Re: c/5914: Infinite loop when compiling with optimisation on of z_log.c of libf2c

[Toon Moene <toon@moene.indiv.nluug.nl>]

The following reply was made to PR c/5914; it has been noted by GNATS.

From: Toon Moene <toon@moene.indiv.nluug.nl>
To: Ulrich Jakobus <jakobus@emss.co.za>
Cc: "gcc-bugs@gcc.gnu.org" <gcc-bugs@gcc.gnu.org>,
 	"gcc-gnats@gcc.gnu.org" <gcc-gnats@gcc.gnu.org>,
 	"gcc-prs@gcc.gnu.org" <gcc-prs@gcc.gnu.org>,
 	"nobody@gcc.gnu.org" <nobody@gcc.gnu.org>,
 	"toon@gcc.gnu.org" <toon@gcc.gnu.org>
Subject: Re: c/5914: Infinite loop when compiling with optimisation on of z_log.c 
 of libf2c
Date: Wed, 13 Mar 2002 21:04:38 +0100

 Ulrich Jakobus wrote:
 
 > I wrote:
 
 > >Ah, now I see - you really meant this a *C* bug report :-)
 
 > Yes, I meant this to be a *C* bug of the optimiser (and I
 > submitted it as a C bug). Obviously the created code with
 > -O2 or -O3 is not correct, resulting in an infinite loop.
 > Compiling exactly the same code with -O works, and also
 > using e.g. gcc-2.95.2 with -O2 or -O3 works fine.
 
 Oh, perhaps you think that this is a regression w.r.t. 2.95.2 - it
 isn't.  The fact that 2.95.2 happened to compile this "correctly" is
 purely by chance.  In general, if you code a floating point
 fixed-point-finder by comparing one estimate with the other for
 equality, you'll lose on such targets as the x86, because it keeps
 variables in registers with a different precision than the corresponding
 memory location.
 
 -ffloat-store is the flag to use to circumvent this problem.
 
 -- 
 Toon Moene - mailto:toon@moene.indiv.nluug.nl - phoneto: +31 346 214290
 Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
 Maintainer, GNU Fortran 77: http://gcc.gnu.org/onlinedocs/g77_news.html
 Join GNU Fortran 95: http://g95.sourceforge.net/ (under construction)