Джером Кунен на открытии памятной таблички IEEE для стандарта 754

[vak]

Про памятную табличку стандарта плавающей точки IEEE 754 я уже писал, а вот еще. Джером мне прислал текст своей речи. Мне кажется, имеет смысл запечатлеть для истории. Тут есть несколько ценных исторических замечаний.

Remarks at the unveiling of the IEEE Milestone plaque for Standard 754

(Jerome Coonen, 3 May 2023)

Thank you, Jim <Demmel>, for those good words <of introduction> and for explaining why it was all worthwhile. In this glittering array of deans, directors, chairs, and professors, I suppose I must represent the graduate student emeritus, having been fortunate to be in the right place in 1977 to be part of this.

I’ll add a couple of points.

First, the roots of this standard reach back long before the formation of the IEEE committee in ‘77. They lie in the decades of sometimes baffling arithmetic of the mainframes and minicomputers of the 50s, 60s and 70s.

In that era, Prof. Kahan made it his business in Toronto, Cambridge, Urbana-Champaign, Stanford, and ultimately Berkeley, to help uncounted engineers and scientists understand their malfunctioning numerical codes.

He and three colleagues advised IBM on fixes to the arithmetic on their new System 360. Replacement floating point boards were installed in the field, resolving at least the worst of the miseries.

One of Kahan’s first graduate students recalls that even as far back as 1964 in Toronto they implemented techniques for handling numerical over/underflow that would later appear in the 1985 standard before us today.

Through it all, Prof. Kahan gained an insider’s understanding of what he did and didn’t want.

Then there is IEEE’s tradition of principles above personalities. They wisely steer plaque citations away from personal credit. Sure, if you coin Moore’s Law or open Shockley Labs, you’re in. But there’s plenty of room for errors cast in bronze, or for omissions not cast. Better to avoid the risk entirely.

Yes, we have Prof. Kahan here, the godfather, and rightfully so. But it’s worth looking further, including some people not with us anymore, who helped all this fit together.

What’s so compelling about IEEE 754 is that, at the cusp of the transition to microprocessors, in one stroke it lifted the industry out of the chaos of the 70s and provided a solid platform that has flourished for decades and now motivates even 8-bit arithmetic, as we just heard. How does such a once-in-a-lifetime opportunity arise?

We can start with John Palmer, who joined Intel in 1976 with the mandate to design a floating point chip for the 16-bit 8086 in development.

Palmer got his Ph.D. at Stanford under Gene Golub, whose pal Kahan from graduate school days would regularly visit Stanford from across the Bay. Palmer knew just where to go for advice about floating point.

Why was Prof. Kahan in Berkeley, so convenient to Silicon Valley, not MIT or Carnegie-Mellon? Well, his friend Prof. Beresford Parlett <present> had actively recruited him to Berkeley from Toronto in 1969, as Computer Science emerged to be a department of its own.

Now, in 1977, with Palmer and Kahan already at work on the Intel 8087, Richard Delp at Signetics was inspired by some first-cut floating point chips from AMD. He approached IEEE standards maven Bob Stewart about standardizing floating point arithmetic on microprocessors while there was still time. The irrepressible Stewart (many of you know what I mean) brought the IEEE 754 committee into being, with Delp as chair.

Intel joined the standards committee with a head start. Their own corporate arithmetic standard was already well underway. Kahan requested and received Intel management’s permission to propose the design as an international standard. All he had to do was recruit some help at the university and write it up. <smile, it’s a joke>

One last link, and thankfully with us today, is David Stevenson, who took over as committee chair midstream. Dave helped navigate the years of discussions of competing proposals and science projects the standards committee attracted. You can Google “IEEE 744 milestone” for more of the tale.

Any way you cut it, the threads that run through this story are the experience, reasoning, persistence, and persuasiveness of Prof. William Kahan.

He applied that same diligence to advising students and colleagues about everything -- from auto tune-ups and use of electronic ignition to speaking clearly and writing the Dickensian way, and even how to dress <open jacket to reveal suspenders> or find a mate. Yes, I was on my own there but fared so well. I know I am not alone in acknowledging the impact on my career, on my life, of having dived into the IEEE 754 project.

The ultimate tribute to IEEE 754 is how little people know about it. Most of them don’t have to know – it just works. Why should there be much surprising about basic arithmetic by “electronic brains”?

Now, it’s finally time to see what lies behind Door Number One. I gladly pass the torch to Prof. Claire Tomlin, Chair of the Department of Electrical Engineering and Computer Sciences...

Comments

[juan_gandhi]

OMG. It's like a funeral.

[vak]

Все живы вроде, дай им бог здоровья. Лучше чествовать героев, пока они еще в доступе. 😀

[spamsink]

Да уж, speaking of persistence and persuasiveness. Кахановская paranoia, запущенная более или менее на любой pre-IEEE 754 машине, поносит реализацию плавающей точки последними словами. В вольном переводе, "как вы можете пользоваться этой бякой, выкиньте её скорее".

[vak]

Меня удивило про IBM System/360, что приходилось заменять платы плавающей точки прямо у юзеров. Настолько плохо всё было.