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

[Simon Wallis via llvm-commits]

simonwallis2 wrote:

> I was looking at this earlier to try and figure out what is right. I think this looks OK, but the NumMicroOps should still be 1. In practice there might be a different resource, but it is not going to be super accurate with it taking a data-dependant number of cycles.

With NumMicroOps set to 1, the output of llvm-mca suggests that machine-scheduler treats these instructions as pipelined.
Setting NumMicroOps to the count of blocking cycles is key to this patch.

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