agat-web

СМ6337

The printer as emulated, and where each detail came from. src/sm6337.js runs the factory firmware on an emulated КР580ВМ80А and turns the board’s outputs into a carriage position, a paper position and needles struck.

node tools/check.js sm6337                    # what the manuals say it does, asserted
node tools/sm6337.js --text 'Hello\r\n'       # print it, draw tmp/sm6337.png
node tools/sm6337.js --key ТЕСТ               # the switch report, its own oracle
node tools/cputest8080.js                     # the 8080, against four test programs

check.js sm6337 is the one to run after a change.

Sources

• РЭ — Дб3.043.005 РЭ, the СМ6337 / СМ6337.06 manual with schematics. §8.1 and рис. 8.1 are the control board, рис. 8.2 its address decoders, табл. 8.1 the ports, §8.4 and рис. 8.10 the mechanism board.
• РЭ.06 — 3.043.017 РЭ, the СМ6337.06 / «Искра» manual. The fullest command table, табл. 7.1.
• ТУ — ТУ 25-7112.0023-88: speeds, pitches, acceptance tests.
• The firmware — SM6337.D27 .D28 .D29, three 8K К573РФ6А, D27 at $0000.

All from oldpc.su/pc/sm/sm6337, and all scans with no text layer: every § and табл. cited here was read off the image, so a disagreement with the code is as likely to be a misreading as a bug.

The models

	paper	interfaces	firmware
СМ6337	A3	ИРПР-М, Стык С2	D27 D28 D29
СМ6337.06	A3	ИРПР-М	015М, 015-01М, and the same D29
СМ6337.02	A4	ИРПР-М, Стык С2
СМ6337.03	A4	ИРПР-М

The D27-D29 set serves both widths: bit 0 of port $80 is a jumper, the firmware takes a line of $0750 steps with it and $03C0 without, and its self-test signs itself СМ6337.02 without.

The control board, БЦУ-1

КР580ВМ80А at 2 MHz from a КР580ГФ24; Ф2ТТЛ clocks the timers too, so a T-state is a timer clock, 0.5 µs.

$0000-$5FFF	ROM. $3000 is the 9×11 font, 12 bytes a character; D29 is the 18×23 one
$6000-$77FF	RAM, three К537РУ10: receive buffer at $6000, line buffer at $7000, variables at $7400, stack from $7800

port
$40 $41	w	КР580ИР82, needles 1-8, D7 the top one, low strikes. The write is СТР.ПЕЧ, the fire strobe. The address bit is needle 9
$48 $49		КР580ВВ51А, Стык С2
$50-$53		КР580ВИ53 D17: 0 counts steps, 1 is the dot clock, 2 the step rate
$58-$5B		КР580ВИ53 D24: 0 baud rate, 1 bell pitch, 2 bell length
$78 $79		КР580ВН59, CALL table at $2E00
$80	r	bit 0 the A3 jumper, 1 paper present, 2 left sensor, 3 /STROBE held
$80-$83	w	ВМА0 = A0 carriage direction, ВМА1 = A1 paper direction
$84-$87	r	DIP switches: A0 reads S6, A1 reads S5
$84	w	ШАГ2, one paper step
$88	r / w	the ИРПР-М byte, and its acknowledge / clear a head fault
$8C	r / w	keys ПС ПФ ТЕСТ КР, low when down / “управление длинной строки”
$9C-$9F	w	“резерв” in the РЭ; $9D as the carriage ramps up, $9C at speed
$AC-$AF	w	lamps: A0 low lights АВР, A1 low lights КР
$BC-$BF	w	A0 lets the step rate through to ШАГ1

The groups from $80 up that are write-only latch address lines, not data: D25 is a К155ТМ7 with MA0, MA1 on its inputs. The firmware never loads A before those OUTs.

IR0	head fault, АВР.ПГ
IR1	SD1, the left sensor
IR2	D17 OUT0: the step count ran out
IR3	D17 OUT1 inverted (D7.3): the next dot column
IR4	a key
IR5	an ИРПР-М byte
IR6	a serial byte

РЭ §8.1’s prose has IR2 and IR3 the other way round; рис. 8.1 and the firmware agree with each other. The inverter on OUT1 is not decoration: on the other edge every line after the first starts with a stray glyph, because the IR3 handler stops the dot clock itself and the stop has to land between requests.

Inches, in millimetres

Every pitch in the manuals is metric and every one of them is an inch fraction: 2,54 mm is 1/10”, 2,117 mm is 1/12”, the 4,23 mm line is 1/6”, and АР2 A n steps “n x 0,353 mm”, which is n/72”. The mechanism is built to the inch and the documentation converts — the command set is Epson’s, and so is the geometry under it. The step sizes here are named in inches because the ratios are exact that way and ugly in millimetres.

The mechanism

Two four-phase 1.8° steppers, ДШИ-200-1-2. Phase switching is a counter and a PROM per motor on the БУМП-1 (РЭ §8.4.4, §8.4.6), so the firmware sends a direction and pulses and nothing else.

