cc/gcc: fix 128-bit long doubles option

Spotted by Arnaud LACOMBE:
http://sourceware.org/ml/crossgcc/2010-10/msg00122.html

Signed-off-by: "Yann E. MORIN" <yann.morin.1998@anciens.enib.fr>

 # Compute IA-64-specific values

 CT_DoArchTupleValues() {

     # The architecture part of the tuple:

     CT_TARGET_ARCH="${CT_ARCH}"

 }