9aa82c6fb9
When the input is a constant, we use a fairly inefficient sequence of fmov+fcvt+dup or, when the double isn't encodable in fmov, adr+ldr+fcvt+dup. Instead, we can use the same lowering as X64 when the input is a constant, and load the vector from memory. However, if the constant is encodable via fmov, we can use a vector fmov instead (which is just one instruction and doesn't need constant space). Fortunately the bit encoding of fmov for 32-bit floating point numbers matches that of 64-bit: the decoding algorithm is a little different because it expands into a larger exponent, but the values are compatible, so if a double can be encoded into a scalar fmov with a given abcdefgh pattern, the same pattern should encode the same float; due to the very limited number of mantissa and exponent bits, all values that are encodable are also exact in both 32-bit and 64-bit floats. This strategy is ~same as what gcc uses. For complex vectors, we previously used 4 instructions and 8 bytes of constant storage, and now we use 2 instructions and 16 bytes of constant storage, so the memory footprint is the same; for simple vectors we just need 1 instruction (4 bytes). clang lowers vector constants a little differently, opting to synthesize a 64-bit integer using 4 instructions (mov/movk) and then move it to the vector register - this requires 5 instructions and 20 bytes, vs ours/gcc 2 instructions and 8+16=24 bytes. I tried a simpler version of this that would be more compact - synthesize a 32-bit integer constant with mov+movk, and move it to vector register via dup.4s - but this was a little slower on M2, so for now we prefer the slightly larger version as it's not a regression vs current implementation. On the vector approximation benchmark we get: - Before this PR (flag=false): ~7.85 ns/op - After this PR (flag=true): ~7.74 ns/op - After this PR, with 0.125 instead of 0.123 in the benchmark code (to use fmov): ~7.52 ns/op - Not part of this PR, but the mov/dup strategy described above: ~8.00 ns/op |
||
|---|---|---|
| .github | ||
| Analysis | ||
| Ast | ||
| bench | ||
| CLI | ||
| CodeGen | ||
| Common/include/Luau | ||
| Compiler | ||
| Config | ||
| extern | ||
| fuzz | ||
| tests | ||
| tools | ||
| VM | ||
| .clang-format | ||
| .gitignore | ||
| CMakeLists.txt | ||
| CMakePresets.json | ||
| CONTRIBUTING.md | ||
| LICENSE.txt | ||
| lua_LICENSE.txt | ||
| Makefile | ||
| README.md | ||
| SECURITY.md | ||
| Sources.cmake | ||
Luau 
Luau (lowercase u, /ˈlu.aʊ/) is a fast, small, safe, gradually typed embeddable scripting language derived from Lua.
It is designed to be backwards compatible with Lua 5.1, as well as incorporating some features from future Lua releases, but also expands the feature set (most notably with type annotations). Luau is largely implemented from scratch, with the language runtime being a very heavily modified version of Lua 5.1 runtime, with completely rewritten interpreter and other performance innovations. The runtime mostly preserves Lua 5.1 API, so existing bindings should be more or less compatible with a few caveats.
Luau is used by Roblox game developers to write game code, as well as by Roblox engineers to implement large parts of the user-facing application code as well as portions of the editor (Roblox Studio) as plugins. Roblox chose to open-source Luau to foster collaboration within the Roblox community as well as to allow other companies and communities to benefit from the ongoing language and runtime innovation. As a consequence, Luau is now also used by games like Alan Wake 2 and Warframe.
This repository hosts source code for the language implementation and associated tooling. Documentation for the language is available at https://luau-lang.org/ and accepts contributions via site repository; the language is evolved through RFCs that are located in rfcs repository.
Usage
Luau is an embeddable language, but it also comes with two command-line tools by default, luau and luau-analyze.
luau is a command-line REPL and can also run input files. Note that REPL runs in a sandboxed environment and as such doesn't have access to the underlying file system except for ability to require modules.
luau-analyze is a command-line type checker and linter; given a set of input files, it produces errors/warnings according to the file configuration, which can be customized by using --! comments in the files or .luaurc files. For details please refer to type checking and linting documentation.
Installation
You can install and run Luau by downloading the compiled binaries from a recent release; note that luau and luau-analyze binaries from the archives will need to be added to PATH or copied to a directory like /usr/local/bin on Linux/macOS.
Alternatively, you can use one of the packaged distributions (note that these are not maintained by Luau development team):
- macOS: Install Homebrew and run
brew install luau - Arch Linux: From the AUR (Arch Linux User Repository), install one of these packages via a AUR helper or manually (by cloning their repo and using
makepkg): luau (manual build), luau-git (manual build by cloning this repo), or luau-bin (pre-built binaries from releases) - Alpine Linux: Enable community repositories and run
apk add luau - Gentoo Linux: Luau is officially packaged by Gentoo and can be installed using
emerge dev-lang/luau. You may have to unmask the package first before installing it (which can be done by including the--autounmask=yoption in theemergecommand).
After installing, you will want to validate the installation was successful by running the test case here.
Building
On all platforms, you can use CMake to run the following commands to build Luau binaries from source:
mkdir cmake && cd cmake
cmake .. -DCMAKE_BUILD_TYPE=RelWithDebInfo
cmake --build . --target Luau.Repl.CLI --config RelWithDebInfo
cmake --build . --target Luau.Analyze.CLI --config RelWithDebInfo
Alternatively, on Linux/macOS you can use make:
make config=release luau luau-analyze
To integrate Luau into your CMake application projects as a library, at the minimum you'll need to depend on Luau.Compiler and Luau.VM projects. From there you need to create a new Luau state (using Lua 5.x API such as lua_newstate), compile source to bytecode and load it into the VM like this:
// needs lua.h and luacode.h
size_t bytecodeSize = 0;
char* bytecode = luau_compile(source, strlen(source), NULL, &bytecodeSize);
int result = luau_load(L, chunkname, bytecode, bytecodeSize, 0);
free(bytecode);
if (result == 0)
return 1; /* return chunk main function */
For more details about the use of host API you currently need to consult Lua 5.x API. Luau closely tracks that API but has a few deviations, such as the need to compile source separately (which is important to be able to deploy VM without a compiler), or lack of __gc support (use lua_newuserdatadtor instead).
To gain advantage of many performance improvements it's highly recommended to use safeenv feature, which sandboxes individual scripts' global tables from each other as well as protects builtin libraries from monkey-patching. For this to work you need to call luaL_sandbox for the global state and luaL_sandboxthread for each new script's execution thread.
Testing
Luau has an internal test suite; in CMake builds it is split into two targets, Luau.UnitTest (for bytecode compiler and type checker/linter tests) and Luau.Conformance (for VM tests). The unit tests are written in C++, whereas the conformance tests are largely written in Luau (see tests/conformance).
Makefile builds combine both into a single target and can be ran via make test.
Dependencies
Luau uses C++ as its implementation language. The runtime requires C++11, whereas the compiler and analysis components require C++17. It should build without issues using Microsoft Visual Studio 2017 or later, or gcc-7 or clang-7 or later.
Other than the STL/CRT, Luau library components don't have external dependencies. The test suite depends on doctest testing framework, and the REPL command-line depends on isocline.
License
Luau implementation is distributed under the terms of MIT License. It is based on Lua 5.x implementation that is MIT licensed as well.
When Luau is integrated into external projects, we ask to honor the license agreement and include Luau attribution into the user-facing product documentation. The attribution using Luau logo is also encouraged.
