0

I am using PFactory (panorama factory) to stitch images in batch mode. Currently it uses multiple threads to do the same on CPU. I want to run it on my nvidia GPU (which has 144 cores). I wrote a kernel program which calls - system(pfactory) - to execute one instance of PFactory. But when I compile it using nvcc, it throws up an error - can't execute a host function in device..."

Is there a way by which I could run multiple instances of this program on GPU cores?

Andro Selva
  • 53,910
  • 52
  • 193
  • 240
pprahul
  • 1
  • 1
  • @talonmies - Ok. I read about calling the host function from kernel. Does that mean the process still runs from host space and not from kernel space? – pprahul Oct 26 '12 at 05:14
  • No it means it nevers runs at all. What you are trying to do is simply impossible. The GPU can't run x86 code, it can't launch processes on the host, or anything else that you are thinking it can. – talonmies Oct 26 '12 at 05:28

1 Answers1

1

CUDA kernels and device functions cannot directly call host functions. system is a host function.

There is no way to do what you ask.

harrism
  • 26,505
  • 2
  • 57
  • 88