NaN result is inconsistent

[Kipling__Michael]

The Sdrnd routine below is part of a much larger program. When the program is run for a particular set of input, which results in the Sdrnd input of X = 0, and N = 4, the result is 0 the first two times the function is called, but the third time and each time after that, the result is NaN.

I fixed a similiar problem with a NaN being produced elsewhere in the program instead of a very small number, by using the compile option /fp:precise for the project.

I can do a work-around by bypassing calculations for an input of zero, but am afraid that a similiar problem may exist elsewhere. Do you have any other suggestions.

I've added the build log for routine below.

Thanks,

Mike

FUNCTION Sdrnd( x, n )

!!!

!-----------------------------------------------------------------------

! PURPOSE: Rounds X to N number of decimal places

! Examples: (49.9938,1) input rounds to 50.0

! (49.9938,2) input rounds to 49.99

!-----------------------------------------------------------------------

! PARAMETERS IN CALL LIST

! NAME TYPE #DIM DESCRIPTION

! ------ ---- ---- -----------

! INPUT:

! x R*8 decimal number

! n I*4 number of decimal points after rounding

!

! OUTPUT:

! Sdrnd R*8 rounded decimal number

!-----------------------------------------------------------------------

!!!

IMPLICIT NONE

INTEGER, PARAMETER :: dbl = SELECTED_REAL_KIND(14)

! List of calling arguments:

REAL(KIND=dbl), INTENT(IN) :: x

INTEGER, INTENT(IN) :: n

REAL(KIND=dbl) :: Sdrnd

! List of local variables:

REAL (KIND=dbl) :: y = 0.0

!*** Begin execution:

Sdrnd = 0.0

IF ( n == 0 ) y = 1.0d0

IF ( n == 1 ) y = 1.0d1

IF ( n == 2 ) y = 1.0d2

IF ( n == 3 ) y = 1.0d3

IF ( n == 4 ) y = 1.0d4

IF ( n == 5 ) y = 1.0d5

IF ( n == 6 ) y = 1.0d6

IF ( n == 7 ) y = 1.0d7

IF ( n == 8 ) y = 1.0d8

IF ( n == 9 ) y = 1.0d9

Sdrnd = ANINT( x * y ) / y

RETURN

END FUNCTION Sdrnd

Build Log

Build started: Project: CLASH_Fsrc, Configuration: Debug|Win32

Output

Compiling with Intel Visual Fortran 11.0.074 [IA-32]...

ifort /nologo /debug:full /Od /fp:precise /module:"Debug\\\\" /object:"Debug\\\\" /traceback /check:bounds

/libs:dll /threads /dbglibs /c /Qvc8

/Qlocation,link,"C:\\Program Files\\Microsoft Visual Studio 8\\VC\\bin" "D:\\ClearCase\\views\\cl_mwk_vs\\clash\\fsrc_f90\\Sdrnd.f90"

ifort: command line warning #10212: /fp:precise evaluates in source precision with Fortran.

CLASH_Fsrc - 0 error(s), 1 warning(s)

[DavidWhite]

Apparently Random NaN results are often due to memory problems - are the arguments properly defined and initialized in your calling program? Are the arguments the same type in the calling program and subroutine.

Regards,

David

[Kipling__Michael]

The calling routine uses initialized variables of the correct type. There may be memory problems somewhere in the 110,000 lines of program Fortran code, or the 140,000 lines of C/C++ code, but this area is OK.

From other posts I've read, it appears that there is sometimes a problem when zero is divided by a number.

Mike