Decimal Exponent Floating like in JOSS 110

At first, I thought that there wasn't that much in common. I was providing an architecture which included a number of features that were on old computers and which could be useful for efficiency on some problems, but, not claiming any expertise in numerical analysis, the upper limit to any features in that respect which I provided was IEEE 754 - and that I only provided for compatibility.

Yes, I did include my own conceit of "extremely gradual underflow", but that was to extend numeric range.

But I am now making *major revisions* to my architecture, thanks to one of your posts. I hadn't realized that unnormalized floating-point arithmetic could be used to allow multi-precision floating-point arithmetic; it wasn't on a 360, but I can add "multiply extensibly unnormalized" and "divide extensibly unnormalized" instructions to make the necessary additional information visible when required; and I had forgotten that unnormalized floating-point was useful for keeping track of significance - and is NOT provided with IEEE 754, since that format suppresses the first bit.

The problem is that I have been using most of the available opcode space; but I can add some additional "short page" modes with plenty of opcode space, and make suitable revisions to opcode translation, and have room for unnormalized floating-point instructions. They won't work with all the defined floating-point modes (they are incompatible with suppressing the first bit of the mantissa, and they are also incompatible with Extremely Gradual Underflow since denormalization affects a number's value when that is used), so that will be explained.

John Savard