# Visionworks - Openvx with cuda user node compilation

**URL:** <https://forums.developer.nvidia.com/t/visionworks-openvx-with-cuda-user-node-compilation/45466>\
**Category:** GPU-Accelerated Libraries\
**Created:** [October 12, 2016, 3:49pm UTC](https://forums.developer.nvidia.com/t/visionworks-openvx-with-cuda-user-node-compilation/45466 "2016-10-12T15:49:42Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Alou](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@Alou](https://forums.developer.nvidia.com/u/Alou)\
**Post date:** [October 12, 2016, 3:49pm UTC](https://forums.developer.nvidia.com/t/visionworks-openvx-with-cuda-user-node-compilation/45466/1 "2016-10-12T15:49:42Z")

</div>

Hi

I am new here and might be posting on the wrong part of the forum. So feel free to move me :)

I am trying to compile openvx code which contain a CUDA kernel based user node. Do I use the windows compiler (cl.exe) or nvcc?

I have already tried both without success.

cl.exe doesn t recognise the CUDA kernel declared with **global** (which I guess is normal). But when I use nvcc, I have the following error on the node registration:

error: argument of type “vx\_status (\*)(vx\_node, const vx\_reference \*, vx\_uint32)” is incompatible with parameter of type “vx\_kernel\_f”.

Anyone has an idea?

Thanks
