Skip to content

Design review, datapath cards (control-path pass) — 2026-09-21

One card of the 2026-09-21 datapath review: the ALU v3.2, findings A1-A3. The other cards, the system-level items S1-S3 and the summary: hardware/reviews/datapath-2026-09-21.md.

Second pass after mechanical-2026-09-21.md (mechanical). Cards: ALU v3.2, Index Register 1.1, Memory v1.3, Mem Switch 1.1, Mem Register 1.0. Method: every select / enable / strobe / clock / reset net was traced pin-to-pin from kicad/<ver>/reports/netlist.net (pin names and types from the Eagle symbols) with chip types from the Eagle .brd, and checked against the way the sequencer microcode actually drives the bus (firmware/microcode/ucode-generator2/), because several hardware choices only work under a microcode convention. Signal polarity: - prefix = active low. Gate delays quoted are LS-TTL typicals from the datasheets in docs/datasheets/.

Severity: HIGH = would stop the card working or damage parts; MED = unreliable / depends on a convention that is not enforced by hardware; LOW = housekeeping or documentation.


ALU v3.2 (hardware/cards/alu/kicad/v3.2/)

A1 — LOW/MED: BR-COND floats when JP1 selects -ALU-FUNC as the mux enable

IC26 (74LS251) G(7) = N$4 = JP1 pin 2; JP1 pin 1 = -ALU-FUNC, pin 3 = GND. Y(5) = N$56 -> IC27.10 (74LS86) XOR AC-LD-INV -> IC27.8 = BR-COND (X1.C24). With JP1 on -ALU-FUNC, N$56 is tri-state whenever -ALU-FUNC is high and has no pull-up, so BR-COND is whatever the floating LS input of IC27 reads. Harmless only if the sequencer's branch flip-flop is set solely while -ALU-FUNC is asserted (Sequencer notes: "SET with and of BRTest and BRCond"). With JP1 on GND the mux always drives. Bench: meter on C24 with -ALU-FUNC high and JP1 on pin 1.

A2 — LOW: -AC-LD-INV doubles as the branch-condition inverter

IC27.9 = AC-LD-INV (IC1.2 = NOT -AC-LD-INV). Any microcode line that asserts -AC-LD-INV for an inverted accumulator load also inverts BR-COND for that line. Only matters if the sequencer samples BR-COND then.

A3 — LOW: brief bus overlaps on internal buses at strobe edges

  • ACI-DATA: IC2 (74LS240, G = -AC-LD-INV) and IC3 (74LS244, G = AC-LD-INV from IC1.2) are complementary through one inverter; both drive for ~10 ns at each transition.
  • BDATA on -AC-RD release: IC10/11 (74LS245) DIR = -AC-RD flips to A->B ~8 ns after release while IC12 (74LS244, G = -AC-RD) is still turning off (~20 ns), and the 245 stays enabled because -ALU-IO-EN = IC7.11 = -ALU-FUNC OR -BUS-EN and -ALU-FUNC is not released in the same line (accumulator.c:350). Tens of ns, LS-tolerated.

Checked, no issue (ALU)

  • Accumulator IC5 (74LS374) CLK = AC-LD = NOT -AC-LD (IC1.12): latches on the leading edge; microcode sets the function one line earlier (accumulator.c:14-22). OC = GND: ACO-DATA always valid.
  • Carry flip-flop IC9A: D = N$7 = IC7.3 = CO/BO OR SHIFT-OUT; CLK = N$5 = IC6.8 = AND(N$3, N$9); N$3 = IC4.3 = NAND(-ADD/SUB, -SHIFT) (= "function is add, sub or shift"); N$9 = JP2 pin 2 = AC-LD per the note "should be ac-ld". The clock is gated by decoder outputs that are stable during the -AC-LD pulse (ALU code is set a line earlier), so no glitch; logic ops preserve carry as intended. CLR = -RESET, polarity correct.
  • CO/BO = IC1.10 = NOT(IC4.6 = NAND(N$15 = ADD/SUB, N$10 = IC27.6 = N$1 (IC36 C4) XOR SUB)): carry for add, borrow for subtract, only during add/sub — matches the v3.2 note.
  • Carry-in IC35 C0(7) = N$2 = IC27.3 = (ALU3 AND C/SHIFT) XOR SUB: plain SUB gets +1 (two's complement with IC33/34 XOR array on BDATA), ADC/SBC (ALU3=1) take the flag with the right sense.
  • IC24/IC25 (74LS85) cascade: IC24 pin 2 = GND, pin 3 = VCC, pin 4 = GND — correct LSB-stage seeding; A = BDATA, B = ACO; outputs to mux D1-D3.
  • IC8 (74LS138) is permanently enabled (G1 = VCC, G2A/B = GND); the eight function 244s (IC13/16/17/20/23/31/37) are mutually exclusive by construction and only feed the internal INV-IN bus.
  • Shift register IC29/30 (74LS194): S0/S1 = ALU0/1, CLK = SR-LD (leading edge of -SR-LD); IC9B (SHIFT-OUT) D = IC32.7 (mux of SRD0/SRD7 by ALU0-1) is clocked on the same edge and samples the pre-shift bit. Fine.
  • V1/V2 (M74HC4078): pin 13 is the NOR output (datasheet 744078.pdf: "all inputs L -> X(13) = H"), so N$6 / N$52 = "byte is zero"; N$55 = both zero. Matches ALUZ (4) and ALU16Z (6) in CodeGen.h.
  • Bus transceivers enabled only with -ALU-FUNC low: the microcode pairs -AC-RD with -ALU-FUNC everywhere it drives the accumulator onto the bus (accumulator.c:343, :377, :513; putBustoRegMem(reg,"-AC-RD") callers are inside blocks that already hold -ALU-FUNC). During -AC-RD, DATA8-15 carry $FF from RN2 (10 k pull-ups on BDATA8-15) through IC11 — a definition, not a conflict.
  • -VMA, -HL-SWAP are connector-only nets on this card (unused), as the note says.

Last updated 26 October 2026