Chapter 39

SANDY - early FEB 62

Two days later, Paul called me into his office.

"I have a new job; I'm the engineering manager for one of the midrange NPL machines; as you know, NPL is the 'New Processor Line' that provides both scientific and commercial capabilities. I want you to work for me until you're done here in Poughkeepsie. Do you think that you could take the work you've done on the Reaper control word structure and work up an unclassified discussion of the techniques? There's a big push to use microprogrammed controls on these machines, and I remember you and Sarah mentioning the similarities."

"Yes, I can do that without a lot of trouble; we'll just say that the fields controlled different parts of the machine, without going into the actual functions. I think it was that structure that made Reaper as easy to build as it was; unlike Stretch, where there were a lot of sequencing controls, most of the sequencing of Reaper was in the program and not the hardware."

"Yes, I know that's some of the motivation for using a microprogrammed approach, along with the fact that it allows a lot of the design to proceed before the instruction architecture is firm. I'll get you a copy of the current architecture document, and have you added to the distribution list for the architecture memos. It looks like your wish to get some exposure to NPL is going to be granted."

Paul called a department meeting for late that afternoon, where he announced his new job, along with the fact that I'd accepted an offer to work with the IBM Space group in Bethesda. He also mentioned that most of the others in the department would be assigned to his new group. He finished by saying that the meeting was being adjourned to the Treasure Chest, and that he would buy the first round.

The next day, I started studying the architecture documents and a file of memos that Paul had managed to get for me. The concept of the NPL series was that there would be a common instruction architecture for all of the machines in the line, from the smallest to the largest, which was supposed to be about half as fast as Stretch. In theory, this would offer significant savings to IBM, in programming systems costs, and to the customers, who would be able to move applications to a larger machine with minimum disruption. The architecture was firm enough in several areas to allow some machine design to proceed. Memory addresses would be byte, not word, addresses. The machines would have binary fixed point operations, floating point operations, and two kinds of variable field length (VFL) operations: decimal and character. The length of a byte was fixed at eight bits. There seemed to be a heavy influence from Stretch, although I was glad to see that VFL operands started on byte boundaries and their length was measured in bytes, not bits.

The memos made it clear that much of the architecture, including the register structure, the character encoding scheme, and the details of memory addressing and instruction formats were still up in the air. I was able to get a short meeting with one of the architects, to get a feel for how it looked like these things would be resolved, and also sat down with Nils Johannsen, the CPU engineering manager, to get a rough idea of how he saw the machine being organized. I spent two days working on my writeup of the Reaper control word scheme, including the WXYZ branching mechanisms.

When I finished, I got back with Paul and gave him a quick summary. He made a few minor suggestions, and called a meeting with Nils and his group. I basically gave them some ideas of how to lay out control fields to enable the microprogram implementation of the instruction set to achieve maximum performance. One key feature was to have each microinstruction specify the address of the next instruction and use a subset of the WXYZ branching mechanism; otherwise any decisions would require at least one additional cycle for a conditional branch. We also discussed the advantages of being able to control and use different parts of the machine independently and concurrently, and the merits of having independent control fields for the various counters and byte selectors. Finally, we started to outline the scheme that we would want to use for fetching and decoding the NPL instructions. We came to the conclusion that we were going to need another kind of branching, where we could do a sixteen-way branch on four bits of the op code in order to do a partial instruction decode in one cycle rather than four. We decided that a workable solution would be to define a control point that would allow four bits of the next microinstruction address to be specified by the contents of one of the registers.

STEVE - early FEB 62

The full explanation of what the primary Reaper application was doing seemed to have inspired Lou and Vlad. For the next few days, they had enthusiastically been crawling through my code, and several times I had to explain what was happening to them. Finally one day, they came to me about a particular section of the code. Lou said, "Vlad and I have been puzzling over this for quite a while, and haven't been able to figure it out."

I studied the listing for a few minutes. Although I knew what I had been trying to do, the code didn't make sense to me.

"Well done. It looks like you have found an error that I made here." I explained what the code was supposed to be doing, and asked them to see if they could suggest how it should have been coded.

The next day, they came back with two approaches. Lou said, "Here's what I came up with, but I think Vlad's approach is better. What do you think?"

I looked at their work. Both of them had come up with excellent approaches. Finally, I said, "These are both excellent, but I think that if you combine these techniques and find a way to use the no-refetch flag, it will be even better. Let's keep all this handy, and once we get it working on the real machine, we can try each version and see which actually is faster."

They rushed off to work together on a combined approach, and I was able to get a few minutes with Sam.

"I think that Vlad and Lou are just about up to speed on Reaper programming by now; yesterday they flushed out a bug in my code. I think we need to do two things. First of all, we need to get them more exposed to the principles and techniques of cryptanalysis. We also should sit down with them and start planning the other applications that we would like to see running on Reaper. Once we get the first one in production, we could also see if there's a way to restructure it and make it more efficient, more effective, or both."

"That all sounds like a good plan. I also need you to start thinking about some procedures for scheduling the daily runs and getting the input data assembled. Talk to Murray about that; he has a tape librarian who could coordinate the process."

That evening, Sandy called me.

"Steve, love, do you get Lincoln's Birthday off?"

"No, it's not a Federal holiday, drat it."

"I do. Why don't I come down Friday and maybe you can go in to work late Monday morning, like last time?"

"Kitten, that's a wonderful idea. I'll plan on coming to Poughkeepsie for the second weekend in March. I'll take Friday the 9^th off, and we can get our marriage license."

"That sounds good. I'll call you when I know when I'm coming Friday. I know it's been less than two weeks, but I've been missing you."

"Dearest, only two more months. The machine is being delivered the day after you'll be going back, and I suspect I'll be busy for the rest of the month. Are you busy these days?"

"My new job with Paul is going very well. I wasn't sure what I'd be doing for the last two months here in Poughkeepsie, but this is perfect. I wrote up an unclassified description of our control word structure, and I've been using it as the basis for defining the microprogram controls for a new computer. It's a lot of fun!"

The next day, Murray told me that Samantha Ishido, a tape librarian, was being assigned to the Reaper coordination task, and had the necessary clearances. I went to the tape library, showed my badge, and went in. It was a large room, with bookcases and cabinets on all the walls, and a large table in the middle. A slender young woman with long black hair, definitely with Oriental blood, was sitting at the table, surrounded by stacks of magnetic tape reels. She looked up, smiled, and said, "Are you Steve?"

I said, "Yes, and you must be Samantha."

"Call me Sam. Everyone does."

I sat down at the table and started to explain what would be needed to coordinate the Reaper input collection and the runs. She took some notes, asked questions from time to time, and it was obvious that she understood it very quickly and would do an excellent job.

"Of course, these procedures will need to be modified as we get some experience. We don't even have the machine here yet, although it's being delivered next week. Then it will take me some time to get my program ready for production; I hope that's done before the end of the month. That way we should have all of March to get the process running smoothly. I'm getting married in early April, and will be gone for a while, although there are two other programmers, Lou Morrison and Vlad Hlavacek who will be also working on this project and will be taking over starting in April. Why don't you think about this for a few days, and I'll get back with you, Lou, and Vlad next week sometime. If you have any questions before then, please call me."

"I will do that, but right now it all looks pretty clear."

After I left Sam, I found Isaac and said, "Let's see if the room is ready for the new machine; it's supposed to be delivered Tuesday." We went to the machine room, and as we walked in, I heard a familiar voice saying, "Hola, Steve! It's good to see you again!"

I turned to the spot where the voice came from, and saw Miguel coming out from a room at one side of the main machine room.