[camperman]

I interviewed Steve Furber for a magazine feature about the ARM many years ago and he confirmed to me that Sophie pretty much designed and implemented the whole ARM instruction set in her head. It worked first time when switched on.

[talideon]

The way I hear it[1], she did the instruction packing on a pub napkin, while IBM were devoting quite a bit of computing horsepower to do the same thing.

[1] Vague recollections of an article in Acorn User. I'd really have to dig out those old magazines to be sure.

[camperman]

That wouldn't surprise me at all. And the really staggering thing is how little the ARM's design and instruction set has changed since then. The fact that you can use Peter Cockerell's 1980-something tutorial on ARM ASM even today shows how well it was designed. I put it up there with such quality-without-a-name pieces of engineering like the SR-71, the HP-12C and the Leica M3.

[engi_nerd]

The SR-71 was so far ahead of its time, it's a little ridiculous. Look down/shoot down capability in the mid 1960s. Synthetic aperture radar mapping. An IMU/astronavigation system almost as capable as the early GPS units of a few decades later. An airframe that actually becomes stronger the more you fly it as designed. Thermal management system techniques that weren't used elsewhere for decades. I can keep going. But basically, for almost every single system in the aircraft, the next comparable system you can find was developed at least 1 to 2 decades later.

[camperman]

Agreed. One of the finest pieces of engineering ever.

[talideon]

Oh, without a doubt. ARM assembly is a joy to code, much more so than any other. It just feels... right. There's no fluff and cruft and, at the same time, it didn't feel like there was anything really missing. RISC processors have a reputation for poor code density, but the ARM, even when it came out originally, was the exception to that rule: it was pretty easy to get code density for the ARM comparable to that of the 680x0 and x86.

It's a wonderful achievement.

[camperman]

Except division of course ;) But then we would have been robbed of those hours poring over the latest 5 cycles per bit division algos.

[tempodox]

...designed and implemented the whole ARM instruction set in her head...

Wow!

With today's CPUs you probably wouldn't stand a chance, no matter how good you are.

[kjs3]

On the contrary. Just because x86 is a baroque monstrosity doesn't mean simple, powerful ISAs are unpossible.

[tempodox]

I can only agree as to the baroque monstrosity. Let me put it another way: Could she do it with comparable ease on today's ARM CPUs?

[kjs3]

Since it's the same ISA, one would assume so.