[AutoTVM] Resnet50 and MobileNetv2 after auto-tvm tuning is much slower than the optimized assembly code on ARM Cortex A53

[merrymercy]

delete -device=arm_cpu in the target string?