Original ticket http://projects.scipy.org/numpy/ticket/362 on 2006-10-22 by trac user bgranger, assigned to unknown.

On AIX, UFUNC_CHECK_STATUS (in ufuncobject.h) always return 8 indicating that an INVALID floating point error has occured. This is happening because the fp flags are not being cleared. The current code tries to clear the flags with the call:

fp_clr_flag( FP_DIV_BY_ZERO | FP_OVERFLOW | FP_UNDERFLOW | FP_INVALID);

But this doesn't work as expected. Instead, after this call is made the FP_INVALID flag is set to true so all further calls to UFUNC_CHECK_STATUS return 8. The correct way to clear the fp flags on AIX is:

fp_swap_flag(0);

the AIX documentation:

http://publib.boulder.ibm.com/infocenter/pseries/v5r3/index.jsp?topic=/com.ibm.aix.basetechref/doc/basetrf1/fp_clr_flag.htm

Makes only a vague reference to this being necessary, but apparently it is.

Could we fix this before 1.0? Thanks!!!!