herrickd@iccgcc.decnet.ab.com (03/15/91)
In article <7571@mentor.cc.purdue.edu>, hrubin@pop.stat.purdue.edu (Herman Rubin) writes: > I would like to write > > exp,mant =UP x, > > where exp is the exponent and mant the mantissa for the floating point x, etc. This is one of my pet peeves in computer science. It is like the use of the word "REAL" in the original FORTRAN (they are a finite subset of the rationals - a lot of bugs come from thinking of them as if they were the real numbers). A mantissa is the fractional part of a logarithm. The entire logarithm is an exponent. The mantissa is that part of the logarithm that determines the significant figures. It determines the significant figures by being an exponent in its own right. The rest of the logarithm only tells where to put the decimal point. A floating point number consists of an exponent and a fraction. The fraction part is not in any way an exponent. I prefer the words exponent and fraction. An interesting historical note: Early 8087 data sheets talked about exponent and mantissa. After a few years, that changed to exponent and significand. Even in the technical notes. The president of Intel had a paper published somewhere (or maybe it was a transcription of a speech) and it talked about exponent and mantissa. The company changed words, but the president didn't. I missed whatever it was that triggered that change. dan herrick herrickd@iccgcc.decnet.ab.com
jlg@cochiti.lanl.gov (Jim Giles) (03/19/91)
From article <3800.27dfaacf@iccgcc.decnet.ab.com>, by herrickd@iccgcc.decnet.ab.com: > [...] > An interesting historical note: Early 8087 data sheets talked about > exponent and mantissa. After a few years, that changed to exponent and > significand. Even in the technical notes. The president of Intel had > a paper published somewhere (or maybe it was a transcription of a > speech) and it talked about exponent and mantissa. The company changed > words, but the president didn't. I missed whatever it was that triggered > that change. The IEEE floating point standard uses the term "significand". It may be that this is the origin of the term. In any case, it is a better term than "fraction". In some floating point representations normalized numbers have a significand that is between 1 and 2 - "fraction" would be an inappropriate term for the significand of such representations. J. Giles