Internal Compiler Error (C00000FD) in DVF6.0

[HTijseling]

Hi,

I'm using Fortran DVF6.0 with AUTO to solve systems of ODE's. When adding more than 10 DoF's the code does not compile anymore ("f90: Fatal: There has been an internal compiler error (C00000FD)"). This can be seen happening when trying to compile the "test2.for" file, included in the zip folder. The "test1.for" file compiles just fine. I've also included all other files (auto, auto libraries, and eispack) necessary for reconstruction of the error.

Hope someone is able to resolve this!

Thanks in advance,

Hessel

[GVautier]

Some limit may have been reached.

This compiler is too old for thinking about a solution. You may deal with that.

[GVautier]

Try CVF 6.6C if you can. Many bugs were corrected in it.

[Steve_Lionel]

Using a 22-year-old compiler, whose development environment is unsupported on current Windows versions. Wow.

The current Intel oneAPI compiler is free. Why struggle with DVF?

[HTijseling]

I'm surprised my (admittedly foolish) question got a response from the famous Steve L! On a serious note - I use this legacy software because my predecessor used it, and his predecessor, etc, etc. I did not know about the oneAPI compiler - I thought all modern options were paid-for. I'll try to implement our code on oneAPI, then report back here with my findings.

[mecej4]

The source files are in fixed form Fortran, and contain many lines that extend beyond 72 columns, and were probably computer-generated.  Some statements have hundreds of continuation lines.

Make sure to specify a compiler option to take this into consideration.

[HTijseling]

The compiler configuration is Win32 Debug with /extend_source:132 added in the Fortran settings. I assumed these settings would be saved in the workspace file (test.dsw) I included in the zip folder. The continuation lines were limited at 500, as to not reach the 511-line limit. If an equation is longer than 511 lines it gets split up into multiple pieces by the code generator we use to write these files.

[mecej4]

For what it is worth:

• With CVF 6.6C running on Windows 10-64, I could build an EXE using the source and object files that you provided.
• With Ifort/Ifx 32-bit, I ran into a number of incompatibilities w.r.t. the math library routines that are referenced in the CVF-compiled OBJ files that you provided. I could not see how to solve this issue other than to obtain the source files autlib1.f ..., autlib5.f and eispack.f. However, with those source files (they are available on Sourceforge), I was able to build an EXE. I suggest that you discard all OBJ and MOD files that were generated by using CVF, and recompile from sources.

[HTijseling]

Many thanks for your effort! In regard to the bulletpoints:

• I could not find a way to obtain CVF 6.6C; could you point me in the right direction? 
• I've tried to replicate your build (recompiling with source files autlib1.f ..., autlib5.f and eispack.f. from sourceforge), but keep running in to lots of "dummy argument" errors (8284, 6634) in Visual Studio 2019 with Fortran. Could you link/zip or screenshot your build? That might help visualize where I'm going wrong.

Again, many thanks for your help!

[mecej4]

The CVF installers may not work on Windows 10-64. I keep a CD backup of the CVF directory from on an old PC that ran XP, and copy that to the HDD on a newer PC when I need it. I don't think that it is permissible for me to redistribute it. Fortunately, that is not necessary.

S:\AUTO97>ifort /O2 /MD /c autlib?.f eispack.f
S:\AUTO97>lib /out:autlib.lib *.obj
S:\AUTO97>ifort /O2 /MD /extend-source:132 test1.f autlib.lib

 These commands in an IFort command window will build  TEST1.EXE.  Similarly for TEST2.EXE.

You probably have some checks enabled in VS. Old codes such as AUTO97 do not play nice with these checks.

If you provide the data files needed by your TEST1 an TEST2, I can run the programs as built by CVF and Ifort and compare the results.