tf.nn.depthwise_conv2d crashes with large strides values when ONEDNN is enabled #97165
Description
Issue type
Bug
Have you reproduced the bug with TensorFlow Nightly?
Yes
Source
source
TensorFlow version
2.20.0-dev20250716
Custom code
Yes
OS platform and distribution
Linux Ubuntu 20.04
Mobile device
No response
Python version
3.12
Bazel version
No response
GCC/compiler version
No response
CUDA/cuDNN version
No response
GPU model and memory
No response
Current behavior?
Running tf.nn.depthwise_conv2d on CPU with large values of strides crashes when ONEDNN is enabled with Aborted (core dumped). Setting TF_ENABLE_ONEDNN_OPTS=0 does not cause the crash and throws an InvalidArgumentError instead (expected behavior).
Reproducible in colab with 2.19.0 (colab)
Standalone code to reproduce the issue
import tensorflow as tf
print(tf.__version__) # 2.20.0-dev20250716
input_tensor = tf.constant([[[[1.],[2.]],[[3.],[4.]]]], dtype=tf.float32)
filter_tensor = tf.constant([[[[1., 2.]]]], dtype=tf.float32)
strides = [4817177250100823153, 5276955028067489600, -6860092642535747309, -915217906097603218]
padding = 'VALID'
data_format = 'NHWC'
dilations = [1, 1]
with tf.device('/cpu:0'):
output_cpu = tf.nn.depthwise_conv2d(input_tensor, filter_tensor, strides=strides, padding=padding, data_format=data_format, dilations=dilations, name='depthwise_conv2d_1')Relevant log output
2025-07-18 03:42:56.115716: F tensorflow/core/common_runtime/mkl_layout_pass.cc:2703] Non-OK-status: GetNodeAttr(orig_node->def(), "strides", &strides)
Status: INVALID_ARGUMENT: Attr strides has value 4817177250100823153 out of range for an int32
Aborted (core dumped)