Intel® Fortran Compiler
Build applications that can scale for the future with optimized code designed for Intel® Xeon® and compatible processors.
27434 Discussions

ICE, debug build with constructed PARAMETER

andrew_4619
Honored Contributor II
1,122 Views

I was creating some new code and a got an ICE in Debug build.....

Configuration: Debug|x64 ------
Compiling with Intel® Fortran Compiler Classic 2021.1.1 [Intel(R) 64]...
graphs_SUBS.f90
fortcom: Fatal: There has been an internal compiler error (C0000005).

So I downloaded and installed the latest compiler Fortran Compiler Classic 2021.4.0 [Intel(R) 64]...
and get the same problem as well as a now get and ICE in 26 other source files. There is clearly a regression. I went back to the older compiler and made a small reproducer (see below). I think the problem is referencing the module parameter that has been 'constructed'. I will have to work around it for now.

    module testmod
        implicit none
        integer, parameter :: WS_OVERLAPPEDWINDOW = iany([int(Z'00000000'),int(Z'00C00000'),int(Z'00080000'),&
                                                          int(Z'00040000'),int(Z'00020000'),int(Z'00010000')] )
        !integer, parameter :: WS_OVERLAPPEDWINDOW = 1 ! no ice
    end module testmod
    subroutine test()
        use testmod
        integer :: ifred
        ifred = WS_OVERLAPPEDWINDOW
    end subroutine test

 

0 Kudos
17 Replies
Barbara_P_Intel
Moderator
1,084 Views

Unfortunately, I cannot duplicate the ICE on Windows or Linux with the 2021.4.0 compiler. Can you attach the BuildLog.htm so I can duplicate your compiler options?



andrew_4619
Honored Contributor II
1,078 Views

ifort /nologo /debug:full /Od /debug-parameters:used test.f90
fortcom: Fatal: There has been an internal compiler error (C0000005).
compilation aborted for test.f90 (code 1)    

I had a play the debug-parameters I think is the trigger

 

Barbara_P_Intel
Moderator
1,070 Views

It's the "-debug-parameters used" that triggers the ICE. Thanks for your compiler options.

Now to file a bug report.



Barbara_P_Intel
Moderator
1,062 Views

I filed CMPLRIL0-34376 for you.  I'll let you know when I know more about its fix.

 

andrew_4619
Honored Contributor II
1,048 Views

Thanks , note that at line 10 ifred=int(WS_OVERLAPPEDWINDOW)  does not crash and ifred gets the correct value but if you hover over WS_OVER...  in the debugger you do not see the correct value.....

Barbara_P_Intel
Moderator
1,022 Views

Thanks for the additional information.



andrew_4619
Honored Contributor II
719 Views
! file1.f90
module fred
    implicit none
     interface
        module subroutine dnodpt( vftime )
            real(8)     :: vftime 
        end subroutine dnodpt
    end interface
end module fred
!file.f90
submodule (fred) fred_SUBS
    implicit none
    contains
    module subroutine dnodpt( vftime )
        real(8) :: vftime
        real(8), parameter :: rzero = 0.0_8 
        vftime = rzero
    end subroutine dnodpt
end submodule fred_SUBS

I did a vs2022 installed and added Intel® Fortran Compiler – toolkit version: 2022.1.0, extension version 22.0.0065.17, Package ID: w_oneAPI_2022.0.0.161. The /debug-parameters:used test.f90
fortcom: Fatal: There has been an internal compiler error (C0000005). problem still exists,  I had forgotten about it but when I pared down a reproducer it looked familiar and I then search and found this thread.  The code above is another variation.

 

Any ideas when this might be fixed?

Devorah_H_Intel
Moderator
697 Views

No updates on the fix yet.

Barbara_P_Intel
Moderator
662 Views

The ICE reported when compiling fred.F90 (the second code snippet) is not at all related to the ICE with test.F90. I filed a new bug report, CMPLRIL0-34563, for Fred's ICE. I'll let you know the progress to a fix for this one, too.

 

andrew_4619
Honored Contributor II
651 Views

Interesting, glad I reported it, I nearly didn't on the assumption it was the same thing but having created the test case I thought why not.....  Thanks for the feedback.

Barbara_P_Intel
Moderator
633 Views

I thought it was the same issue, too, and reported it as a second reproducer. But the compiler engineer quickly determined it was not. The second one has something to do with the symbol table.



Steve_Lionel
Black Belt Retired Employee
627 Views

And this is exactly why I kept trying to tell people that they were probably wrong in thinking that they had "the same problem" as someone else just because the symptom was the same. I'd always insist on a reproducer, and more often than not it was in fact a different problem.

Also, while minimal examples are great, also supply the original case if you can. Many times, I told a customer that a bug was fixed, because the cut-down test case now passed, but their full application still failed because of a different bug.

ereisch
New Contributor II
532 Views

This might be related, or might be something different entirely:

        MODULE API
        INTERFACE
                MODULE SUBROUTINE INTEGRATE ( XDOT ) BIND(C)
                        USE, INTRINSIC          :: ISO_C_BINDING
                        IMPLICIT                NONE
                        REAL(KIND=C_FLOAT),     INTENT(IN)      :: XDOT
                END SUBROUTINE INTEGRATE
        END INTERFACE

        END MODULE API

        SUBMODULE (API) INTEGRATE__
        CONTAINS
        MODULE PROCEDURE INTEGRATE
        IMPLICIT        NONE

        STRUCTURE       /SOME_ARRAY/
                LOGICAL*1       TAG
        END STRUCTURE
        INTEGER*4       A
        INTEGER*4       B
        PARAMETER       (A=100)
        PARAMETER       (B=16)

        STRUCTURE       /FOO_ARRAY/
                RECORD/SOME_ARRAY/      FOO(A,B)
        END STRUCTURE

        END PROCEDURE INTEGRATE
        END SUBMODULE INTEGRATE__

 Compile with "ifort -c -g -debug-parameters" and you get an ICE.  Using ifort version 2021.5.0 (also fails with 2021.4.0).

jimdempseyatthecove
Black Belt
526 Views

Try:

        integer, parameter :: WS_OVERLAPPEDWINDOW_int = & 
[Z'00000000',Z'00C00000',Z'00080000',Z'00040000',Z'00020000',Z'00010000']
        integer, parameter :: WS_OVERLAPPEDWINDOW = &
iany(WS_OVERLAPPEDWINDOW_int)

There was a case sometime back regarding BOZ and initialization via argument to intrinsic function. Initialization is OK, argument was not.

Jim Dempsey

Barbara_P_Intel
Moderator
507 Views

@ereisch  Your code compiles ok with ifx 2022.1.0 that is part of oneAPI HPC 2022.2. Unfortunately, ifort 2021.6.0 which is also part of oneAPI HPC 2022.2 still ICEs.

But there is some good news! I compiled it successfully with an internal version of ifort! That fix should be part of the next compiler release.

 

andrew_4619
Honored Contributor II
491 Views

Yes the two original issues CMPLRIL0-34376 and CMPLRIL0-34563 are still open having just tested in fort 2021.6.0. It is good to know that the end is in sight and debug-parameters can go back on.

Does anyone know why debug-parameters even exists, excluding symbols in debug by default is a strange thing to do. You get a hint from the options of a large performance penalty if there are large number of parameter constants though I suspect that is now a historical thing on no real relevance.

Barbara_P_Intel
Moderator
289 Views

The issues you reported while using the compiler option -debug-parameters are fixed in the latest Fortran compiler, 2021.7.0. That version was released last week as part of oneAPI 2022.3.

Enjoy!


Reply