Decimal Exponent Floating like in JOSS 348

Terje,

It really depends on the details of the respective representations. If you have a bit somewhere that says whether it is an integer or FP and the two are handled very differently, then you are right. However, this doesn't have to be the case.

Suppose for example the exponent is biased such that with a zero exponent, the value in the mantissa is interpreted as though it were the integer that it plainly is, then integers would simply be a type of unnormalized FP. This is basically what Burroughs did in their mainframes back in the 1960s and 1970s. Now, add a bit to indicate whether the mantissa is precise (ala '754's stated presumption) or whether it is +-- 1-2 LSB, and you have an integer representation that will cost NO additional time over what would normally be required for FP, except that it would have to be normalized when adding a FP number to it to keep the significance correct, a problem that Burroughs didn't deal with because theirs was not a significance system.

The concept of "type" was obsolete before C was ever even conceived, as the Burroughs design pretty clearly demonstrates! Types are useful for programming small antique microcomputers but have no place in modern CPUs, except for the fact that there is still a large dose of antiquity in "modern" CPUs.

It is only "just for fun" until it develops beyond current systems, then it becomes a proposal for people to throw money at.

Steve Richfie1d