Join 10350+ others. No spamming.
I promise!

Follow us at github.



weld-project / weld


Weld is a runtime and language for accelerating data analytics frameworks



Build Status

Weld is a language and runtime for improving the performance of data-intensive applications. It optimizes across libraries and functions by expressing the core computations in libraries using a common intermediate representation, and optimizing across each framework.

Modern analytics applications combine multiple functions from different libraries and frameworks to build complex workflows. Even though individual functions can achieve high performance in isolation, the performance of the combined workflow is often an order of magnitude below hardware limits due to extensive data movement across the functions. Weld’s take on solving this problem is to lazily build up a computation for the entire workflow, and then optimizing and evaluating it only when a result is needed.

You can join the discussion on Weld on our Google Group or post on the Weld mailing list at



To build Weld, you need Rust 1.13 or higher and LLVM 3.8 or newer.

To install Rust, follow the steps here. You can verify that Rust was installed correctly on your system by typing rustc into your shell.

MacOS LLVM Installation

To install LLVM on macOS, first install Homebrew. Then:

$ brew install llvm@3.8

Weld's dependencies require llvm-config, so you may need to create a symbolic link so the correct llvm-config is picked up (note that you might need to add sudo at the start of this command):

$ ln -s /usr/local/bin/llvm-config-3.8 /usr/local/bin/llvm-config

To make sure this worked correctly, run llvm-config --version. You should see 3.8.x or newer.

Ubuntu LLVM Installation

To install LLVM on Ubuntu :

$ sudo apt-get install llvm-3.8
$ sudo apt-get install llvm-3.8-dev
$ sudo apt-get install clang-3.8

Weld's dependencies require llvm-config, so you may need to create a symbolic link so the correct llvm-config is picked up:

$ ln -s /usr/bin/llvm-config-3.8 /usr/local/bin/llvm-config

To make sure this worked correctly, run llvm-config --version. You should see 3.8.x or newer.

Building Weld

With LLVM and Rust installed, you can build Weld. Clone this repository, set the WELD_HOME environment variable, and build using cargo:

$ git clone
$ cd weld/
$ export WELD_HOME=`pwd`
$ cargo build --release

Note: If you are using a version of LLVM newer than 3.8, you will have to change the llvm-sys crate dependency in easy_ll/Cargo.toml to match (e.g. 40.0.0 for LLVM 4.0.0). You may also need to create additional symlinks for some packages that omit the version suffix when installing the latest version, e.g. for LLVM 4.0:

$ ln -s /usr/local/bin/clang /usr/local/bin/clang-4.0
$ ln -s /usr/local/bin/llvm-link /usr/local/bin/llvm-link-4.0

Weld builds two dynamically linked libraries (.so files on Linux and .dylib files on macOS): libweld and libweldrt. Both of these libraries are found using WELD_HOME. By default, the libraries are in $WELD_HOME/target/release and $WELD_HOME/weld_rt/target/release.

Finally, run the unit and integration tests:

$ cargo test


The docs/ directory contains documentation for the different components of Weld.

  • describes the syntax of the Weld IR.
  • describes the low-level C API for interfacing with Weld.
  • gives an overview of the Python API.
  • contains a tutorial for how to build a small vector library using Weld.


Grizzly is a subset of Pandas integrated with Weld. Details on how to use Grizzly are in python/grizzly. Some example workloads that make use of Grizzly are in examples/python/grizzly.

Running an Interactive REPL

  • cargo test runs unit and integration tests. A test name substring filter can be used to run a subset of the tests:

    cargo test <substring to match in test name>
  • The target/release/repl program is a simple "shell" where one can type Weld programs and see the results of parsing, macro substitution and type inference.

Example repl session:

> let a = 5 + 2; a + a
Raw structure: [...]

After macro substitution:
let a=((5+2));(a+a)

After inlining applies:
let a=((5+2));(a+a)

After type inference:
let a:i32=((5+2));(a:i32+a:i32)

Expression type: i32

> map([1, 2], |x| x+1)
Raw structure: [...]

After macro substitution:

After inlining applies:

After type inference:

Expression type: vec[i32]


cargo bench runs benchmarks under the benches/ directory. The results of the benchmarks are written to a file called benches.csv. To specify specific benchmarks to run:

$ cargo bench [benchmark-name]

If a benchmark name is not provided, all benchmarks are run.