Where should the type information be 343
Peter, et al,
It's possible that that was mine, although I did it at the University of Alberta, not the University of Arizona. We had a new fancy smart terminal hooked up to the PDP...
Peter, et al, Here in the real (backwards) world, things have been divided into: 1. Programs, that have no metadata (and which was what...
I'm not sure anyone else sees it, but this discussion is on-topic ...
But of course IEEE-754 now draws a needless barrier and effectively says that any information buttociated with a measure is to be represented and computed in a particular (known to be rather defective) manner. Yes, IEEE-754 WAS useful when it was first proposed, but that was then and this is now. This should be just another part - the lowest part of the data structure design. Sure, we could probably specify a fixed ~128-bit representation that would show the full exponent range of variable-length exponent representation, maintain significance information, show the dimensionality, keep some guard bits, etc., so there would be no need to design the FP for the application, and maybe that is the way things will be in another few decades of HW development.
However, for the present we are in "the middle" between having more HW than we need for a bare bones FP while wanting more than a bare bones FP; and having the HW to throw at the problem to eliminate the need for designing the FP to fit the application.
Steve Richfie1d
Where should the type information be 344
Alt Folklore Computers from Newsgroups