set default for llvm-ptx to False?

[jwaldmann]

README says This will use optimised backend implementations where available. This sounds as if availability is checked in some automated way. Does not seem to work, as the llvm-ptx flag is True by default, and consequently (?) cabal install breaks if I don't have CUDA. Yes, I was able to figure out

cabal install  accelerate-fft --flags="-llvm-ptx" 

but you might make this flag default to False. - I guess llvm-cpu: True is fine since most people will have this installed.

[tmcdonell]

It certainly would be nice if this were more automated. I'm not sure if this can be done in a better way... maybe if the Setup.hs guesses whether CUDA is available, before committing to add that dependency (if not explicitly set either way)? It'd have to be a guess though, or provide all the same options that the cuda package has for locating your installation (which feels a bit heavy-weight).

Are there better ways to do it?