Category: Computers

Since the last post, I have begun working on inputs – switches, in particular. The input subsystem is much simpler than output, at least for things like switches and console input- there is no need for any kind of common FIFO or arbitration. Instead the module IBM1410_UART_INPUT_SUBSYSTEM receives data from the FPGA development board UART, and then places that character onto one of N (currently 0..7) input FIFOs (one each for receiving data for switches, console input, card reader, and so on), and N write data flags.

Then, each input module (for example, module IBM1410_CONSOLE_SWITCHES_RECEIVER) reads data off of its FIFO, and does the appropriate thing with that data. Specifically for switches, this means receiving the entire vector of switches into a temporary vector, and once fully received, replacing the one that the 1410 itself is using.

Testing the various switches has led to some interesting discoveries:

• Adding switches made a number of multi-page combinatorial loops pop up, that were previously hidden as a result of optimization by the Xlinx Vivado toolset. The 1401 compatibility switch was the second (and I hope last) of the loops involving large numbers (more than 300) gates. I could have change the design rule check to ignore this issue, but I decided instead to modify the SMS generation application to look in the logic block Notes field for “DFLIPFLOP” causing it to add a D Flip Flop after the identified logic block.
• I like to be able to run the FPGA CPU on its own. This means that it’s initial switch vector (the one that gets copied into when an entire switch vector is received from the host support program) needs to be reasonably initialized. To do this there are a few SWITCH_VECTOR(….._INDEX) => ‘1’ entries along with an OTHERS => ‘0’ at the end to set things up right. This fact is what actually led to the replace with a buffered copy concept described above. (The first time ANY switch is activated on the PC host program, the entire switch vector gets replaced by the transmitted one; indexing in hardware is hard. 😉
• While testing the “DISPLAY” function of the console’s MODE switch, when one presses START, a “D” should appear. However, currently, instead an “F” with invalid parity currently appears. This is occurring because the printing of “B” is also enabled, causing a “pick” of the “2” bit. It looks like my ALD is out of date – the “B” should really be being suppressed under these conditions – in other words, a hardware “bug”. Sure hope there are not too many of those! A similar thing happens with the MODE switch set to C.E. A “#” should be printed, but instead I get a “.” – same bit pick, for the same reason.
• Pressing CHECK TEST SWITCH #3 (and holding – which the C# user interface host program makes easy by holding on a click, and then releasing on the next click), and then pressing START should light up all the PROCESS CHECK lamps. This works fine for check test #1 and #2, but for #3 it does not light up the ADDRESS CHANNEL error as expected. This scenario uses the DC SET input to set the flip flop, but it appears that there is a signal using collector pull down causing the flip flop to reset (DCRFORCE in the SMS gate’s module) at the same time – and that reset it taking priority in the VHDL. There are not too many of that particular gate type in the system, so I may try reversing the priority of those things, so that a DC SET takes priority and seeing if that breaks anything else.
• The SCAN GATE switch on page 14.17.18.1 (part of the ADDRESS STOP circuitry) has a note that says it is a “CIRCUIT OPENINING” switch – in other words, all outputs are -36V except the selected position. In most cases in the IBM 1410, a connected switch means ‘0’, that is, -36V, so this behavior is (to be verified) a natural outcome of how I implemented switches – I “NOT” switches in the VHDL all – because an “on” switch electrically connects to -36V. For this switch, a given unselected position has to be a ‘1’, which is then “NOT-ed” in the VHDL to a ‘0’ – which then forces the output of the gate it is connected to to a ‘1’ – taking it out of the filtering criteria for address stop – it’s scan position can be either on or off. This, along with a failure to initialize the SCAN GATE switch and the address top address switches had me tied up in “NOTs” for a little while (pun intended.)
• On this same page there are a couple of optional SYNC COND inputs – at least one of which need to be tied to logic ‘0’ in order for address stop to work. They were intended to give CE’s a way to do an address stop on arbitrary conditions inside the machine.

Having the 1410 sending output to the console a few months back was great, but lots more is left to do. Over the past few months I have been working on a couple of needed capabilities:

The console is only one of many devices that will need to send output out from the FPGA development board to a support application. Others will include lamp display information, tapes, disks, printing and punching. In order to make that happen, I needed to do several things:

• There needs to be a FIFO, since at times multiple devices may need to send output to the console support program, and they are not “aware” of each other – they could conceivably overlap in activity. Also, some devices (like lamp output) would not necessarily be constrained to IBM 1410 cycles – they could generate output more quickly. That means that some kind of FIFO is required.
• Since devices could possibly send at the same time. There needs to be some kind of arbitration to decide which device has priority and gets to go first.
• The arbitration and transfer of a character to send to the support host would not be instantaneous, which means that each device that can generate output needs to have at least a single character buffer, and some kind of signal to tell it whether that buffer is full or empty.
• A multiplexor to select which source is currently providing the next character to be placed into the FIFO to send out.

