Xref:	news2.mv.net comp.os.msdos.djgpp:4433
From:	Thomas Demmer <demmer AT LSTM DOT Ruhr-UNI-Bochum DOT De>
Newsgroups:	comp.os.msdos.djgpp
Subject:	More buggy libm
Date:	Fri, 31 May 1996 14:21:39 +0100
Organization:	Lehrstuhl fuer Stroemungsmechanik
Lines:	44
Message-ID:	<31AEF263.167E@LSTM.Ruhr-UNI-Bochum.De>
NNTP-Posting-Host:	bvb.lstm.ruhr-uni-bochum.de
Mime-Version:	1.0
To:	djgpp AT delorie DOT com
DJ-Gateway:	from newsgroup comp.os.msdos.djgpp

Investigating deeper into libm, I found paranoia, a program that
checks the basic math performance of a compiler. 
This showed some interesting problems, which may be due to the
optimizer. If I run the unmodified version of paranoia, gcc fails badly.
Adding 
 _control87(1372,0xffff);   /* That what BC sets */
 _control89(MCW_EM,MCW_EM);
helps a bit:

Optimization     libm      Result
  none            no        1 flaw
  none            yes       1 flaw (same as above)
   -O1            no        hung
   -O1            yes       hung
   -O2            no        hung
   -O2            yes       hung

Keeping the default value of _control87() gives a stop in the
middle of the program. 

I just cross-checked the program with gcc.2.6.3 on an R6000, that 
one passes all the tests. So, if anybody has an R6000 with 2.7.2 might 
try paranoia to see if it is a bug in the optimizer. And I would like
to find a x86 compiler that passes the test. Neither TC2.0 nor BC3.1
passed....

Oh yes: Libm doesn't seem to change anything about the results so far,
too bad.
-- 
Ciao
Tom



*************************************************************
* Thomas Demmer                                             *
* Lehrstuhl fuer Stroemungsmechanik                         *
* Ruhr-Uni-Bochum                                           *
* Universitaetsstr. 150                                     *
* D-44780  Bochum                                           *
* Tel: +49 234 700 6434                                     *
* Fax: +49 234 709 4162                                     *
* http://www.lstm.ruhr-uni-bochum.de/~demmer                *
*************************************************************

- Raw text -