ci.yaml

# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0

name: "CI"

on:
  push:
    branches:
      - "main"
      - "pull-request/[0-9]+"

jobs:
  lint:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v5
      - uses: actions/setup-python@v6
        with:
          python-version: '3.10'
      - name: "Running Lint"
        run: |
          pip install hatch
          hatch run lint:check

  build-docs:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v5
      - uses: actions/setup-python@v6
        with:
          python-version: '3.10'
      - name: "Building Docs"
        run: |
          pip install hatch
          hatch run docs:build

  test-ncu:
    name: test-ncu (${{ matrix.platform.runner }})
    runs-on: ${{ matrix.platform.runner }}
    strategy:
      matrix:
        pytorch_image: ["nvcr.io/nvidia/pytorch:26.09-py3"]
        platform:
          - runner: linux-amd64-gpu-h100-latest-1
            ncu_installer_url: "https://developer.nvidia.com/downloads/assets/tools/secure/nsight-compute/2026_2_1/nsight_compute-linux-x86_64-2026.2.1.6.run"
          - runner: linux-arm64-gpu-l4-latest-1
            ncu_installer_url: "https://developer.nvidia.com/downloads/assets/tools/secure/nsight-compute/2026_2_1/nsight_compute-linux-sbsa-2026.2.1.6.run"
    container:
      image: ${{ matrix.pytorch_image }}
      options: "--cap-add=SYS_ADMIN --gpus all"
    steps:
      - uses: actions/checkout@v5
      - name: "Install NCU 2026.2.1.0 (CUDA 13.3 Update 1)"
        run: |
          curl -L -o ncu_installer.run "${{ matrix.platform.ncu_installer_url }}"
          chmod +x ncu_installer.run
          ./ncu_installer.run -- -noprompt
          rm ncu_installer.run
          echo "/usr/local/NVIDIA-Nsight-Compute-2026.2" >> $GITHUB_PATH
      - name: "Running Tests"
        run: |
          pip install hatch
          hatch run test_cu13:run

  test-cupti:
    name: test-cupti (${{ matrix.platform.runner }})
    runs-on: ${{ matrix.platform.runner }}
    strategy:
      matrix:
        pytorch_image: ["nvcr.io/nvidia/pytorch:26.09-py3"]
        platform:
          - runner: linux-amd64-gpu-h100-latest-1
          - runner: linux-arm64-gpu-l4-latest-1
    container:
      image: ${{ matrix.pytorch_image }}
      options: "--cap-add=SYS_ADMIN --gpus all"
    steps:
      - uses: actions/checkout@v5
      - name: "Running Tests (CUPTI backend)"
        run: |
          pip install hatch
          hatch run test_cu13:run -- --tool-init=cupti

  test-ncu-windows:
    name: test-ncu (windows-amd64-gpu-a100-latest-1)
    runs-on: windows-amd64-gpu-a100-latest-1
    steps:
      - uses: actions/checkout@v5
      - uses: actions/setup-python@v6
        with:
          python-version: '3.12'
      - name: "Install NCU 2026.3.0 (CUDA 13.4)"
        shell: pwsh
        run: |
          curl.exe -fL -o ncu_installer.msi "https://developer.nvidia.com/downloads/assets/tools/secure/nsight-compute/2026_3_0/nsight_compute-windows-x86_64-2026.3.0.13.msi"
          if ($LASTEXITCODE) { exit $LASTEXITCODE }
          Start-Process msiexec.exe -Wait -ArgumentList '/i ncu_installer.msi /quiet /norestart'
          Remove-Item ncu_installer.msi
          "C:\Program Files\NVIDIA Corporation\Nsight Compute 2026.3.0" | Out-File -FilePath $env:GITHUB_PATH -Encoding utf8 -Append
      - name: "Install MSVC Redistributable (required by torch)"
        shell: pwsh
        run: |
          curl.exe -fL -o vc_redist.x64.exe https://aka.ms/vc14/vc_redist.x64.exe
          if ($LASTEXITCODE) { exit $LASTEXITCODE }
          Start-Process vc_redist.x64.exe -Wait -ArgumentList '/install /quiet /norestart'
          Remove-Item vc_redist.x64.exe
      - name: "Create hatch test environment"
        shell: pwsh
        run: |
          pip install hatch
          hatch env create test_cu13
      - name: "Install PyTorch (CUDA 13.2)"
        shell: pwsh
        run: |
          hatch run test_cu13:pip install torch --index-url https://download.pytorch.org/whl/cu132
      - name: "Install cuTile"
        shell: pwsh
        run: |
          hatch run test_cu13:pip install cuda-tile[tileiras]
      - name: "Running Tests"
        shell: pwsh
        env:
          # On Windows matplotlib defaults to the TkAgg backend, and tkinter
          # is not properly configured on the runner. This is a headless
          # runner and no GUI is required, so force the non-interactive
          # backend. Headless Linux falls back to Agg already.
          MPLBACKEND: Agg
        # WAR for a known NCU issue on Windows: the Python process aborts
        # with 0xC0000409 (FAIL_FAST_FATAL_APP_EXIT) at exit, after pytest
        # has finished. Save pytest's own exit code before that happens and
        # report it instead. Activate the test_cu13 hatch environment and run
        # python directly rather than through `hatch run`: before Python
        # 3.13.1, passing the 0xC0000409 exit code on through sys.exit() on
        # Windows exits with -1 instead (https://github.com/python/cpython/pull/125896).
        run: |
          . (Join-Path (hatch env find test_cu13) 'Scripts\activate.ps1')
          python -c "import pytest, sys; rc = int(pytest.main(['-v', '--junitxml=report.xml'])); open('pytest_rc.txt', 'w').write(str(rc)); sys.exit(rc)"
          $code = $LASTEXITCODE
          Write-Host ("python exit code: {0} (0x{0:X8})" -f $code)
          if ($code -eq 0xC0000409 -and (Test-Path pytest_rc.txt)) {
            $code = [int](Get-Content pytest_rc.txt)
            Write-Host "Ignoring exit code 0xC0000409 (known NCU issue); pytest exit code: $code"
          }
          exit $code