stack smashing when using visionworks 1.6 and cuda 9.0

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Hi,

We are using DeepStream 1.5, but we want to using visionworks together, the question is: DeepStream 1.5 depend on CUDA 9.0 but visionworks 1.6 depend on CUDA 8.0. We tried write some code just using CUDA 9.0 and let the libvisionworks.so link to CUDA 8.0, but "stack smashing detected" reported.
We made a minimum code base to see is there any one can help us.

//main.cpp
#include <iostream>

using namespace std;

#include <cuda_runtime_api.h>
#include <NVX/nvx.h>

int main()
{
    cout << "Hello world!" << endl;

    void* data;
    cudaMalloc(&data, 3 * 448 * 640 * 4); //<b>link to CUDA 9.0</b>

    vx_context ctx = vxCreateContext(); //<b>*** stack smashing detected ***</b>
    vxReleaseContext(&ctx);

    cudaFree(data);
    return 0;
}

The attachments is code project and crash snapshot, and the driver is 390.30

Thanks.
vx_cuda9.zip (1.11 KB)

Screenshot from 2018-04-27 17-24-59.png
Screenshot from 2018-04-27 17-24-59.png1372×917 129 KB

2

Hi,

VisionWorks for x86 Ubuntu doesn’t support CUDA 9.0 yet.
Please use CUDA 8.0 and DeepStream-1.0 with VisionWorks-1.6.

Thanks.

3

Hi,

We are trying to use some other “Two-Frame Motion Estimation” library, any advise?
It seems NVIDIA doing some research with CNNs for Optical Flow, Is there any more pages to share?

PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume
http://research.nvidia.com/publication/2018-02_PWC-Net%3A-CNNs-for

Thanks.

4

Hi,

Here is a relevant topic for your reference:

Thanks.

5

Hi,

We will do some test with flownet2-pytorch when we have time.

Thanks.