It works!!! Eureka! And on the very first attempt, no less. I have achieved computation from a big mess o’ wires, and a couple of dozen basic logic chips. I can now say confidently that fibonacci(12) = 144. Check out the last line of the logic analyzer data listing:

Each line shows the state for a single clock cycle, with RESET (active low, so 1 means normal operation), OPCODE (in hex), and the X register (in decimal). Opcode FF means the machine has halted. You can see the last few terms of the fibonacci sequence on the preceding lines, although they’re not in order, and there’s a random value of 110 there too.

Did I mention that it works? Holy cow. The best part was that mere moments after the successful bootup, a friend called to ask me about something else, so I talked his ear off about the machine.

Since the hardware is still far from complete, running the fibonacci program required quite a bit of chicanery. At the moment there are only two 8-bit registers, and no RAM. The T register was intended to be used for temporary storage by the microcode, and wasn’t meant to be user-visible at all, so I had to add some additional instructions to expose it temporarily. I also added an instruction to add the X and T registers. Then I had to write microcode for a conditional absolute jump instruction, since the hardware needed for a relative branch isn’t finished yet. I had to modify the absolute jump instruction to work only within the first 256 bytes of memory, to avoid disturbing the T register. And finally, since I didn’t have any place to store a running count of how many fibonacci sequence terms had been generated, I resorted to cheating: the program terminates when the sign bit of the X register (bit 7) is 1. So it’s not really computing fibonacci(12), but rather the first fibonacci number >= 128, which happens to be fibonacci(12).

Here is BMOW’s first program:

Honestly, I’m fairly amazed that it worked on the first try. Yes, I’d been testing the subsystems as much as I could as I built them, but this was the first real integration test. What’s more, it was the first test of any kind that tried to modify the program counter, or use the ALU, condition codes, data registers, databus, or memory bus to data bus interface. I fully expected to spend a long time working out all sorts of problems before getting to the first successful program run. Heck, I must have run into 10 different logic and microcode bugs while testing the fibonacci program on the simulator, and the potential for errors in the impenetrable mass of wires the composes the BMOW hardware is far greater.

Here’s a look at the testing setup for my moment of glory:

So now I’ve got a very rudimentary computer, with two 8-bit registers and no RAM, running at a blazing 470 kHz. What’s next? I’ll probably write a few more test programs to exercise the hardware in its current state, to make sure everything’s really working as it ought to. Next, I think I’ll try to tackle integrating the LCD module. Checking the progress of the computer with all those logic analyzer probes is not much fun, so it would be great to display “fib(12) = 144” instead. Once I’m able to check the machine’s health without connecting up the logic analyzer every time, I’ll probably move on to the remaining data registers, RAM, stack pointer, and other hardware devices. There’s still a tremendous amount left to do, but as of today, I can finally say I’ve built a working homebrew CPU.