Path: utzoo!utgpu!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!mailrus!ames!pasteur!agate!ucbvax!CIM-VAX.HONEYWELL.COM!derstad From: derstad@CIM-VAX.HONEYWELL.COM ("DAVE ERSTAD") Newsgroups: comp.sys.apollo Subject: Re: Fortran 9.95 problem Message-ID: <8811280001.AA27361@umix.cc.umich.edu> Date: 27 Nov 88 22:49:00 GMT Sender: daemon@ucbvax.BERKELEY.EDU Organization: The Internet Lines: 21 > Yes, I think the option is -frnd The -frnd option affects a different, but related problem. frnd forces rounding to occur before comparing two floating pointer numbers (Otherwise, a 64 bit quantity might be compared to an 80 bit quantity). In this particular instance, a 64 bit quantity is being subtracted from an 80 bit quantity, then compared to zero. I actually did try the -frnd switch to see if there was any difference, and there wasn't. When the comparcomparison to zero occurs, there comparison to zero occurs, the quantity does get rounded, but it is still non-zero (since the mantissa bits getting chopped off are in a different place than if you round the operands). Put another way, round(a - b) != round(a) - round(b) I admit I don't like the way spice implements these conditional branches either, but on the other hand how many sites run spice or a variant without any problems?