I want to compute the generalized EVs of the matrix pair (A,B) (e.g. general symmetric real 3x3 matrices).
Therefore I wanted to use the xGGEV (the DGGEV for instance). I am using the LAPACK routines in C++.
I have recognized, that if I activate the floating point exception handler (by
- Code: Select all
feenableexcept(FE_DIVBYZERO | FE_INVALID | FE_OVERFLOW);
using the fenv.h lib) that I ALWAYS get immediately a floating point exception after calling the LAPACK-routine.
If I deactivate this mechanism then everything works fine.
My question is now: Does this anything have to do with the IEEE arithmetic which the environment has to be capable of (i.e. NaN and Inf arithmetic)?
I found a comment on this in the LAPACK user guide (Further Details: Floating Point Arithmetic - http://www.netlib.org/lapack/lug/node74.html ).
Is anyone experienced in such a problem?
Regards, Erwin
