CDC 6600 emulation - was Re: How do they make Verilog code for unknown ICs?

[Paul Koning]

> On Jun 20, 2016, at 4:22 PM, Toby Thain <toby at telegraphics.com.au> wrote:
> 
> On 2016-06-20 4:17 PM, Paul Koning wrote:
>> ...A hardware model can
> be used to replicate what old hardware did; for example, I have a
> partial CDC 6600 model that shows how it boots, and that model includes
> propagation delays on some signals (which are critical to correct
> operation in certain spots).
>> 
> 
> This is probably of great interest to more than just me.
> 
> Any more details? Going to publish?

Sure.  You can see it at svn://akdesign.dyndns.org/dtcyber/trunk; most is in the vhdl subdirectory but it uses some pieces from the top level.  Very much work in progress.

It's derived from the 6600 wire lists on bitsavers.  I OCRed them (what a pain), wrote VHDL models (gate level structural models) of each 6000 series module, and a Python script builds a structural model out of those elements interconnected by the wires from the wire lists.  "Sufficiently long" wires have their delay explicitly modeled.

I currently have chassis 1 (PPUs) working, at least to the point where I can do a deadstart, have that load from the deadstart panel, and execute some instructions.  I've used it to grok some magic related to deadstart that's part of the lore but not documented.  I also have the 6612 display controller working, so I can see how the text display on the console tubes works.  Unfortunately I don't have a good SPICE model for the analog parts, so I don't yet know why the character shapes are changed by the circuitry in the manner that the photos show.

The CPU is work in progress.  I'm trying to get exchange to work.  It's hard; even more than the PPUs, the CPU relies on hairy timing in certain spots.  It appears that the wire lists are not all the same rev, and there are also some typos in them.

Some day, if things go really well, this may result in a gate level accurate FPGA 6600.  I'm thinking I'll stop trying OCR and simply type the wire lists instead; that's likely to be faster in the long run.

	paul