Hi all,
I’m trying to get some information about register usage in my CUDA kernels using NVCC option
–ptxas-options=v and while with global functions everything is ok, I’m having some difficulties with the device ones since the
ptxas info : Used N registers
line is missing in the output. I tried to use the noinline keyword and to keep them in another file, with respect to the calling global function, since I thought that NVCC was reporting the full register usage of the global function including the called device ones after the inline but nothing changes. I can get the information about register usage of the device functions only defining them as global.
Do you have any suggestions?
Thanks!