Clarify what function-parallel passes can do, and fix an asm2wasm bug (#1627)

The problem this fixes is that we made precompute look at globals in #1622, while asm2wasm was creating globals while adding functions and optimizing them - which could race. This was caught by threadSanitizer (with low frequency, so we missed it on the initial landing).

The underlying issue is that function-parallel passes should be able to read global state, just not modify it, and not read other functions' contents (which is why the Call node has a name, not a pointer to a function). This PR clarifies that in the docs, and fixes asm2wasm by not handling function bodies in parallel to creating globals.
3 files changed
tree: 5095d28a3d14a3d55cc86eeff1e79633f2f3f1c1
  1. bin/
  2. media/
  3. scripts/
  4. src/
  5. test/
  6. .clang-format
  7. .flake8
  8. .gitattributes
  9. .gitignore
  10. .gitmodules
  11. .travis.yml
  12. appveyor.yml
  16. CMakeLists.txt
  20. setup.cfg
  22. ubsan.blacklist

Build Status Windows CI


Binaryen is a compiler and toolchain infrastructure library for WebAssembly, written in C++. It aims to make compiling to WebAssembly easy, fast, and effective:

  • Easy: Binaryen has a simple C API in a single header, and can also be used from JavaScript. It accepts input in WebAssembly-like form but also accepts a general control flow graph for compilers that prefer that.
  • Fast: Binaryen‘s internal IR uses compact data structures and is designed for completely parallel codegen and optimization, using all available CPU cores. Binaryen’s IR also compiles down to WebAssembly extremely easily and quickly because it is essentially a subset of WebAssembly.
  • Effective: Binaryen's optimizer has many passes that can improve code very significantly (e.g. local coloring to coalesce local variables; dead code elimination; precomputing expressions when possible at compile time; etc.). These optimizations aim to make Binaryen powerful enough to be used as a compiler backend by itself. One specific area of focus is on WebAssembly-specific optimizations (that general-purpose compilers might not do), which you can think of as wasm minification, similar to minification for JavaScript, CSS, etc., all of which are language-specific (an example of such an optimization is block return value generation in SimplifyLocals).

Compilers built using Binaryen include

  • asm2wasm which compiles asm.js to WebAssembly
  • AssemblyScript which compiles TypeScript to Binaryen IR
  • wasm2asm which compiles WebAssembly to asm.js
  • Asterius which compiles Haskell to WebAssembly

Binaryen also provides a set of toolchain utilities that can

  • Parse and emit WebAssembly. In particular this lets you load WebAssembly, optimize it using Binaryen, and re-emit it, thus implementing a wasm-to-wasm optimizer in a single command.
  • Interpret WebAssembly as well as run the WebAssembly spec tests.
  • Integrate with Emscripten in order to provide a complete compiler toolchain from C and C++ to WebAssembly.
  • Polyfill WebAssembly by running it in the interpreter compiled to JavaScript, if the browser does not yet have native support (useful for testing).

Consult the contributing instructions if you're interested in participating.

Binaryen IR

Binaryen's internal IR is designed to be

  • Flexible and fast for optimization.
  • As close as possible to WebAssembly so it is simple and fast to convert it to and from WebAssembly.

There are a few differences between Binaryen IR and the WebAssembly language:

  • Tree structure
    • Binaryen IR is a tree, i.e., it has hierarchical structure, for convenience of optimization. This differs from the WebAssembly binary format which is a stack machine.
    • Consequently Binaryen‘s text format allows only s-expressions. WebAssembly’s official text format is primarily a linear instruction list (with s-expression extensions). Binaryen can't read the linear style, but it can read a wasm text file if it contains only s-expressions.
  • Types and unreachable code
    • WebAssembly limits block/if/loop types to none and the concrete value types (i32, i64, f32, f64). Binaryen IR has an unreachable type, and it allows block/if/loop to take it, allowing local transforms that don't need to know the global context.
    • Binaryen ignores unreachable code when reading WebAssembly binaries. That means that if you read a wasm file with unreachable code, that code will be discarded as if it were optimized out (often this is what you want anyhow, and optimized programs have no unreachable code anyway, but if you write an unoptimized file and then read it, it may look different). The reason for this behavior is that unreachable code in WebAssembly has corner cases that are tricky to handle in Binaryen IR (it can be very unstructured, and Binaryen IR is more structured than WebAssembly as noted earlier). Note that Binaryen does support unreachable code in .wat text files, since as we saw Binaryen only supports s-expressions there, which are structured.
  • Blocks
    • Binaryen IR has only one node that contains a variable-length list of operands: the block. WebAssembly on the other hand allows lists in loops, if arms, and the top level of a function. Binaryen's IR has a single operand for all non-block nodes; this operand may of course be a block. The motivation for this property is that many passes need special code for iterating on lists, so having a single IR node with a list simplifies them.
    • As in wasm, blocks and loops may have names. Branch targets in the IR are resolved by name (as opposed to nesting depth). This has 2 consequences:
      • Blocks without names may not be branch targets.
      • Names are required to be unique. (Reading .wat files with duplicate names is supported; the names are modified when the IR is constructed).
    • As an optimization, a block that is the child of a loop (or if arm, or function toplevel) and which has no branches targeting it will not be emitted when generating wasm. Instead its list of operands will be directly used in the containing node. Such a block is sometimes called an “implicit block”.

