Significance Arithmetic and Unnormalization 717

On 22 May 2005 18:18:44 -0700, I wrote, in part:

And here is something else I've turned up in my research.

In the beginning, there was the IBM 704 computer, which was wrought from vacuum tubes, and which did have a 36-bit word length.

It had a floating-point format consisting of a sign bit, an eight-bit exponent in excess-128 format, and 27 bits which did contain the same information as the mantissa of the number's base-2 logarithm would contain, which we shall therefore call a 27-bit mantissa though the purists declaim.

And it begat the IBM 709, and the IBM 7090, and the IBM 7094.

And, lo, it came to pbutt that the Digital Equipment Corporation of Maynard, Mbuttachussets, when they wrought the PDP-6 computer of discrete transistors, would also use the very same floating-point format, with the exception that negative numbers were two's complemented, even as the philosopher Max Palevsky would later do unto the IBM System-360 floating-point format in his Sigma computers.

And, thus, when the Digital Equipment Corporation came out with their famous PDP-11 computer, they decided to also use an eight-bit exponent in excess-128 format, so that numbers might have the same ranges on both the PDP-11 and the PDP-10 (which was a successor to the PDP-6). As it had a 16-bit word, two of which made 32 bits, being smaller than 36 bits, they decided to get one of those bits back.

And this idea was so brilliant that it became part of the IEEE-754 floating-point standard. Which used an excess-126 exponent for no doubt some very good reasons.

But this means that the IEEE-754 format actually has an ancestry that goes past the PDP-11 *all the way to the IBM 704*!

John Savard Usenet Zone Free Binaries Usenet Server More than 120,000 groups Unlimited download