Hi,
When I write an API code in CUDA GPU, for example, calculate contact force with calculateForce(). I noticed that I need to define the main function for both CPU cpp and GPU cu files. Do both of the main functions calculate contact force or only GPU code is calculating? Besides, what shall I do if I want only my CPU code to run for a certain timestep and GPU code keep sleeping for that timestep?
Kindly regards,
Qi
Hi Qi,
If you are using the CUDA GPU solver (which you would be if it's a .cu file) then the physics model is run 100% on the GPU. You can leave the physics section of the CPU blank if you like, all this would mean is that it would be GPU compatible only, not able to run on CPU.
You do still need the 'extras' in the CPU code, not for the physics but some of the main things the CPU code does is:
Passes the name of the .cu file to EDEM
Passes the version of the code to EDEM
Reads in any parameters from a .txt file in and passes these to the GPU
It is a lot easier to debug in CPU with something like Visual Studio, so often code is written in the CPU and then any debugging/testing can be done on the CPU and the code the copied/ported to the GPU. Not necessary but if you just write it for the GPU and have any problems with the physics you'd have to debug using printf.
Regards
Stephen