CHARMM Development Project Forums CHARMM Community Bug Reports & Fixes -ffortran-bounds-check

Hi,
i am having a problem extending charmm code. my code performs rmsd clustering and uses extensivly ROTLS1 subroutine. in charmm c30a2 everything was fine. in c33a2 and c33b1 the code hangs at some point. if i restart the job, the code hangs exactly at the same point. if i change a source slightly, the code hangs also but at a different point. seems like a memory leak or initialization problem. the write(*,*) statements showed that the bug is not in my part of the code.

To catch the bug, i decided to recompile charmm with array boundary checking flag "-ffortran-bounds-check". when i started it, it showed a lot of places with invalid indexes. Did anybody ever checked this or am i doing something wrong?

Because of the CHARMM memory management scheme using the HEAP data structure, it is not possible to apply bounds checking to all Fortran files; those with references to the HEAP array must be excluded.

could be but for example: subroutine corcom in corman.src in the begining gives

write(*,*) islct,jslct,rwork ! my modification
stop !

CALL CORMAN('MAIN', X, Y, Z, WMAIN, XCOMP, YCOMP, ZCOMP,
& WCOMP, NATIML, HEAP(ISLCT), HEAP(JSLCT), TYPE,
& RESID, RES, IBASE, SEGID, NICTOT, NSGIML, AMASS,
& CG, NTRANS, IMNAME, IMTRNS, XUCELL,
& HEAP(RWORK))


-31415927 -31415927 -31415927

is it correct? can heap use negative indexes? or it is just printed wrong due to the different integer kind?
thanks.

Because CHARMM uses malloc() to acquire more memory from the OS, the indices into the HEAP can be outside the initial dimensioned size of the HEAP array, and could be negative depending on the memory mapping scheme of the OS. That's exactly why you cannot use array bounds checking on any routine that contains indices into the HEAP.

OK, thanks a lot for the answer.

Your problem is probably in the diagonalization routine, DIAGQ.

Check the precision variables and the level of optimization. If these are not correct, the sturm bisection iterative code may never converge.

You can certainly use the program which check memory problems. It is called valgrind from www.valgrind.org. To run this you do not need to recompile CHARMM, it works with the executable!