name: PR Test (XPU)

on:
  push:
    branches: [ main ]
  pull_request:
    branches: [ main ]
  workflow_dispatch:
  workflow_call:
    inputs:
      ref:
        description: 'Git ref (branch, tag, or SHA) to test. If not provided, uses the default branch.'
        required: false
        type: string
        default: ''
      run_all_tests:
        description: "Run all tests (for releasing or testing purpose)"
        required: false
        type: boolean
        default: false

concurrency:
  group: pr-test-xpu-${{ inputs.ref || github.ref }}
  cancel-in-progress: ${{ github.event_name != 'workflow_call' }}

jobs:
  # ==================== Check Changes ==================== #
  check-changes:
    runs-on: ubuntu-latest
    outputs:
      changes_exist: ${{ steps.filter.outputs.main_package == 'true' || steps.filter.outputs.multimodal_gen == 'true' || steps.run-mode.outputs.run_all_tests == 'true' }}
      main_package: ${{ steps.filter.outputs.main_package == 'true' || steps.run-mode.outputs.run_all_tests == 'true' }}
      multimodal_gen: ${{ steps.filter.outputs.multimodal_gen == 'true' || steps.run-mode.outputs.run_all_tests == 'true' }}
    steps:
      - name: Checkout code
        uses: actions/checkout@v4
        with:
          ref: ${{ inputs.ref || github.ref }}

      - name: Determine run mode
        id: run-mode
        run: |
          # Run all tests for workflow_call (when ref input is provided)
          # Note: github.event_name is inherited from caller, so we detect workflow_call by checking inputs.ref
          if [[ "${{ inputs.run_all_tests }}" == "true" ]]; then
            echo "run_all_tests=true" >> $GITHUB_OUTPUT
            echo "Run mode: ALL TESTS (run_all_tests=${{ inputs.run_all_tests }})"
          else
            echo "run_all_tests=false" >> $GITHUB_OUTPUT
            echo "Run mode: FILTERED (triggered by ${{ github.event_name }})"
          fi
      - name: Detect file changes
        id: filter
        uses: dorny/paths-filter@v3
        if: steps.run-mode.outputs.run_all_tests != 'true'
        with:
          filters: |
            main_package:
              # Extend test/registered/ entries when adding a non-nightly register_xpu_ci.
              - "python/sglang/!(multimodal_gen|kernels|cli|test)/**/!(*.md)"
              - "python/sglang/test/*.py"
              - "python/sglang/test/!(ascend|observability|mock_model|manual|external_models|kernels)/**/!(*.md)"
              - "python/pyproject_xpu.toml"
              - "test/registered/xpu/**/!(*.md)"
              - "test/registered/disaggregation/test_disaggregation_xpu.py"
              - "test/registered/attention/test_chunk_gated_delta_rule.py"
              - "test/registered/attention/test_deterministic.py"
              - "test/registered/lora/test_moe_lora_info.py"
              - "test/registered/lora/test_virtual_experts_kernels.py"
              - "test/registered/unit/sampling/test_sampling_params.py"
              - "test/registered/unit/spec/test_adaptive_spec_params.py"
              - "test/run_suite.py"
              - "test/pytest.ini"
              - ".github/workflows/pr-test-xpu.yml"
              - "docker/xpu.Dockerfile"
              - "scripts/ci/xpu/**"
            multimodal_gen:
              # Only paths imported by the 1-gpu-xpu suite; other-vendor and off-suite files are excluded.
              - "python/sglang/multimodal_gen/*.py"
              - "python/sglang/multimodal_gen/runtime/!(pipelines)/**/!(*.md|*.ipynb)"
              - "python/sglang/multimodal_gen/runtime/*.py"
              - "python/sglang/multimodal_gen/runtime/pipelines/__init__.py"
              - "python/sglang/multimodal_gen/runtime/pipelines/@(zimage_pipeline|flux_2_klein|flux_2|wan_pipeline|diffusers_pipeline).py"
              - "python/sglang/multimodal_gen/configs/**/!(*.md|*.ipynb)"
              - "python/sglang/multimodal_gen/apps/webui/**/!(*.md|*.ipynb)"
              - "python/sglang/multimodal_gen/benchmarks/compare_perf.py"
              - "python/sglang/multimodal_gen/test/*.py"
              - "python/sglang/multimodal_gen/test/runner/**/!(*.md|*.ipynb)"
              - "python/sglang/multimodal_gen/test/server/!(test_server_1_gpu_5090|test_server_4_gpu_h100|test_server_b200|test_server_2_gpu|test_request_logger).py"
              - "python/sglang/multimodal_gen/test/server/common/**/!(*.md|*.ipynb)"
              - "python/sglang/multimodal_gen/test/server/perf_baselines/xpu_b60.json"
              - "python/pyproject_xpu.toml"
              - ".github/workflows/pr-test-xpu.yml"
              - "docker/xpu.Dockerfile"
              - "scripts/ci/xpu/**"

