wrong value access in a user-defined type

[under7]

Here is a sample of code that have different outputs depending on the degree of optimisation.
Output :
$ ./compil_sh
$ ./a.out
fin_init_vitesse = 300000.000000000
LumC = 300000.000000000

$ ./compil_sh_bug
$ ./a.out
fin_init_vitesse = 300000.000000000
LumC = 2.00000000000000

Function LumC here points to the imaginary part of Const%CI if it was compiled with "-O0" or "-O1" (thus with "-g"), but gives the expected value with "-O2" or at least "-O1 -ip".

Furthermore, LumC points to the expected value without specifying "kind=8" for the complex CI.

It was tested on several computers, from ifort 9.023 to 9.1.032.
Seems like a bug of the compiler to me.


Sources :

- compile_sh

#!/bin/bash


ifort -module ./ -c -O2 mod_bidon.f90

ifort -module ./ -c -O2 prog_bidon.f90

ifort -module -O2 mod_bidon.o prog_bidon.o

- compile_sh_bug

#!/bin/bash

ifort -module ./ -c -O1  mod_bidon.f90

ifort -module ./ -c -O1 prog_bidon.f90

ifort -module -O1 mod_bidon.o prog_bidon.o

- prog_bidon.f90

program bidon


  use constantes


  implicit none


  real(kind=8)  :: lum


  call InitialiserConst()


  lum = LumC()


  print *, 'LumC = ', lum


end program bidon

- mod_bidon.f90

module constantes


  implicit none


  type t_constantes


     real(kind=8)    :: Pi


     real(kind=8)    :: Vitesse


     complex(kind=8) :: Ci


  end type t_constantes



  type(t_constantes)  :: Const


contains


  subroutine InitialiserConst()


    Const%Pi = 3.14


    Const%Vitesse = 300000


    Const%Ci = (0.0, 2.0)


    print *, 'fin_init_vitesse = ', Const%Vitesse


  end subroutine InitialiserConst


  real(kind=8) function LumC()


    LumC = Const%Vitesse


  end function LumC


end module constantes

Message Edited by under7 on 05-16-200601:42 PM

Message Edited by under7 on 05-16-200601:42 PM

Message Edited by under7 on 05-16-200601:43 PM

Message Edited by under7 on 05-16-200601:47 PM

[Steven_L_Intel1]

I agree that it's a bug. Please submit it to Intel Premier Support.