they didn’t attempt to achieve it. As a result, despite nearly universal conformance to (most of) the IEEE 754 standard throughout the computer industry, programmers of portable software must continue to cope with unpredictable floatingpoint arithmetic.
If programmers are to exploit the features of IEEE 754, they will need programming languages that make floatingpoint arithmetic predictable. The C99 standard improves predictability to some degree at the expense of requiring programmers to write multiple versions of their programs, one for each
FLT_EVAL_METHOD
. Whether future languages will choose instead to allow programmers to write a single program with syntax that unambiguously expresses the extent to which it depends on IEEE 754 semantics remains to be seen. Existing extendedbased systems threaten that prospect by tempting us to assume that the compiler and the hardware can know better than the programmer how a computation should be performed on a given system. That assumption is the second fallacy: the accuracy required in a computed result depends not on the machine that produces it but only on the conclusions that will be drawn from it, and of the programmer, the compiler, and the hardware, at best only the programmer can know what those conclusions may be.^{1} Examples of other representations are floating slash and signed logarithm [Matula and Kornerup 1985; Swartzlander and Alexopoulos 1975].^{2} This term was introduced by Forsythe and Moler [1967], and has generally replaced the older term mantissa.^{3} This assumes the usual arrangement where the exponent is stored to the left of the significand.^{4} Unless the number z is larger than +1 or smaller than . Numbers which are out of range in this fashion will not be considered until further notice.^{5} Let z‘ be the floatingpoint number that approximates z. Then d.d…d – (z/^{e})^{p1} is equivalent to z‘z/ulp(z‘). A more accurate formula for measuring error is z‘z/ulp(z). – Ed.^{6} 700, not 70. Since .1 – .0292 = .0708, the error in terms of ulp(0.0292) is 708 ulps. – Ed.^{7} Although the expression (x – y)(x + y) does not cause a catastrophic cancellation, it is slightly less accurate than x^{2} – y^{2} if or . In this case, (x – y)(x + y) has three rounding errors, but x^{2} – y^{2 }has only two since the rounding error committed when computing the smaller of x^{2} and y^{2} does not affect the final subtraction.^{8} Also commonly referred to as correctly rounded. – Ed.^{9} When n = 845, x_{n}= 9.45, x_{n} + 0.555 = 10.0, and 10.0 – 0.555 = 9.45. Therefore, x_{n} = x_{845 }for n > 845.^{10} Notice that in binary, q cannot equal . – Ed.^{11} Left as an exercise to the reader: extend the proof to bases other than 2. – Ed.^{12} This appears to have first been published by Goldberg [1967], although Knuth ([1981], page 211) attributes this idea to Konrad Zuse.
^{13} According to Kahan, extended precision has 64 bits of significand because that was the widest precision across which carry propagation could be done on the Intel 8087 without increasing the cycle time [Kahan 1988].
^{14} Some arguments against including inner product as one of the basic operations are presented by Kahan and LeBlanc [1985].^{15} Kirchner writes: It is possible to compute inner products to within 1 ulp in hardware in one partial product per clock cycle. The additionally needed hardware compares to the multiplier array needed anyway for that speed.
^{16} CORDIC is an acronym for Coordinate Rotation Digital Computer and is a method of computing transcendental functions that uses mostly shifts and adds (i.e., very few multiplications and divisions) [Walther 1971]. It is the method additionally needed hardware compares to the multiplier array needed anyway for that speed. d used on both the Intel 8087 and the Motorola 68881.
^{17} Fine point: Although the default in IEEE arithmetic is to round overflowed numbers to , it is possible to change the default (see Rounding Modes)^{18} They are called subnormal in 854, denormal in 754.^{19} This is the cause of one of the most troublesome aspects of the standard. Programs that frequently underflow often run noticeably slower on hardware that uses software traps.
^{20} No invalid exception is raised unless a “trapping” NaN is involved in the operation. See section 6.2 of IEEE Std 7541985. – Ed.^{21} may be greater than if both x and y are negative. – Ed.^{22} It can be in range because if x < 1, n < 0 and x^{–}^{n} is just a tiny bit smaller than the underflow threshold , then , and so may not overflow, since in all IEEE precisions, –e_{min} < e_{max.}^{23} This is probably because designers like “orthogonal” instruction sets, where the precisions of a floatingpoint instruction are independent of the actual operation. Making a special case for multiplication destroys this orthogonality.^{24} This assumes the common convention that3.0
is a singleprecision constant, while3.0D0
is a double precision constant.^{25} The conclusion that 0^{0} = 1 depends on the restriction that f be nonconstant. If this restriction is removed, then letting f be the identically 0 function gives 0 as a possible value for lim _{x 0} f(x)^{g}^{(}^{x}^{)}, and so 0^{0} would have to be defined to be a NaN.^{26} In the case of 0^{0}, plausibility arguments can be made, but the convincing argument is found in “Concrete Mathematics” by Graham, Knuth and Patashnik, and argues that 00 = 1 for the binomial theorem to work. – Ed.
^{27} Unless the rounding mode is round toward –, in which case x – x = 0.^{28} The VMS math libraries on the VAX use a weak form of inline procedure substitution, in that they use the inexpensive jump to subroutine call rather than the slowerCALLS
andCALLG
instructions.^{29} The difficulty with presubstitution is that it requires either direct hardware implementation, or continuable floatingpoint traps if implemented in software. – Ed.^{30} In this informal proof, assume that = 2 so that multiplication by 4 is exact and doesn’t require a _{i.}^{31} This is the sum if adding w does not generate carry out. Additional argument is needed for the special case where adding w does generate carry out. – Ed.^{32} Rounding gives ^{k}x + w^{k} – r^{k }only if^{ (}^{k}x + w^{k}^{) }keeps the form of^{ }^{k}x^{. }–^{ Ed. }
Sun Microsystems, Inc. Copyright information. All rights reserved. Feedback 
Library  Contents  Previous  Next  Index 
These Python Tutor users are asking for help right now. Please volunteer to help!

