[llvm] [FMV][GlobalOpt] Statically resolve calls to versioned functions. (PR #87939)

[Alexandros Lamprineas via llvm-commits]

labrinea wrote:

Verdict for the llvm-test-suite regression: What I am thinking is that the compiler assumes that we have `predres` because of `mcpu=neoverse-v2`, but in reality we don't have it on the target, and since the optimization is skipping the runtime check, the test fails. Perhaps it's only safe to perform the optimization when the caller is a versioned function. In this example the caller is main, which is a regular function.

https://github.com/llvm/llvm-project/pull/87939