vulkan.yml

name: Test Vulkan Backend (specialized runners)

# Vulkan CI jobs that require special runners (an NVIDIA GPU, or a Windows
# MSVC toolchain). These are separate from test-backend-vulkan.yml (which runs the
# default SwiftShader coverage on standard runners) so that they're only run
# when Vulkan related files change.

on:
  push:
    branches:
      - main
      - release/*
    tags:
      - ciflow/nightly/*
  pull_request:
    paths:
      - .github/workflows/vulkan.yml
      - backends/vulkan/**
      - examples/vulkan/**
      - .ci/scripts/setup-vulkan-linux-deps.sh
      - .ci/scripts/setup-vulkan-windows-deps.ps1
      - .ci/scripts/setup-windows-msvc-vulkan.ps1
  workflow_dispatch:

concurrency:
  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
  cancel-in-progress: true

permissions:
  contents: read

jobs:
  changed-files:
    name: Get changed files
    uses: ./.github/workflows/_get-changed-files.yml
    with:
      include-push-diff: true # so push commits can also be path-filtered

  run-decision:
    name: CI run decision
    uses: ./.github/workflows/_ci-run-decision.yml

  test-vulkan-nvidia:
    needs: [changed-files, run-decision]
    # Path-filtered: skip commits that don't touch Vulkan-relevant paths, except
    # on sampled full runs (see _ci-run-decision.yml).
    if: |
      contains(needs.changed-files.outputs.changed-files, 'backends/vulkan/') ||
      contains(needs.changed-files.outputs.changed-files, 'examples/vulkan/') ||
      contains(needs.changed-files.outputs.changed-files, '.ci/scripts/setup-vulkan-linux-deps.sh') ||
      contains(needs.changed-files.outputs.changed-files, '.github/workflows/vulkan.yml') ||
      needs.run-decision.outputs.is-full-run == 'true'
    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
    permissions:
      id-token: write
      contents: read
    with:
      timeout: 120
      runner: linux.g5.4xlarge.nvidia.gpu
      gpu-arch-type: cuda
      gpu-arch-version: "12.6"
      use-custom-docker-registry: false
      submodules: recursive
      ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
      script: |
        set -eux

        # Install a glibc-compatible glslc (conda-forge), the Vulkan loader and
        # tools, and select a real system ICD. The CUDA runner image is
        # manylinux_2_28 (glibc 2.28), where the LunarG SDK's prebuilt glslc does
        # not run, so setup-vulkan-linux-deps.sh sources those from conda-forge and
        # the system package manager instead. The NVIDIA container runtime mounts
        # the driver's Vulkan library but not its ICD manifest, so the script
        # synthesizes one and pins the loader to it; if no NVIDIA library is found
        # it falls back to SwiftShader.
        # NOTE: first-run check - inspect the vulkaninfo output below to confirm a
        # real NVIDIA device is selected (not llvmpipe/SwiftShader).
        source .ci/scripts/setup-vulkan-linux-deps.sh real-gpu
        vulkaninfo --summary || true

        # Full from-source install. Unlike the SwiftShader jobs in pull.yml, the
        # CUDA runner image does not pre-install ExecuTorch's dependencies, so
        # setup-linux.sh's "deps already in the image" assumption does not hold.
        # CMAKE_ARGS enables Vulkan in the pybindings so the model --test runs and
        # the pt2e/torchao e2e tests below execute on the GPU (default is OFF).
        CMAKE_ARGS="-DEXECUTORCH_BUILD_VULKAN=ON" PYTHON_EXECUTABLE=python ./install_executorch.sh

        # Model coverage (mirrors test-vulkan-models-linux, on real hardware).
        PYTHON_EXECUTABLE=python bash backends/vulkan/test/scripts/test_model.sh --build

        models="mv2 mv3 edsr resnet18 resnet50 dl3 w2l ic3 ic4"
        for model in $models; do
          python -m examples.vulkan.export --model_name=$model --test
        done

        # For selected vision models, test with dynamic shapes
        models="mv2 resnet18 resnet50 ic3 densenet161"
        for model in $models; do
          python -m examples.vulkan.export --model_name=$model --test -d
        done

        # Operator coverage (mirrors test-vulkan-operators-linux, on real hardware).
        # The custom-op prototyping binaries are GPU microbenchmarks that rely on
        # GPU timestamp queries; they need a real device and crash on the
        # SwiftShader software fallback. Always build them (compile coverage), but
        # only run them when a real GPU was selected (setup-vulkan-linux-deps.sh
        # exports ETVK_USING_SWIFTSHADER when it falls back to SwiftShader).
        PYTHON_EXECUTABLE=python bash backends/vulkan/test/custom_ops/build_and_run.sh
        if [ -z "${ETVK_USING_SWIFTSHADER:-}" ]; then
          ./cmake-out/backends/vulkan/test/custom_ops/test_add
          ./cmake-out/backends/vulkan/test/custom_ops/test_q8csw_linear
          ./cmake-out/backends/vulkan/test/custom_ops/test_q8csw_conv2d
          ./cmake-out/backends/vulkan/test/custom_ops/test_q4gsw_linear
          ./cmake-out/backends/vulkan/test/custom_ops/test_choose_qparams_per_row
          ./cmake-out/backends/vulkan/test/custom_ops/test_q8ta_qdq
          ./cmake-out/backends/vulkan/test/custom_ops/test_q8ta_clone
          ./cmake-out/backends/vulkan/test/custom_ops/test_q8ta_binary
        else
          echo "SwiftShader fallback active: built custom-op benchmarks but skipping execution (they require real-GPU timestamp queries)."
        fi

        PYTHON_EXECUTABLE=python bash backends/vulkan/test/scripts/test_op.sh --build

        # Run e2e testing for selected operators.
        python -m unittest backends/vulkan/test/test_vulkan_delegate.py -k "*pt2e*"
        python -m unittest backends/vulkan/test/test_vulkan_delegate.py -k "*torchao*"

  build-vulkan-windows-msvc:
    needs: [changed-files, run-decision]
    if: |
      contains(needs.changed-files.outputs.changed-files, 'backends/vulkan/') ||
      contains(needs.changed-files.outputs.changed-files, '.ci/scripts/setup-vulkan-windows-deps.ps1') ||
      contains(needs.changed-files.outputs.changed-files, '.ci/scripts/setup-windows-msvc-vulkan.ps1') ||
      contains(needs.changed-files.outputs.changed-files, '.github/workflows/vulkan.yml') ||
      needs.run-decision.outputs.is-full-run == 'true'
    name: build-vulkan-windows-msvc
    uses: pytorch/test-infra/.github/workflows/windows_job.yml@main
    with:
      ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
      timeout: 90
      script: |
        git config --global http.sslBackend openssl
        git submodule update --init backends/vulkan/third-party/Vulkan-Headers backends/vulkan/third-party/volk backends/vulkan/third-party/VulkanMemoryAllocator
        git submodule update --init
        conda init powershell
        powershell -Command "& {
          Set-PSDebug -Trace 1
          \$ErrorActionPreference = 'Stop'
          \$PSNativeCommandUseErrorActionPreference = \$true
          .ci/scripts/setup-windows-msvc-vulkan.ps1
        }"