I'm working on a Motorola CPU, but this may be just the ticket. If it works for 80->64, I'll see about inverting it as well. An Intel "extended double" is a valid form of IEEE754 extended-double-precision, which requires at least 80 bits to hold it. The 128 bit "long double" that Sun uses is another conforming implementation, as that standard simply sets minimum bit sizes for exponent and mantissa.

Thanks!