[clang] [llvm] [IR] Don't set strictfp on irrelevant calls (PR #122735)

[Kevin P. Neal via cfe-commits]

kpneal wrote:

> To handle the case where a block isn't owned by a function, we need the attribute at the call site. I don't know the specifics of how that case arises, but if we remove the attribute from the call site, we would have to do something to add it again when the block gets detached from the function (possibly during cloning?).

And then remove it again when reinserting the BB into a function body? That sounds like a complication, said complication may be missed by someone in the future who isn't focused on FP, and I haven't yet heard a reason we need all of this.

@spavloff -- Why do we need to change how we handle the strictfp attribute? What's the benefit?

https://github.com/llvm/llvm-project/pull/122735