matter@milton.u.washington.edu (Matthew Trunnell) (05/14/91)
We have been noticing some very buggy behavior with the exp() library routine
in DEC Fortran. The problem is that, when passed a large negative exponent,
the routine generates either an OVERFLOW or an arbitrary number depending
on the -fpe option used. Here is some sample output:
Script started on Mon May 13 15:36:11 1991
(killer)1> cat t.f
program t
do 100 i=80, 100
fi = float(i)
a = exp(-fi)
print *, i, a
100 continue
end
(killer)2> f77 -fpe0 t.f
(killer)3> a.out
80 1.8048513E-35
81 6.6396770E-36
82 2.4426007E-36
83 8.9858262E-37
84 3.3057005E-37
85 1.2160993E-37
86 4.4737793E-38
87 1.6458115E-38
forrtl: error: floating overflow <- OVERFLOW, not UNDERFLOW,
Illegal instruction (core dumped) as it should be
(killer)4> f77 -fpe1 t.f
(killer)5> a.out
80 1.8048513E-35
81 6.6396770E-36
82 2.4426007E-36
83 8.9858262E-37
84 3.3057005E-37
85 1.2160993E-37
86 4.4737793E-38
87 1.6458115E-38
88Infinity <- The number was too SMALL
89Infinity to handle--so it was set
90Infinity to INFINITY!
91Infinity
92Infinity
93Infinity
94Infinity
95Infinity
96Infinity
97Infinity
98Infinity
99Infinity
100Infinity
(killer)6> f77 -fpe2 t.f
(killer)7> a.out
80 1.8048513E-35
81 6.6396770E-36
82 2.4426007E-36
83 8.9858262E-37
84 3.3057005E-37
85 1.2160993E-37
86 4.4737793E-38
87 1.6458115E-38
forrtl: error: floating underflow
forrtl: error: floating underflow
forrtl: error: floating overflow
88 -862.7500 <- ???
forrtl: error: floating overflow
89 1.8626075E-35 <- ???
90 -1.2258180E+15 <- ???
91 1.3533067E-24 <- Where the heck are these
92 -3.7006078E+19 numbers coming from??
93 177568.0 <-
94 -3.3480398E+25 <-
95 -14.99609 <-
96 3.8368579E-23 <-
97 7.7316458E-31 <-
98 1.8650276E-19 <-
99 1.3840992E+28 <-
100 1.3298483E+08 <-
forrtl: info: 26 floating underflow traps
forrtl: info: 13 floating overflow traps
(killer)10> ^D
script done on Mon May 13 15:39:23 1991
This is DEC Fortran for ULTRIX v3.0 External Field Test 5, running on a
DECstation 5000 under ULTRIX v4.0
_____________________________________________________________________________
Matthew Trunnell Ec-2 matter@ocean.washington.edu
School of Oceanography, WB-10
University of Washington
Seattle, Washington 98195 206-543-0686
_____________________________________________________________________________