home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Newsgroups: bit.listserv.sas-l
- Path: sparky!uunet!paladin.american.edu!auvm!uvvm!klassen
- References: <9209111646.AA17115@slc10.INS.CWRU.Edu>
- Message-ID: <92255.163248KLASSEN@UVVM>
- Date: Fri, 11 Sep 1992 16:32:48 PDT
- Reply-To: Melvin Klassen <KLASSEN@UVVM.BITNET>
- Sender: "SAS(r) Discussion" <SAS-L@UGA.BITNET>
- Comments: Warning -- original Sender: tag was NETNEWS@UVVM.UVIC.CA
- From: Melvin Klassen <KLASSEN@UVVM.BITNET>
- Subject: Re: Last (??) word on variable lengths
- Lines: 18
-
- "Paul A. Thompson" <pat@PO.CWRU.EDU> writes:
- >I looked at the algorithm. It was all written in single precision,
- >and these values were being used to minimize this complex function.
- >Essentially, this was the entire problem!
-
- You didn't indicate how/where you obtained the FORTRAN code,
- but if the program was written for the Control Data Corporation CDC-6000
- series computer, "short-precision" floating-point used 60-bit words,
- and "long-precision" used 120-bit words.
-
- In this case, single-precision would have been entirely adequate,
- especially when compared to the 64-bit words used for "long-precision"
- on IBM computers, where 3 of those 64 bits are likely to be zero,
- due to the hexadecimal-normalization performed by the IBM hardware.
-
- So, take a look at the 'PRECISION= type-param-value' clause as part
- of the 'REAL' statement in the Fortran 90 standard. Use it as a way of
- writing **portable** FORTRAN code. A case of "programmer, know thy hardware".
-