[llvm] [Attributor] Change allocation size and load/store offsets using AAPointerInfo for Alloca instructions (PR #72029)

Vidush Singhal via llvm-commits llvm-commits at lists.llvm.org
Mon Jul 8 15:27:40 PDT 2024


https://github.com/vidsinghal updated https://github.com/llvm/llvm-project/pull/72029

>From 5a52f5bc9bf440d35464087ed54232d7c963cd21 Mon Sep 17 00:00:00 2001
From: vidsinghal <vidush.sl at gmail.com>
Date: Mon, 24 Jun 2024 11:00:52 -0400
Subject: [PATCH 1/2] Store the full chain of instructions that make up the
 access.

---
 llvm/include/llvm/Transforms/IPO/Attributor.h | 202 ++++++++-
 .../Transforms/IPO/AttributorAttributes.cpp   | 306 +++++++++-----
 .../pointer-info-track-access-chain.ll        | 387 ++++++++++++++++++
 3 files changed, 784 insertions(+), 111 deletions(-)
 create mode 100644 llvm/test/Transforms/Attributor/pointer-info-track-access-chain.ll

diff --git a/llvm/include/llvm/Transforms/IPO/Attributor.h b/llvm/include/llvm/Transforms/IPO/Attributor.h
index f70fd02ca573c..e2065b496e210 100644
--- a/llvm/include/llvm/Transforms/IPO/Attributor.h
+++ b/llvm/include/llvm/Transforms/IPO/Attributor.h
@@ -103,7 +103,9 @@
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SetOperations.h"
 #include "llvm/ADT/SetVector.h"
+#include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/SmallSet.h"
+#include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/iterator.h"
 #include "llvm/Analysis/AssumeBundleQueries.h"
 #include "llvm/Analysis/CFG.h"
@@ -320,6 +322,10 @@ inline bool operator==(const RangeTy &A, const RangeTy &B) {
   return A.Offset == B.Offset && A.Size == B.Size;
 }
 
+inline bool operator<(const RangeTy &A, const RangeTy &B) {
+  return A.Offset < B.Offset;
+}
+
 inline bool operator!=(const RangeTy &A, const RangeTy &B) { return !(A == B); }
 
 /// Return the initial value of \p Obj with type \p Ty if that is a constant.
@@ -5785,6 +5791,139 @@ struct AAPointerInfo : public AbstractAttribute {
     AK_MUST_READ_WRITE = AK_MUST | AK_R | AK_W,
   };
 