These capabilities were created in VHDL in the module IBM1410_UART_OUTPUT_SUBSYSTEM, which is comprised of up to eight (currently) “requesters” (that one character buffer and flag signal, so one for the console, for now – but eventually more of them), the “arbiter” which decides who goes first, the “mux” (the multiplexor), “mux to fifo” which manages the transfer of characters from the multiplexor to the fifo, and “fifo to uart” which manages the transfer of characters to the UART when the UART is available to send more characters.

I am also in the process of modifying the SMS HDL Generation Application to generate vectors for lamp bits for transmission to the support host program, and switch bits to accept switch information from the host program. The lamp vector will be sent out from the FPGA periodically (maybe as often as 20 times/s) via a state machine to display lights on the emulated console in the support host program, and the switch vector provides a convenient means to number each switch for the purposes of host/FPGA communication.

Eventually there will be a serial input subsystem as well, to support the aforementioned switches, as well as input devices, like the card reader, tape, disk and so on.

Having completed the pass through the ALDs, the next thing I undertook was some cleanup work. The goal was to be able to generate an HDL “group” that contained ALL of the usable HDL pages.

The first step of that was to mark pages I didn’t want as no HDL generation – a relatively recent addition to the application. Then I went through and found all of the switches, and regenerated the affected pages to use update code that used notations in what is ordinarily the signal level fields of the ALD logic blocks associated with the switches (but show up as thinks like DK 2 – meaning switch deck two of a multi-layer rotary switch). I also had to fix up the associated test benches and groups to use the new signal names.

Then I took a shot at generating the whole shebang. I ran into three issues. There were a couple of pages I had skipped that had to do with auto-restart. Dealing with those was simply a matter of marking them to not generate. Next, I ran into an issue with page 13.72.06.1 – which turned out to be a missing page having to do with compute disable cycles for systems with “fast” file (disk) systems. Finally, the ALDs had several cases where the same IBM 1415 console lamp location was specified. I separated them arbitrarily – I don’t seem to have any specific information to help me decide.

Simulation Begins

With those fixes I was able to begin simulation. At first the power on reset wasn’t working right – the 20ms single shot on page 12.65.01.1 wasn’t triggering. This is really a simulation issue, I think – putting in a 30ns wait at the beginning of the simulation dealt with that.

The next issue was that after the reset, pressing the start key did not start the clock – the STOP LATCH stayed on. That in turn kept the clock from starting (ILD Figure 1). That in turn led to noticing that there was no START KEY PULSE when the start key was pressed in the simulation.

Part of the issue was tracked down to the fact that I don’t yet have a console printer emulation in place, and the console contact signals were inconsistent (both C1 and C2 normally closed and normally open signals were false), so I “fudged” it so the cam contact signals would indicate it was not in a cycle. That helped – and was the sort of thing I expect might need attention.

The next issue with the START KEY PULSE signal was that the CONS STOP PRINT COMPLETE signal was 0 – as though it was in the middle of a console stop print out. That was traced down to the Console STARTPRINT switch (page 44.10.01.1) which is unusual in that its N output (usually reserved for an active HIGH signal) is active LOW in this instance. That got me to the point where the START KEY PULSE appeared, and the machine went from Logic Gate B (perhaps it should have been at A?) through Logic Gate E, and which point Master Error was asserted and things stopped.

The error lamps were multitudinous, but the lamp that was first in line in the time line was the B Register Reset Error. The B register gets the contents from memory, and if it does not reset before then, that is an error (otherwise it might return the previous character, or some combination of the previous character and the current one).

Have I ….

By now, if you have been following along, you know what comes next. Yup. Triggers. In particular, card type DEY, used for generating the clock pulses and also for the logic clock (A, B, C, D, ….). Based on the waveforms I saw in the System Fundamentals manual, I had supposed that the AC input signals for set and reset would remain at logic zero until the gate had been turned on, and then pulsed. However, that turned out to be wrong, and it caused the system to advance into Logic Gate B immediate after reset, which caused the missing reset of the B register, among other things. What the system actually does is that the AC Set signal may already be at logic one when the gate on signal comes along – but since the AC Set signal is capacitively coupled, in that case the AC Set does NOT trigger the trigger. That does not happen until, while the gate is on, the AC Set signal transitions from zero to one.

