[flang-commits] [flang] [llvm] [flang][cuda] Do not use double descriptor for unified allocatables (PR #190713)

[Valentin Clement バレンタイン クレメン via flang-commits]

clementval wrote:

> > This is only because you have the save attribute on it.
> 
> 
> 
> Yes our test uses explicit save. But managed and unified are lowered differently at program scope, managed is lowered as cuf.alloc while unified becomes fir.global internal. Should we fix in the lowering instead for unified so that it's lowered as cuf.alloc?

They should be lowered the same way 

https://github.com/llvm/llvm-project/pull/190713