This is a package containing a Rust frontend for Enzyme. This is very much a work in progress and bug reports/discussion is greatly appreciated!
Enzyme is a plugin that performs automatic differentiation (AD) of statically analyzable LLVM. It is highly-efficient and its ability perform AD on optimized code allows Enzyme to meet or exceed the performance of state-of-the-art AD tools.
We are working on adding support for dyn trait objects, slices and enums.
Adding Generics to your types or implementing traits is already working fine.
First you have to get an adequate rustc/llvm/enzyme build here: enzyme_build.
Afterwards for your convenience you should export this path for LLVM_SYS
$ export LLVM_SYS_130_PREFIX=$HOME/.cache/enzyme/rustc-1.59.0-src/build/x86_64-unknown-linux-gnu/llvm
and tell Enzyme about your library locations:
$ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$HOME/.cache/enzyme/Enzyme-0.0.29/enzyme/build/Enzyme:$HOME/.cache/enzyme/rustc-1.59.0-src/build/x86_64-unknown-linux-gnu/llvm/build/lib/
As an alternative you can also run
$ ninja install
inside of your enzyme and llvm build directory.
Afterwards you can execute the following lines in oxide-enzyme/example
, in order to compile the example.
$ cargo enzyme
You will find your executable in ./target/$TARGET/debug/
We generate gradient functions based on LLVM-IR code. Therefore we currently need two compilation runs. The first to generate a llvm-bc file with the LLVM-IR code, the second to process the bc file, generate the gradients, and build the entire crate. You can do that manually using
RUSTFLAGS="--emit=llvm-bc" cargo +enzyme -Z build-std rustc --target x86_64-unknown-linux-gnu -- --emit=llvm-bc -g -C opt-level=3 -Zno-link && RUSTFLAGS="--emit=llvm-bc" cargo +enzyme -Z build-std rustc --target x86_64-unknown-linux-gnu -- --emit=llvm-bc -g -C opt-level=3
We have created a wrapper for this command which you can call with:
cargo enzyme
Please be aware that our wrapper will ignore all additional commands.
This approach won't work on dependencies since cargo doesn't support such a build process.
More information on installing and using Enzyme directly (not through Rust) can be found on our website: https://enzyme.mit.edu.
To get involved or if you have questions, please join our mailing list.
If using this code in an academic setting, please cite the following paper to appear in NeurIPS 2020
@inproceedings{NEURIPS2020_9332c513,
author = {Moses, William and Churavy, Valentin},
booktitle = {Advances in Neural Information Processing Systems},
editor = {H. Larochelle and M. Ranzato and R. Hadsell and M. F. Balcan and H. Lin},
pages = {12472--12485},
publisher = {Curran Associates, Inc.},
title = {Instead of Rewriting Foreign Code for Machine Learning, Automatically Synthesize Fast Gradients},
url = {https://proceedings.neurips.cc/paper/2020/file/9332c513ef44b682e9347822c2e457ac-Paper.pdf},
volume = {33},
year = {2020}
}
Dual-licensed to be compatible with the Rust project.
Licensed under the Apache License, Version 2.0 http://www.apache.org/licenses/LICENSE-2.0 or the MIT license http://opensource.org/licenses/MIT, at your option. This file may not be copied, modified, or distributed except according to those terms.