swarmlang/swarm

Swarm is still a work-in-progress and is frequently in a broken or half-functional state. For a rough list of ongoing and future work, see TODO.md. For a tutorial on the features of the language that should be working, see the swarm documentation

Swarm is a parallel/distributed programming language with modular, swappable parallel drivers. It is intended for everything from multi-threaded local computation on a single machine to large scientific computations split across many nodes in a cluster.

The Swarm language provides built-in shared variables with guaranteed access safety. It abstracts parallel and background execution so the programmer need not reason about it explicitly.

Architecture

The Swarm programming language is parsed into the Swarm AST using a Bison-generated parser. This AST is then compiled to SVI, Swarm's virtual ISA. The SVI executes on the Swarm Virtual Machine (SVM), which provides the abstractions for parallel & asynchronous function calls, shared variables, and synchronization.

Examples

Here's a contrived example of a sum computed in parallel:

enumeration<number> valuesToSum = [6, 4, 55, 21098, 38, 5858, 33, 4, 7, 58];
shared number sum = 0;

enumerate valuesToSum as value {
    sum += value;
}

The SVM unrolls the enumerate block into 10 jobs which are executed in parallel. The writes to sum are synchronized automatically.

For more complex and less contrived examples (written in both the Swarm language and SVI), see the demo/ directory.

Building

To build Swarm, you need clang, make, bison, and git.

# initialize and build external libraries
make binn

# build the production binary, `swarmc`
make

# build the debugging binary, `swarmc_debug`
make debug

# run the test suite
make test

License

This project is licensed under the terms of the MIT license. See the LICENSE file for more information.