The CPU Executes Its First Instruction

Upon fixing that, a whole bunch of stuff worked. Upon power up restart, the memory address register – the MAR – and the IAR – are already set to 1. So when START is pressed, that is the character read from storage. The instruction I am force feeding to it is the HALT instruction. Then the IAR gets incremented by 1, loaded into the MAR and the following character is read out (and, since I am force feeding it, the character is exactly the same). Since it has a word mark (WM), the system knows that this is a plain halt, and, lo and behold, the system halts.

This success, though, means that I have to go back and test (though not regenerate) all of the pages that use card type SMS_DEY. They should all be fine.

The IBM 1410 executes its first halt instruction

Well, actually, not all of the SMS_DEY usages were fine – there were four test benches with issues. In each case the AC Set or AC Reset signal was changed simultaneously with a collector pull off signal (DCRFORCE in the VHDL) which caused problems given how I coded SMS_DEY.vhdl. Adding a slight delay between the two (which is realistic) fixed the test benches.

Try and Fit onto the FPGA

With that, I made a couple of tweaks to a copy of the Integration2_tb.vhdl to become Integration2_fpga.vhdl and tried to synthesize it. There were a few issues:

1. A couple of signals had no drivers because they did not appear as outputs from the sheet: -S F CH FILE OP is used on 12.12.67.1, ostensibly coming from 13.74.04.1 . 13.74.04.1 has the latter ECO (251849 vs. 251824). Also, the card location chart on page 12.12.67.1 does not show the gates that appear on the ALD. For now, I changed 12.12.67.1 to use signal -S FILE OP instead, and changed 13.74.01.1 to output that signal to 12.12.67.1
2. Similarly, on the same two pages, 13.47.01.1 outputs -S FILE OP.D CY.EXTENSION but this got mangled into -S FILE OP.DLY.EXTENSION on page 12.12.67.1. It goes to the same gate as #2 above, so it may turn out to be bogus.
3. A signal +S BLOCK USE A CH WM is used on page 15.49.02.1, ostensibly originating from page 15.41.12.1. It does not originate there (or anywhere else). Here I changed the signal to “LOGIC ONE” on the destination page, and removed it from the origin page (which has the later ECO, at 253469 vs.251829).

Fixing numbers 2 through 4 was easy, if tedious – I also regenerated and ran the associated test benches. For a little while I thought #1 was just a matter of Vivado not recognizing the my signal named FPGA_CLK was a clock, but found it was indeed recognized. So I dropped that for easier prey…

There were a bunch of references in the Vivado synthesis log complaining about unconnected signal “GROUND”. This was because it appears as an input to sheets – most of them using it for capacitor input, or, in once case, a switch input that the generated HDL ignores in both cases. So, I changed the app to ignore that signal name, as it does with “LOGIC ONE” and “LOGIC ZERO”. Some of the triggers map a signal named GROUND to an *output* so that they can use it as a logic one internally – those I have left — for now. Interestingly, there were two more pages that had this issue that Vivado did NOT call out with a warning – but then failed when I regenerated IntegrationTest2.

The rest of the unconnected inputs I ignored for now – mostly they are cases where a bit vector is input to a page that does not use all the bits. Heck, that really should not be issuing a warning!

Next up was InvShiftRegister.vhdl. The way I had done the shift, one bit was “undefined” for a while – or conceivably could be. I changed the assignment and tested the affected pages again (39.10.01.1, 02.1, 03.1). No page HDL changes required for that. But what was weird was that this exact same assignment is used in ShiftRegister.vhdl, but there it did not generate the warning – but was not called out (see the discussion regarding GROUND, above, where the same kind of thing occurred).

Loop “D” Loops

Another issue that cropped up during synthesis, and became roadblocks when I tried to generate the bitstream for the FPGA involved a couple of apparent combinatorial loops. The application prevents those from being an issue on a single page, but there is no check across pages. Two such loops showed up, one involving +S CONSOLE PRINTER NOT BUSY and the other involving the A Channel WM bit.

It took me a couple of days to learn how to get the tooling to leave enough information in during synthesis so that I could see the problem. Setting the Xilinx Vivado “-flatten_hierarchy” to “none” made things much easier.

