floating point exception in tf.nn.max_pool3d when value of ksize is too large.

[xl1043237213]

Issue type

Bug

Have you reproduced the bug with TensorFlow Nightly?

Yes

Source

binary

TensorFlow version

tf 2.15

Custom code

No

OS platform and distribution

Linux Ubuntu 20.04

Mobile device

No response

Python version

3.10

Bazel version

No response

GCC/compiler version

No response

CUDA/cuDNN version

No response

GPU model and memory

No response

Current behavior?

floating point exception in tf.nn.max_pool3d when value of ksize is too large, which will cause a crash. This crash occurs in the CPU execution environment.

Standalone code to reproduce the issue

https://colab.research.google.com/drive/1S9D9zdhknu9ReoPahd0_kw2OEuRxLAvB?usp=sharing

Relevant log output

Floating point exception (core dumped)

[tilakrayal]

@xl1043237213,
It's not being killed because of TF. You are literally allocating so much memory that the OS is killing the process. It seems like you're giving large negative value for multiple arguments to the function. so due to Integer overflow to buffer overflow or due to insufficient memory (RAM), code is crashing or the process getting killed and I was able to replicate the issue on Google colab. Could you please specify the usecase where you are trying with the large ksize.

input: A 5-D Tensor of the format specified by data_format.
ksize: An int or list of ints that has length 1, 3 or 5. The size of the window for each dimension of the input tensor.

Thank you!

[xl1043237213]

Thanks for your response. : )
I found the problem with a self-designed fuzzy testing tool, and have not yet found a relevant application scenario that requires setting ksize to such a large value. However, when the value of ksize is accessible to the user, packed in the model, it can cause the program to hang or even crash.

[tilakrayal]

I was able to reproduce the issue on tensorflow v2.15 and v2.16. Kindly find the gist of it here.