------- Comment #6 from jvdelisle at gcc dot gnu dot org 2007-05-30 03:37
-------
I modified the program given in comment #4 to display the exact values at the
end of the cycle along with the three methods. Method "a" and method "c" give
identical results. Method "b" has a different phase and its the end point
sooner. The exact true value is shown in the fourth column of the output
below.
-- snippet --
a = 3600.0*real(m)+ 60.0*real(l) + real(k) + 0.001*real(j)
b = 0.001*real(j) + real(k) + 60.0*real(l) + 3600.0*real(m)
c = real(3600*m + 60*l + k)
c = c + real(j)/1000.0
exact1 = 3600*m + 60*l + k
exact2 = j
write(6,100)a, b, c, exact1, exact2
100 format (3(f11.5,2x),i5,".",i3)
Result:
86399.98438 86399.98438 86399.98438 86399.988
86399.99219 86399.99219 86399.99219 86399.989
86399.99219 86399.99219 86399.99219 86399.990
86399.99219 86399.99219 86399.99219 86399.991
86399.99219 86399.99219 86399.99219 86399.992
86399.99219 86399.99219 86399.99219 86399.993
86399.99219 86399.99219 86399.99219 86399.994
86399.99219 86399.99219 86399.99219 86399.995
86399.99219 86400.00000 86399.99219 86399.996
86400.00000 86400.00000 86400.00000 86399.997
86400.00000 86400.00000 86400.00000 86399.998
86400.00000 86400.00000 86400.00000 86399.999
a and c are actually a little more accurate at the end. Its also interesting
that this deviation between the methods appears to happen only near the end of
the sequence. I have visually scanned in quite a few places and all three are
identical until this point.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32057
