/wizard-engine

Research WebAssembly Engine

Primary LanguageWebAssembly

The Wizard Research Engine

The Wizard Research Engine is a fully-featured WebAssembly engine (virtual machine) designed for teaching and research. Its implementation is designed to be flexible and easy to grasp, ideal for instrumentation, experimentation and modification. Built with the future in mind, it is written in a fast and lightweight, safe, garbage-collected programming language, Virgil.

Features

Wizard supports Wasm standard features, including:

  • Wasm MVP features
  • Multi-values
  • Reference types
  • Bulk-memory operations
  • SIMD

Wizard fully supports these Wasm proposals:

  • Exception handling
  • Tail-call
  • GC
  • Multi-memory
  • Function References
  • Relaxed section order

Other features under development:

  • Threads
  • Stack-switching

Wizard can run testcases specified in the .bin.wast format, like the specification tests that are part of the Wasm spec repo and proposal repos. Wizard supports a small embedding environment suitable for running simple programs. WASI support in Wizard is a work in progress.

Supported Targets

Because Wizard is written in Virgil, it runs on all the targets that Virgil currently supports, including:

  • x86-darwin : 32-bit Darwin kernels (MacOS)
  • x86-linux : 32-bit Linux kernels
  • x86-64-linux : 64-bit Linux kernels
  • jar : JAR files for the Java Virtual Machine
  • wasm : WebAssembly module for any Wasm engine (!)

In fact, because Wizard can itself be compiled to Wasm, it fully self-hosts. This means that Wizard can run a copy of itself compiled to Wasm!

Wizard has special support on the x86-64-linux target, with a fast, hand-written interpreter, nearly 40x faster than the simple interpreter, and as fast as interpreter tiers in other engines.

Design and Implementation

Wizard is simple! As opposed to engines focused on performance, Wizard is just a few thousand lines of code. Instead, its architecture is focused on flexibility and readability, making it suitable for language and VM research.

Wizard is implemented in Virgil, a fast and lightweight programming language. Learning Virgil is easy; you can pick it up in no time! Using Virgil allows Wizard to compile to a single native binary of just a few hundred kilobytes. Development with Wizard is very quick turnaround, as a full production build takes less than a second. Virgil is garbage-collected, thus Wasm proposals such as GC reuse the collector of the underlying language, which keeps Wizard small and easy to understand.

Wizard is the first Wasm engine to do fast in-place interpretation of Wasm bytecode. Wizard currently has two interpreters:

  • an easy-to-read, straightforward interpreter which runs on all the supported targets
  • a fast interpreter in hand-written x86-64 assembly, which only runs on x86-64-linux

Both interpreters execute bytecode in-place (no code rewriting) for simplicity and low memory overhead. Later, Wizard will have a baseline compiler and optimization compiler, to match production engines' performance.

Documentation

The most up-to-date documentation is, as always, the implementation in this repository!

See the Design for a closer look at Wizard's internals.

Read how to Build and Test Wizard.

Read how to use tracing options in Wizard.

Read how to use built-in monitors for analysis in Wizard.

Read how to build your own custom monitors for instrumentation in Wizard.

Research Projects

Currently, Wizard supports three closely related research projects:

  • Fast-int: the fast, in-place interpreter for Wasm
  • Stack-switching: a fast implementation of WasmFX for Wizard

License

Licensed under the Apache License, Version 2.0. (rt/LICENSE or https://www.apache.org/licenses/LICENSE-2.0)