[llvm] [LV][VPlan] Extract the implementation of transform Recipe to EVLRecipe into a small function. NFC (PR #119510)

[Florian Hahn via llvm-commits]

================
@@ -1727,7 +1727,7 @@ define float @fmuladd(ptr %a, ptr %b, i64 %n, float %start) {
 ; IF-EVL-NEXT:    [[TMP14:%.*]] = getelementptr inbounds float, ptr [[B:%.*]], i64 [[TMP11]]
 ; IF-EVL-NEXT:    [[TMP15:%.*]] = getelementptr inbounds float, ptr [[TMP14]], i32 0
 ; IF-EVL-NEXT:    [[VP_OP_LOAD1:%.*]] = call <vscale x 4 x float> @llvm.vp.load.nxv4f32.p0(ptr align 4 [[TMP15]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP10]])
-; IF-EVL-NEXT:    [[TMP16:%.*]] = call reassoc <vscale x 4 x float> @llvm.vp.fmuladd.nxv4f32(<vscale x 4 x float> [[VP_OP_LOAD]], <vscale x 4 x float> [[VP_OP_LOAD1]], <vscale x 4 x float> [[VEC_PHI]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP10]])
----------------
fhahn wrote:

This is changing because we aren't passing the underlying instruction to the VPWIdenIntrinsic constructor any longer, right?

I think it is fine to change it, but the patch isn't NFC because of it. Maybe keep passing the underlying instruction in this patch, and then change the constructor separately to keep this just a NFC refactoring?

https://github.com/llvm/llvm-project/pull/119510