Created by: wanghaoshuang

The unittest of warp-CTC op will fail if use_cudnn was true with cuda10. It seems that the gradients given by cudnnCTCloss is different between cuda8 and cuda10.

The cudnn kernel of warpctc has been removed in https://github.com/PaddlePaddle/Paddle/pull/19730. And the implement of baidu/research is enough for applications in speech or NLP fields.

But it is also necessary to check the reason of failure in cuda10.