Kahan on creating IEEE Standard Floating Point
Вставка
- Опубліковано 5 лип 2024
- William Kahan, winner of the Association for Computing Machinery's A.M. Turing Award, describes his leadership of the process to create the IEEE 754 Standard for floating point. The standard incorporated key ideas developed in his then-ongoing work with Intel to produce what became the 8087 chip. This clip is taken from an interview conducted with Kahan by Thomas Haigh for the ACM on 12 March, 2016 in Berkeley, California. Video of the full interview is available as part of Kahan’s ACM profile at amturing.acm.org/award_winner....
These interviews are priceless pieces of history. Thank you for preserving them for future generations.
It's hard to comprehend how important and influential Prof. Kahans work on floating-point arithmetic was. The fruits of his work are now everywhere, touching almost everyone's life in one way or another, mostly going unnoticed.
Thank you for the finite precision that we know today. YES there would be concerns, but that's what we get when we tend TO FIT INFINITY to 64 bits!
We got through it, but this was not a very smooth innovation. 1 | 8 | 23 is a totally weird bit layout. On the other side, the importance of this to numerical and scientific computation was immense.
What about the more precise BCD arithmetic?
I would assume it's much slower.
@@OpenGL4ever It is but a bee's-dick slower.
Maybe this is what the engineers of the time were thinking. Each digit uses 4 bits (nibble), but then we would not have a fixed number of digits and it would be harder to achieve what was possible with float and double, systems that used a fixed number of bits. I think the key idea that helped adoption was that in engineering they use significant digits anyway... After those digits they do do not care about precision.
And to make worse BCD also had different encodings...