[Mlir-commits] [flang] [mlir] [openmp] [Flang][OpenMP] Initial mapping of Fortran pointers and allocatables for target devices (PR #71766)

llvmlistbot at llvm.org llvmlistbot at llvm.org
Wed Jan 3 04:25:39 PST 2024


agozillon wrote:

> I am back in the office only on 8th.
> 
> Thanks for making progress here.
> 
> Can we move the addition of the extra MapInfos to a pass in Flang?
> 
> It might be good to document the present flow and details in flang/docs. This is probably easy to do since you have given a lot of explanations and these can be copypasted.

Hi Kiran, thank you very much for the quick reply, especially as you're on vacation! Please do not feel rushed to answer any other prompts on this PR, they can wait till you're back, I hope you have an awesome well deserved vacation! :-) 

I'm happy to look into trying to add it as an extra pass, as well as adding some documentation. In this case, I was thinking about continuing to generate the extra MapInfos in the OpenMP lowering, but generate the block arguments that link to them later in the pass? But I can attempt to generate both the extra MapInfo and BlockArg in the opt pass and see how it goes as well!

https://github.com/llvm/llvm-project/pull/71766


More information about the Mlir-commits mailing list