+  /// A helper containing a list of offsets computed for a Use. Ideally this
+  /// list should be strictly ascending, but we ensure that only when we
+  /// actually translate the list of offsets to a RangeList.
+  struct OffsetInfo {
+    using VecTy = SmallVector<AA::RangeTy>;
+    // A map to store depth 1 predecessors per offset.
+    using OriginsTy = SmallVector<SmallPtrSet<Value *, 4>>;
+    using const_iterator = VecTy::const_iterator;
+    OriginsTy Origins;
+    VecTy Ranges;
+
+    const_iterator begin() const { return Ranges.begin(); }
+    const_iterator end() const { return Ranges.end(); }
+
+    bool operator==(const OffsetInfo &RHS) const {
+      return Ranges == RHS.Ranges && Origins == RHS.Origins;
+    }
+
+    bool operator!=(const OffsetInfo &RHS) const { return !(*this == RHS); }
+
+    // Insert a new Range and Origin
+    void insert(AA::RangeTy Range, Value &V) {
+      auto *It = std::find(Ranges.begin(), Ranges.end(), Range);
+      // Offset exists in Offsets map
+      if (It != Ranges.end()) {
+        size_t Index = It - Ranges.begin();
+        if (Index < Origins.size())
+          Origins[Index].insert(&V);
+      } else {
+        Ranges.push_back(Range);
+        Origins.emplace_back();
+        Origins.back().insert(&V);
+      }
+    }
+
+    // Set the size of the offset for all ranges.
+    void setSizeAll(uint64_t Size) {
+      for (auto &Range : Ranges)
+        Range.Size = Size;
+    }
+
+    // Helper function to get just the offsets from Ranges.
+    void getOnlyOffsets(SmallVector<int64_t> &Offsets) {
+      for (auto &Range : Ranges)
+        Offsets.push_back(Range.Offset);
+      // ensure unique
+      sort(Offsets.begin(), Offsets.end());
+      Offsets.erase(std::unique(Offsets.begin(), Offsets.end()), Offsets.end());
+    }
+
+    bool isUnassigned() const { return Ranges.empty(); }
+
+    bool isUnknown() const {
+      if (isUnassigned())
+        return false;
+      if (Ranges.size() == 1)
+        return Ranges.front().Offset == AA::RangeTy::Unknown;
+      return false;
+    }
+
+    void setUnknown(Value &V) {
+      Ranges.clear();
+      Origins.clear();
+      insert(AA::RangeTy{AA::RangeTy::Unknown, AA::RangeTy::Unknown}, V);
+    }
+
+    // Increment all ranges by Inc.
+    // Add an origin V to all offsets.
+    void addToAll(int64_t Inc, Value &V) {
+      for (auto &Range : Ranges)
+        Range.Offset += Inc;
+
+      if (!Origins.empty()) {
+        for (auto &Origin : Origins)
+          Origin.insert(&V);
+      } else {
+        for (size_t Index = 0; Index < Ranges.size(); Index++) {
+          Origins.emplace_back();
+          Origins[Index].insert(&V);
+        }
+      }
+    }
+
+    // Increment all ranges by Inc.
+    void addToAll(int64_t Inc) {
+      for (auto &Range : Ranges)
+        Range.Offset += Inc;
+    }
+
+    /// Copy offsets from \p R into the current list.
+    ///
+    /// Ideally all lists should be strictly ascending, but we defer that to the
+    /// actual use of the list. So we just blindly append here.
+    void merge(const OffsetInfo &R) {
+      Ranges.append(R.Ranges);
+      // ensure elements are unique.
+      sort(Ranges.begin(), Ranges.end());
+      Ranges.erase(std::unique(Ranges.begin(), Ranges.end()), Ranges.end());
+
+      OriginsTy ToBeMergeOrigins = R.Origins;
+      for (auto &Origin : ToBeMergeOrigins)
+        Origins.emplace_back(Origin);
+    }
+
+    // Merge two OffsetInfo structs.
+    // takes an additional origin argument
+    // and adds it to the corresponding offset in the
+    // origins map.
+    void mergeWithOffset(const OffsetInfo &R, Value &CurPtr) {
+      Ranges.append(R.Ranges);
+      // ensure elements are unique.
+      sort(Ranges.begin(), Ranges.end());
+      Ranges.erase(std::unique(Ranges.begin(), Ranges.end()), Ranges.end());
+      auto &ROffsets = R.Ranges;
+      for (auto Offset : ROffsets) {
+        auto *It = std::find(Ranges.begin(), Ranges.end(), Offset);
+        if (It == Ranges.end())
+          continue;
+        size_t Index = It - Ranges.begin();
+        if (Index >= Origins.size()) {
+          Origins.emplace_back();
+          Origins.back().insert(&CurPtr);
+        } else {
+          Origins[Index].insert(&CurPtr);
+        }
+      }
+    }
+  };
+
+  using OffsetInfoMapTy = DenseMap<Value *, OffsetInfo>;
+  using AccessPathTy = SmallVector<Value *, 4>;
+  using AccessPathSetTy = SmallPtrSet<AccessPathTy *, 4>;
+
   /// A container for a list of ranges.
   struct RangeList {
     // The set of ranges rarely contains more than one element, and is unlikely
@@ -5939,15 +6078,17 @@ struct AAPointerInfo : public AbstractAttribute {
   /// An access description.
   struct Access {
     Access(Instruction *I, int64_t Offset, int64_t Size,
-           std::optional<Value *> Content, AccessKind Kind, Type *Ty)
+           std::optional<Value *> Content, AccessKind Kind, Type *Ty,
+           AccessPathSetTy *AccessPaths)
         : LocalI(I), RemoteI(I), Content(Content), Ranges(Offset, Size),
-          Kind(Kind), Ty(Ty) {
+          Kind(Kind), Ty(Ty), AccessPaths(AccessPaths) {
       verify();
     }
     Access(Instruction *LocalI, Instruction *RemoteI, const RangeList &Ranges,
-           std::optional<Value *> Content, AccessKind K, Type *Ty)
+           std::optional<Value *> Content, AccessKind K, Type *Ty,
+           AccessPathSetTy *AccessPaths)
         : LocalI(LocalI), RemoteI(RemoteI), Content(Content), Ranges(Ranges),
-          Kind(K), Ty(Ty) {
+          Kind(K), Ty(Ty), AccessPaths(AccessPaths) {
       if (Ranges.size() > 1) {
         Kind = AccessKind(Kind | AK_MAY);
         Kind = AccessKind(Kind & ~AK_MUST);
@@ -5956,9 +6097,9 @@ struct AAPointerInfo : public AbstractAttribute {
     }
     Access(Instruction *LocalI, Instruction *RemoteI, int64_t Offset,
            int64_t Size, std::optional<Value *> Content, AccessKind Kind,
-           Type *Ty)
+           Type *Ty, AccessPathSetTy *AccessPaths)
         : LocalI(LocalI), RemoteI(RemoteI), Content(Content),
-          Ranges(Offset, Size), Kind(Kind), Ty(Ty) {
+          Ranges(Offset, Size), Kind(Kind), Ty(Ty), AccessPaths(AccessPaths) {
       verify();
     }
     Access(const Access &Other) = default;
@@ -5966,7 +6107,8 @@ struct AAPointerInfo : public AbstractAttribute {
     Access &operator=(const Access &Other) = default;
     bool operator==(const Access &R) const {
       return LocalI == R.LocalI && RemoteI == R.RemoteI && Ranges == R.Ranges &&
-             Content == R.Content && Kind == R.Kind;
+             Content == R.Content && Kind == R.Kind &&
+             checkAccessPathsAreSame(R.AccessPaths);
     }
     bool operator!=(const Access &R) const { return !(*this == R); }
 
@@ -6078,11 +6220,53 @@ struct AAPointerInfo : public AbstractAttribute {
       }
     }
 
