[llvm] [AArch64] Fix #94909: Optimize vector fmul(sitofp(x), 0.5) -> scvtf(x, 2) (PR #141480)

[David Green via llvm-commits]

================
@@ -5407,6 +5433,102 @@ class BaseIntegerToFPUnscaled<bits<2> rmode, bits<3> opcode,
   let Inst{4-0}   = Rd;
 }
 
+multiclass IntegerToFPVector<
----------------
davemgreen wrote:

This looks like it is defining an instruction. We just need a Pat to convert the `mul(sint_to_fp(` to a.. SCVTFv2i32_shift I think. The new pattern will need to be defined after the instruction is, which mean somewhere after the `defm SCVTF: SIMDVectorRShiftToFP<..>`.

https://github.com/llvm/llvm-project/pull/141480