The console issue was easy to fix: I just manually inserted a “D” flip flop on page 45.50.08.1. The A Channel one was harder. The first time I looked at the schematic +S A CH NOT WM BIT was in the loop, so I put a “D” flip flop on that signal on page 15.39.08.1. But then the issue moved to +S CH WM BIT, so I put a second flip flop there. But then the issue moved again, this time to +B A CH WM BIT. Looking at the page I quickly saw that these were all fed from the logic block at 4D, so I took out the other two flip flops and put one at the output of block 4D, which cured the problem, both during synthesis and when generating the bitstream.

With normal settings, the design takes up 2858 LUT (look up table) cells and 1628 flip flops. With flatten_hierarcy set to non, it takes up 4001 LUTs and 1805 flip flops (two of which are the ones I added to fix the loops).

Doing some “greps” on the source, I found I had 1200 D flip flops that I had inserted (generated for on-page combinatorial loops — almost all latches), about 128 triggers, not counting the registers needed for implementing one-shots and delays, so the flip flop of 1807 count seems about right.

At one point the flip flop count had been as high as 2046 with a smaller LUT count, but while looking for the combinatorial loop issue I made a change to the VHDL for trigger card type DEY to bring what it thinks of as a DC Set or Reset into the section edge triggered by the 100MHz FPGA clock which actually made things smaller.

The FPGA on the Digilent Nexys 4 is an Artix 7 XCA100T, which can has 15,850 slices – each of which has 8 FF, and which has 101,440 cells. The report says I am using under 6% of the LUT’s and under 1.5% of the flip flops. I hadn’t realized this chip was as big as it is. I should have plenty of room for memory (BRAM with some ROM, too, maybe), channels, console, reader/punch and printer along with a debug core and either or a microBlaze co-processor or an I2C interface (or some such) to connect an Arduino to talk IP to my PC).

A Halting Start

Programming the results into the FPGA, with some code to map the START and COMPUTER RESET buttons sort of works – but unlike the simulation, where all that was necessary after a power on-reset was to press START, here I have to do START, then COMPUTER RESET, then START again – so something is not quite right (and it might be one of the other trigger types needs fixing). [The START KEY PULSE is indeed present during the first START.] Still, it apparently successfully executes the HALT instruction, which is a good step forward.

I also learned about a new tool: the Vivado Integrated Logic Analyzer, which can capture digital signals, and then communicate them back to the PC using a JTAG connection. Handy as heck.

End of Phase II

With the completion of Volume X, I have completed Phase II of this project. Phase I was to develop an application to enter data from the Automated Logic Diagrams (ALDs) and generate VHDL from that.

Phase II was to test the generated VHDL for each page, to make sure it was “reasonable”. Early on in the process, I took to using the Instructional Logic Diagrams (ILDs) to guide the testing – which was a tremendous help, though there were some errors and omissions in the ILDs. I also used the CE manuals, particularly the System Fundamentals manual, the Preliminary Core Storage manual and the IBM 1415 console manuals to guide testing as well.

Next Steps

1. Post the VHDL for the generated pages and test benches to a new project on Github
2. Update the Xilinx Vivado software on my machine – something I was hesitant to do mid-stream to either the latest release of Vivado, or their new tool, Vitis.
3. Create an HDL entity that combines all the pages, and throw that at the Xilinx toolchain to see how much space it will need on the FPGA – if it will even fit on the development board I already own.
4. Add a feature to the a application to assign each page to a subsystem, for collective/group generation, based on the existing group generation: console memory, registers, CPU Channels (A, B, C, D), CPU Instruction execution, and I/O Channels (E, F) are some possibilities. (Integration above that initial level will probably be manual)
5. Fix up any multi-deck rotary switches so that the deck is identified for all instances, so they can fit together. (For step 3 I may have to “fake” it.) Otherwise, when brining together pages with different decks of the same switch, they compete. The application has already been modified to do most of the work.
6. Decide if I am going to snag console characters internally, or via the actual IBM 1410 / IBM 1415 interface signals – hard to test an IBM 1410 without a console typewriter, and build a simulation to try that out. That might be enough to use for simulated testing of other components.
7. Translate the console interface onto a PC via embedded “IP” as they call it.
8. Do something about print/punch/card read/tape – probably based on the ILD’s. That may be a while – I can test an awful lot with just the console.

Things will proceed much more slowly now as I deal with holiday activities and some fun things I put on hold to focus on testing the ALDs.

