CUDA disable L1 cache only for one variable

Question

Is there any way on CUDA 2.0 devices to disable L1 cache only for one specific variable? I know that one can disable L1 cache at compile time adding the flag -Xptxas

Answer 1

If you declare the variable to be volatile, then it will only be cached in the L2 cache on Fermi GPUs. Note that some compiler optimizations, such as removing repeated loads, are not performed on volatile variables because the compiler assumes they may be written by another thread.