[flang-commits] [flang] [llvm] [flang][cuda] share runtime type info between host and device under managed memory (PR #229213)

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

clementval wrote:

> Does a type-bound procedure call from an attributes(global) kernel still work with -gpu=mem:managed? And if cudaHostRegister fails in the constructor, does the whole program abort at startup?

No, that is a restriction of this mode. The reference compiler does not support TBP on the device so it's not a regression but we will need to do some work to support it in managed mode. 

>  if cudaHostRegister fails in the constructor, does the whole program abort at startup?

yes it should abort when `cudaHostRegister` fails

https://github.com/llvm/llvm-project/pull/229213