[flang] [llvm] [mlir][llvm][OpenMP] Emit llvm.stacksave and llvm.stackrestore in __atomic_compare_exchange (PR #132888)

[via llvm-commits]

NimishMishra wrote:

> stacksave/stackrestore is for dynamically-sized allocas[1](#user-content-fn-1-c2e54219bfae97730cb05f87a9346239). If you don't have dynamically-sized allocas, don't use them. Normal allocas **must** be emitted into the AllocaInsertPt (as done in #101966).

Oh. Okay, understood. I'll revert the changes to the PR then. 

Again, this is new info to me; could you tell why is it that stack save/restore is not for allocas of known size at compile time? I am assuming performance reasons, but is there something else too?



https://github.com/llvm/llvm-project/pull/132888