LLNL
Our var of CUDA_SEPARABLE_COMPILATION should be BLT_ CUDA_SEPARABLE_COMPILATION
CUDA_SEPARABLE_COMPILATION is the per target prop name, CMAKE_CUDA_SEPARABLE_COMPILATION is the global setting
but we are using: CUDA_SEPARABLE_COMPILATION in the blt macros
On the subject of CUDA related variable, should there be a BLT_CUDA_ARCH variable that does the right thing? I have seen multiple ways to set this.
Adding BLT_ to the cmake variables that BLT handles would help users understand what BLT is doing and what it is not.
I agree that the naming is a little confusing. I recently noticed that we are passing -DCUDA_SEPARABLE_COMPILATION to our dependencies that use BLT, but it seems like we should be using -DCMAKE_CUDA_SEPARABLE_COMPILATION instead, though I'm not sure what BLT is actually expecting.