As a result, you might notice that round-trip conversions (wasm => Binaryen IR => wasm) change code a little in some corner cases.

Notes when working with Binaryen IR:

  • As mentioned above, Binaryen IR has a tree structure. As a result, each expression should have exactly one parent - you should not “reuse” a node by having it appear more than once in the tree. The motivation for this limitation is that when we optimize we modify nodes, so if they appear more than once in the tree, a change in one place can appear in another incorrectly.
  • For similar reasons, nodes should not appear in more than one functions.


This repository contains code that builds the following tools in bin/:

  • wasm-shell: A shell that can load and interpret WebAssembly code. It can also run the spec test suite.
  • wasm-as: Assembles WebAssembly in text format (currently S-Expression format) into binary format (going through Binaryen IR).
  • wasm-dis: Un-assembles WebAssembly in binary format into text format (going through Binaryen IR).
  • wasm-opt: Loads WebAssembly and runs Binaryen IR passes on it.
  • asm2wasm: An asm.js-to-WebAssembly compiler, using Emscripten‘s asm optimizer infrastructure. This is used by Emscripten in Binaryen mode when it uses Emscripten’s fastcomp asm.js backend.
  • wasm2asm: A WebAssembly-to-asm.js compiler (still experimental).
  • wasm-merge: Combines wasm files into a single big wasm file (without sophisticated linking).
  • wasm-ctor-eval: A tool that can execute C++ global constructors ahead of time. Used by Emscripten.
  • wasm-emscripten-finalize: Takes a wasm binary produced by llvm+lld and performs emscripten-specific passes over it.
  • wasm.js: wasm.js contains Binaryen components compiled to JavaScript, including the interpreter, asm2wasm, the S-Expression parser, etc., which allow you to use Binaryen with Emscripten and execute code compiled to WASM even if the browser doesn't have native support yet. This can be useful as a (slow) polyfill.
  • binaryen.js: A standalone JavaScript library that exposes Binaryen methods for creating and optimizing WASM modules. For builds, see binaryen.js on npm (or download it directly from github, rawgit, or unpkg).

Usage instructions for each are below.


cmake . && make

Note that you can also use ninja as your generator: cmake -G Ninja . && ninja

  • A C++11 compiler is required.
  • The JavaScript components can be built using, see notes inside. Normally this is not needed as builds are provided in this repo already.

If you also want to compile C/C++ to WebAssembly (and not just asm.js to WebAssembly), you‘ll need Emscripten. You’ll need the incoming branch there (which you can get via the SDK), for more details see the wiki.

Visual C++

  1. Using the Microsoft Visual Studio Installer, install the “Visual C++ tools for CMake” component.

  2. Generate the projects:

    mkdir build
    cd build
    "%VISUAL_STUDIO_ROOT%\Common7\IDE\CommonExtensions\Microsoft\CMake\CMake\bin\cmake.exe" ..

    Substitute VISUAL_STUDIO_ROOT with the path to your Visual Studio installation. In case you are using the Visual Studio Build Tools, the path will be “C:\Program Files (x86)\Microsoft Visual Studio\2017\BuildTools”.

  3. From the Developer Command Prompt, build the desired projects:

    msbuild binaryen.vcxproj

    CMake generates a project named “ALL_BUILD.vcxproj” for conveniently building all the projects.




bin/wasm-opt [.wasm or .wat file] [options] [passes, see --help] [--help]

The wasm optimizer receives WebAssembly as input, and can run transformation passes on it, as well as print it (before and/or after the transformations). For example, try

bin/wasm-opt test/passes/lower-if-else.wast --print

That will pretty-print out one of the test cases in the test suite. To run a transformation pass on it, try

bin/wasm-opt test/passes/lower-if-else.wast --print --lower-if-else

The lower-if-else pass lowers if-else into a block and a break. You can see the change the transformation causes by comparing the output of the two print commands.

It's easy to add your own transformation passes to the shell, just add .cpp files into src/passes, and rebuild the shell. For example code, take a look at the lower-if-else pass.

