forums.silverfrost.com

[arctica]

Hello.

There does not appear to be support for ieee_arithmetic in the current ftn95 e.g.

[PaulLaidler]

Lester

Thank you for the feedback. I have made a note of your suggestion.

This is a part of the 2003 Standard that has not yet been implemented in FTN95.

At the moment computing z = 0.0 then v = 1.0/z will raise a runtime time exception as will v = z/z. This default exception handling will need to be changed when implementing the ieee_arithmetic module and this could be unsafe if the programmer is not fully aware if the implications.

[PaulLaidler]

The next release of FTN95 and its associated DLLs will provide support for the Fortran 2003 intrinsic IEEE_ARITHMETIC module.
This is particularly for x64, with some support for Win32.

A new routine TRAP_FP_EXCEPTIONS@ will be released for x64 but not for Win32.

-------------------------------------------------------------------------------------------------------------------
SUBROUTINE TRAP_FP_EXCEPTIONS@(trap)
INTEGER trap
A trap value of zero masks all floating point runtime exceptions. A non-zero value restores the default state where all
floating point exceptions lead to a runtime failure.

This routine is not provided for Win32 because the the existing TRAP_EXCEPTION@ function does not enable
progression (i.e. the CONTINUE_EXECUTION return state) after handling a floating point exception.
--------------------------------------------------------------------------------------------------------------------

1) TRAP_FP_EXCEPTIONS@ is implemented independently of the IEEE_ARITHMETIC module.
2) Standard PRINT and WRITE statements will display "NaN", "Infinity", or "-Infinity" where appropiate (depending on the
field width) and this is independent of IEEE_ARITHMETIC. The current output is "????....".
3) Users who want to mask floating point exceptions must use x64 and must call TRAP_FP_EXCEPTIONS@(0) for related sections
of their code.

Example
---------
real t,z,v(6)
z = 0.0
t = tiny(z)
call trap_fp_exceptions@(0)
v = [z/z,z/1.0,-z/1.0,1.0/z,-1.0/z,t/10]
call trap_fp_exceptions@(1)
do i = 1,6; print*, v(i); end do
end

Output:
--------
NaN
0.00000
0.00000
Infinity
-Infinity
0.00000