Cuda 10.0 to 10.2

Hi,

I upgraded to Jetpack 4.4DP from Jetpack 4.2.2 rev1. This updated Cuda from 10.0 to 10.2. An application that used to run under 8ms now takes 45ms. Anything that I need to update for this application to revert back to previous speeds?

Thanks.

Hi,

May I know which application do you use?
Please noticed that you will need to recompile a CUDA related app since the toolkit version is upgraded.

Thanks.

Hi,

It is a cuda based application. I did recompile with cuda10.2. Without recompiling I got CuSparse library not found error. Does it by default compile in debug mode? Should I explicitly mention release mode? I did not do anything like that in cuda10.0.

Thanks.

Hi,

nvcc default is release mode. Debug mode need to be trigger with this --debug flag.
https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html#options-for-altering-compiler-linker-behavior

Thanks.

Thank you. I am still unable to match the compiles between 10.0 and 10.2.

Hi, any suggestions on this?