I am able to test your program with IFort 11.1.070 (MKL 10.2.7), 126.96.36.199 and 188.8.131.52.
Only the oldest of these aborts inside MKL during the first call to DSS_SOLVE_REAL(), with an unhandled exception. The later compilers (with later versions of MKL) build EXEs that run to completion. Is it feasible for you to update your MKL installation?
Perhaps you have some VS settings that are culpable. It is also possible that there is a hard-to-reproduce bug that surfaces in your runs but not mine.
Here is the transcript of my "debug" compilation and run.
S:\lang\MKL>ifort /Qmkl /Od /traceback /extend_source MKL_problem.for Intel(R) Visual Fortran Compiler XE for applications running on IA-32, Version 184.108.40.206 Build 20140726 Copyright (C) 1985-2014 Intel Corporation. All rights reserved. Microsoft (R) Incremental Linker Version 12.00.30723.0 Copyright (C) Microsoft Corporation. All rights reserved. -out:MKL_problem.exe -subsystem:console -incremental:no "-libpath:C:\Program Files (x86)\Intel\Composer XE 2015\mkl\lib\ia32" MKL_problem.obj S:\lang\MKL>mkl_problem S:\lang\MKL>
Luis, I have created a simpler reproducer and posted it, see https://software.intel.com/en-us/forums/topic/535447 . The difficulty with this is that the bug does not always occur. When we compile with different compiler options, then run it and experience a crash, we tend to see the options as the probable cause of the crash. However, what I found is that a single EXE, when run multiple times, will run fine and produce correct results most of the time, but will crash now and then.
I think that we have to rely on the Intel people to help with this problem, because it occurs deep inside the MKL library.