[llvm] [LV][EVL] Restore inbounds flag in VPVectorEndPointerRecipe when using EVL tail folding (PR #145306)

[Florian Hahn via llvm-commits]

================
@@ -2171,6 +2171,29 @@ static VPRecipeBase *createEVLRecipe(VPValue *HeaderMask,
       .Default([&](VPRecipeBase *R) { return nullptr; });
 }
 
+// Replace VF operands with EVL in recipes that use VF.
+static void replaceVFWithEVL(VPValue &VF, VPValue &EVL) {
+  for (VPUser *U : to_vector(VF.users())) {
+    if (auto *VEP = dyn_cast<VPVectorEndPointerRecipe>(U)) {
+      VPValue *Ptr = VEP->getPtr();
+      GEPNoWrapFlags NewFlags = VEP->getGEPNoWrapFlags();
+      // Replacing VF with EVL ensures that VPVectorEndPointer will not compute
+      // out-of-bounds, as long as the original pointer had the inbounds flag.
+      if (!NewFlags.isInBounds()) {
+        if (auto *GEP = dyn_cast<GetElementPtrInst>(
----------------
fhahn wrote:

We shouldn't rely on the the underlying GEP to re-set `inbounds`. There might be other reasons for why the underlying inbounds may not be valid any longer, e.g. if the original GEP + memory op where executed conditionally and the loop and are unconditionally executed now.

How critical is this? If it is important would be good to think of a better way to preserve the inbounds until EVL recipes are added or recover it w/o looking at the underlying IR

https://github.com/llvm/llvm-project/pull/145306