SIGSEGV error in mixed language code

[Evan_P_]

SIGSEGV error in mixed language code

---

I have a mostly Fortran code which I can compile successfully in Windows without any problems, however when I try to compile it on Linux I get a SIGSEGV error. The code contains a small portion written in C, and this is where the problems are caused. I did not write the code, and I have little experience with C, so I am having major issues fixing the error.

The section of the FORTRAN code (in uvss_in_lib.f) that is giving me problems is as follows:

subroutine uvss_lib(kneq,kncoeff,values,irow,icol,rhs,iflag)
include 'sord.inc'
dimension values(1),rhs(1)
dimension irow(1),icol(1)
INTEGER(KIND=INT_PTR_KIND()) :: i_kpt
pointer(i_kpt ,kpt(1))
INTEGER(KIND=INT_PTR_KIND()) :: i_maxa
pointer(i_maxa,maxa(1))
INTEGER(KIND=INT_PTR_KIND()) :: i_row2
pointer(i_row2,irow2(1))
INTEGER(KIND=INT_PTR_KIND()) :: i_values2
pointer(i_values2,values2(1))

if(iflag.eq.1) then
open(79,file='summary',status='scratch')
write(79,*) ' neq =',kneq
write(79,*) ' ncoeff =',kncoeff
endif

c if(iflag.eq.1) then
c write(79,*) '0- lic is fine ' ,kneq,kncoeff
c call send_file()
c endif
if(iflag.eq.1) then
i_maxa = mallocw((kneq+10)*nb,110)
i_kpt = mallocw((kneq+10)*nb,110)
i_values2 = mallocw((kncoeff+10)*nb,110)
i_row2 = mallocw((kncoeff+10)*nbi,110)
endif

call uvss_convert(kneq,kncoeff,values,irow,icol,rhs,maxa,
& irow2,values2,kpt,iflag)
if(iflag.eq.3) then
call free(i_maxa)
call free(i_kpt )
call free(i_values2 )
call free(i_row2 )
endif
return
end

subroutine uvss_convert(kneq,kncoeff,values,irow,icol,rhs,
& maxa,irow2,values2,kpt,iflag)
include 'sord.inc'
dimension values(1),rhs(1)
dimension irow(1),icol(1),maxa(1),kpt(1)
dimension values2(1),irow2(1)

do i = 1, kneq

kpt(i) = 0

enddo

...
...
...

return
end

where mallocw (mallocw.c) is C code:

#include
#include
#if defined (_CRAY) || defined (_WIN32)
# define mallocw MALLOCW
#else
#if !defined(_AIX) && !defined(__hpux)
# define mallocw mallocw_
#endif
#endif
void* mallocw(n,ii)
int *n;
int *ii;
{
long *block;
void *idx;
/* if (*n <= 0)
{
printf("N is negative or zero n");
exit(0);
} */
/* fprintf(stderr,"Bir = %i n",*n ); */
block=(long *)malloc((unsigned)*n);
/* if (block == (long *)NULL)
{
printf("Error in MALLOC - PANIC n");
exit(0);
} */

idx = block;
/* fprintf(stderr," ip = %i n",idx ); */
return(idx);
}

The SIGSEGV error comes when I try to assign a value to "kpt".

I changet the mallocw statements to the intrinsic function MALLOC, but I would then get an error in the memory.c when intCalloc is called.

I have a slightly different version of the code which I compile into a library and link to the commercial finite element software Abaqus, and I have been able to do that without any issues. The two codes only differ by a couple of files (i.e. main.F90 is absent from the library). Perhaps there are some linking flags that Abaqus uses that fixes the problem I am encountering when I try to compile and link on my own. I have tried numerous compiler and linker options including -cxxlib, as well as others.

One other thing. When I make the library version, I compile all the C code into a library (uvss_lib_lib.a). I then compile uvss_inv_lib.f and archive it with the C library into a new library (uvss_lib.a). Then when I compile the rest of the FORTRAN only objects (none of which is given here), I archive the mixed language library with it:

ar cr main_code.a *.o uvss_lib.a

When I try this procedure to build the executeable I link the mixed language library when I am compiling the executeable:

ifort -fPIC *.o -L./ uvss_lib.a -o main_code

However, when I do this it is unable to find any of the functions within the C library (uvss_lib_lib.a). Therefore, I am forced to compile and archive all the source within uvss_lib.a at once. The procedure I follow to compile the executeable successfully is as follows:

%Compile C code

>> icc -c -cxxlib -fPIC -fp-speculationsafe -traceback -debug -g *.c

%Compile uvss_inv_lib.f

>> ifort -c -fPIC -traceback -g -debug uvss_inv_lib.f

%Archive files

>>ar cr uvss_lib.a *.o *.h *.inc

%I then copy uvss_lib.a to another directory contain the rest of my already compile fortran objects and I compile the code

>> -ifort -fPIC -traceback -g -debug -cxxlib main.F90 *.o -L./ uvss_lib.a -lifport -lifcore -o main_code

I am not sure if this has anything to do with my error, but I am confused as to why if I try to link an .a archive that is composed of a .o object and a .a archive the compiler cannot find any of the functions with in the sub-archive.

Any help would be greatly appreciated, as I need to get this code working on Linux so I can finish my dissertation.

Thanks,

-Evan

[Ron_Green]

I don't have 'sord.inc' to test this.

BUT you may have a problem with pointer sizes. What are you expecting for the sizeof( void * ) ? You may investigate the -mcmodel=medium option on both Ifort and ICC to see if that helps.

But the bigger question: WHY mallocw and these old Cray pointers? 25 years ago before Fortran90 ALLOCATE, yes, probably would have used Cray pointers and calling C to do the allocation. But now, why? It may be time to recode this to modern standards and get rid of those Cray pointers and malloc() calls.

ron

[Evan_P_]

Thanks Ron,

This is not my code, but rather a code that was given to me that I am making some modifications. It was definitely something with the pointers. I re-wrote the FORTRAN wrapper so that all the memory allocation was handled in FORTRAN, rather than C, and it works fine now.