A BEAM-native systolic array simulator.

ex_systolic is a clocked spatial dataflow simulator: explicit time (ticks), explicit data movement (links), and local processing elements (PEs). Data pulses through a grid of simple processors in a regular rhythm, like blood through a heart.

This is not a spreadsheet engine, a DAG executor, or a reactive system.

Architecture

Every tick follows the Bulk Synchronous Parallel (BSP) model:

┌─────────┐  INJECT   ┌─────────┐  READ    ┌─────────┐  EXECUTE
│  Input  │──────────>│  Links   │─────────>│   PEs   │──────────┐
│ Streams │           │ (FIFOs)  │          │ (step)  │          │
└─────────┘          └─────────┘          └─────────┘          │
                                                              │ COLLECT
                                                              ▼
┌─────────┐  RECORD   ┌─────────┐  WRITE   ┌─────────┐  ┌─────────┐
│  Trace  │<──────────│  Clock  │<─────────│  Links  │  │Outputs  │
│ Events  │           │ driver  │          │ (FIFOs) │  │  map    │
└─────────┘           └─────────┘          └─────────┘  └─────────┘

All backends (interpreted, partitioned) execute the same six-step sequence. The partitioned backend parallelises step 3 (EXECUTE) across tiles but preserves determinism via the BSP barrier.

Tutorial & Examples

The full tutorial, quick-start guide, real-world examples (image convolution, shortest paths), architecture diagrams, roadmap, and references are in the README.

One-minute quick start

alias ExSystolic.{Array, Clock, PE.MAC}

array =
  Array.new(rows: 2, cols: 2)
  |> Array.fill(MAC)
  |> Array.connect(:west_to_east)
  |> Array.connect(:north_to_south)
  |> Array.input(:west, [{{0, 0}, [1, 2]}, {{1, 0}, [3, 4]}])
  |> Array.input(:north, [{{0, 0}, [5, 7]}, {{0, 1}, [6, 8]}])

result = Clock.run(array, ticks: 5)

Module overview