+    // Merge two access paths into one.
+    void mergeAccessPaths(const AccessPathSetTy *AccessPathsNew) const {
+      for (auto *Path : *AccessPathsNew)
+        if (!existsChain(Path))
+          AccessPaths->insert(Path);
+    }
+
+    // Check if the given access paths are same.
+    bool checkAccessPathsAreSame(const AccessPathSetTy *AccessPathsR) const {
+      bool IsSame = true;
+      if (AccessPaths->size() != AccessPathsR->size())
+        return false;
+
+      for (auto *Path : *AccessPathsR) {
+        if (!existsChain(Path))
+          IsSame = false;
+      }
+      return IsSame;
+    }
+
+    // Check if the chain exists in the AccessPathsSet.
+    bool existsChain(const AccessPathTy *NewPath) const {
+      for (auto *OldPath : *AccessPaths)
+        if (*OldPath == *NewPath)
+          return true;
+
+      return false;
+    }
+
+    void dumpAccessPaths(raw_ostream &O) const {
+      O << "Print all access paths found:"
+        << "\n";
+      for (auto *It : *AccessPaths) {
+        O << "Backtrack a unique access path:\n";
+        for (Value *Ins : *It) {
+          O << *Ins << "\n";
+        }
+      }
+    }
+
+    const AccessPathSetTy *getAccessChain() const { return AccessPaths; }
     const RangeList &getRanges() const { return Ranges; }
 
     using const_iterator = RangeList::const_iterator;
     const_iterator begin() const { return Ranges.begin(); }
     const_iterator end() const { return Ranges.end(); }
+    size_t size() const { return Ranges.size(); }
 
   private:
     /// The instruction responsible for the access with respect to the local
@@ -6105,6 +6289,10 @@ struct AAPointerInfo : public AbstractAttribute {
     /// The type of the content, thus the type read/written, can be null if not
     /// available.
     Type *Ty;
+
+    /// The full chain of instructions that participate in the Access.
+    /// There may be more than one access chain.
+    AccessPathSetTy *AccessPaths;
   };
 
   /// Create an abstract attribute view for the position \p IRP.
diff --git a/llvm/lib/Transforms/IPO/AttributorAttributes.cpp b/llvm/lib/Transforms/IPO/AttributorAttributes.cpp
index 2816a85743faa..ca7b6e555111c 100644
--- a/llvm/lib/Transforms/IPO/AttributorAttributes.cpp
+++ b/llvm/lib/Transforms/IPO/AttributorAttributes.cpp
@@ -11,6 +11,8 @@
 //
 //===----------------------------------------------------------------------===//
 
+#include "llvm/Analysis/TargetLibraryInfo.h"
+#include "llvm/IR/GlobalVariable.h"
 #include "llvm/Transforms/IPO/Attributor.h"
 
 #include "llvm/ADT/APInt.h"
@@ -850,8 +852,13 @@ struct AA::PointerInfo::State : public AbstractState {
   ChangeStatus addAccess(Attributor &A, const AAPointerInfo::RangeList &Ranges,
                          Instruction &I, std::optional<Value *> Content,
                          AAPointerInfo::AccessKind Kind, Type *Ty,
+                         AAPointerInfo::OffsetInfoMapTy &OffsetInfoMap,
                          Instruction *RemoteI = nullptr);
 
+  AAPointerInfo::AccessPathSetTy *
+  findAllAccessPaths(AAPointerInfo::OffsetInfoMapTy &OffsetInfoMap,
+                     Instruction *LocalI);
+
   AAPointerInfo::const_bin_iterator begin() const { return OffsetBins.begin(); }
   AAPointerInfo::const_bin_iterator end() const { return OffsetBins.end(); }
   int64_t numOffsetBins() const { return OffsetBins.size(); }
@@ -926,10 +933,95 @@ struct AA::PointerInfo::State : public AbstractState {
   BooleanState BS;
 };
 
