Segmentation fault during call to DPOSV routine from MKL 11.3 library

I am trying to run the following code, compiling with IFORT 2016, linking MKL 11.3 library:

program bug
  implicit none
  INCLUDE 'mkl.fi'
  integer*4, parameter :: Npart=25
  real(kind=8) :: u(1:Npart), lapackab(1:Npart,1:Npart)
  integer*4 :: i

  u=0.2d0
  lapackab=0d0
  do i=1,Npart
      lapackab(i,i)=2d0
  enddo

  call DPOSV('L',Npart,1,lapackab,Npart,u,Npart,i)
  write(*,*) "i=",i
end program

using the following command:

ifort -O0 -g -openmp -o file.o -c file.f90 -I/opt/share/INTEL/mkl/include
ifort -O0 -g -openmp file.o -o run -L/opt/share/INTEL/mkl/lib/  -I/opt/share/INTEL/mkl/include  -lmkl_intel_ilp64 -lmkl_intel_thread -lmkl_core -liomp5 -lpthread -lm -ldl

for some obscure reason, the code segfaults during the DPOSV. I have no idea why. Seriously, even Valgrind does not report anything weird.. Is there anybody that had the same issue with this particular routine?

EDIT: the code works perfectly with LAPACK 3.6.1 but not with MKL 11.3. I sent a post on the Intel Support forum, but it still has to be approved by moderators..

Solution

You are linking the ILP64 version of MKL but that one is used for 64-bit integers. You are using integer*4 so your integers are 32-bit (and the default integers as well). Use the MKL link advisor to link MKL correctly. You likely need the LP64 version.