yamt/toywasm

What's this?

A WebAssembly interpreter written in C.

Goals

• Learn the spec by implementing it
• Have a fun
• Correctness
• Clean code
• Small footprint
• Reasonable performance
• Portability
  • Core wasm: C11
  • WASI: POSIX-like
  • Clang/LLVM extentions are ok where an alternative implmentation in pure C is reasonably easy.

Non-goals

• Top-notch performance
• Stable API/ABI

What are implemented?

Feature	cmake config	Notes
WebAssembly		2.0 (Draft 2022-06-27)
extended-const	TOYWASM_ENABLE_WASM_EXTENDED_CONST
exception-handling	TOYWASM_ENABLE_WASM_EXCEPTION_HANDLING	See the top comment in insn_impl_eh.h
multi-memory	TOYWASM_ENABLE_WASM_MULTI_MEMORY
tail-call	TOYWASM_ENABLE_WASM_TAILCALL
threads	TOYWASM_ENABLE_WASM_THREADS
custom-page-sizes	TOYWASM_ENABLE_WASM_CUSTOM_PAGE_SIZES
wasi_snapshot_preview1	TOYWASM_ENABLE_WASI	See the top comment in wasi.c
wasi-threads	TOYWASM_ENABLE_WASI_THREADS
dynamic-linking	TOYWASM_ENABLE_DYLD
littlefs for WASI	TOYWASM_ENABLE_WASI_LITTLEFS

Note: As this runtime is relatively new, all proposals which had finished when I started this implementation are just included in the WebAssembly in the above table. It includes mutable-global, nontrapping-float-to-int-conversions, sign-extension-ops, multi-value, reference-types, bulk-memory-operations, simd.

Where can this run?

Platform	Tested on CI	Notes
macOS/amd64	Yes
wasm32-wasi	Yes (on toywasm)
Ubuntu/amd64	Yes
Ubuntu/i386	Yes	32-bit, smaller alignment
Ubuntu/arm64	Yes (on qemu)
Ubuntu/armhf	Disabled	32-bit
Ubuntu/s390x	Disabled	Big endian
Ubuntu/riscv64	Yes (on qemu)
wasm32-wasi-threads	No	Occasionally tested manually
NuttX/esp32	No	Occasionally tested manually
NuttX/sim on macOS/amd64	No	Occasionally tested manually
NetBSD/amd64	No	Occasionally tested manually

Use as a command

See toywasm command help message.

Usage:
	toywasm [OPTIONS] [--] <MODULE> [WASI-ARGS...]
Options:
	--allow-unresolved-functions
	--disable-jump-table
	--disable-localtype-cellidx
	--disable-resulttype-cellidx
	--dyld
	--dyld-bindnow
	--dyld-dlfcn
	--dyld-path LIBRARY_DIR
	--dyld-stack-size C_STACK_SIZE_FOR_PIE_IN_BYTES
	--invoke FUNCTION[ FUNCTION_ARGS...]
	--load MODULE_PATH
	--max-frames NUMBER_OF_FRAMES
	--max-memory MEMORY_LIMIT_IN_BYTES
	--max-stack-cells NUMBER_OF_CELLS
	--repl
	--repl-prompt STRING
	--print-build-options
	--print-stats
	--timeout TIMEOUT_MS
	--version
	--wasi
	--wasi-dir HOST_DIR[::GUEST_DIR]
	--wasi-env NAME=VAR
	--wasi-littlefs-dir LITTLEFS_IMAGE_PATH::LFS_DIR[::GUEST_DIR]
Examples:
	Run a wasi module
		toywasm --wasi module
	Load a module and invoke its function
		toywasm --load module --invoke "func arg1 arg2"

Use as a library

See example apps:

• Investigate module

• Load and execute WASI module

Toywasm provides cmake config files for its libraries. If your app is using cmake, you can use find_package to find toywasm libraries as it's done in the CMakeLists.txt of the above example apps.

Release binaries

Warning Toywasm version numbers are NOT a semver. It's just an increasing number, which doesn't imply anything about compatibilities or features.

• Release binaries are built with clang, with ThinLTO where it's available.

• For macOS, we ship a universal binary. (amd64 and arm64)

• We ship binaries with two configurations for each platforms/architectures. The assets with full- in the filename have most features enabled. Another one is with the default configuration.

• Each asset contains the toywasm cli command, static libraries, and cmake config files.

Build from source

• To build toywasm, we recommend to use clang for the following features, which some parts of toywasm codebase is kind of assuming to have. They are not required for the correctness though.

  • musttail attribute (LLVM 13 and later)

  • LTO

• For macOS and similar posix-like environment, you can build it with cmake in a usual way.

  % cmake -B build
  % cmake --build build
  

  For build-time options, see the ToywasmConfig.cmake.

  By default, it uses ThinLTO if available. You might want to use regular LTO instead to get a bit better optimization:

  % cmake -B build . -DUSE_IPO=OFF -DCMAKE_C_FLAGS=-flto=full
  % cmake --build build
  

  By default, it requires wabt and cmocka for tests. If you don't want to install them, you can disable tests.

  % cmake -B build -D BUILD_TESTING=OFF
  % cmake --build build
  

  Otherwise, after building it, you can run tests with:

  % cd build
  % ctest
  

• For wasm32-wasi and wasm32-wasi-threads targets, we provide a convenient script build-wasm32-wasi.sh to download wasi-sdk and build toywasm with it.

• For NuttX, it's probably easiest to use interpreters/toywasm in NuttX apps. There are a few sample defconfig with toywasm enabled, including sim:toywasm and esp32s3-devkit:toywasm.

How slow/fast is this?

• Computation benchmark

• Startup benchmark

Why is this slow?

• Unlike many of interpreters, toywasm aims to execute wasm bytecode directly where possible. That is, it doesn't "compile" wasm bytecode into intermediate code.

  Unfortunately, wasm bytecode is simply not efficient to execute that way. It's basically designed to be somehow "compiled" at the load time.

  Many of interpreters out there translate wasm bytecode to their internal bytecode for performance reasons. Wasm3 and WAMR "fast" interpreter work exactly that way. Even WAMR "classic" interpreter replaces some critical instructions in-place.

  While toywasm maps wasm modules read-only and never modifies them in-place, it still generates a few types of offline annotations on the bytecode to avoid being too slow. While they are smaller than a full translation, you might consider them a kind of translation. See Annotations for details.

• I don't like to use huge-switch statements or labels as values GNU C extension, which are well-known techniques to implement efficient interpreters.

Internals

• Annotations

• check_interrupt and restartable errors

References

• Wasm3

• WAMR "fast" interpreter

• labels as values GNU C extension