https://github.com/TG9541/stm8ef/wiki/BoardW1209
The W1209 thermostat module has a STM8S controller, a sensor input, keys, LED display, and a relay. It can be reprogrammed for control, timing, or counting. At a price below $1.50 it can be used for “set and forget” applications.
Note: there have been reports of more or less compatible W1209 clones. Please refer to the identifying W1209 boards page for more information!
A GitHub project for a W1209 DataLogging Thermostat provides an alternative firmware: it’s based on STM8 eForth, and interactively programming other control applications, e.g. using level, pressure, or light sensors, is easy!
The W1209 module (also XHW1209) from SmartClima or XINHE/SINHE) has the following features:
Instead of the NTC temperature sensor other types of resistive transducers can be used (e.g. for measuring light, pressure, strain, or angle). Hardware details and schematics are discussed in technical description.
The W1209 thermostat was one of the first STM8S Value Line Gadgets supported by TG9541/stm8ef. The W1209 runs its own interactive programming system, which allows code to be compiled into RAM or Flash ROM, and tested, directly on the target!
If you’re new to STM8 or to Forth, please refer to the following sections
A HEX file for the W1209, optimized for embedded control tasks, is part of the binary release. The default configuration provides a complete interactive Forth system with sufficient free memory for typical control applications (special configurations provide more free memory). The binary can be programmed to the W1209 board using an inexpensive STLINK adapter.
STM8EF v2.2.13 provides a fullduplex interface through the pushbutton pins. Alternatively, with the help of a simple hardware modificatio, the sensor header can be turned into a halfduplex serial interface.
https://zeptobars.com/en/read/openmicrochipasicwhatinsideIImsp430picz80
In this article we’ll continue cracking microchips open. If you’ve missed first article – it’s here.
Cut a small hole in the center of the chip with a dremel:
WARNING!: All operations with acids must be performed in a fume hood, with proper protection (gloves, gas mask, e.t.c).
https://www.pcjs.org/devices/ti57/machine/
This is a demonstration of the PCjs TI57 Emulator with no diagnostic features enabled. Visit the TI57 Resources page for more options.
The TI57 Emulator is also available as a Web Application.
https://github.com/fcambus/jsemu/blob/master/README.md#calculatoremulators
A list of emulators written in the JavaScript programming language.
This list started as a compilation of JavaScript emulators posted to Echo JS over the years. If you know about any missing emulators, please consider adding them to the collection: the source for this page is available on GitHub. Thank you in advance.
This resource collection is maintained by Frederic Cambus.
To the extent possible under law, Frederic Cambus has waived all copyright and related or neighboring rights to this work.
https://blogs.msdn.microsoft.com/ashleyf/2010/09/25/microcodelevelhp35emulatorinjavascript/
★★★★★
★★★★
★★★
★★
★
Ashley Nathan FenielloSeptember 25, 20105
[Part 1 of the HP Calc
series]
I recently started a supergeeky side hobby of collecting vintage calculators and got my hands on a pair of HP calcs. The more I learned about the internals of the devices, the more intrigued I was. Jacques Laporte has an absolutely wonderful site going into crazy detail.
In 1972, HewlettPackard amazed the engineering world and killed the slide rule with the release of the HP35 calculator; the very first handheld electronic scientific calculator. Most astonishing was that apparently even cute girls loved RPN and were impressed by the device’s ability to do transcendental logarithm and trigonometry functions.
HP12c Emulator Software

01232014, 09:45 PM
Post: #1




HP12c Emulator Software
I don’t know if this is the right place for this question or not, but at least it appears to be in the ballpark.A couple of years ago I bought a copy of the HP12c emulator software sold by HP for use on a Win7 notebook I no longer own. I now want to install this emulator on my Win8 desktop. However, I am unable to activate the software.
When I enter my registration key, the emulator tries to connect to activate.hpcalculating.com but that domain does not appear in the Google DNS. This is quite frustrating. Digital River still offers this software but I don’t want to pay for another copy unnecessarily. Can anyone tell me how to get this software registered? Any help that anyone can give is greatly appreciated. Many thanks. 

