[Big-Tensor Fix] Add error throwing check of ctc_loss

[HeyDavid633]

PR Category

Operator Mechanism

PR Types

Bug fixes

Description

修复 paddle.nn.functional.ctc_loss 的大Tensor问题

该实现调用了第三方库 warp-ctc 其中关于 kernel 启动的配置 Paddle/third_party/warpctc/include/detail/gpu_ctc.h 也决定了该API不支持大Tensor，属于第三方库不支持的范围。对于测试case的修改见 PFCCLab/PaddleAPITest#432

在启动 Kernel 之前，对输入的张量尺寸进行检查，并抛出非法错误

PADDLE_ENFORCE_LT( num_sequences * sequence_width * max_sequence_length, std::numeric_limits<int>::max(), errors::InvalidArgument( "The total number of elements in Input(Logits) should be less than " "%d, " "but received %d", std::numeric_limits<int>::max(), num_sequences * sequence_width * max_sequence_length));

[CLAassistant]

All committers have signed the CLA.

[paddle-bot]

你的PR提交成功，感谢你对开源项目的贡献!
请关注后续CI自动化测试结果，详情请参考Paddle-CI手册。
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

[wanghuancoder]

LGTM