Update nightly report track

.github/actions/inductor-xpu-e2e-test/action.yml

@@ -108,14 +108,18 @@ runs:
               do
                 contains "accuracy,performance" $scenario
                 $contains_status
-                bash inductor_xpu_test.sh ${suite} ${dt} ${mode} ${scenario} xpu 0 static 8 0 &
-                bash inductor_xpu_test.sh ${suite} ${dt} ${mode} ${scenario} xpu 1 static 8 1 &
-                bash inductor_xpu_test.sh ${suite} ${dt} ${mode} ${scenario} xpu 2 static 8 2 &
-                bash inductor_xpu_test.sh ${suite} ${dt} ${mode} ${scenario} xpu 3 static 8 3 &
-                bash inductor_xpu_test.sh ${suite} ${dt} ${mode} ${scenario} xpu 4 static 8 4 &
-                bash inductor_xpu_test.sh ${suite} ${dt} ${mode} ${scenario} xpu 5 static 8 5 &
-                bash inductor_xpu_test.sh ${suite} ${dt} ${mode} ${scenario} xpu 6 static 8 6 &
-                bash inductor_xpu_test.sh ${suite} ${dt} ${mode} ${scenario} xpu 7 static 8 7 &
+                if [ "${MODEL_ONLY_NAME}" == "" ];then
+                  bash inductor_xpu_test.sh ${suite} ${dt} ${mode} ${scenario} xpu 0 static 8 0 &
+                  bash inductor_xpu_test.sh ${suite} ${dt} ${mode} ${scenario} xpu 1 static 8 1 &
+                  bash inductor_xpu_test.sh ${suite} ${dt} ${mode} ${scenario} xpu 2 static 8 2 &
+                  bash inductor_xpu_test.sh ${suite} ${dt} ${mode} ${scenario} xpu 3 static 8 3 &
+                  bash inductor_xpu_test.sh ${suite} ${dt} ${mode} ${scenario} xpu 4 static 8 4 &
+                  bash inductor_xpu_test.sh ${suite} ${dt} ${mode} ${scenario} xpu 5 static 8 5 &
+                  bash inductor_xpu_test.sh ${suite} ${dt} ${mode} ${scenario} xpu 6 static 8 6 &
+                  bash inductor_xpu_test.sh ${suite} ${dt} ${mode} ${scenario} xpu 7 static 8 7 &
+                else
+                  bash inductor_xpu_test.sh ${suite} ${dt} ${mode} ${scenario} xpu 0 static 1 0 ${MODEL_ONLY_NAME} &
+                fi
                 wait
               done
             done

.github/workflows/inductor_xpu_e2e_nightly.yml

@@ -36,18 +36,28 @@ on:
         type: string
         default: 'accuracy'
         description: accuracy,performance. Delimiter is comma
+      model:
+        required: false
+        type: string
+        default: ''
+        description: If set, will only launch this one
 
 
 permissions: read-all
 
 concurrency:
-  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}-${{ github.event_name == 'schedule' }}-${{ inputs.suite }}-${{ inputs.dt }}-${{ inputs.mode }}-${{ inputs.secnario }}
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}-${{ github.event_name == 'schedule' }}-${{ inputs.suite }}-${{ inputs.dt }}-${{ inputs.mode }}-${{ inputs.scenario }}-${{ inputs.triton }}-${{ inputs.model }}
   cancel-in-progress: true
 
