MLIR Generator

Rationale

This repository is meant to help MLIR pass writers to generate MLIR files that often come from ML/HPC sources, to allow testing and developing their passes (as well as creating unit tests) with the current state of the known front-ends.

How To Use

Warning: This is very much work-in-progress, don't expect half of it to work.

Each front-end has a prepare script, for example:

cd torch-script
./prepare.sh

It also has models ready to use (after prepared):

source venv/bin/activate
python models/linear.py

Results are in <front-end-name>/mlir.

Development

Adding a new front-end

If you want to build an existing front-end, you need to:

1. Add a new directory for the front-end
2. Add a prepare.sh script that installs it and prepares the virtualenv
3. Add a generate.sh script that generates all the MLIR files
4. Add models as Python files in models, run the export, and save them as MLIR files in mlir