Serious bug in SELECT CASE - Intel Fortran Compiler on Intel 64

[e745200]

I think I came across a serious bug in how Intel Fortran Intel 64 Compiler Professional deals with SELECT CASE. 

The problem arises when using a CASE in the form :       
   CASE (:max,value,min:) 
None of the specified values will be matched correctly. 
The error does not occur when using        
   CASE (:max,min:)  
A former 32 bit version I used to work with, works fine. 

Here is my sample test, with the different results given by the two versions of the compiler.

$ cat tcase2.f90
INTEGER, DIMENSION(5) :: ival=(/ -5,-3,0,3,5 /)
INTEGER :: i , ik
  DO i=1,5
    ik=ival(i)
    SELECT CASE(ik)
      CASE (:-5,0,5:)
           write (*,*) ik,' special case '
      CASE DEFAULT
           write (*,*) ik,'       default case '
    END SELECT
  END DO
END

$ ifort -logo  tcase2.f90 -o tcase2 && tcase2
Intel Fortran Compiler Professional for applications running on IA-32, Version 11.0    Build 20081105 Package ID: l_cprof_p_11.0.074
...
 Intel Fortran 11.0-1558
GNU ld version 2.17.50.0.6-12.el5 20061020
          -5  special case
          -3        default case
           0  special case
           3        default case
           5  special case


$ ifort64 -logo  tcase2.f90 -o tcase2 && tcase2
Intel Fortran Intel 64 Compiler Professional for applications running on Intel 64, Version 11.1    Build 20091012 Package ID: l_cprof_p_11.1.059
...
 Intel Fortran 11.1-2620
GNU ld version 2.17.50.0.6-12.el5 20061020
          -5        default case
          -3        default case
           0        default case
           3        default case
           5        default case


Has this been fixed since the date of the build I have used  (20091012) ?
Thanks.


G.M.

[Kevin_D_Intel]

Appears to be an optimization related bug introduced in 11.1. The program produces expected results with 11.1 with no optimizations (-O0). I'll report to the appropriate compiler component Development team and keep this post updated as I learn more.

(Internal tracking id: DPD200148788)

(Resolution Update on 4/21/2010): This defect is fixed in the Intel Fortran Compiler Professional Edition 11.1 Update 5 (11.1.069 - Linux).


$ ifort -V u70738.f90 && a.out
Intel Fortran Intel 64 Compiler Professional for applications running on Intel 64, Version 11.1 Build 20091130 Package ID: l_cprof_p_11.1.064
Copyright (C) 1985-2009 Intel Corporation. All rights reserved.

Intel Fortran 11.1-2649
GNU ld version 2.17.50.0.6-5.el5 20061020
-5 default case
-3 default case
0 default case
3 default case
5 default case

$ ifort -V -O0 u70738.f90 && a.out
Intel Fortran Intel 64 Compiler Professional for applications running on Intel 64, Version 11.1 Build 20091130 Package ID: l_cprof_p_11.1.064
Copyright (C) 1985-2009 Intel Corporation. All rights reserved.

Intel Fortran 11.1-2649
GNU ld version 2.17.50.0.6-5.el5 20061020
-5 special case
-3 default case
0 special case
3 default case
5 special case

[e745200]

Thanks, for your prompt answer.

You are right, I tried with no optimizations and it worked fine. Good to know.

Not to renounce to a minimum level of optimization, while waiting for the fix, I'll try to split the cases so that the misfunction does not occur.

However, I would have never expected such a kind of regression trouble with a Professional Compiler by Intel ...

Regards
G.M.