[flang-commits] [flang] [mlir] [Flang][MLIR][OpenMP] - Add support for firstprivate when translating omp.target ops from MLIR to LLVMIR (PR #131213)

[Pranav Bhandarkar via flang-commits]

bhandarkar-pranav wrote:

> @ergawy and @tblah - Update: Like I said before, I have some end-to-end testcases that exposed some problems. These problems are not all in `firstprivate`. In fact, I have found problems in the handling of `maps` for certain types too (See #134912) . I have now, what I believe, the last two of these problems on my plate now. One is a wrong answer (bad run) and one is a runtime crash because of a bad device pointer. Both of these are related to `fir.boxchar`. I am working on these fixes and would like to merge this PR only after they are done.

@tblah  and @ergawy  - Could you please review this PR once again. My fixes for `fir.boxchar` are still having some runtime issues. I have verified and convinced myself that the bitcode generated is correct, but there is something that I am doing that is running afoul of the expectation of the runtime library. So, while I plug away at that, could you please review this PR. I'll open a new PR once I have fixed character allocatables which, I don't think, are critically important for `omp.target` because I do not imagine people dying to send strings over to the device ;-)

https://github.com/llvm/llvm-project/pull/131213