Mariano Guerra (@warianoguerra)

Co-founder event-fabric.com

StuttgartJS, August 2016

The real one, often abbreviated asm:

• Low-level programming language

• Very strong (generally 1 to 1) correspondence between

  • The language

  • and the architecture's machine code instructions

• Transport Tycoon

• Roller Coster Tycoon

• KolibriOS

• MenuetOS

• Collection of modular and reusable compiler and toolchain technologies

• Used to develop compiler front ends and back ends

llvm.org

Source Code -> LLVM IR*

* IR: intermediate representation

LLVM IR -> CPU specific binary*

* ARM, x86, MIPS, PowerPC etc.

Optimizable low-level subset of JavaScript*

* Not for humans

• Portable

• Size- and load-time-efficient format

• Suitable for compilation to the web

• AngryBots

• BananaBread

• Oryol 3D Engine Demos in asm.js and wasm

• It's not a replacement for js

• It's not a programming language

• It's not a separate VM

• It's a complement to JS

Getting to MVP

• Future of Binaryen in a stack machine world?

  • Slides where it was decided to switch from being an AST to being a stack machine

  • spec branch with stack changes

  • Patch to chromium

• Structured code for the stack machine

• LLVM backend upstream

• Lots of tools

• Reference implementation (spec) in Ocaml

• 3 Browser engines have native support in various stages

  • Google Chrome: fully spec compliant on all architectures, behind a flag

  • Mozilla Firefox: optimized for ia32 and x64, behind a flag

  • Microsoft Edge: support in an experimental build

• MVP (Version 1.0) expected to be shipped this summer

• Standardization expected by the end of the year

(as of August 10th, 2016)

[1] Custom Format (js-like, not s-expression based)

• MVP

• Post-MVP

• Future

• Will contain features which are available today in modern web browsers

• Which perform well even on mobile devices

• Leads to roughly the same functionality as asm.js

• Threads

• SIMD

• Zero-cost Exception Handling

• Feature Testing

• Finer-grained control over memory

• Large page support

• More expressive control flow

  • goto, tail calls

• GC

• Source maps integration

• Coroutines

• Platform-independent JIT compilation

• Multiprocess support

• Additional float operators

• Additional integer operators

  • min

  • max

  • abs

• Integer Overflow Detection

• Better feature testing support

• Mutable global variables

• Streaming Compilation

• Multiple Tables and Memories

• Module

• AST

• Binary format

• Text format

• To be implemented by

  • Web browsers

  • Completely different execution environments

• Can be natively decoded much faster than JavaScript can be parsed

  • Experiments show more than 20X faster

• Must be

  • Portable

  • Stable

  • Small

  • Fast to decode

  • Fast to compile

• For "view source"

• For tools, experiments, debugging, optimization

• Equivalent and isomorphic to the binary format

• Standardized only for tooling purposes

• Nested statements

  • Instead of linear list like in asm

• Prototypes use s-expression based syntax

• Currently no final, official, text format

  • JS-like proposal from FF Nightly

From Ben Titzer @ VMSS16's slides

• void

• i32

• i64

• f32

• f64

• Flat, single global table

• Static binding

• Indirect calls through table

• Linear memory

  • Large, bounds-checked array

• i32: + - * / % << >> >>> etc

• i64: + - * / % << >> >>> etc

• f32: + - * / sqrt ceil floor

• f64: + - * / sqrt ceil floor

• conversions

• load store

• call_direct call_indirect

• if

• loop

• block

• br

• switch

• Memory

• Data

• Imports

• Exports

• Start function

• Global variables

• Tables

• Elements

• Functions

• Code

• Provided, at instantiation time, by the host environment

• Similar to a native syscall

• Wasm doesn't know about Javascript or the DOM

• Can import functions, globals, memory, tables

• Returned at instantiation time to the host environment

• Can export functions, globals, memory, tables

• Can share with another Wasm instance

• Called after module loading and before any call to the module function

• Definition of zero or more linear memories

• in the MVP it's limited to 1

• Analogous to the .data section of native executables

• Initializes memory

• Zero or more definitions of distinct tables

• In the MVP it's limited to 1

• Table of pointer to provided opaque functions

  • called with call_indirect

• Like data section but for tables

• Declares the signatures of each internal function

• Contains the function body of each function declared by the function section

• Function Index Space

• Global Index Space

• Linear Memory Index Space

• Table Index Space

C/C++ -> llvm* -> asm.js

* LLVM Emscripten backend

• Compiler and toolchain infrastructure library for Wasm

• Written in C++

• wast -> wasm

• Interpret wasm

• asm.js -> wasm

• Polyfill wasm

Binaryen (C++) -> Emscripten -> asm.js wasm polyfill

Let's translate hello.c to asm.js and Wasm.

• Attempt 1: apt install emscripten

  • Ubuntu 16.04

  • FAIL: weird version errors

• Attempt 2: emsdk

  • FAIL: no support for binaryen yet

• Attempt 3: compile from source

  • FAIL: no support for binaryen yet (WAT)

• Attempt 4: compile from source (incoming branch of emscripten)

  • SUCCESS!

  • Thanks to gh/qis for the tip

  • Instructions for linux and for windows

• I like minimal simple things

• I used to code in assembly

• I convert any problem into a "let's implement a programming language" problem

A sometimes minimal FORTH compiler and tutorial for Linux but in WebAssembly!

No easy translation from asm to Wasm

• Code is not stored on the same memory as data

• Can't implement the threaded code technique

Translate the C version mentioned in the comments.

The code was a little unreadable: http://ftp.funet.fi/pub/doc/IOCCC/1992/buzzard.2.orig.c

First deobfuscated it

• Compile the C version to asmjs and wasm

• Translate it to Javascript by hand

• Translate the C to the text format of wasm by hand

Ricardo Forth

• A Forth dialect base on buzzard.2

• Implemented in C, Javascript and WebAssembly

• Compiled from C to

  • asm.js (using emscripten)

    • run, def_word, r, inlines append_to_dict

  • Wasm (using binaryen)

    • $run, $def_word, $r, inlines $append_to_dict

[1] I close parenthesis the lisp way, binaryen doesn't

[2] "almost wasm"

• Prototype Spec in Markdown

• WebAssembly Explorer

• ast.run WebAssembly playground

• WASM Pilot Text Format Playground

• gh/drom wasm/wast related projects

• emscripten's emsdk/src/settings.js contains docs about the flags that can be passed to emcc

• Luke Wagner -on- WebAssembly: A New Compiler Target For The Web

• Ben Titzer @ VMSS16: A Little on V8 and WebAssembly

• Emscripten and WebAssembly by Alon Zakai / @kripken

• A Particle System experiment designed to benchmark web technologies from a non trivial piece of code: ES6, Emscripten and Web Assembly

• SHA-3 (Keccak padding) in WebAssembly WAST

• Experimental programming language that compiles to both WebAssembly and JavaScript

• Build Your First Thing With WebAssembly

• Wasm example and instructions to build on windows