[llvm-branch-commits] [llvm] [LV] Reduce register usage for scaled reductions (PR #133090)

[David Sherwood via llvm-branch-commits]

================
@@ -5039,10 +5039,25 @@ calculateRegisterUsage(VPlan &Plan, ArrayRef<ElementCount> VFs,
             // even in the scalar case.
             RegUsage[ClassID] += 1;
           } else {
+            // The output from scaled phis and scaled reductions actually have
+            // fewer lanes than the VF.
+            ElementCount VF = VFs[J];
+            if (auto *ReductionR = dyn_cast<VPReductionPHIRecipe>(R))
----------------
david-arm wrote:

I realise it may be less efficient, but perhaps it's better to commonise these into the same block? If for some reason we need to update this logic in future it's easier to fix it only once, i.e.

```
  if (isa<VPReductionPHIRecipe, VPPartialReductionRecipe>(R)) {
    auto *ReductionR = dyn_cast<VPReductionPHIRecipe>(R);
    auto *PartialReductionR = dyn_cast<VPPartialReductionRecipe>(R);
    unsigned ScaleFactor = ReductionR ? ReductionR->getVFScaleFactor() : PartialReductionR->getVFScaleFactor();
    VF = VF.divideCoefficientBy(ScaleFactor);
  }
```

If `getVFScaleFactor` becomes available to a common base class then it should simplify further. What do you think?

https://github.com/llvm/llvm-project/pull/133090