Floating point exceptions C++ / Fortran

[albert]

Dear All,

I have a mixed cpp / fortran project and I want that (in debug mode) the program stops as soon as a floating point exception occurs. Probably a classic problem but I cannot find a solution.

When I have only a Fotran program -fpe0 works perfectly (with -g and -traceback I get the correct location)

program tst
double precision dd
double precision ee

dd = 1
ee = 0
write(*,*) dd / ee
end

Compile with: ifort aaa.f90 -g -traceback -fpe0 -o aaa

Run: ./aaa

forrtl: error (73): floating divide by zero
Image PC Routine Line Source
aaa 000000000040295C MAIN__ 7 aaa.f90
aaa 0000000000402902 Unknown Unknown Unknown
libc.so.6 00000036C301D994 Unknown Unknown Unknown
aaa 0000000000402829 Unknown Unknown Unknown
Abort

When I have a cpp main program (bb.cpp) and a subroutine (aa.f90) I cannot get similar behaviour:

bb.cpp:

extern "C" {
void tst_(void);
}
main()
{
tst_();
}

aa.f90:

subroutine tst
double precision dd
double precision ee

dd = 1
ee = 0
write(*,*) dd / ee
end

Compilation:

ifort -c aa.f90 -g -traceback -fpe0
icpc -c bb.cpp -g -traceback
icpc -g -traceback -shared-intel ./bb.o ./aa.o -lguide -lpthread -lrt -ldl -lifcoremt /usr/lib64/libstdc++.so.5 -o aa

run: ./aa

Infinity

Compiler information:

> ifort --version
ifort (IFORT) 10.1 20080112
Copyright (C) 1985-2007 Intel Corporation. All rights reserved.

> icpc --version
icpc (ICC) 10.1 20080112
Copyright (C) 1985-2007 Intel Corporation. All rights reserved.

System: 64-bit linux

Has anybody a suggestion to solve this problem ?

Best Regards,

Albert

[Ron_Green]

With the 10.1 compiler, no. With this compiler, fpe0 sets up FP exception handling at the Fortran MAIN program startup. In your mixed language case, you do not have a Fortran MAIN and thus fpe0 has no effect.

This was solved in the 11.1 compiler with the -fpe-all option. This option does this: on entry to every Fortran procedure, the existing FP mask is saved, the FP mask is set to the configuration set up by the argument to fpe-all, the routine is entered and run. On exit, the FP mask is restored to it's previous value.

Note: this allows fp trapping in the Fortran routines.

ron

[albert]

Thanks for the answer, one more reason to push the migration to the 11.1 compiler.

Is there with the Intel 10.1 compiler a possibility to accomplish something that is done automatically in version 11.1 i.e. can I make a call in the C / Cpp function before calling the Fortran routine or in the Fortran routine so the floating point exception is caught.

Best Regards,

Albert

[Steven_L_Intel1]

Look in the on-disk documentation under Building Applications > Exception and Error Handling. There is material there about calling FOR_SET_FPE from C code.

[albert]

Thank you for the answer.

I get a message now: "Floatingpoint exception", with gdb I can find the place but is it also possible to have a traceback straight away (like with the Fortran only program ) ?

Albert

[Martyn_C_Intel]

Fortran supports the nice feature -traceback, but I'm afraid there is no complete equivalent for the C/C++ compiler. The C/C++ compiler does support a -traceback option for use in mixed language programs, (see documentation), so that tracebacks get propagated through a mixed C and Fortran stack, but I'm not sure whether it works when the main program is in C, you'd have to experiment. I also suspect that it might only give the detailed function name and line number for the Fortran routines in the traceback.

You might get more suggestions for generating stack traces in the C/C++ compiler forum...