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

[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.