fix for div zero

[zh794390558]

PR types

Bug fixes

PR changes

OPs

Describe

fix for div zero error

functional.ctc_loss存在除0异常
调用paddle.nn.functional.ctc_loss进行计算的时候，在sequence_padding.cc中存在除0异常。

PoC:

import paddle
from paddle.nn import TransformerEncoderLayer
import numpy as np

input_arr = np.array([], dtype=np.float32)
log_probs = paddle.to_tensor(np.reshape(input_arr, (3, 0, 19)), dtype='float32')

labels_arr = np.array([], dtype=np.float32)
labels = paddle.to_tensor(np.reshape(input_arr, (0, 0)), dtype='float32')

input_lengths = paddle.to_tensor([], dtype='float32')

label_lengths = paddle.to_tensor([], dtype='float32')

paddle.nn.functional.ctc_loss(log_probs, labels, input_lengths, label_lengths, blank=0)
ASAN Info:

==162441==ERROR: AddressSanitizer: FPE on unknown address 0x7f3bb9b98db4 (pc 0x7f3bb9b98db4 bp 0x7ffc8f998740 sp 0x7ffc8f9985c0 T0)
#0 0x7f3bb9b98db4 in paddle::operators::math::UnpaddingLoDTensorFunctor<paddle::platform::CPUDeviceContext, int>::operator()(paddle::platform::CPUDeviceContext const&, paddle::framework::LoDTensor const&, paddle::framework::LoDTensor*, int, int, bool, paddle::operators::math::PadLayout) /home/work/yakun/py-paddle/Paddle/paddle/fluid/operators/math/sequence_padding.cc:152:42
#1 0x7f3bb118672d in paddle::operators::WarpCTCKernel<paddle::platform::CPUDeviceContext, float>::Compute(paddle::framework::ExecutionContext const&) const /home/work/yakun/py-paddle/Paddle/paddle/fluid/operators/warpctc_op.h:321:9
#2 0x7f3bb1181fae in paddle::framework::OpKernelRegistrarFunctor<paddle::platform::CPUPlace, false, 0ul, paddle::operators::WarpCTCKernel<paddle::platform::CPUDeviceContext, float>, paddle::operators::WarpCTCKernel<paddle::platform::CPUDeviceContext, double> >::operator()(char const*, char const*, int) const::'lambda'(paddle::framework::ExecutionContext const&)::operator()(paddle::framework::ExecutionContext const&) const /home/work/yakun/py-paddle/Paddle/paddle/fluid/framework/op_registry.h:182:25
#3 0x7f3bb1181fae in std::_Function_handler<void (paddle::framework::ExecutionContext const&), paddle::framework::OpKernelRegistrarFunctor<paddle::platform::CPUPlace, false, 0ul, paddle::operators::WarpCTCKernel<paddle::platform::CPUDeviceContext, float>, paddle::operators::WarpCTCKernel<paddle::platform::CPUDeviceContext, double> >::operator()(char const*, char const*, int) const::'lambda'(paddle::framework::ExecutionContext const&)>::_M_invoke(std::_Any_data const&, paddle::framework::ExecutionContext const&) /usr/bin/../lib/gcc/x86_64-linux-gnu/8/../../../../include/c++/8/bits/std_function.h:297:2
#4 0x7f3bbb5b56dc in std::function<void (paddle::framework::ExecutionContext const&)>::operator()(paddle::framework::ExecutionContext const&) const /usr/bin/../lib/gcc/x86_64-linux-gnu/8/../../../../include/c++/8/bits/std_function.h:687:14
Python version: 3.8

paddlepaddle版本：Develop （2021-07-14 pull）

漏洞代码
paddlepaddle develop中，代码位置：https://sourcegraph.com/github.com/PaddlePaddle/Paddle@develop/-/blob/paddle/fluid/operators/math/sequence_padding.cc?L152

问题代码：

int step_width = seq_tensor->numel() / seq_tensor_dims[0];
其中，由于没做check，seq_tensor_dims[0]为0，触发异常

[paddle-bot-old]

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

[XiaoguangHu01]

LGTM