The Power of the NORC 3757
in part:
In that case, you *will* know more about this sort of thing than I do...
CONTROL's program was functionally equivalent to an empty program. It was simply the optimizer to work it out. Oh you had not seen a similar...
The 6, yes. The 4, 7, 9, and 15 can be left out.
Generally, in terms of architecture, of course you're right. But the floating-point format of the 6 and 10, rather than one from the architecturally-closer 5-8 or 4-7-9-15 series, is that which the floating-point format of the PDP-11 most resembles.
Of course, IBM did have other choices.
It could have gone to 48 bits, or even just 40 bits, for something greater than 36 but divisible by eight.
It's true that some floating-point formats were limited to 10^38; on the other hand, Cray's floating-point format used a 16-bit exponent.
Yes, but that's hardly an excuse for forcing the programmer to waste time on a double-precision operation where it isn't needed.
Of course, the 8087 solved that problem; while a 360 performed a single-precision floating operation in less time than a double-precision floating operation, the 8087 just did everything internally in its 80-bit format.
John Savard Usenet Zone Free Binaries Usenet Server More than 140,000 groups Unlimited download
Alt Folklore Computers from Newsgroups