[Openmp-commits] [PATCH] D41817: [OMPT] Enable OMPT on 32-bit ARM machines

[Joachim Protze via Phabricator via Openmp-commits]

protze.joachim added a comment.

In https://reviews.llvm.org/D41817#973066, @pawosm01 wrote:

>   Library built with: Clang  | GCC
>   Testing with:              |
>   Clang               works  | fails
>   GCC                 fails  | fails
>  
>


I think, this is how we understood that. But at the end this means, that a tool might see an strange address, i.e. neither the proper frame address nor NULL, for any of these failing combinations. As I understood Jonas, we should therefore return NULL as the frame address, independent of the compiler in use.

I can see three approaches:

- Generally return NULL on arm32:

`#define OMPT_GET_FRAME_ADDRESS(level) NULL`

- Generally return NULL on arm32, if level>0:

`#define OMPT_GET_FRAME_ADDRESS(level) ((level==0)?__builtin_frame_address(level):NULL)`

- go into the runtime and use NULL for the cases, where the tests break.

The tricky part for testing any of these changes is how NULL is printed :(
I would feel better if we could provide something like (~(0L)) for the case where the frame address cannot be determined.


Repository:
  rOMP OpenMP

https://reviews.llvm.org/D41817