Kirin X90 and Kirin 9030 do not support L2 cache, and the GlobalTensor::SetL2CacheHint API does not take effect. The operator code does not need to be modified. The performance is affected, but the function is not affected.