Are there any CPU's with less than 32 bit floating point hardware, triggering the error? There are lots without floating point hardware, but the software routines would emulate >= 32 bit.