GitHub

Manual Overview

This manual is split into three layers:

  • Overview (this page): mental model plus one runnable example.
  • Concepts and Semantics: detailed rules for effects, regions, and conflict detection.
  • API Reference: generated docs for all exported types, functions, and macros.

If you are new to Parables, start here, run the example, then continue to the Concepts page.

Who this manual is for

This manual assumes you are comfortable with Julia and want to:

  • express task-level parallel work,
  • make data dependencies explicit,
  • and debug scheduling behavior with predictable rules.

The mental model

Parables separates:

  • Task code: the computation you want to run.
  • Access metadata: what each task reads/writes/reduces.

From that metadata, Parables builds a dependency DAG and executes it safely with either a serial or threaded backend.

A complete example

using Parables

x = rand(100)
y = similar(x)
blocks = eachblock(length(x), 25)

dag = Parables.@dag begin
    for (i, r) in enumerate(blocks)
        Parables.@spawn Parables.@task "scale-$i" begin
            Parables.@access x Read() Block(r)
            Parables.@access y Write() Block(r)
            @inbounds for idx in r
                y[idx] = 2 * x[idx]
            end
        end
    end
end

# Optional: inspect the inferred dependency structure.
print_dag(dag)

# Run serially first for correctness/debuggability.
execute_serial!(dag)

# Then run threaded for parallel execution.
execute_threads!(dag)

Why these tasks can run in parallel:

  • Every task only reads from x.
  • Every task writes to a non-overlapping Block(r) of y.

If the writes overlapped, Parables would introduce ordering edges automatically.

Where to go next