[8121] in Athena Bugs
decmips 7.3M: f77
daemon@ATHENA.MIT.EDU (ellis@ATHENA.MIT.EDU)
Fri Sep 6 16:02:32 1991
From: ellis@ATHENA.MIT.EDU
To: bugs@ATHENA.MIT.EDU
Cc: nschmidt@ATHENA.MIT.EDU
Date: Fri, 06 Sep 91 16:02:36 EDT
System name: m33-222-9
Type and version: KN01 7.3M
Display type: PMAX-MFB
What were you trying to do?
Compile and run Numerical Recipes Library
What's wrong:
DEC has seriously screwed up floating point exception handling in
DEC Fortran 3.0. Consider the following example cut from
a Numerical Recipes program.
data am,x,s,y/ 9.000000,133.8781,4.358899,28.64901/
c data am,x,s,y/ 9.000000,242.1776,4.358899,53.49461/
x1 =AM*LOG(X/AM)-S*Y
t3 = EXP(X1)
write(6,*) x1,t3
100 stop
end
What should have happened:
In this example, the value of the exponent is -100.58, so e^(-100.58)
should produce a floating underflow. What actually happens depends on
the compilation option:
a) When compiled with "f77 tmp1.f", this program runs to completion,
and prints out a value that is supposed to be outside the range of
a real number( min value ~10^-38):
-100.5808 2.1019477E-44
b) When compiled with no optimization, "f77 -O0 tmp1.f" a floating
overflow message and program termination result:
forrtl: error: floating overflow
Illegal instruction
In addition, this behabvior is not consistent for all values.
The commented out data, does not produce any problem for the example
regardless of compilation option. The result is that e^y (exp) produces
floating overflow for only some values of y.
