Issues with openMP offload, with NVC. CUDA_EXCEPTION_14, Warp Illegal Address

I have an issue with my code. I have offload 2 functions of the code with OpenMP.

he first work well on the device. But with the second who call the first I have an issue and I don’t understand it.

In fact this function have a call to the other function (who wok well). Nut when I call it from the second I have an error: CUDA_EXCEPTION_14, Warp Illegal Address.

The first functin is call before the offload part of the second. The second function work well when I call the CPU version of the first.

I have try to use Intel integreted GPU and it work well. And I don’t see were the error can be.

I can post some code but it’s a big code, I can give you acces to the source code on github.

Thanks for your help !