I’ve been trying to run auto-scheduling on an AGX Xavier device.
However, I cannot get a speedup. Auto-scheduling runs and finishes, but produces no usable output. If I look at the execution, I get messages saying MeasureResult(error_type:RuntimeDeviceError, error_msg:Traceback (most recent call last)
.
This exact same code works fine on other NVidia devices, and I can run non-tuned programs on the device, suggesting that my CUDA libraries are fine. Has anyone experienced a similar issue, or been able to tune on the AGX Xavier before?