+AAPointerInfo::AccessPathSetTy *AA::PointerInfo::State::findAllAccessPaths(
+    AAPointerInfo::OffsetInfoMapTy &OffsetInfoMap, Instruction *LocalI) {
+  AAPointerInfo::AccessPathSetTy *AccessPathsSet =
+      new AAPointerInfo::AccessPathSetTy();
+
+  // Store the instruction and its storage (i.e, which path it belongs to)
+  // on the stack.
+  // We also store the visited map on the stack.
+  // Since we want to find new paths, we want to make sure an instruction is
+  // not visited twice on the same path. However, we can visit the same
+  // instruction more that once if it exists on different paths.
+  using VisitedTy = SmallPtrSet<Value *, 4>;
+  using StackElementTy =
+      std::tuple<Value *, AAPointerInfo::AccessPathTy *, VisitedTy>;
+
+  SmallVector<StackElementTy, 16> Stack;
+
+  // Populate the stack with elements.
+  for (auto *It = LocalI->op_begin(); It != LocalI->op_end(); It++) {
+    Value *V = cast<Value>(It);
+    if (!OffsetInfoMap.contains(V))
+      continue;
+
+    SmallPtrSet<Value *, 4> LocalVisitedMap;
+    AAPointerInfo::AccessPathTy *NewPath = new AAPointerInfo::AccessPathTy();
+    AccessPathsSet->insert(NewPath);
+    NewPath->push_back(LocalI);
+    Stack.push_back(std::make_tuple(V, NewPath, LocalVisitedMap));
+  }
+
+  while (!Stack.empty()) {
+    auto Entry = Stack.pop_back_val();
+    Value *Top = std::get<0>(Entry);
+    AAPointerInfo::AccessPathTy *CurrentChain = std::get<1>(Entry);
+    auto &Visited = std::get<2>(Entry);
+
+    if (!OffsetInfoMap.contains(Top))
+      continue;
+
+    if (!Visited.insert(Top).second)
+      continue;
+
+    CurrentChain->push_back(Top);
+    auto OI = OffsetInfoMap.lookup(Top);
+    auto &Origins = OI.Origins;
+
+    SmallPtrSet<Value *, 16> Successors;
+    for (auto &Origin : Origins) {
+      for (auto *Val : Origin) {
+        // Since we store depth 1 predecessors in our Origins map
+        // We can be sure that we hit termination condition if the
+        // Successor is the current instruction.
+        if (Val != Top)
+          Successors.insert(Val);
+      }
+    }
+
+    if (Successors.empty())
+      continue;
+
+    // Create new paths to be forked
+    SmallVector<AAPointerInfo::AccessPathTy *> NewPaths;
+    NewPaths.push_back(CurrentChain);
+    for (size_t Index = 1; Index < Successors.size(); Index++) {
+      AAPointerInfo::AccessPathTy *NewPath = new AAPointerInfo::AccessPathTy(
+          CurrentChain->begin(), CurrentChain->end());
+      NewPaths.push_back(NewPath);
+    }
+
+    int Index = 0;
+    // Traverse the successors
+    for (auto *Successor : Successors) {
+      AAPointerInfo::AccessPathTy *NextChain = NewPaths[Index];
+      AccessPathsSet->insert(NextChain);
+      // Push successors to traverse and their corresponding storage on
+      // stack.
+      VisitedTy NewVisitedSet(Visited.begin(), Visited.end());
+      Stack.push_back(std::make_tuple(Successor, NextChain, NewVisitedSet));
+      Index++;
+    }
+  }
+
+  return AccessPathsSet;
+}
+
 ChangeStatus AA::PointerInfo::State::addAccess(
     Attributor &A, const AAPointerInfo::RangeList &Ranges, Instruction &I,
     std::optional<Value *> Content, AAPointerInfo::AccessKind Kind, Type *Ty,
-    Instruction *RemoteI) {
+    AAPointerInfo::OffsetInfoMapTy &OffsetInfoMap, Instruction *RemoteI) {
   RemoteI = RemoteI ? RemoteI : &I;
 
   // Check if we have an access for this instruction, if not, simply add it.
@@ -956,7 +1048,11 @@ ChangeStatus AA::PointerInfo::State::addAccess(
   };
 
   if (!AccExists) {
-    AccessList.emplace_back(&I, RemoteI, Ranges, Content, Kind, Ty);
+    AAPointerInfo::AccessPathSetTy *AccessPaths =
+        AA::PointerInfo::State::findAllAccessPaths(OffsetInfoMap, &I);
+    AccessList.emplace_back(&I, RemoteI, Ranges, Content, Kind, Ty,
+                            AccessPaths);
+
     assert((AccessList.size() == AccIndex + 1) &&
            "New Access should have been at AccIndex");
     LocalList.push_back(AccIndex);
@@ -966,13 +1062,18 @@ ChangeStatus AA::PointerInfo::State::addAccess(
 
   // Combine the new Access with the existing Access, and then update the
   // mapping in the offset bins.
-  AAPointerInfo::Access Acc(&I, RemoteI, Ranges, Content, Kind, Ty);
+  AAPointerInfo::AccessPathSetTy *AccessPaths =
+      AA::PointerInfo::State::findAllAccessPaths(OffsetInfoMap, &I);
+  AAPointerInfo::Access Acc(&I, RemoteI, Ranges, Content, Kind, Ty,
+                            AccessPaths);
   auto &Current = AccessList[AccIndex];
   auto Before = Current;
   Current &= Acc;
   if (Current == Before)
     return ChangeStatus::UNCHANGED;
 
+  // Merge the newly generated access paths with the old access paths.
+  Before.mergeAccessPaths(Acc.getAccessChain());
   auto &ExistingRanges = Before.getRanges();
   auto &NewRanges = Current.getRanges();
 
@@ -1002,60 +1103,19 @@ ChangeStatus AA::PointerInfo::State::addAccess(
 
 namespace {
 
-/// A helper containing a list of offsets computed for a Use. Ideally this
-/// list should be strictly ascending, but we ensure that only when we
-/// actually translate the list of offsets to a RangeList.
-struct OffsetInfo {
-  using VecTy = SmallVector<int64_t>;
-  using const_iterator = VecTy::const_iterator;
-  VecTy Offsets;
-
-  const_iterator begin() const { return Offsets.begin(); }
-  const_iterator end() const { return Offsets.end(); }
-
-  bool operator==(const OffsetInfo &RHS) const {
-    return Offsets == RHS.Offsets;
-  }
-
-  bool operator!=(const OffsetInfo &RHS) const { return !(*this == RHS); }
-
-  void insert(int64_t Offset) { Offsets.push_back(Offset); }
-  bool isUnassigned() const { return Offsets.size() == 0; }
-
-  bool isUnknown() const {
-    if (isUnassigned())
-      return false;
-    if (Offsets.size() == 1)
-      return Offsets.front() == AA::RangeTy::Unknown;
-    return false;
-  }
-
-  void setUnknown() {
-    Offsets.clear();
-    Offsets.push_back(AA::RangeTy::Unknown);
-  }
-
-  void addToAll(int64_t Inc) {
-    for (auto &Offset : Offsets) {
-      Offset += Inc;
-    }
-  }
-
-  /// Copy offsets from \p R into the current list.
-  ///
-  /// Ideally all lists should be strictly ascending, but we defer that to the
-  /// actual use of the list. So we just blindly append here.
-  void merge(const OffsetInfo &R) { Offsets.append(R.Offsets); }
-};
-
 #ifndef NDEBUG
-static raw_ostream &operator<<(raw_ostream &OS, const OffsetInfo &OI) {
+static raw_ostream &operator<<(raw_ostream &OS,
+                               const AAPointerInfo::OffsetInfo &OI) {
   ListSeparator LS;
-  OS << "[";
+  int I = 0;
   for (auto Offset : OI) {
-    OS << LS << Offset;
+    OS << LS << "[Offset, Size]: " << Offset << "\n";
+    auto &Origin = OI.Origins[I];
+    for (auto *Val : Origin)
+      OS << "Origin: " << *Val << "\n";
   }
-  OS << "]";
+  OS << "\n";
+
   return OS;
 }
 #endif // NDEBUG
@@ -1365,7 +1425,8 @@ struct AAPointerInfoImpl
 
