[flang-commits] [flang] [flang][test] Experimental support of MemoryEffectOpInterface for fir.call. (PR #191580)

[Slava Zakharin via flang-commits]

vzakhari wrote:

> This is a good idea, but how could it be used currently? Would we need to add attributes to fir.call to carry this information where it is known (e.g. for runtime calls)?

Yes, attaching some attributes to `fir.call` early in FE and then deducing the memory effects from them is what I am anticipating. But before I can propose a reasonable and sufficient set of attributes to cover all the important cases I need to experiment with many tests, which basically implies that I recognize the functions by name in a downstream pass, specify some memory effects for them and see how it affects different optimizations in MLIR.

https://github.com/llvm/llvm-project/pull/191580