I was trying to use this function on a C470 card. Whenever I used it, an “invalid device function” error was given. I used -arch sm_20. Is there any other options that should be specified?
What form are you using? I’ve had luck with cudaFuncSetCacheConfig(kernel_function, …). If you use the form cudaFuncSetCacheConfig(“kernel_function”, …) you have to put the fully-qualified C++ mangled name in the quotes - something that is not easy to determine.
Improved in 3.2!
Oh, yes, the other formats works.
Thank you all!