Question

I’m getting some strange results from the log function in gfortran 4.5 (OSX) and gfortran 4.7 (OSX).

The following program gives the unexpected results:

program test_log
  real(8) :: e = 2.7182818284590451_8
  write(*,*) 'log(2.7..)', log(2.7182818284590451_8)
  write(*,*) 'log(e)', log(e)
end program test_log

I run it like this:

gfortran-mp-4.5 ./test.f90  && ./a.out 
log(2.7..)   1.6249753165355076     
log(e)   1.0000000000000000

I expected both to be 1.0

Update:

program test_log
  real(8) :: e = 2.7182818284590451_8
  real(8) :: e2 = 2.7182818284590451D0
  real(8) :: e3 = exp(1.0)
  write(*,*) 'log(2.7..)', log(2.7182818284590451_8)
  write(*,*) 'log(e)', log(e)
  write(*,*) 'log(e2)', log(e2)
  write(*,*) 'log(e3)', log(e3)
end program test_log

gives

gfortran-mp-4.5 ./test.f90  && ./a.out 
 log(2.7..)   1.6249753165355076     
 log(e)   1.0000000000000000     
 log(e2)   1.0000000000000000     
 log(e3)   1.0188423211430429

Answer 1

program test_log
  integer, parameter :: Dbl_K = selected_real_kind (14)
  real(Dbl_K) :: e = 2.7182818284590451_Dbl_K
  real(Dbl_K) :: e2 = 2.7182818284590451D0
  real(Dbl_K) :: e3 = exp(1.0)
  real(Dbl_K) :: e4 = exp(1.0_Dbl_K)
  write(*,*) 'log(2.7..)', log(2.7182818284590451_Dbl_K)
  write(*,*) 'log(e)', log(e)
  write(*,*) 'log(e2)', log(e2)
  write(*,*) 'log(e3)', log(e3)
  write(*,*) 'log(e4)', log(e4)
end program test_log

Uses selected_real_kind instead of relying on 8 being the value for double.

Output with gfortran 4.6 and 4.7 on a MacBook Pro with Snow Leopard is:

 log(2.7..)   1.0000000000000000     
 log(e)   1.0000000000000000     
 log(e2)   1.0000000000000000     
 log(e3)  0.99999996963214000     
 log(e4)   1.0000000000000000 

The value for e3 is different from 1 because “exp(1.0)” has 1 as a single precision real.

On my new Mac Air with Lion I get incorrect results. I have observed other problems gfortran on this machine.

gfortran 4.6

 log(2.7..)   1.6249753165355076     
 log(e)   1.6249753165355076     
 log(e2)   1.6249753165355076     
 log(e3)   1.6438056251294211     
 log(e4)   1.6438055841322456

gfortran 4.7

log(2.7..)   1.6249753165355076     
 log(e)   1.0000000000000000     
 log(e2)   1.0000000000000000     
 log(e3)   1.0188423211430429     
 log(e4)   1.0188422801197137

So the problem isn’t unique. Not sure what it is.

P.S. On the MacAir compiler options change the output. I’m not sure which option or options matter.

P.P.S.

program test2_log
  real:: e = 2.7182818284590451
  real:: e2 = 2.7182818284590451D0
  real:: e3 = exp(1.0)
  real:: e4 = exp(1.0)
  write(*,*) 'log(2.7..)', log(2.7182818284590451)
  write(*,*) 'log(e)', log(e)
  write(*,*) 'log(e2)', log(e2)
  write(*,*) 'log(e3)', log(e3)
  write(*,*) 'log(e4)', log(e4)
end program test2_log

On the MacAir, gfortran 4.6 gives:

 log(2.7..)   1.3489696    
 log(e)  0.99999994    
 log(e2)  0.99999994    
 log(e3)   1.0188423    
 log(e4)   1.0188423 

with the compiler option -fdefault-real-8 it changes to

 log(2.7..)   1.6249753165355076     
 log(e)   1.0000000000000000     
 log(e2)   1.0000000000000000     
 log(e3)   1.0188422801197137     
 log(e4)   1.0188422801197137

It’s some defect in the installation.

Edit: all gfortran versions on both machines are from MacPorts. Is the different the OS version or something else? Works on Snow Leopard, not on Lion.