A Special Latch

The beginning of Volume X is the B data register. This is actually 4 8-bit registers for a basic IBM 1410 – one for each 10K of core storage. These registers are a set of latches with open collector outputs, only one (for a given 10K) of which has its outputs (for data readout and inhibit, independently) enabled at any one time. Most latches are comprised of the same kind of gate, typically NAND and NOR. The DGM card type used for the B data register is specific to the B data register. But instead of being two NAND gates or two NOR gates, this one is an OR gate with a NAND gate which has an additional inverter (so, and AND gate), electrically. (The System Fundamentals manual depicts it as an OR gate followed by a second OR gate).

Internally, a diode OR circuit is followed by a diode AND circuit and two inverters after the diode AND circuit (T9 and T8). The output of T8 is the input to the first OR gate, as well as feeding two NOR circuits to gate the output. (The “OR” shows up as a “-A” logic block on the ALD and the “AND circuit with its two inverters shows up as a “-O” logic block. -A ==> +O ==> OR and -O ==> +A ==> AND). At least, that is how I modeled and test it, based on my analysis of the card’s circuits.

Inhibitions tied up in knots

I took way way too long to test the generation of inhibit signals out of the combination of loading a character into core from the assembly channel and any inhibit that would otherwise occur due to character regen (ALD pages 36.10.01.1 – 36.10.04.1, Info Input Gating)

The ILD depicted this equation: Do NOT inhibit if the output of the regen is a 1 bit OR the assembly channel is being loaded AND the assembly channel has a 1 bit.

This makes perfect sense. What got me tied up in knots is that all of the inputs are -Y inputs (inverted) and the output is also inverted. Normally I have dealt with that very easily, but not this time. I ended up thinking the logic through myself in order to realize what the (inverted) output ought to be given what the inhibit line actually does, testing that, and then, lo and behold, confirmed that it really did match the ILD.

More on the 1410’s Inhibitions

Page 37.11.01.1 – 37.11.04.1 also confused me for a bit. These ALDs take the inhibit lines and then gate them – but all with negative logic.

From a “what ought to happen” perspective, the inhibit line should only be driven if 1) writing the bit is to be inhibit (set to 0) AND 2) that the corresponding core plane area (0-5K or 5K-10K) is being gated.

I had initially entered card type, DFW, into the database as an “AND” because that is what the waveforms in the system fundamentals manual depicted. (In other words, inhibit the writing of a “1” if and only if conditions #1 AND #2 above are present.) On the other hand, the system fundamentals refers to this card as a -A (negative AND) gate, which is usually actually referring to NOR logic (IBM at this stage in time, disregarded the output inverter when considering these things).

However, when I looked at it, that confused me. That would mean that the -Y output line would be 1 (no inhibit) only if that bits inhibit input was off (logic 1 0 no inhibit) AND that section of core was NOT being gated. That made no sense to me. Furthermore, the ILD figure 49 is confusing, because it shows the inhibit output being an AND of NOT driving the OTHER section of core AND’ed the corresponding inhibit line for that bit. But it would not make sense to drive the inhibit line when no write cycle was taking place (no Z Pulse).

So, I analyzed the circuit. It starts off with a diode OR gate (anodes facing the input signals – so either one being logic 1 (0 volts) – results in the output junction between the two being close to 0v. This is then followed ay an inverter, turning it into a NOR gate – more or less standard SDTDL diode-transistor logic. But then that part is followed by an inverting power driver – so we have an OR gate.

Electrically, this means that the -V output is a 1 (NO inhibit) if either a) the input bit is a 1 [since in that case the earlier read of that location left it at 0 and we want it to change it] OR b) that section of core isn’t being gated (in which case we don’t want to be driving anything, including its inhibit winding).

The output is 0 (inhibit writing a 1 bit) only if a) the input bit is a 1 bit – the input is at a -V level, so negative voltage AND b) the corresponding section of core is being driven – this input is also -V level – so a negative voltage – resulting in a negative voltage on the driven inhibit output.

Conclusion: the system fundamentals diagrams with waveforms showing an AND gate from an electrical perspective is incorrect: this is an OR gate, electrically speaking (i.e., the waveforms are upside down) – which makes sense because this gate has -V inputs and outputs.

I don’t know if this really matters, because it will be needed to know which bits to write as “1” bits into the simulated memory if I use a memory structure like the actual IBM 1410. If, on the other hand, I just use the read/write signals and the data bits, the inhibit line signals won’t be used.