   ChangeStatus translateAndAddStateFromCallee(Attributor &A,
                                               const AAPointerInfo &OtherAA,
-                                              CallBase &CB) {
+                                              CallBase &CB,
+                                              OffsetInfoMapTy &OffsetInfoMap) {
     using namespace AA::PointerInfo;
     if (!OtherAA.getState().isValidState() || !isValidState())
       return indicatePessimisticFixpoint();
@@ -1388,15 +1449,16 @@ struct AAPointerInfoImpl
         AK = AccessKind(AK & (IsByval ? AccessKind::AK_R : AccessKind::AK_RW));
         AK = AccessKind(AK | (RAcc.isMayAccess() ? AK_MAY : AK_MUST));
 
-        Changed |= addAccess(A, RAcc.getRanges(), CB, Content, AK,
-                             RAcc.getType(), RAcc.getRemoteInst());
+        Changed |=
+            addAccess(A, RAcc.getRanges(), CB, Content, AK, RAcc.getType(),
+                      OffsetInfoMap, RAcc.getRemoteInst());
       }
     }
     return Changed;
   }
 
   ChangeStatus translateAndAddState(Attributor &A, const AAPointerInfo &OtherAA,
-                                    const OffsetInfo &Offsets, CallBase &CB) {
+                                    const OffsetInfo &Ranges, CallBase &CB) {
     using namespace AA::PointerInfo;
     if (!OtherAA.getState().isValidState() || !isValidState())
       return indicatePessimisticFixpoint();
@@ -1409,16 +1471,16 @@ struct AAPointerInfoImpl
     for (const auto &It : State) {
       for (auto Index : It.getSecond()) {
         const auto &RAcc = State.getAccess(Index);
-        for (auto Offset : Offsets) {
-          auto NewRanges = Offset == AA::RangeTy::Unknown
+        for (auto Range : Ranges) {
+          auto NewRanges = Range.Offset == AA::RangeTy::Unknown
                                ? AA::RangeTy::getUnknown()
                                : RAcc.getRanges();
           if (!NewRanges.isUnknown()) {
-            NewRanges.addToAllOffsets(Offset);
+            NewRanges.addToAllOffsets(Range.Offset);
           }
           Changed |=
               addAccess(A, NewRanges, CB, RAcc.getContent(), RAcc.getKind(),
-                        RAcc.getType(), RAcc.getRemoteInst());
+                        RAcc.getType(), OffsetInfoMap, RAcc.getRemoteInst());
         }
       }
     }
@@ -1449,9 +1511,11 @@ struct AAPointerInfoImpl
           else
             O << "       - c: <unknown>\n";
         }
+        Acc.dumpAccessPaths(O);
       }
     }
   }
+  OffsetInfoMapTy OffsetInfoMap;
 };
 
 struct AAPointerInfoFloating : public AAPointerInfoImpl {
@@ -1462,8 +1526,8 @@ struct AAPointerInfoFloating : public AAPointerInfoImpl {
   /// Deal with an access and signal if it was handled successfully.
   bool handleAccess(Attributor &A, Instruction &I,
                     std::optional<Value *> Content, AccessKind Kind,
-                    SmallVectorImpl<int64_t> &Offsets, ChangeStatus &Changed,
-                    Type &Ty) {
+                    OffsetInfo &OI, ChangeStatus &Changed, Type &Ty,
+                    OffsetInfoMapTy &OffsetInfoMap) {
     using namespace AA::PointerInfo;
     auto Size = AA::RangeTy::Unknown;
     const DataLayout &DL = A.getDataLayout();
@@ -1472,16 +1536,23 @@ struct AAPointerInfoFloating : public AAPointerInfoImpl {
       Size = AccessSize.getFixedValue();
 
     // Make a strictly ascending list of offsets as required by addAccess()
-    llvm::sort(Offsets);
-    auto *Last = llvm::unique(Offsets);
-    Offsets.erase(Last, Offsets.end());
+    auto Ranges = OI.Ranges;
+    auto Origins = OI.Origins;
+
+    llvm::sort(Ranges);
+    auto *Last = llvm::unique(Ranges);
+    Ranges.erase(Last, Ranges.end());
+
+    SmallVector<int64_t> OffsetsOnly;
+    OI.getOnlyOffsets(OffsetsOnly);
 
