Generic CPU discussion

Before it was IEEE754 it was the floating point implementation he specified for Intel's 8087 floating point co-processor for the 8086 family.

Professor Kahan's work was a marvel for accuracy improvement, but was a commercial disaster for Intel.  The 8087 was an enormous chip, thus terribly expensive to build, and the market did not much thank Intel for the fine contribution it made to numeric accuracy.

Personal footnote: the 8087 was designed at Intel Haifa design center in Israel, and I flew to Israel a couple of weeks after the birth of my first daughter in July 1979 to look over the 8087 design for interface compatibility issues with the 8087 (of which I was one of the four design engineers).

Only at the 80486 generation were the terribly complex requirements of Kahan's version compatible with a reasonable silicon implementation.  But we have all benefited ever since.

Wow, I didn't know that the 8087 was the reference model for IEEE754. Interesting to read about your experience from first hand knowledge. Thanks for that.

At that time, I think, the "real" computers were still built in New England, for example the series of minis DEC PDP-11/ -15, -10, VAX, with significantly more memory, wider data words (16, 18, 36 bits), more registers. That have been the work horses for scientific computing, development of UNIX, C, or the ARPANET. Meanwhile, microprocessors initially enabled cheap home computers (Homebrew Computer Club etc.) and office computers. The 8087 was not necessary for the intended purpose of IBM's PC, AT, XT. Amazing that it was Intel of all the companies who implemented this first precisely (deterministic) calculating FPU and set the IEEE754 standard. You would have expected that from one of the big players back then.

As a teenager, I was fascinated by using computers to graphically display fractals (e.g. Mandelbrot set)... these beautiful mathematical structures. On my first 8 bit home computer (some Zilog Z80 replica) without FPU and programmed in BASIC it took almost a week of computing time. After optimizing critical repeated computations using external libraries which emulated required floating point computations directly in assembler (reading lots of magazines, books, adapting examples before), it was still 24..48 hours. Sometime when I was 16 I got my first PC, an i486DX2/66 MHz with integrated 8087 FPU. That was a revelation. The same graphics took seconds with hardware FPU and one could create much more detailed graphics within minutes. Back then I was using Borland's PASCAL compiler (learned at school). There you could easily insert inline assembler code into PASCAL code to replace individual loops or sub-routines to call your own ASM code instead. So you haven't to use Microsoft's Macro Assembler. The 8087 instruction set is very clear and easy to learn which you can't say from 8086/286/386 (if compared to e.g. Motorola 68000: clear by design). The critical inner loop of the Mandelbrot iteration is quickly replaced by 8087 assembler (stack use planned on a sheet of paper, a few ASM lines looked up in a book). And wow, what a speedup. That was really cool back then.

It is also interesting that the 8087 internally uses wider registers for extended 80-bit floating points. There are no rounding errors unexpectedly piling up if you actually use 64 bit (double precision) floats in your higher level code - an ingenious design.

We knew what you meant from context. :-)

Even if not intended for the general market, or a new market for home PCs, the 8086 was a substantial game changer. However my very first PC ( after a Commodore 64 ) was an 8088 based machine which I looked upon as a thing of wonder. It did consume a decent amount of a pay packet though and my wife thought me mad, but at least I was not down the pub wasting it on beer was my weak justification. It was a time of my life when I hardly slept anyway ( hospital resident doctor ), it consumed what little spare time I had. I think I first played a game called Impossible Mission. Then I was introduced to programming it, Basic I think was my first try and then later assembler, C and much later C++.

Now where would the internet be now if it were not for a plethora of devices derived from the 8086, and it's initial opening up of a new niche for such a product type ? Computing for the masses followed from IBMs crucial decision to have an open hardware plan ( compare with Apple here and boo if you like ).

I for one am pleased that some one of your calibre participates here.

BTW does anyone remember the Byte magazine ? I have a copy of one devoted entirely to the 8086/88 and it's ecosystem. I have one book full of detail about the ROM code for boot up and another the machine code. The list goes on .... but it was way ahead of the DEC PDP-11 in it's accessibility and with regard to that : myself and a colleague from the Engineering Faculty got into decent strife for using one, well not illegally per se, but inappropriately. We flogged this text based game call Adventure over several late nights ( if you must know, you have to throw the rabbit at the dragon to win ). We were dressed down by our respective heads of faculty, but became minor legends on campus for logging our time against other user's accounts. For shame. ;-)

Cheers, Mike.