[flang-commits] [flang] [flang][AliasAnalysis] Cray pointers/pointees might alias with anything (PR #170900)

[via flang-commits]

https://github.com/jeanPerier commented:

> I would be surprised if this leads to regressions because for most cases this will produce the same result as current HEAD.

That is true at the LLVM level, but not the MLIR level where the alias analysis currently returned no alias between a cray pointer and non pointer/targets (your reverted change only touched the TBAA generation, not the AliasAnalysis itself). Although I agree that given cray pointer usages, the LLVM side probably makes a big difference.

I am also OK with having this change under a flag, it is well done, and it could be useful outside of the gfortran test.

When you add the switch, can you mention it in the AliasAnalysis doc next to the section about cray pointers?

https://github.com/llvm/llvm-project/pull/170900