 jobs:
   Inductor-XPU-E2E-Nightly-Tests:
     runs-on: pvc_e2e
     timeout-minutes: 900
+    outputs:
+      TORCH_BRANCH_ID: ${{ steps.pinned.outputs.TORCH_BRANCH_ID }}
+      TORCH_COMMIT_ID: ${{ steps.pinned.outputs.TORCH_COMMIT_ID }}
+      DKMS_VERSION: ${{ steps.pinned.outputs.DKMS_VERSION }}
+      DPCPP_VERSION: ${{ steps.pinned.outputs.DPCPP_VERSION }}
     steps:
       - name: Checkout torch-xpu-ops
         uses: actions/checkout@v4
@@ -75,20 +85,26 @@ jobs:
           # Workaround for torch-xpu-ops ci test
           sed -i "s/checkout --quiet \${TORCH_XPU_OPS_COMMIT}/log -n 1/g" caffe2/CMakeLists.txt
       - name: Identify pinned versions
+        id: pinned
         run: |
           cd ../pytorch
           if [ -z ${{ inputs.triton }} ]; then
             echo "TRITON_COMMIT_ID=$(<.ci/docker/ci_commit_pins/triton-xpu.txt)" >> "${GITHUB_ENV}"
           else
             echo "TRITON_COMMIT_ID=${{ inputs.triton }}" >> "${GITHUB_ENV}"
           fi
-          echo "TORCH_COMMIT_ID=$(git rev-parse HEAD)" >> "${GITHUB_ENV}"
+          echo "TORCH_BRANCH_ID=$(git rev-parse --abbrev-ref HEAD)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
+          echo "TORCH_COMMIT_ID=$(git rev-parse HEAD)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
           echo "TORCHBENCH_COMMIT_ID=$(<third_party/torch-xpu-ops/.github/ci_commit_pins/torchbench.txt)" >> "${GITHUB_ENV}"
           echo "TORCHVISION_COMMIT_ID=$(<.github/ci_commit_pins/vision.txt)" >> "${GITHUB_ENV}"
           echo "TORCHTEXT_COMMIT_ID=$(<.github/ci_commit_pins/text.txt)" >> "${GITHUB_ENV}"
           echo "TORCHAUDIO_COMMIT_ID=$(<.github/ci_commit_pins/audio.txt)" >> "${GITHUB_ENV}"
           echo "TRANSFORMERS_VERSION=$(<.ci/docker/ci_commit_pins/huggingface.txt)" >> "${GITHUB_ENV}"
           echo "TIMM_COMMIT_ID=$(<.ci/docker/ci_commit_pins/timm.txt)" >> "${GITHUB_ENV}"
+          echo "MODEL_ONLY_NAME=${{ inputs.model }}" >> "${GITHUB_ENV}"
+          source /opt/intel/oneapi/compiler/latest/env/vars.sh
+          echo "DKMS_VERSION=$(dkms status 2>&1 |grep 'intel-i915-dkms' |sed 's/.*\///;s/,.*//')" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
+          echo "DPCPP_VERSION=$(dpcpp --version 2>&1 |grep 'DPC++/C++' |sed 's/.*(//;s/).*//')" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
           echo ${GITHUB_ENV}
       - name: Triton Installation
         run: |
@@ -214,18 +230,37 @@ jobs:
     needs: Inductor-XPU-E2E-Nightly-Tests
     steps:
       - name: Report github issue for XPU OPS nightly
-        if: ${{ needs.Inductor-XPU-E2E-Nightly-Tests.result == 'success' }} && ${{ !inputs.suite }}
-        run: |
-          build_url="${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
-          repo="${{ github.repository }}"
-          gh --repo $repo issue comment 304  \
-              --body "XPU OPS Nightly Successful $(date +'%F'), See: $build_url"
-      - name: Report github issue when failure detected
-        if: ${{ needs.Inductor-XPU-E2E-Nightly-Tests.result == 'failure' }} && ${{ !inputs.suite }}
         run: |
+          set -xe
+          if [ "${GITHUB_EVENT_NAME}" == "workflow_dispatch" ];then
+            report_issue_id=426
+            test_type="On-demand"
+          else
+            report_issue_id=432
+            test_type="Nightly"
+          fi
+          if [ "${{ needs.Inductor-XPU-E2E-Nightly-Tests.result }}" == "success" ];then
+            test_status=Success
+          elif [ "${{ needs.Inductor-XPU-E2E-Nightly-Tests.result }}" == "failure" ];then
+            test_status=Failure
+            cc_comment="CC ${{ secrets.NIGHTLY_EMAIL_LIST }}"
+          else
+            test_status=None
+          fi
+          if [ "${GITHUB_EVENT_NAME}" == "workflow_dispatch" ];then
+            cc_comment="CC @${GITHUB_TRIGGERING_ACTOR}"
+            test_scope="Inputs: suite=${{ inputs.suite }}; dt=${{ inputs.dt }}; mode=${{ inputs.mode }}; scenario=${{ inputs.scenario }}; triton=${{ inputs.triton }}; model=${{ inputs.model }}"
+            test_sw="SW info: XPU=${GITHUB_WORKFLOW_SHA} on ${GITHUB_REF_NAME}, "
+            test_sw+="Torch=${{ needs.Inductor-XPU-E2E-Nightly-Tests.outputs.TORCH_COMMIT_ID }} on ${{ needs.Inductor-XPU-E2E-Nightly-Tests.outputs.TORCH_BRANCH_ID }}"
+            test_hw="HW info: Run on ${RUNNER_NAME} with DKMS=${{ needs.Inductor-XPU-E2E-Nightly-Tests.outputs.DKMS_VERSION }} and DPCPP=${{ needs.Inductor-XPU-E2E-Nightly-Tests.outputs.DPCPP_VERSION }}"
+          fi
           build_url="${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
-          notification_list="${{ secrets.NIGHTLY_EMAIL_LIST }}"
           repo="${{ github.repository }}"
-          gh --repo $repo issue comment 304  \
-              --body "XPU OPS Nightly Failed $(date +'%F'), See: $build_url
-                      cc $notification_list"
+          if [ "${test_status}" != "None" ];then
+            gh --repo $repo issue comment $report_issue_id --body "$cc_comment
+                  $(date +'%F') $test_type Test Status: ${test_status}
+                  See: $build_url
+                  $test_scope
+                  $test_sw
+                  $test_hw"
+          fi