Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!caen!ox.com!emory!gatech!wrdis01!nstn.ns.ca!clyde.concordia.ca!altitude!matrox!IRO.UMontreal.CA!JSP.UMontreal.CA!gaumondp
From: gaumondp@JSP.UMontreal.CA (Gaumond Pierre)
Newsgroups: comp.sys.mips
Subject: Floating point internal format
Message-ID: <1991Apr26.154307.28913@jsp.umontreal.ca>
Date: 26 Apr 91 15:43:07 GMT
Sender: news@jsp.umontreal.ca (Administration de C news)
Organization: Universite de Montreal
Lines: 31
Nntp-Posting-Host: spt3.jsp.umontreal.ca

I would like to know the internal (binary) format of floating point numbers
(single AND double precision) on the SGI 4D/380S (MIPS R3000 processors).

   - Mantissa (M) and exponent (E) positions and sizes.
   - Is there any bias or offset to the exponent?
   - Is F=M*2^E or F=M*16^E ?
   - What are the rules on normalisation (shifting left the most significant 
     bit of the mantissa)?
   - Are there two exponent fields in double precision?
   - negative numbers.
   - Negative exponent (small numbers).
   - Are there some invalid representations (giving arithmetic faults when 
     used)?
   - etc ...

I will use this information to migrate binary files containing floating point 
numbers to the SGI machine mentionned above. I already know the format of the 
floating point numbers on the source machine.

I want these informations to keep as much precision as possible. I don't want 
to convert numbers in decimal format and translate back in binary on the SIG 
because some precision is lost in these conversions.

Thanks.

Pierre Gaumond.
-- 
Pierre Gaumond.                                 | gaumondp@JSP.UMontreal.CA
Services Informatiques, Universite de Montreal. | gaumondp@centrcn.UMontreal.CA
C.P. 6128, Succursale "A", Montreal,            |
Quebec, Canada.   H3C 3J7                       |