[llvm] [SLP]Fix graph traversal in getSpillCost (PR #124984)

[Luke Lau via llvm-commits]

================
@@ -12252,25 +12254,28 @@ InstructionCost BoUpSLP::getSpillCost() const {
       };
 
       // Debug information does not impact spill cost.
-      if (isa<CallBase>(&*PrevInstIt) && !NoCallIntrinsic(&*PrevInstIt) &&
-          &*PrevInstIt != PrevInst)
+      // Vectorized calls, represented as vector intrinsics, do not impact spill
+      // cost.
+      if (const auto *CB = dyn_cast<CallBase>(&*PrevInstIt);
+          CB && !NoCallIntrinsic(CB) && !isVectorized(CB))
----------------
lukel97 wrote:

It think vectorized calls might still impact spill cost. E.g. `call <2 x i64> @llvm.sin(<2 x i64> %x)` will get scalarized during codegen to:

```asm
	vs1r.v	v8, (a0)                        # Unknown-size Folded Spill
	vslidedown.vi	v8, v8, 1
	vfmv.f.s	fa0, v8
	call	tanh
	fmv.d	fs0, fa0
	fld	fa0, 16(sp)                     # 8-byte Folded Reload
	call	tanh
```

https://github.com/llvm/llvm-project/pull/124984