Optimization options in CUDA

Accelerated Computing CUDA CUDA Programming and Performance
#1

Hi all:

Does any body know what optimizations one can do with -O level with in cuda? Can we switch on or off these optimization options?

Thanks,

AM

#2

In theory you can pass optimisation options to ptxas like:
–ptxas-options="-O#"
where # stands for a number {0 … 4}. However I didn’t notice any difference. On the other hand, you can also pass optimisation parameter into nvopencc and that has an impact on the final code.
–opencc-options="-O#"
Be careful with that however, after disabling optimisation on this level compiler was impared in detecting when a pointer points to shared memory instead of global one and resulted in incorrect code!