  # ==================== PR Gate ==================== #
  pr-gate:
    needs: check-changes
    if: needs.check-changes.outputs.changes_exist == 'true'
    uses: ./.github/workflows/pr-gate.yml
    secrets: inherit

  # Runs the entire per-commit XPU test suite in one job (single install, single queue).
  stage-a-test-1-gpu-xpu:
    needs: [check-changes, pr-gate]
    if: needs.check-changes.outputs.main_package == 'true'
    runs-on: intel-bmg
    env:
      DOCKERHUB_INTEL_USERNAME: ${{ secrets.DOCKERHUB_INTEL_USERNAME }}
      DOCKERHUB_INTEL_TOKEN: ${{ secrets.DOCKERHUB_INTEL_TOKEN }}
    steps:
      - name: Reset workspace ownership
        run: |
          # Prior runs install sglang as root inside the CI container against
          # the bind-mounted workspace, leaving root-owned egg-info on the host.
          # actions/checkout's pre-clean step can't unlink those as the runner
          # user, so reset ownership before checkout runs.
          docker run --rm -v "${{ github.workspace }}:/w" busybox:latest \
            chown -R "$(id -u):$(id -g)" /w || true

      - name: Checkout code
        uses: actions/checkout@v4
        with:
          fetch-depth: 0
          ref: ${{ inputs.ref || github.ref }}

      - name: Start CI container
        run: |
          export HF_TOKEN="$(cat ~/huggingface_token.txt)"
          bash scripts/ci/xpu/xpu_ci_start_container.sh
        env:
          GITHUB_WORKSPACE: ${{ github.workspace }}

      - name: Install Dependency
        timeout-minutes: 60
        run: |
          # Bridge fix: install libssl-dev in the running container so the
          # JIT build of hicache_hash_cpp (needs <openssl/sha.h>) succeeds.
          # Can be dropped once intel/sglang-dev:latest is rebuilt from a
          # docker/xpu.Dockerfile that already ships libssl-dev.
          docker exec -e DEBIAN_FRONTEND=noninteractive ci_sglang_xpu apt-get update
          docker exec -e DEBIAN_FRONTEND=noninteractive ci_sglang_xpu apt-get install -y libssl-dev
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --upgrade pip
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install pytest expecttest ray huggingface_hub tabulate "lmcache>=0.3.9" accelerate
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip uninstall -y flashinfer-python sgl-kernel sglang
          docker exec ci_sglang_xpu cp /sglang-checkout/python/pyproject_xpu.toml /sglang-checkout/python/pyproject.toml
          # Fetch tags so setuptools_scm resolves a real version instead of
          # falling back to 0.0.0 on a shallow/tag-less checkout.
          docker exec -w /sglang-checkout ci_sglang_xpu git fetch origin '+refs/tags/*:refs/tags/*' --force
          docker exec -w /sglang-checkout/python ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir . --extra-index-url https://download.pytorch.org/whl/xpu
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir --no-deps xgrammar==0.1.33
          docker exec ci_sglang_xpu /bin/bash -c '/opt/venv/bin/hf auth login --token ${HF_TOKEN}'

