vxLoadKernels() in VisionWorks 1.6

Accelerated Computing GPU-Accelerated Libraries
1

Hello,
While trying to load an OpenVX user kernel using vxLoadKernels() in cuda project using Nsight, an error is occurred,
‘[NVX LOG] context=(nil),ref=0x80e8f0,status=-11 : vxLoadKernels: failed to load symbol vxPublishKernels in module my_user_kernel’.

This error does NOT occur in the following cases:

  1. OpenVX user kernel implementation without VisionWorks. The user kernel, in shared object, gets loaded using vxLoadKernel() in c++ application. Then, vxGetKernelByName() followed by vxCreateGenericNode() creates user node.
  2. Nsight cuda project compiled with source files of OpenVX user kernel implementation.
  3. An OpenVX user kernel is generated as a shared object. It is linked to a cuda project, and the project is compiled with an associated header file.