     VectorType *VT = dyn_cast<VectorType>(&Ty);
     if (!VT || VT->getElementCount().isScalable() ||
         !Content.value_or(nullptr) || !isa<Constant>(*Content) ||
         (*Content)->getType() != VT ||
         DL.getTypeStoreSize(VT->getElementType()).isScalable()) {
-      Changed = Changed | addAccess(A, {Offsets, Size}, I, Content, Kind, &Ty);
+      Changed = Changed | addAccess(A, {OffsetsOnly, Size}, I, Content, Kind,
+                                    &Ty, OffsetInfoMap);
     } else {
       // Handle vector stores with constant content element-wise.
       // TODO: We could look for the elements or create instructions
@@ -1493,7 +1564,8 @@ struct AAPointerInfoFloating : public AAPointerInfoImpl {
       int64_t ElementSize = DL.getTypeStoreSize(ElementType).getFixedValue();
       auto *ConstContent = cast<Constant>(*Content);
       Type *Int32Ty = Type::getInt32Ty(ElementType->getContext());
-      SmallVector<int64_t> ElementOffsets(Offsets.begin(), Offsets.end());
+      SmallVector<int64_t> ElementOffsets;
+      OI.getOnlyOffsets(ElementOffsets);
 
       for (int i = 0, e = VT->getElementCount().getFixedValue(); i != e; ++i) {
         Value *ElementContent = ConstantExpr::getExtractElement(
@@ -1501,7 +1573,8 @@ struct AAPointerInfoFloating : public AAPointerInfoImpl {
 
         // Add the element access.
         Changed = Changed | addAccess(A, {ElementOffsets, ElementSize}, I,
-                                      ElementContent, Kind, ElementType);
+                                      ElementContent, Kind, ElementType,
+                                      OffsetInfoMap);
 
         // Advance the offsets for the next element.
         for (auto &ElementOffset : ElementOffsets)
@@ -1520,7 +1593,7 @@ struct AAPointerInfoFloating : public AAPointerInfoImpl {
   /// \return true iff \p UsrOI is updated.
   bool collectConstantsForGEP(Attributor &A, const DataLayout &DL,
                               OffsetInfo &UsrOI, const OffsetInfo &PtrOI,
-                              const GEPOperator *GEP);
+                              GEPOperator *GEP, Value *CurPtr);
 
   /// See AbstractAttribute::trackStatistics()
   void trackStatistics() const override {
@@ -1528,11 +1601,9 @@ struct AAPointerInfoFloating : public AAPointerInfoImpl {
   }
 };
 
-bool AAPointerInfoFloating::collectConstantsForGEP(Attributor &A,
-                                                   const DataLayout &DL,
-                                                   OffsetInfo &UsrOI,
-                                                   const OffsetInfo &PtrOI,
-                                                   const GEPOperator *GEP) {
+bool AAPointerInfoFloating::collectConstantsForGEP(
+    Attributor &A, const DataLayout &DL, OffsetInfo &UsrOI,
+    const OffsetInfo &PtrOI, GEPOperator *GEP, Value *CurPtr) {
   unsigned BitWidth = DL.getIndexTypeSizeInBits(GEP->getType());
   MapVector<Value *, APInt> VariableOffsets;
   APInt ConstantOffset(BitWidth, 0);
@@ -1542,7 +1613,7 @@ bool AAPointerInfoFloating::collectConstantsForGEP(Attributor &A,
          "determined to be unknown.");
 
   if (!GEP->collectOffset(DL, BitWidth, VariableOffsets, ConstantOffset)) {
-    UsrOI.setUnknown();
+    UsrOI.setUnknown(*CurPtr);
     return true;
   }
 
@@ -1551,7 +1622,9 @@ bool AAPointerInfoFloating::collectConstantsForGEP(Attributor &A,
                     << *GEP << "\n");
 
   auto Union = PtrOI;
-  Union.addToAll(ConstantOffset.getSExtValue());
+  // clear the origins since we just want to keep only one predecessor.
+  Union.Origins.clear();
+  Union.addToAll(ConstantOffset.getSExtValue(), *CurPtr);
 
   // Each VI in VariableOffsets has a set of potential constant values. Every
   // combination of elements, picked one each from these sets, is separately
@@ -1560,7 +1633,7 @@ bool AAPointerInfoFloating::collectConstantsForGEP(Attributor &A,
     auto *PotentialConstantsAA = A.getAAFor<AAPotentialConstantValues>(
         *this, IRPosition::value(*VI.first), DepClassTy::OPTIONAL);
     if (!PotentialConstantsAA || !PotentialConstantsAA->isValidState()) {
-      UsrOI.setUnknown();
+      UsrOI.setUnknown(*CurPtr);
       return true;
     }
 
@@ -1579,14 +1652,16 @@ bool AAPointerInfoFloating::collectConstantsForGEP(Attributor &A,
     OffsetInfo Product;
     for (const auto &ConstOffset : AssumedSet) {
       auto CopyPerOffset = Union;
-      CopyPerOffset.addToAll(ConstOffset.getSExtValue() *
-                             VI.second.getZExtValue());
+      CopyPerOffset.addToAll(
+          ConstOffset.getSExtValue() * VI.second.getZExtValue(), *CurPtr);
       Product.merge(CopyPerOffset);
     }
     Union = Product;
   }
 
   UsrOI = std::move(Union);
+  TypeSize Size = DL.getTypeAllocSize(GEP->getResultElementType());
+  UsrOI.setSizeAll(Size);
   return true;
 }
 