• Carriage: 1/120″ a step. D17 counter 2 makes the pulses, counter 0 counts them. Homing is 600 steps/s. A line ramps 400 → top speed in 32 steps up a table at $2C0D and back down.
• A stage of N is N steps. The firmware moves the carriage in stages on counter 0 and plans each as N steps: six spaces one way is a stage of 72, the same cells back a skip of 60 and a glyph of 12. The stage is N+1 pulses long, so the pulse that loads the count does not step.
• The dot clock is fixed, 2200 Hz, and pitch is carriage speed: 2200 steps/s is 12 steps a character, 2,54 mm; 1833 is 10, 2,117 mm; 1284 is 7, the narrow font. 2200 / 12 = 183 characters a second against the ТУ’s “не менее 180”.
• Quality print is two passes. §4.9’s 18x23 matrix against the 9x11 is twice the resolution each way: the dot clock runs at twice the step rate, so columns fall half a step apart, and the row is struck again two paper steps lower. Its glyphs come from D29, the 9x11 font from $3000 in D27.
• Paper: 1/240″ a step, 40 to the 4,23 mm line. АР2 A n feeds n × 240/72 steps, exactly — 6, 12 and 24 measure 20, 40 and 80 — so the firmware’s own arithmetic is 240 steps to the inch and nothing else.
• A form is a step count at $746D, and a form feed advances to the next multiple of it. АР2 C n stores n × the line spacing in force at the time, so later spacing changes do not move it (РЭ §7.9.40, and measured: 66 lines is 2640 steps and 11″, but 66 lines after АР2 A 24 is 5280). Power-on loads the ROM constant $0B64 at $024E — 2916 steps, 12.15″.
• Head: nine needles 1/72″ apart, energized for 290 µs by a one-shot on the БУМП-1 (РЭ §8.4.2). Printing is in both directions.

On the Agat’s cable

printer=sm6337 puts this printer on the printer card’s cable (SM6337.Cable), the wide machine with the A3 jumper fitted. It runs on the Agat’s clock: before every access to the card, and once a frame, it is run up to the Agat’s cycle at 2 MHz against CPU_HZ. /STROBE falling is a byte exactly when the capture counts one, and BUSY is the firmware’s D5.1. The cable carries no INIT, so resetting the Agat does not reset the printer. The printer, its paper included, carries over when the machine is rebuilt. A snapshot does not hold the printer’s state: a restored machine finds the printer as it was left.

check.js printer runs the factory’s ТЕСТ ‘ПЕЧАТЬ’ into it: the bytes are the golden ones and the sheet has the test’s five lines on it.

Measured

check.js sm6337 asserts all of this; it is 25 checks and about a minute.

• The 8080 passes TST8080, 8080PRE, CPUTEST and 8080EXM.
• Power-on: turns twice, comes to rest at the left reference, lights КР at 0.63 s, raises no fault — РЭ §7.7.
• ТЕСТ held at power-on prints «СОСТОЯНИЕ ПЕРЕКЛЮЧАТЕЛЕЙ РЕЖИМОВ РАБОТЫ» (РЭ рис. 11.1) in quality print, fifteen rows, the same every run, and exactly one row of it moves when one switch does.
• Pitch is 12, 10 and 7 steps for the three S6 settings, which is 2,54 and 2,117 mm and the narrow font.
• АР2 A n feeds n/72”, and a plain line 12/72” = 4,23 mm.
• A form feed advances to the next multiple of the form length, so pages come out of the strike stream without being told where they are. src/paper.js turns those strikes into A4, A3, 15×11 fanfold or a roll. src/pdf.js writes them out, a page at a time as each one finishes, so a hundred-page job costs the size of its output and not of its length.
• АР2 j n winds the paper back, 255 × 0,118 mm at a time — 283 paper steps — and nothing stops a program repeating it. Paper keeps the last few pages open for it and counts a strike that reaches further back as jammed.
• A line shorter than the one before, and a backspace, leave the next line starting in the same column.
• Direction follows §4.3. Text alternates; graphics prints one way only; АР2 U 1 asks text to do the same and АР2 U 0 gives it back.
• No chip feature is touched that the ВИ53 and ВН59 here do not implement.

Not settled

• Why the loading pulse does not step. On рис. 8.1 ШАГ1 is OUT2 gated by D30.2, and nothing there drops a pulse. At N+1 steps a stage the head comes home a step further left after every two-stage pass; at N pulses as well, the dot clock loses or adds glyphs. The likeliest place is how the ВИ53 loads a mode 0 count, where only the 8254 data sheet says anything.
• Where a dot lands. A strike is reported at the carriage position the needle reaches, interpolated between steps, plus flight microseconds of travel. flight defaults to 0, because the manuals time the energizing pulse (290 µs, §8.4.2) and the current rise (100-300 µs) but never the flight, and the gap it crosses is 0,30-0,45 mm (РЭ.06 п.12). 290 µs is its upper bound — a needle that has not struck by then will not — which at 2200 steps/s is 0,64 of a step, 0,135 mm, and twice that between one direction and the next. A scan of real output would settle it: print a vertical rule bidirectionally, measure the zigzag, halve it. The firmware does not compensate and does not try to; §4.3 prints graphics one way instead.
• Paper step rate. РЭ §8.4.6 says ШАГ2 is a constant 400 Hz; the firmware’s own timing here comes to 650. It loads counter 0 with 714 and counter 2 with 4, and 714 × 7 T-states would be 400 Hz exactly, so CLK0 may not be OUT2 while the paper moves.
• Why the factory chose $0B64 for the default form. Everything the printer does with it is pinned; the value itself is not a standard form length, is not a whole number of 1/6″ lines (72.9 of them), and the РЭ never states a default to check it against. Only intent is missing.
• Where the left sensor sits against step 0: the РЭ has the head come to rest right of it, and here it rests at 0. Nothing downstream depends on it.
• $9C/$9D and the $8C write: logged, not understood.
• The serial port is a stub, and nothing raises a head fault.
• The СМ6337.06 firmware has not been run, and the .06 schematic album (CM6337_06_sch.pdf, 17 sheets) has not been read.
• Reading the printed page back as text is most of the way there and not here: the printer’s own glyphs make the alphabet, but a row printed right to left sits a fraction of a step off one printed left to right, so cells have to be found per row. It would make the switch report an oracle in words.

This site is open source. Improve this page.