[ci][xpu] Record device time in the multimodal_gen perf lane (#39956)

2026-09-21 15:39:29 +08:00
parent b63f8416b3
commit c2f860af1c
2 changed files with 161 additions and 154 deletions
@@ -251,7 +251,9 @@ jobs:
      - name: Run diffusion server tests (1-GPU)
        timeout-minutes: 60
        run: |
-          docker exec 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"
+          # 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()