http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22473
--- Comment #4 from John David Anglin <danglin at gcc dot gnu.org> 2012-01-14 04:54:23 UTC --- This is a binutils bug. Single word conversions need to use the 0E opcode rather than the 0C opcode that's currently generated for fcnv,t,sgl,uwd. This doesn't fail for PA1.1 because the instruction isn't available. On the 64-bit target, the right half of the floating point register is always used for word operands forcing the 0E opcode.