@@ -1595,9 +1670,27 @@ ChangeStatus AAPointerInfoFloating::updateImpl(Attributor &A) {
   ChangeStatus Changed = ChangeStatus::UNCHANGED;
   const DataLayout &DL = A.getDataLayout();
   Value &AssociatedValue = getAssociatedValue();
-
-  DenseMap<Value *, OffsetInfo> OffsetInfoMap;
-  OffsetInfoMap[&AssociatedValue].insert(0);
+  OffsetInfoMap.clear();
+
+  uint64_t Size;
+  Function *F = getAssociatedFunction();
+  TargetLibraryInfo *TLI = nullptr;
+  if (F)
+    TLI = A.getInfoCache().getTargetLibraryInfoForFunction(*F);
+
+  if (TLI && getObjectSize(&AssociatedValue, Size, DL, TLI)) {
+    OffsetInfoMap[&AssociatedValue].insert(AA::RangeTy(0, Size),
+                                           AssociatedValue);
+  } else if (isa<GlobalVariable>(AssociatedValue)) {
+    auto &Glob = cast<GlobalVariable>(AssociatedValue);
+    TypeSize SizeOfType = DL.getTypeAllocSize(Glob.getValueType());
+    OffsetInfoMap[&AssociatedValue].insert(AA::RangeTy(0, SizeOfType),
+                                           AssociatedValue);
+  } else {
+    TypeSize SizeOfType = DL.getTypeAllocSize(AssociatedValue.getType());
+    OffsetInfoMap[&AssociatedValue].insert(AA::RangeTy(0, SizeOfType),
+                                           AssociatedValue);
+  }
 
   auto HandlePassthroughUser = [&](Value *Usr, Value *CurPtr, bool &Follow) {
     // One does not simply walk into a map and assign a reference to a possibly
@@ -1616,7 +1709,13 @@ ChangeStatus AAPointerInfoFloating::updateImpl(Attributor &A) {
     auto &PtrOI = OffsetInfoMap[CurPtr];
     assert(!PtrOI.isUnassigned() &&
            "Cannot pass through if the input Ptr was not visited!");
-    UsrOI.merge(PtrOI);
+    if (isa<PHINode>(Usr) || isa<SelectInst>(Usr)) {
+      UsrOI.mergeWithOffset(PtrOI, *CurPtr);
+    } else {
+      UsrOI = PtrOI;
+      UsrOI.Origins.clear();
+      UsrOI.addToAll(0, *CurPtr);
+    }
     Follow = true;
     return true;
   };
@@ -1649,11 +1748,11 @@ ChangeStatus AAPointerInfoFloating::updateImpl(Attributor &A) {
 
       if (PtrOI.isUnknown()) {
         Follow = true;
-        UsrOI.setUnknown();
+        UsrOI.setUnknown(*GEP);
         return true;
       }
 
-      Follow = collectConstantsForGEP(A, DL, UsrOI, PtrOI, GEP);
+      Follow = collectConstantsForGEP(A, DL, UsrOI, PtrOI, GEP, CurPtr);
       return true;
     }
     if (isa<PtrToIntInst>(Usr))
@@ -1677,7 +1776,7 @@ ChangeStatus AAPointerInfoFloating::updateImpl(Attributor &A) {
         LLVM_DEBUG(dbgs() << "[AAPointerInfo] PHI operand offset unknown "
                           << *CurPtr << " in " << *PHI << "\n");
         Follow = !UsrOI.isUnknown();
-        UsrOI.setUnknown();
+        UsrOI.setUnknown(*CurPtr);
         return true;
       }
 
@@ -1688,7 +1787,6 @@ ChangeStatus AAPointerInfoFloating::updateImpl(Attributor &A) {
         LLVM_DEBUG(dbgs() << "[AAPointerInfo] PHI is invariant (so far)");
         return true;
       }
-
       // Check if the PHI operand can be traced back to AssociatedValue.
       APInt Offset(
           DL.getIndexSizeInBits(CurPtr->getType()->getPointerAddressSpace()),
@@ -1700,7 +1798,7 @@ ChangeStatus AAPointerInfoFloating::updateImpl(Attributor &A) {
         LLVM_DEBUG(dbgs() << "[AAPointerInfo] PHI operand is too complex "
                           << *CurPtr << " in " << *PHI
                           << " (base: " << *CurPtrBase << ")\n");
-        UsrOI.setUnknown();
+        UsrOI.setUnknown(*CurPtr);
         Follow = true;
         return true;
       }
@@ -1717,7 +1815,7 @@ ChangeStatus AAPointerInfoFloating::updateImpl(Attributor &A) {
               *PHI->getFunction());
       if (mayBeInCycle(CI, cast<Instruction>(Usr), /* HeaderOnly */ true)) {
         auto BaseOI = It->getSecond();
-        BaseOI.addToAll(Offset.getZExtValue());
+        BaseOI.addToAll(Offset.getZExtValue(), *CurPtr);
         if (IsFirstPHIUser || BaseOI == UsrOI) {
           LLVM_DEBUG(dbgs() << "[AAPointerInfo] PHI is invariant " << *CurPtr
                             << " in " << *Usr << "\n");
@@ -1727,12 +1825,12 @@ ChangeStatus AAPointerInfoFloating::updateImpl(Attributor &A) {
         LLVM_DEBUG(
             dbgs() << "[AAPointerInfo] PHI operand pointer offset mismatch "
                    << *CurPtr << " in " << *PHI << "\n");
-        UsrOI.setUnknown();
+        UsrOI.setUnknown(*CurPtr);
         Follow = true;
         return true;
       }
 
