[llvm] [LV] Use VPReductionRecipe for partial reductions (PR #144908)

[Sander de Smalen via llvm-commits]

================
@@ -2861,6 +2886,12 @@ tryToMatchAndCreateMulAccumulateReduction(VPReductionRecipe *Red,
 
   VPValue *VecOp = Red->getVecOp();
   VPValue *A, *B;
+  // Some chained partial reductions used for complex numbers will have a
+  // negation between the mul and reduction. This extracts the mul from that
+  // pattern to use it for further checking.
+  if (Red->isPartialReduction())
+    match(VecOp,
+          m_Binary<Instruction::Sub>(m_SpecificInt(0), m_VPValue(VecOp)));
----------------
sdesmalen-arm wrote:

It still needs to somehow encode that this operation is a sub-reduction instead of an add-reduction. At the moment, this just generates a regular add reduction, which is not correct.

https://github.com/llvm/llvm-project/pull/144908