Calling a host function from within the device

Accelerated Computing CUDA CUDA Programming and Performance
1

Exactly one year ago I used the current at the time CUDA version (don’t remember which one) and I wrote some CUDA kernels that made use of my vector class. This class has some functions, mostly overloaded operators. And they all worked fine. But today I downloaded that newest CUDA toolkit, try to compile my old project and…

error: calling a host function("vec3::operator +") from a __device__/__global__ function("rasterizePixel") is not allowed

Is there a way to fix that? And I would prefer a way that avoids any modifications to vec3 class.

2

I have a same problem when I tried to compile julia_gpu.cu (cuda_by_example/chapter04) with CUDA 7.5

Since Cuda By Example is a very old textbook, I guess CUDA of higher version has changed a lot and is not compatible with the old codes in the textbook.

Does anybody know how to solve this problem?

3
  1. Try decorating the function in question with host device
  2. Check the cuda by example sample code, I think it may have the issue fixed:

[url]https://developer.nvidia.com/cuda-example[/url]

4

Well, you mean the codes in CUDA By Example have been updated?

5

Unfortunately, however, the codes that I am using are the latest ones.