      - name: Install checkpoint-engine extra (optional; tests skip if unavailable)
        continue-on-error: true
        run: |
          docker exec -w /sglang-checkout/python ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir ".[checkpoint-engine]" --extra-index-url https://download.pytorch.org/whl/xpu

      - name: Run tests
        # Old stage-a (30) + stage-b (120) = 150; +20% headroom for --enable-retry.
        timeout-minutes: 180
        run: |
          docker exec ci_sglang_xpu bash -c "source /opt/venv/bin/activate && cd /sglang-checkout/test && python3 run_suite.py --hw xpu --suite stage-a-test-1-gpu-xpu,stage-b-test-1-gpu-xpu --enable-retry"

      - name: Cleanup container
        if: always()
        run: |
          # pip install ran as root inside the container against the
          # bind-mounted workspace, so build artifacts are root-owned on
          # the host. Chown them back before rm to avoid Permission denied.
          docker run --rm -v "${{ github.workspace }}:/w" busybox:latest \
            chown -R "$(id -u):$(id -g)" /w || true
          # Wipe everything the run wrote into the workspace so the next
          # job starts from a clean tree.
          rm -rf \
            python/build \
            python/dist \
            python/sglang.egg-info \
            python/sglang/*.egg-info \
            test/result.jsonl \
            test/results \
            test/.pytest_cache \
            .pytest_cache || true
          find . -type d -name "__pycache__" -prune -exec rm -rf {} + || true
          find . -type f -name "*.pyc" -delete || true
          # SIGTERM sglang and drain GPU context before `docker rm -f`;
          # SIGKILL leaves the xe/GuC exec queue registered and triggers a
          # GT reset (+ devcoredump) on B580.
          if docker ps --format '{{.Names}}' | grep -qx ci_sglang_xpu; then
            docker exec ci_sglang_xpu bash -c '
              pkill -TERM -f "sglang|run_suite|python3.*test_" 2>/dev/null || true
              for _ in $(seq 1 30); do
                pgrep -f "sglang::|sglang.launch_server" >/dev/null || break
                sleep 1
              done
              pkill -KILL -f "sglang|run_suite" 2>/dev/null || true
            ' || true
          fi
          docker rm -f ci_sglang_xpu || true
          if [[ -n "${CI_SGLANG_XPU_IMAGE:-}" ]]; then
            docker rmi -f "${CI_SGLANG_XPU_IMAGE}" || true
          fi

  # ==================== Multimodal Gen ==================== #
  multimodal-gen-test-1-gpu-xpu:
    needs: [check-changes, pr-gate]
    if: needs.check-changes.outputs.multimodal_gen == 'true'
    runs-on: bmg-multigen-models
    env:
      DOCKERHUB_INTEL_USERNAME: ${{ secrets.DOCKERHUB_INTEL_USERNAME }}
      DOCKERHUB_INTEL_TOKEN: ${{ secrets.DOCKERHUB_INTEL_TOKEN }}
    steps:
      - name: Reset workspace ownership
        run: |
          docker run --rm -v "${{ github.workspace }}:/w" busybox:latest \
            chown -R "$(id -u):$(id -g)" /w || true

      - name: Checkout code
        uses: actions/checkout@v4
        with:
          fetch-depth: 0
          ref: ${{ inputs.ref || github.ref }}

      - name: Start CI container
        run: |
          export HF_TOKEN="$(cat ~/huggingface_token.txt)"
          bash scripts/ci/xpu/xpu_ci_start_container.sh
        env:
          GITHUB_WORKSPACE: ${{ github.workspace }}

