The cexp function has undue overflows arising from overflow in its internal call to exp. Example (tested x86): cexp (709.8125 + 0.75 * I) overflows, although both real and imaginary parts of the result are actually within the range of double. (Examples with somewhat larger real part of the argument can also show undue overflows where one part of the result should be finite and the other infinite but instead both are infinite at present.)
I'm testing a patch for this.
Author: Joseph Myers <firstname.lastname@example.org>
Date: Thu Mar 22 19:38:09 2012 +0000
Fix cexp overflow (bug 13892).