Optimize the bce_loss op to avoid copy input back to CPU.

[Xreki]

PR types

Performance optimization

PR changes

OPs

Describe

bce_loss OP要求输入数据每个元素的值在[0, 1]范围内，因此将输入数据从GPU拷贝到CPU上检查，比较低效。

优化方法：在CUDA Kernel中使用PADDLE_ENFORCE进行检查。

[paddle-bot-old]

Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

[Xreki]

2021-04-14 15:42:51 0. The error message you wrote in PADDLE_ENFORCE{**} or PADDLE_THROW does not meet our error message writing specification. Possible errors include 1. the error message is empty / 2. the error message is too short / 3. the error type is not specified. Please read the specification [ https://github.com/PaddlePaddle/Paddle/wiki/Paddle-Error-Message-Writing-Specification ], then refine the error message. If it is a mismatch, please request chenwhql (Recommend), luotao1 or lanxianghit review and approve.
2021-04-14 15:42:51 The PADDLE_ENFORCE{**} or PADDLE_THROW entries that do not meet the specification are as follows:
2021-04-14 15:42:51 PADDLE_ENFORCE( + (x >= static_cast(0)) && (x <= one), + "Input is expected to be within the interval [0, 1], but recieved %f.", + x);
2021-04-14 15:42:51 There are 1 approved errors.

经确认，CUDA Kernel里面只能使用PADDLE_ENFORCE进行错误检查，且不能使用CPU端定义的error类型。

[chenwhql]

LGTM for PADDLE_ENFORCE