      - name: Install Dependency
        timeout-minutes: 60
        run: |
          # Bridge fix: install libssl-dev in the running container so the
          # JIT build of hicache_hash_cpp (needs <openssl/sha.h>) succeeds.
          # Can be dropped once intel/sglang-dev:latest is rebuilt from a
          # docker/xpu.Dockerfile that already ships libssl-dev.
          docker exec -e DEBIAN_FRONTEND=noninteractive ci_sglang_xpu apt-get update
          docker exec -e DEBIAN_FRONTEND=noninteractive ci_sglang_xpu apt-get install -y libssl-dev
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --upgrade pip
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install pytest expecttest ray huggingface_hub tabulate "lmcache>=0.3.9"
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip uninstall -y flashinfer-python sgl-kernel sglang
          docker exec ci_sglang_xpu cp /sglang-checkout/python/pyproject_xpu.toml /sglang-checkout/python/pyproject.toml
          # Fetch tags so setuptools_scm resolves a real version instead of
          # falling back to 0.0.0 on a shallow/tag-less checkout.
          docker exec -w /sglang-checkout ci_sglang_xpu git fetch origin '+refs/tags/*:refs/tags/*' --force
          docker exec -w /sglang-checkout/python ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir . --extra-index-url https://download.pytorch.org/whl/xpu
          docker exec ci_sglang_xpu /opt/venv/bin/python3 -m pip install --no-cache-dir --no-deps xgrammar==0.1.33
          docker exec ci_sglang_xpu /bin/bash -c '/opt/venv/bin/hf auth login --token ${HF_TOKEN}'

      - name: Run diffusion server tests (1-GPU)
        timeout-minutes: 60
        run: |
          # xpu_b60.json is seeded with this flag on; without it the SYCL queue
          # is deep enough that step timings record host enqueue, not device time.
          docker exec -e SGLANG_DIFFUSION_SYNC_STAGE_PROFILING=1 ci_sglang_xpu bash -c "source /opt/venv/bin/activate && cd /sglang-checkout/python && python3 sglang/multimodal_gen/test/run_suite.py --suite 1-gpu-xpu"

      - name: Cleanup container
        if: always()
        run: |
          docker run --rm -v "${{ github.workspace }}:/w" busybox:latest \
            chown -R "$(id -u):$(id -g)" /w || true
          rm -rf \
            python/build \
            python/dist \
            python/sglang.egg-info \
            python/sglang/*.egg-info \
            test/result.jsonl \
            test/results \
            test/.pytest_cache \
            .pytest_cache || true
          find . -type d -name "__pycache__" -prune -exec rm -rf {} + || true
          find . -type f -name "*.pyc" -delete || true
          # SIGTERM sglang and drain GPU context before `docker rm -f`;
          # SIGKILL leaves the xe/GuC exec queue registered and triggers a
          # GT reset (+ devcoredump) on B580.
          if docker ps --format '{{.Names}}' | grep -qx ci_sglang_xpu; then
            docker exec ci_sglang_xpu bash -c '
              pkill -TERM -f "sglang|run_suite|python3.*test_" 2>/dev/null || true
              for _ in $(seq 1 30); do
                pgrep -f "sglang::|sglang.launch_server" >/dev/null || break
                sleep 1
              done
              pkill -KILL -f "sglang|run_suite" 2>/dev/null || true
            ' || true
          fi
          docker rm -f ci_sglang_xpu || true
          if [[ -n "${CI_SGLANG_XPU_IMAGE:-}" ]]; then
            docker rmi -f "${CI_SGLANG_XPU_IMAGE}" || true
          fi

  finish:
    if: always()
    needs: [stage-a-test-1-gpu-xpu, multimodal-gen-test-1-gpu-xpu, pr-gate]
    runs-on: ubuntu-latest
    steps:
      - name: Check job status
        run: |
          stage_a="${{ needs.stage-a-test-1-gpu-xpu.result }}"
          multimodal_gen="${{ needs.multimodal-gen-test-1-gpu-xpu.result }}"
          if [ "$stage_a" != "success" ] && [ "$stage_a" != "skipped" ]; then
            echo "stage-a failed with result: $stage_a"
            exit 1
          fi
          if [ "$multimodal_gen" != "success" ] && [ "$multimodal_gen" != "skipped" ]; then
            echo "multimodal-gen failed with result: $multimodal_gen"
            exit 1
          fi
          echo "All jobs completed successfully"
          exit 0