[llvm] [LV] Fix MVE regression from #132190 (PR #141736)

[David Sherwood via llvm-commits]

================
@@ -4349,15 +4358,24 @@ VectorizationFactor LoopVectorizationPlanner::selectVectorizationFactor() {
   for (auto &P : VPlans) {
     ArrayRef<ElementCount> VFs(P->vectorFactors().begin(),
                                P->vectorFactors().end());
-    auto RUs = calculateRegisterUsageForPlan(*P, VFs, TTI, CM.ValuesToIgnore);
-    for (auto [VF, RU] : zip_equal(VFs, RUs)) {
+
+    SmallVector<VPRegisterUsage, 8> RUs;
+    if (CM.useMaxBandwidth(TargetTransformInfo::RGK_ScalableVector) ||
+        CM.useMaxBandwidth(TargetTransformInfo::RGK_FixedWidthVector))
+      RUs = calculateRegisterUsageForPlan(*P, VFs, TTI, CM.ValuesToIgnore);
+
+    for (unsigned I = 0; I < VFs.size(); I++) {
+      ElementCount VF = VFs[I];
       // The cost for scalar VF=1 is already calculated, so ignore it.
       if (VF.isScalar())
         continue;
 
       /// Don't consider the VF if it exceeds the number of registers for the
       /// target.
-      if (RU.exceedsMaxNumRegs(TTI))
+      if (CM.useMaxBandwidth(VF.isScalable()
----------------
david-arm wrote:

nit: This is just a suggestion, but you could add an overloaded version of CM.useMaxBandwidth that takes an ElementCount and does something like:

```
  bool useMaxBandwidth(ElementCount VF) {
    return useMaxBandwidth(VF.isScalable() ? ...);
  }
```

and use that here and in `computeBestVF`?

https://github.com/llvm/llvm-project/pull/141736