[llvm] [LV] Fix issue in VPFirstOrderRecurrencePHIRecipe::usesFirstLaneOnly (PR #179977)

[David Sherwood via llvm-commits]

================
@@ -654,6 +654,9 @@ void VPlanTransforms::replicateByVF(VPlan &Plan, ElementCount VF) {
       ToRemove.push_back(DefR);
     }
   }
-  for (auto *R : reverse(ToRemove))
+  for (auto *R : reverse(ToRemove)) {
+    assert(!cast<VPSingleDefRecipe>(R)->getNumUsers() &&
+           "Attempting to remove a VPSingleDefRecipe with users!");
----------------
david-arm wrote:

OK fair enough. I'm happy to remove it. I'm just concerned that this code is quite fragile due to this:

```
      DefR->replaceUsesWithIf(LaneDefs[0], [DefR](VPUser &U, unsigned) {
        return U.usesFirstLaneOnly(DefR);
      });
```

because if any instance `U.usesFirstLaneOnly` actually returns false we may trigger the assert in `~VPRecipeValue`. I plan to look at this a bit more after this PR lands because I wonder if this should just be something like:

```
      DefR->replaceUsesWithIf(LaneDefs[0], [DefR](VPUser &U, unsigned) {
        assert(U.usesFirstLaneOnly(DefR));
        return true;
      });
```


https://github.com/llvm/llvm-project/pull/179977