I parallelized a C++ code using OpenACC in one computer under Ubuntu and with a NVIDIA GTX1080. To compile the code I used
pgc++ -Minform=inform -Minfo=all,ccff -std=c++11 -ta=tesla:cc60 -ta=time *.cpp
To test for the correctness of the Open ACC pragmas I compared the code run in the device against the outputs obtained after compiling using -ta=host.
As expected, after a number of time steps both outputs differ, that’s way I used short simulations first and longer simulations later. However, for the short simulations (100 time steps, sampling every 10 time steps), the match between both versions was almost perfect.
Now, I have to use another computer with the same code. The computer also uses Ubuntu and a NVIDIA GTX1080 (Not sure if the drivers are the same version).
The point is that, compiling and running the same code in the new computer, I’m getting an error much more relevant when comparing the host and tesla compilations.
Can you provide some help? (I looked for existing posts but I’ve not been able to find one, which does not mean it doesn’t exist already).
Thank you very much for your help,