[llvm] [AArch64][SelectionDAG] Fold fp_to_int(fmul) for FPR results (PR #210987)

[Kieran B via llvm-commits]

================

----------------
kieroxide wrote:

I moved these patterns to a separate multiclass. This was due to the fact that the existing multiclass used a Instrinsic parameter which when changed to SDPatternOperator (which allows passing any_fp_to_int) would not compile due to duplicated pattern matches for FastIsel. So I decided to keep these split into FPToIntegerFmulPats. I moved the existing fmul pats into this class and changed the name from round to OpN which matched my already existing multiclass.

https://github.com/llvm/llvm-project/pull/210987