Lot 02 · Shift · In design
Barrel shifter and the multiply loop that stretched my brain.
Rotate, logical/arithmetic shift, and a priority-encoder-optimized multiply—built from roughly forty 74ACT157 2-to-1 muxes stitched into a 32-bit barrel.
A naive add-and-shift multiply would burn ~32 cycles for 16×16. Tomato’s loop uses a priority encoder to skip entire chunks of zero bits in the multiplier operand. Worst case drops to 16 cycles; best case zero; amortized average about 4 cycles per operation. The ALU was the deep combinational problem—this lot is sequencing plus a very wide mux tree made tolerable in 74ACT.
Role in the machine
The shift board feeds the datapath beside 07_alu: barrel operations for shifts/rotates, and the multiply FSM that reuses the adder through controlled shifts. Control lives on shift-mul-control.kicad_sch with the barrel on the main shift sheet. Local EEPROM decode (same 10-bit opcode philosophy as every other slice) keeps propagation predictable at bench clocks.
- Status
- In design · schematic started
- Key ICs
- ~40× 74ACT157 · priority encoder glue
- Multiply
- 16 max · ~4 avg cycles
- Project
hardware/kicad/boards/02_shift_encoder/
Related journal: Multiply engine · Mul/div dispatch.