[llvm] [LV] Create in-loop sub reductions (PR #147026)

[Sander de Smalen via llvm-commits]

================
@@ -810,10 +810,15 @@ Value *VPInstruction::generate(VPTransformState &State) {
         Value *RdxPart = RdxParts[Part];
         if (RecurrenceDescriptor::isMinMaxRecurrenceKind(RK))
           ReducedPartRdx = createMinMaxOp(Builder, RK, ReducedPartRdx, RdxPart);
-        else
-          ReducedPartRdx = Builder.CreateBinOp(
-              (Instruction::BinaryOps)RecurrenceDescriptor::getOpcode(RK),
-              RdxPart, ReducedPartRdx, "bin.rdx");
+        else {
+          // Interleaved/unrolled sub reductions should still be added together
----------------
sdesmalen-arm wrote:

nit: I think this requires some rationale to explain why `RecurKind::Sub` uses `add`, since the comment above on line 814 doesn't describe that.

Perhaps something like this is clearer:
```
  Instruction::BinaryOps Opcode;

  // For sub-recurrences, each UF's reduction variable is already
  // negative, we need to do: reduce.add(-acc_uf0 + -acc_uf1)
  if (RK == RecurKind::Sub)
    Opcode = Instruction::Add;
  else
    Opcode = RecurrenceDescriptor::getOpcode(RK);
```

https://github.com/llvm/llvm-project/pull/147026