Here we go with Triggers — Again

Page 39.10.0.1 is straightforward enough – it is the memory read timing. When a cycle is started, first the B register is reset. 425 ns that reset signal is removed, and the Y Read line is driven. 280 ns after that, the X Read line is driven. 1100 ns after that (with some possible additional delays), the sense line is strobed.

All of this involves -yup, you guessed it – triggers – and on a new card type, DEM. I set that up based on triggers I had already done – very little effort. I chose the inputs and outputs based on the little timing diagram and block diagram shown in the system fundamentals manual. Below I also have noted whether a given signal is on the “ON” side or “OFF” side.

Pin	Side	System Fund (DEM)	Schematic (DEM)
A	ON	On	On
B	ON	AC Set	AC Set
C	ON	Gate Off	Gate Off
D	ON	DC Reset	DC Reset
G	OFF	AC Reset	AC Reset
P	OFF	Gate On	Gate On
Q	OFF	Off	Off
R	OFF	DC Set	DC

One thing is odd right away: The Gate Off signal is on the “ON” side but the corresponding “AC Reset” is on the OFF side, and vice versa. That is clearly not correct. But because this page and the following two – the only places where this card type is used in the IBM 1410 – all have both gate signals tied to logic one, that of itself made no difference.

Now, compare that to what we see for card type DEZ, which has a common AC Set line, but otherwise has a very similar circuit (cross connected PNP transistors) but with a -6V to 0V swing for card type DEM. Here I ordered them by function for easy comparison with the DEM table, above.

Pin	Side	System Fund (DEZ)	Schematic (DEZ)
D	ON	On	(Not labeled)
R	N/A	AC Set/Reset (Common)	AC Set
L	ON	Gate Off	Gate Off
P	OFF	DC Reset	Reset
Q	OFF	Gate On	(Not labeled)
C	OFF	Off	(Not labeled)
K	ON	DC Set	(Not Labeled)

Notice that DEZ has the negative going DC reset on the OFF side on the schematic. This makes sense – the DC reset is an active negative signal. It turns the corresponding transistor ON, causing its output to be pulled close to 0V (logic ONE). Same issue with DC Set. It also has the Gate signals on the opposite side, as those signals are positive going, and so turn the corresponding PNP transistor OFF resulting in a logic ZERO on that side.

At first I tried tried switching the assignments for pins A and Q (the On and Off outputs). That almost worked – the circuit is essentially symmetric. However, this memory timing page has a little wrinkle: if the address is more than 39,999 the memory timing is suppressed in the B frame, because memory 40K and above is in the Z frame. It does that via a pull-off condition on the trigger at 5F via the output of the block at 5G being logic ONE. That meant that “Q” had to be the “OFF” output. So, in the end, I flipped the outputs back to their original pins, and swapped the various inputs from side to side (fixing the obvious gate on/off and AC set/reset signals to match), ending up with the following, which passed the test I had coded based on the ILD perfectly.

Pin	Side	My DEM Implementation – corrected	Schematic (DEM)
A	ON	On	On
B	ON	AC RESET	AC Set
C	ON	Gate Off	Gate Off
D	ON	DC SET	DC Reset
G	OFF	AC SET	AC Reset
P	OFF	Gate On	Gate On
Q	OFF	Off	Off
R	OFF	DC RESET	DC

So now we have the negative going DC Reset signal on the OFF side where it belongs, and the positive going Reset signals on the ON side where they belong.

Here is the resulting timing diagram during simlation. The test code does a Computer Reset at the beginning of each iteration of the test 0 the first 3 do not set the conditions to start a read cycle, but the fourth one does. This matches up reasonably well with the timing diagram shown on figure 48 in the ILDs. (Not shown is the suppressed read with the Ten Thousands position 8 bit set.)

Note the long long delays here – about 425 ns for the B register reset (and activating the Y Read line), another 280 ns before activating the X read line, and then 1100 ns before deactivating the X and Y read lines. At 10ns per cell, that means 42 + 28 + 110 shift register cells. The Sense timing will add another 310 to 580 + 425 ns, and the write cycle will add 150 ns + 280 ns + 1200 ns – the grand total being 474 cells. Hopefully, I will have room without having to fudge them to preserve “slices” (each slice contains 8 flip flops and there are over 15,000 slices available in the chip on my development board.)

Oh, and have I … never mind. 8)

Reset and what was that?

