Created by: pengwangucla

    Matrix::resizeOrCreate(
        targetCpu, target.getHeight(), target.getWidth(), false, false);
    Matrix::resizeOrCreate(
        outputCpu, output.getHeight(), output.getWidth(), false, false);
    Matrix::resizeOrCreate(labelCpu,
                           label.value->getHeight(),
                           label.value->getWidth(),
                           false,
                           false);
    targetCpu->copyFrom(target);
    outputCpu->copyFrom(output);
    labelCpu->copyFrom(*label.value);
    targetCpu->smoothL1(*outputCpu, *(labelCpu));
    target.copyFrom(*targetCpu);

It is not right to do Gpu computing by copy gpu memory to cpu and compute at cpu, moreover, the gpu implementation is not difficult.

Someone should change this.