01232014, 09:50 PM
Post: #2




RE: HP12c Emulator Software
That should not still be availiable for sale on digital river per my knowledge. Could you please send me a message with where you are seeing it (so we can try to get it removed), and your email address and I will put you in contact with someone who can get you up and running again.Thanks!
If desried, email directly by putting myname at hp.com. Else use the forum software. TW Although I work for the HP calculator group, the views and comments I post here are my own. 

01232014, 10:11 PM (This post was last modified: 01232014 10:18 PM by MotoTom.)
Post: #3




RE: HP12c Emulator Software
Tim,
Here’s the link to the DR offering:https://shopper.mycommerce.com/checkout/…ZKRB4EHLX My email is MotoTom800 at gmail 

01222015, 10:37 AM
Post: #4




RE: HP12c Emulator Software
I’m having exactly the same problem! But the activation site http://activate.hpcalculating.com/ is also dead.Id love to get my 12c emulator running again on Win 8.1.
Roger 

01222015, 02:39 PM
Post: #5




RE: HP12c Emulator Software
The current versions of all HP emulators don’t need any registration/activation (or entering a serial number) anymore. Only the HP10BII+ is missing, but this one can be downloaded from EduCalc: And the HP39gII emulator from this french site is not the latest one (a newer one is available from the original HP FTPserver. Franz 

01222015, 04:04 PM
Post: #6




RE: HP12c Emulator Software
Lots of emulators… 12C … 15C … alas no HP10C. Seems even emulated/simulated a collectors item.Ciao…..Mike


01232015, 02:49 PM
Post: #7




RE: HP12c Emulator Software
Franz – Thank you for this link. I had not seen the HP17BII+ emulator before now, a machine I like and use daily. I can now use it virtually in other places as well. Appreciated! Its my guess that the French website had an unusually large spike in traffic yesterday… –Bob Prosperi 

01242015, 01:04 PM
Post: #8




RE: HP12c Emulator Software
I saw the HP 50G Virtual Calculator here!!!
I am playing with HP Prime but the HP 50G is very famous in Brazilian universities… The 50G emulator works very well in Windows 8.1!!! Thanks for link!!! 

« Next Oldest  Next Newest »

http://www.hpmuseum.org/forum/printthread.php?tid=2668
Kahan summation algorithm – Printable Version
+ HP Forums (http://www.hpmuseum.org/forum) 
Kahan summation algorithm – Thomas Klemm – 12192014 10:12 PM For details please consult Wikipedia. Program for the HP42S: Code:
00 { 19Byte Prgm } Worked example: \(10^{10}+\pi+e\) CLST y: 10,000,000,005.9 This is the correctly rounded result. The classic addition is not the same: E10 y: 0 Cheers RE: Kahan summation algorithm – Gerald H – 12202014 10:47 AM Thank you. Never seen this before. RE: Kahan summation algorithm – Werner – 12212014 09:52 AM What I have never seen mentioned is the fact that this algorithm is meant for binary floating point. Try adding x=y= 1e122 BTW my version reads: Code:
*LBL "K+" same size. Werner 
by Eric hpcalc.org
http://blogs.msdn.com/b/ashleyf/archive/tags/hpcalcs/
> > >
http://www.jacqueslaporte.org/sim65.htm
http://www.jacqueslaporte.org/forum/viewtopic.php?id=180
Project VTQ13c – Part 1 – Introduction
By the end of april 2013, the VTQ13 project came to a sussessful end: The new calculator is now in use with it’s full functionality. The first conceptual works begun in september 2011, so the project lasted for 20 months. Removing holiday times and phases, when I got no time to work on that project, there remains a networkingtime of about 11 months, one month for hardware issues, 10 months implementing the calculator’s operating system.
1.1 VTQ13 Model description
Originally planned to be calculator similar to the beloved HP45, it turned out to be more like a HP27++.
The image below shows a HP27 (image is courtesy of The HPMuseum), the shining example of my calculator in concerns of functionality.
The VTQ13c’s catalog of features reads like this:
While most of the older HP calculators were designed as portrait models, the VTQ13c comes in landscape format with it’s keyboard layed out similar to a mixture of an HP97 an a voyager machine:
I went for this decision primarily in order to be able to place the display module optimally. But it looks nice this way doesn’t it?
There are more differences to the predecessor VTQ9 concerning the layout: Instead of the clumsy computer’s keyboard switches I used small microswitches and homemade key caps that give a lookandfeel similar to original HPcalculators and recuded dimensions of the case dramatically.