Little Endian 1326

I can't speak to how or by whom, but why I can explain.

Many computers before the PDP-11 were "little-endian" to a limited extent.

For example, the Honeywell DDP-516 and Honeywell 316 computers, when they operated on 32-bit integers, stored the least significant 16 bits in the word at the lower address.

But when character strings were stored in 16-bit words, the first of the two characters stored in a word was still stored in the most significant part of the integer.

The reason for storing the least significant part first - or at the point to which the address points (the big-endian IBM 705 addressed strings representing zoned decimal numbers by their units digit, which also had zone bits indicating the sign) - is so that addition can start with the first fetch from each operand, as carries propagate forwards (and has been noted on the thread).

The PDP-11 had byte addressing, inspired by the IBM 360. It was desired to make it elegant and consistent; it was a sharp break with past DEC designs, like the PDP-8 or PDP-9. The Data General NOVA was similar to a design that was rejected for it. If DEC had wanted to make a minicomputer that was "more of the same", it would have come up with something like a Honeywell 316 or an HP 2116. But the PDP-11 is not at all like that.

But it was still a mini - it couldn't afford the extra gates that the IBM 360 used with abandon. So it had to fetch the least significant part of an integer first.

How to avoid the ugliness of having characters snake their way through integers? If going from MSB to LSB is out in long integers... how about going from LSB to MSB?

The PDP-11, envisaged as a modern architecture, a clean break from the past, had to be elegant and symmetrical. But it also had to be economical.

And, thus, the idea of being *consistently* little-endian was born. The National Semiconductor 16032-32016 was a much later embodiment of this.

John Savard