Page 39.10.13.1 is pretty simple: a negative going read condition signal goes through two delay lines of 310ns and 0ns (which for now I have modeled as 10ns) to the AC set input of a trigger.

But there is a quirk: The first delay line is non inverting fed by a negative going input that goes through two inverters to get there – so it starts out as a 1. But the delay line initializes itself to 0 – it takes 310 ns for it to reflect that original input. During that time the output goes through the second delay line and manages to fire the trigger and activate the sense strobe (say what?). In the real world that would just result in garbage input to the sense amps.

The fix, in the test, was to wait for that all to settle out before activating the computer reset signal, but it took 15 minutes to figure out what was going on. In the original machine there is a delay of 25 microseconds between power good and issuing the system wide reset, which would take care of this just fine in the real world.

One issue I have seen in all of the memory timings is that the setup time for the triggers is quite long, and so pulses end up being longer than they should be. The Sense pulse should be somewhere just about 500ns according to the waveform diagram on ILD figure 48 and the . Mine is 525 ns. I am going to have to watch the relationship between some of these and the logic gates to make sure they don’t last too long or start too late.

With this, the memory pages are done, leaving just the IBM 1415 Console – the “fun” there will be the Selectric I/O portions.

The “Ghosts” in the 1410 Mode Switch

The IBM 1410 Mode switch had 6 labeled positions: C.E. (at the bottom), I/E Cycle, Address Set, Run, Display and Alter. Naturally there were 6 detents on the rotary switch that accompanied those positions.

Now, when entering the 1410 bootstrap, the usual sequence was to set the mode switch to address set, press Start, enter 0000 with the Load switch to +1, turn the Mode switch to display, press Start, enter 00000, wait for a bunch of “bbbbbbbbbb” to type out, press margin release (which allowed one to then enter multiple word marks later), turn the Mode Switch to Alter, press Start, enter the tape boot sequence (A WordMark+L % B 0 0 0 0 1 2 $ WordMark+N), turn the Mode switch to Run, press Computer Reset and then Start.

Note that I never mentioned pressing the Stop button. Well today, in looking at the ALD, I noticed that the switch actually has twelve positions (one of which is not actually shown on the ALD as it should be on block 5E). Every other switch position generates signal “-Y CONSOLE MODE SW STOP POS”. So, whenever the Mode switch is turned between two positions that are labeled (and have detents) it passes through another un-detented “ghost” position which activates a Stop condition. Clever.

Fits with Switches

Switches have been kind of problematic all along. Most of them signal active low when they are ON – so much so that rotary switches take that as their default. But momentary, toggle and “ALT” (which act like toggle) switches are mixed bag.

On top of that, switches, especially rotary switches, have multiple decks, and it is important to keep the switch positions aligned, bit wise, to make life easier later. So it took a bit to track down the switch decks on other pages and make sure they were in sync. Fortunately, only one of them was not: the Check Control switch deck on page 13.42.10.1 .

This sheet has toggles set up three different ways: The “Write Inhibit” toggle (which is actually unlabeled) creates an active LOW signal (logic 0, negative voltage) when OFF. It is the only switch in the bunch which works with the default settings. The “Auto Start” toggle creates an active LOW signal from the N pin when ON – in other words, active low. The “Inhibit PO” (print out) and “Asterisk Insert” switches create active LOW signals from their T output when OFF – also a sign of active low. It took me more than an hour to get my head around all of the toggle switches.

Page 40.10.03.1 as a bunch of switches. Most of the rotary switch decks and even the toggle switches (except 1 of each switch type) on this page have the grand task of turning on the “OFF NORMAL” lamp when any switch is not in what IBM was deemed the normal operating position. Two of the rotary switch decks (address entry deck 2 and storage scan deck 3) connect up in a kind of “and” gate via an isolating diode.

The Off Normal Lamp is also tricky: it lights when it’s input is logic 0 – being driven directly by the switches which are connected together. This ended up being the first “DOT” function that needed to be something other than AND or OR (besides the fact that it is fed from a mixed bag, so the HDL generation code punts that to an OR). What it in fact needs to be is a NAND: any logic 0 input (negative voltage) lights the lamp.

All told it took me more than four hours to get code generation patched up to segregate switch decks based on the info in their logic blocks (which IBM puts in the input and output level fields (which normally contain things like “S”, “B”, and “Y”)), fix other pages, write the test code and get my understanding of the switches operations to get it all working right.

I’d bet they will give me fits when I try and integrate multiple toggle switch decks together, too.

