[clang] [llvm] [Clang][AArch64] Implement widening FMMLA intrinsics (PR #165282)

[via cfe-commits]

Lukacma wrote:

So the discussion you linked, seems to be about whether to add data dependency between set.fpmr and its use and whether the preserve memory side effect of it.  I think in the current design we decided use memory effects to express the dependency, but @CarolineConcatto is implementing it, so she  might be able to explain better why we went in that direction. 

>  It seems problematic to allow users to explicitly control the state at the IR level, as opposed to letting the compiler configure it in the backend.

I don't think I quite follow what you mean here ? I might be completely wrong here, but there is no way for compiler to know how to setup fields in that register, so we need to rely on user to do it himself corretly ? 

https://github.com/llvm/llvm-project/pull/165282