onboard-arch-precheck

To prevent running onboard tests with the wrong Ascend silicon, invoke an arch precheck script before pytest or task-submit to verify the --platform value matches the detected host architecture and block mismatched invocations.

Ascend silicon detection works by reading npu-smi output and CANN platform_config Short_SoC_version to map the host hardware to a repo arch, caching the detection result for one hour to speed up subsequent onboard test checks.

You do not need to run silicon arch detection for a2a3sim or a5sim platform tests because these sim variants are silicon-agnostic and automatically skip the precheck to keep workflows fast.

When the --platform value does not match the detected Ascend arch, the precheck exits with a code of 2 to block the mismatched invocation, exiting with 1 if detection fails or 0 on a successful match.

You should use an Ascend platform precheck before submitting tasks when invoking onboard hardware tests, reproducing flaky-test reports, or wrapping pytest in task-submit to ensure hardware matches before any device is locked.

Yes, you can verify CANN platform_config before running pytest by using a precheck that reads npu-smi output and Short_SoC_version to confirm your host Ascend silicon matches the specified --platform.