Some more notes:

  • See bin/wasm-opt --help for the full list of options and passes.
  • Passing --debug will emit some debugging info.



bin/asm2wasm [input.asm.js file]

This will print out a WebAssembly module in s-expression format to the console.

For example, try

$ bin/asm2wasm test/hello_world.asm.js

That input file contains

function () {
  "use asm";
  function add(x, y) {
    x = x | 0;
    y = y | 0;
    return x + y | 0;
  return { add: add };

You should see something like this:

example output

By default you should see pretty colors as in that image. Set COLORS=0 in the env to disable colors if you prefer that. On Linux and Mac, you can set COLORS=1 in the env to force colors (useful when piping to more, for example). For Windows, pretty colors are only available when stdout/stderr are not redirected/piped.

Pass --debug on the command line to see debug info, about asm.js functions as they are parsed, etc.

C/C++ Source ⇒ asm2wasm ⇒ WebAssembly

When using emcc with the BINARYEN option, it will use Binaryen to build to WebAssembly. This lets you compile C and C++ to WebAssembly, with emscripten using asm.js internally as a build step. Since emscripten's asm.js generation is very stable, and asm2wasm is a fairly simple process, this method of compiling C and C++ to WebAssembly is usable already. See the emscripten wiki for more details about how to use it.



(or python will run wasm-shell, wasm-opt, asm2wasm, wasm.js, etc. on the testcases in test/, and verify their outputs.

The script supports some options:

./ [--interpreter=/path/to/interpreter] [TEST1] [TEST2]..
  • If an interpreter is provided, we run the output through it, checking for parse errors.
  • If tests are provided, we run exactly those. If none are provided, we run them all.
  • Some tests require emcc or nodejs in the path. They will not run if the tool cannot be found, and you'll see a warning.
  • We have tests from upstream in tests/spec, in git submodules. Running ./ should update those.

Design Principles

  • Interned strings for names: It's very convenient to have names on nodes, instead of just numeric indices etc. To avoid most of the performance difference between strings and numeric indices, all strings are interned, which means there is a single copy of each string in memory, string comparisons are just a pointer comparison, etc.
  • Allocate in arenas: Based on experience with other optimizing/transformating toolchains, it's not worth the overhead to carefully track memory of individual nodes. Instead, we allocate all elements of a module in an arena, and the entire arena can be freed when the module is no longer needed.


  • How does asm2wasm relate to the new WebAssembly backend which is being developed in upstream LLVM?

This is separate from that. asm2wasm focuses on compiling asm.js to WebAssembly, as emitted by Emscripten's asm.js backend. This is useful because while in the long term Emscripten hopes to use the new WebAssembly backend, the asm2wasm route is a very quick and easy way to generate WebAssembly output. It will also be useful for benchmarking the new backend as it progresses.

  • How about compiling WebAssembly to asm.js (the opposite direction of asm2wasm)? Wouldn't that be useful for polyfilling?

Experimentation with this is happening, in wasm2asm.

This would be useful, but it is a much harder task, due to some decisions made in WebAssembly. For example, WebAssembly can have control flow nested inside expressions, which can't directly map to asm.js. It could be supported by outlining the code to another function, or to compiling it down into new basic blocks and control-flow-free instructions, but it is hard to do so in a way that is both fast to do and emits code that is fast to execute. On the other hand, compiling asm.js to WebAssembly is almost straightforward.

We just have to do more work on wasm2asm and see how efficient we can make it.

  • Can asm2wasm compile any asm.js code?

Almost. Some decisions made in WebAssembly preclude that, for example, there are no global variables. That means that asm2wasm has to map asm.js global variables onto locations in memory, but then it must know of a safe zone in memory in which to do so, and that information is not directly available in asm.js.

asm2wasm and do some integration with Emscripten in order to work around these issues, like asking Emscripten to reserve same space for the globals, etc.

  • Why the weird name for the project?

“Binaryen” is a combination of binary - since WebAssembly is a binary format for the web - and Emscripten - with which it can integrate in order to compile C and C++ all the way to WebAssembly, via asm.js. Binaryen began as Emscripten's WebAssembly processing library (wasm-emscripten).

“Binaryen” is pronounced in the same manner as “Targaryen”: bi-NAIR-ee-in. Or something like that? Anyhow, however Targaryen is correctly pronounced, they should rhyme. Aside from pronunciation, the Targaryen house words, “Fire and Blood”, have also inspired Binaryen's: “Code and Bugs.”

  • Does it compile under Windows and/or Visual Studio?

Yes, it does. Here's a step-by-step tutorial on how to compile it under Windows 10 x64 with CMake and Visual Studio 2015. Help would be appreciated on Windows and OS X as most of the core devs are on Linux.