Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
49 lines (35 loc) · 1.63 KB

CONTRIBUTING.md

File metadata and controls

49 lines (35 loc) · 1.63 KB
Raw

Contributing

Thanks for looking at TinyCompiler. It is a small compiler for a small tensor language, and the bar for changes is that every result stays checked against a numpy oracle on the real chips.

Set up

git submodule update --init
python3 -m venv .venv && .venv/bin/pip install -e ".[dev]" && .venv/bin/pip install -e targets/tiny-gpu --no-deps
source scripts/env.sh

For the Compiler Explorer:

cd web && pnpm install && pnpm dev

pnpm dev regenerates the Python bundle and the example list first (scripts/build_web.py).

Before you open a pull request

Run the same checks CI runs:

ruff check . && black --check .
pytest -q
cd web && pnpm lint && pnpm typecheck && pnpm test && pnpm e2e
  • Golden files (tests/golden/) pin the emitted IR and assembly. If you change emitted output on purpose, review the diff and run pytest --update-golden.
  • Measured numbers in docs/ARCHITECTURE.md come from scripts/opt_table.py --write. A test fails if they drift.
  • Do not format targets/ (the hardware submodules); the formatter config excludes it.

Branches and pull requests

  • Work on a feat/*, fix/* or docs/* branch. main is the default branch and is not pushed to directly.
  • Keep a pull request to one change. Describe what changed and how you checked it.
  • Add a regression test for each bug you fix. A minimized failing program can go in tests/regressions/.

Reporting bugs

Open an issue with the .tiny program that fails, the command you ran, and what you expected. A minimal program makes the fix much faster. For security problems, see SECURITY.md.