regarding cuda performance optimization

1
votes

For those cuda functions which are only going to be called from device, does it matter in terms of performance if we declare them __ global __ or __ device __ ?

1
performancecuda
Your question makes no sense - you can't call __global__functions from device code in CUDA. Kernels cannot launch other kernels. - talonmies
Thanks for the clarification. I see. - user1118148

1 Answers

1
votes

As of CUDA 4.1, __global__ functions cannot be called from __device__ functions.