Fun with Latches

Most of the latches in the IBM 1410 are pretty straight-forward: the left side is usually the “Reset” side – a zero on an input to that side usually resets the latch, and the right side is usually the “Set” side – a zero on an input on that side usually sets the latch. Of course, one ought to avoid having 0 inputs on both sides, which would result in undefined behavior.

But sometimes latches can be reset another way: by having all ONE inputs on the SET side – which results in a zero on the reset side. Such is the case with the Console Stop Cond Latch on page 44.10.01.1. There is a “DOT” connection on one input of the “set” side – which happens to also be the output of the reset side. Two signals can reset the latch: Program Reset or the Inhibit Print Out switch. Easy enough. But testing this page got tricky because that DOT-ed connection is also an output (+S CONS STOP PRINT COMP COND).

When I first started testing, I just made that the NOT’ed output of the latch – since it was the output of the reset side of the latch. HOWEVER, that particular signal is also asserted if either of the two reset lines mentioned above is asserted – regardless of the state of the latch. Took me waaaaaaay too long to figure that out until common sense kicked in: that signal should be asserted if either the latch is reset OR either of the reset lines is asserted, stopping the printout.

“There You Go Again”

The quip from Ronald Reagan seems to apply to triggers. In this case, card type TAM, on page 45.20.02.1 – the console Y matrix (state machine) for dealing with 5 character addresses. In this case, the card type has its own VHDL file to implement it, because it is a trigger. Generally, I derived those from the logic and timing diagrams in the System Fundamentals – which for TAM is figures 113 and 114 on page 95. But something seemed odd, looking at the following connections for Y2, using those figures as a guide

Gate	Signal	AC Pulse	Signal
Gate ON 1	Y1	AC Set 1	MX Y Drive
Gate ON 2 (Pin F -really E)	MX ADDR DRIVE	AC Set 2 (Pin E – really F)	Y1
Gate OFF 1	Alter . Y2 . MX X6 OR Y2 . MX X1A	AC Reset 1	MX Addr Drive
Gate OFF 2	Set Y3 (MX X6 . Y2) OR (Not Alter . Not Read . Not Write . MX x6 . Y2)	AC Reset 2	MX Y Drive

As I was trying to make my tests work, I actually constructed the table above. It seemed “off”. The AC inputs are capacitator coupled (and not just for speed up – there is no parallel resist0r). Checking the Y3 trigger showed the same pattern, so it wasn’t just that the ALD had them reversed. Then, reading through the description in the IBM 1415 console manual, S223-2648, there was no hint of this situation.

Finally, I looked at the circuit diagram. Sure enough, E is the GATE input and F is the AC Pulse input – the System Fundamentals manual had reversed them. A quick use of my application’s card report revealed that this page was also the first time I had seen Gate On 2 / AC Set 2 used.

Fortunately, trivial to fix: all I had to do was go into my application card pin editor and swap the HDL signal name mappings for pins E and F (ACSET2 and GATEON2).

Have I… (“well…. never mind” — Paul Simon).

The 1415 Console I/O Selectric

Triggers and Latches and Cams – Oh My! (And latches driving triggers, driving latches – yeesh). Lots going on here, that made for some slow testing.

I will have some choices to make as to where I decide to “pick off” console output characters – either at the BCD level or emulate the various cams and solenoids. Of course, eventually I’d like to do the latter – frankly, I am a bit surprised that nobody has done that before. The I/O Selectric interface for inputs from cams happens on various pages, but the output magnet selection can be found on ILD figure 85. But for expediency, during my early testing I may decide to do it at the console character level.

However, while entering page 45.50.06.1 I found an error in the ILD. The IBM 1415 console manual, S223-2648 and the ALD agree that magnet R2A is “picked” when the input bits include either the 8 bit or not the 4 bit. The ILD incorrectly shows it being not the 8 bit (incorrect) or not the 4 bit (but the little chart at the bottom right of figure 85 does show R2A correctly.)

More errors were found on the ILD while working on page 45.50.07.1. For solenoid T1, the logic diagram shows a term in the equation of not 4 bit. The ALD and the table show that as not the A bit. Similarly, the logic diagram shows not 8 bit as a term for solenoid T2, but the ALD and the table shows that as not B bit. Finally, the ILD and the table on the ILD show the C bit as a term in the equation for the CHK solenoid, however the ALD and the IBM 1415 Console manual show this is really not the C bit.

The rest of the pages for the IBM 1415 were uneventful.