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

[via Openmp-commits]

agozillon wrote:

Back from vacation and ready to push this PR further forward! If I could have some input on what needs to be done to get this into a good state for landing I'd very much appreciate it. 

I think the main question currently is if we're (primarily @kiranchandramohan in this case I think) okay leaving the addition of new MapInfo in the OpenMPToLLVMIRtranslation step for this PR and updating it in a subsequent PR related to mapping of members for derived types. The likely change in the update would be to incorporate a mapping of block arguments to MapInfo in the relevant map holding operations, similarly to how @TIFitis wished to do it in the original creation of his PR that added BlockArgs. This is because not all MapInfo will have there own symbol and BlockArg (at least in the case of the descriptor mapping and I also believe some cases for derived types), and we currently process all MapInfo as if it has a BlockArg (to do some minor processing in OpenMPToLLVMIRTranslation but also to print out the arguments when printing MLIR), and we cannot skip MapInfo that does not hold a BlockArg (e.g. just loop through and first come first serve for BlockArgs) based on ordering as we're not sure which BlockArg corresponds to which MapInfo at this stage without some kind of record of which BlockArg corresponds to which MapInfo. However, I'd like to investigate the implementation of member mapping a bit more before this is committed too if at all possible, but if we wish to do it in this PR I am also more than happy to do that, just need to know what we'd like to do!

There may be some other open changes I've missed so please do mention if there is any, but as far as I am aware most are closed and a few are left for later work in follow up PRs in the future based on disucssion. 

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