[llvm] [LV] Add cost for VPInstruction::Not. (PR #198445)

[Florian Hahn via llvm-commits]

================
@@ -1285,6 +1285,17 @@ InstructionCost VPInstruction::computeCost(ElementCount VF,
     return Ctx.TTI.getIndexedVectorInstrCostFromEnd(Instruction::ExtractElement,
                                                     VecTy, Ctx.CostKind, 0);
   }
+  case VPInstruction::Not: {
+    Type *ValTy = Ctx.Types.inferScalarType(this);
+    // InstCombine will fold `xor` to the conditional branch.
+    if (auto *U = const_cast<VPUser *>(getSingleUser()))
+      if (match(U, m_BranchOnCond(m_VPValue())))
+        return 0;
+    if (!vputils::onlyFirstLaneUsed(this))
+      ValTy = toVectorTy(ValTy, VF);
----------------
fhahn wrote:

It looks like we are missing test coverage for the recent changes, would it be possible to cover  via a cost-printing test? I guess it would need a case where the vector type does not fit into a single register, so there's a difference between scalar and vector cost?

https://github.com/llvm/llvm-project/pull/198445