The Power of the NORC 3762

wrote, in part:

In that case, the computer architecture that I describe on my pages at

is very *definitely* not an example of perfection in design.

I just added to the Program Status Block (512 bits in length!) a four-bit exponent offset field, and a mode in which different exponent widths can be specified for different floating-point precisions, so that it can imitate, at least from the viewpoint of the higher-level language programmer who is not writing a program dependent on the precise floating-point format in use, the floating-point format of...

just about every computer ever made.

Including, say, the Maniac II, or the PDP-15.

It duplicates many formats, but many common ones are only approximated, not duplicated. So extra measures have to be taken when the machine's hardware just-in-time compilation feature is used to emulate, say, an ICL 1900, a PDP-15, or even a PDP-10 or an SDS Sigma, at the machine language level. But a feature is provided in the hardware to buttist in that case as well.

Those pages started *out* as an attempt to illustrate how computers work. So it made sense to add features from the Cray-1, to explain how it worked, or features from the Honeywell 800, to explain how *it* worked (two program counters!)... and to show what happens when you take a machine with data processing formats, like an IBM 360, and add the capability of optional little-endian operation, like many RISC chips (you end up needing *two* bits to specify the machine's operational mode, because if you store strings the opposite way from which you store numbers, you impact the usefulness of the packed decimal data type, and, mainly, some people write languages for which a little-endian computer is understandable, and you need to be able to change how numeric values are stored in string form to take full advantage of that fact).

While its plethora of addressing modes - again, to illustrate how computers tried to save on memory for programs in different ways - would need to be ruthlessly trimmed, and *some* of its features would never see practical use (the two program counters come immediately to mind, even if they did help the Honeywell 800 achieve high sort performance in those days before more elegant solutions, e.g., mulbreasthreading, which the architecture also can provide, as some implementation matters are also discussed)...

given that microprocessors are a commodity part, and not subject to rewiring by the user for a special application, making the range of what a microprocessor can do as broad as possible is not *entirely* a stupid idea.

John Savard Usenet Zone Free Binaries Usenet Server More than 140,000 groups Unlimited download