codspeed.yml

name: codspeed-benchmarks

on:
  # Run on pushes to the main branch
  push:
    branches:
      - "master" # or "main"
  # Run on pull requests
  pull_request:
  # `workflow_dispatch` allows CodSpeed to trigger backtest
  # performance analysis in order to generate initial data.
  workflow_dispatch:

jobs:
  benchmarks:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v3
      # ...
      # Setup your environment here:
      #  - Configure your Python/Rust/Node version
      #  - Install your dependencies
      #  - Build your benchmarks (if using a compiled language)
      # ...
      - uses: actions/setup-python@v3
        with:
          python-version: "3.10"

      - name: cuda-toolkit
        uses: Jimver/cuda-toolkit@v0.2.11
        with:
          cuda: "12.3.0"

      - run: echo "Installed cuda version is: ${{steps.cuda-toolkit.outputs.cuda}}"
      - run: echo "Cuda install location: ${{steps.cuda-toolkit.outputs.CUDA_PATH}}"
      - run: nvcc -V  

      - name: Install project dependencies
        run: pip install -e . && pip install pytest

      - name: Run benchmarks
        uses: CodSpeedHQ/action@v2
        with:
          token: ${{ secrets.codspeed_token }}
          run: "pytest tests/kernels/test_layernorm.py --codspeed"