What is asm.js?

One early attempt to achieve native-like speed in the web browser using JavaScript was asm.js. Although that goal was reached and asm.js was adopted by all the major browser vendors, it never achieved widespread adoption by developers. The beauty of asm.js is that it still runs in most browsers, even in those that do not optimize for it. The idea behind asm.js was that typed arrays could be used in JavaScript to fake a C++ memory heap. The browser simulates pointers and memory allocation in C++, as well as types. A well-designed JavaScript engine can avoid dynamic type checking. Using asm.js, browser makers could get around many of the optimization problems created by the dynamic nature of JavaScript, by just pretending that this version of JavaScript is not dynamically typed. Emscripten, designed as a C++-to-JavaScript compiler, quickly adopted asm.js as the subset of JavaScript that it would compile to because of its improved performance in most browsers. The performance improvements driven by asm.js lead the way to WebAssembly. The same engine modifications used to make asm.js perform well could be used to bootstrap the WebAssembly MVP. Only the addition of a bytecode-to-bytecode compiler was required to take the WebAssembly bytecode and directly convert it into the IR bytecode used by the browser.

At the time of writing, Emscripten does not compile directly from LLVM to WebAssembly. Instead, it compiles to asm.js and uses a tool called Binaryen to convert the asm.js output from Emscripten into WebAssembly.