Thanks to visit codestin.com
Credit goes to github.com

Skip to content

[CUDA] illegal memory read on Transpose #104360

@kokol16

Description

@kokol16

Issue type

Bug

Have you reproduced the bug with TensorFlow Nightly?

No

Source

source

TensorFlow version

tf 2.20

Custom code

Yes

OS platform and distribution

Ubuntu 22.04

Mobile device

No response

Python version

3.10

Bazel version

No response

GCC/compiler version

No response

CUDA/cuDNN version

CUDA 12.5.1, cuDNN 9.2.1

GPU model and memory

No response

Current behavior?

Compute-Sanitizer reports an out of bounds read on SwapDimension1And2InTensor3UsingTiles

Standalone code to reproduce the issue

import tensorflow as tf
import numpy as np

def test_transpose_op_raw():
    with tf.device('/job:localhost/replica:0/task:0/device:GPU:0'):
        # Create tensors
        x = tf.ones([46341, 46341], dtype=tf.float16)
        perm = tf.constant([1, 0], dtype=tf.int32)
        
        # Raw ops call with exact parameters from mutation
        result = tf.raw_ops.Transpose(
            x=x,
            perm=perm
        )
        
        return result

if __name__ == "__main__":
    
    try:
        # Run the main test
        result = test_transpose_op_raw()
        print("Test completed successfully")
        
    except Exception as e:
        print(f"Test failed with error: {e}")
        print("This may be the expected behavior for this mutation")

Relevant log output

========= Invalid __global__ read of size 2 bytes
=========     at void tensorflow::functor::SwapDimension1And2InTensor3UsingTiles<unsigned short, (int)256, (int)32, (int)32, (bool)0>(const T1 *, tensorflow::functor::Dimension<(int)3>, T1 *)+0xcd0
=========     by thread (128,0,0) in block (2099518,0,0)
=========     Address 0x7f66d60015a8 is out of bounds

Metadata

Metadata

Assignees

Labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions