[llvm] [AArch64][machine-scheduler][Neoverse-N2] fdiv is blocking (PR #119206)

[David Green via llvm-commits]

davemgreen wrote:

OK. I went and figured out how I think they should work. I believe 1MOp is closer but I don't know if we will get anything perfect here, especially with the instructions taking a variable number of cycles so either is probably fine.

> > Can we change the vector and sve divides at the same time?
> 
> The scope of this patch is limited to the blocking vfp instructions. I do not have a motivating case for the vector and sve divides. Do you have such a case? They should be addressed in a separate patch

Can we change them at the same time? I think they should work in (roughly) the same way, still being recirculating and can hopefully use the same SchedWriteRes's. I just don't love that we are only doing half of the instructions, I'd prefer to keep them in sync. With that I think this sounds OK.

https://github.com/llvm/llvm-project/pull/119206