Wrong result of logical expression evaluation

Hello,

I'm facing a very bizarre problem: a logical expression of the form (A .and. B) .or. (C .and. D) has to be evaluated in a do while construct. But if gives the wrong result. I mean: when printing the expressions one by one, they are correct. When printing the left part, and the right part, they are also correct. When the full expression is used, its wrong.

A, B, C, D are : T, F, T, F

(A .and. B) gives F

(C .and. D) gives F

BUT: (A .and. B) .or. (C .and. D) gives T !!!!

Here is a copy of the code:

100   IF ( (Direction.GT.0).AND.((T-Tend)+Roundoff.LE.ZERO)&     .OR. (Direction.LT.0).AND.((Tend-T)+Roundoff.LE.ZERO) ) THEN

      print*,'*** start of time loop**'
      print*,'T, Tend',T,Tend
      print*,'test result',(Direction.GT.0).AND.((T-Tend)+Roundoff.LE.& ZERO) .OR. (Direction.LT.0).AND.((Tend-T)+Roundoff.LE.ZERO)

! (list of statements, which increase T towards Tend)

      print'(a,2f25.18)','T, Tend',T,Tend
      print*,'Roundoff =',Roundoff
      AA = (Direction.GT.0).AND.((T-Tend)+Roundoff.LE.ZERO)
      BB = (Direction.LT.0).AND.((Tend-T)+Roundoff.LE.ZERO)

      print*,'test result',& (Direction.GT.0).AND.((T-Tend)+Roundoff.LE.ZERO)& .OR.& (Direction.LT.0).AND.((Tend-T)+Roundoff.LE.ZERO)

      print*,'test with previous evalution',AA.OR.BB
      print*,'each part A,B (for A .OR. B test)',AA,BB
      print*,""
      print*,'subtest p1',& (Direction.GT.0).AND.((T-Tend)+Roundoff.LE.ZERO)
      print*,'subtest p2',& (Direction.LT.0).AND.((Tend-T)+Roundoff.LE.ZERO)

      print*,'split 1',(Direction.GT.0),((T-Tend)+Roundoff.LE.ZERO)
      print*,'split 2',(Direction.LT.0),((Tend-T)+Roundoff.LE.ZERO)

      print*,'*** END of time loop**'

      GOTO 100

      END IF

and here is the output:

 *** start of time loop**
 T, Tend   19.3310000000000        20.0000000000000
 test result T
T, Tend    20.000000000000000000    20.000000000000000000
 Roundoff =  2.220446049250313E-016
 test result T
 test with previous evalution T
 each part A,B (for A .OR. B test) T F

 subtest p1 F
 subtest p2 F
 split 1 T F
 split 2 F T
 *** END of time loop**
 *** start of time loop**
 T, Tend   20.0000000000000        20.0000000000000
 test result T
forrtl: error (73): floating divide by zero

Look: when evaluation first the logical expressions, and storing them in logical variables AA and BB, they are said to be T (which is wrong!) and F (which is correct).

Now, when evaluated inside the print*, command (see "subtest p1/2"), they are F and F (which is now correct).

How is that possible ?

Note that "Roundoff" is defined as epsilon(1.d0).

The ifort version used is 9.0.

Many thanks for your help.