-      UsrOI.merge(PtrOI);
+      UsrOI.mergeWithOffset(PtrOI, *CurPtr);
       Follow = true;
       return true;
     }
@@ -1746,8 +1844,8 @@ ChangeStatus AAPointerInfoFloating::updateImpl(Attributor &A) {
       else
         AK = AccessKind(AK | AccessKind::AK_MAY);
       if (!handleAccess(A, *LoadI, /* Content */ nullptr, AK,
-                        OffsetInfoMap[CurPtr].Offsets, Changed,
-                        *LoadI->getType()))
+                        OffsetInfoMap[CurPtr], Changed, *LoadI->getType(),
+                        OffsetInfoMap))
         return false;
 
       auto IsAssumption = [](Instruction &I) {
@@ -1831,9 +1929,9 @@ ChangeStatus AAPointerInfoFloating::updateImpl(Attributor &A) {
         Content =
             A.getAssumedSimplified(*Assumption.first, *this,
                                    UsedAssumedInformation, AA::Interprocedural);
-      return handleAccess(
-          A, *Assumption.second, Content, AccessKind::AK_ASSUMPTION,
-          OffsetInfoMap[CurPtr].Offsets, Changed, *LoadI->getType());
+      return handleAccess(A, *Assumption.second, Content,
+                          AccessKind::AK_ASSUMPTION, OffsetInfoMap[CurPtr],
+                          Changed, *LoadI->getType(), OffsetInfoMap);
     }
 
     auto HandleStoreLike = [&](Instruction &I, Value *ValueOp, Type &ValueTy,
@@ -1859,8 +1957,8 @@ ChangeStatus AAPointerInfoFloating::updateImpl(Attributor &A) {
       if (ValueOp)
         Content = A.getAssumedSimplified(
             *ValueOp, *this, UsedAssumedInformation, AA::Interprocedural);
-      return handleAccess(A, I, Content, AK, OffsetInfoMap[CurPtr].Offsets,
-                          Changed, ValueTy);
+      return handleAccess(A, I, Content, AK, OffsetInfoMap[CurPtr], Changed,
+                          ValueTy, OffsetInfoMap);
     };
 
     if (auto *StoreI = dyn_cast<StoreInst>(Usr))
@@ -1983,8 +2081,8 @@ struct AAPointerInfoCallSiteArgument final : AAPointerInfoFloating {
       } else {
         auto Kind =
             ArgNo == 0 ? AccessKind::AK_MUST_WRITE : AccessKind::AK_MUST_READ;
-        Changed =
-            Changed | addAccess(A, {0, LengthVal}, *MI, nullptr, Kind, nullptr);
+        Changed = Changed | addAccess(A, {0, LengthVal}, *MI, nullptr, Kind,
+                                      nullptr, OffsetInfoMap);
       }
       LLVM_DEBUG({
         dbgs() << "Accesses by bin after update:\n";
@@ -2004,8 +2102,8 @@ struct AAPointerInfoCallSiteArgument final : AAPointerInfoFloating {
       auto *ArgAA =
           A.getAAFor<AAPointerInfo>(*this, ArgPos, DepClassTy::REQUIRED);
       if (ArgAA && ArgAA->getState().isValidState())
-        return translateAndAddStateFromCallee(A, *ArgAA,
-                                              *cast<CallBase>(getCtxI()));
+        return translateAndAddStateFromCallee(
+            A, *ArgAA, *cast<CallBase>(getCtxI()), OffsetInfoMap);
       if (!Arg->getParent()->isDeclaration())
         return indicatePessimisticFixpoint();
     }
@@ -2022,7 +2120,7 @@ struct AAPointerInfoCallSiteArgument final : AAPointerInfoFloating {
     auto Kind =
         ReadOnly ? AccessKind::AK_MAY_READ : AccessKind::AK_MAY_READ_WRITE;
     return addAccess(A, AA::RangeTy::getUnknown(), *getCtxI(), nullptr, Kind,
-                     nullptr);
+                     nullptr, OffsetInfoMap);
   }
 
   /// See AbstractAttribute::trackStatistics()
diff --git a/llvm/test/Transforms/Attributor/pointer-info-track-access-chain.ll b/llvm/test/Transforms/Attributor/pointer-info-track-access-chain.ll
new file mode 100644
index 0000000000000..b7c3f1f33191e
--- /dev/null
+++ b/llvm/test/Transforms/Attributor/pointer-info-track-access-chain.ll
@@ -0,0 +1,387 @@
+; RUN: opt -aa-pipeline=basic-aa -passes=attributor -attributor-manifest-internal -debug-only=attributor  -attributor-annotate-decl-cs  -S < %s 2>&1 | FileCheck %s
+; RUN: opt -aa-pipeline=basic-aa -passes=attributor-cgscc -attributor-manifest-internal -debug-only=attributor  -attributor-annotate-decl-cs -S < %s 2>&1  | FileCheck %s
+; REQUIRES: asserts
+
+
+ at globalBytes = internal global [1024 x i8] zeroinitializer, align 16
+
+; CHECK: Accesses by bin after update:
+; CHECK: [8-12] : 1
+; CHECK:      - 5 -   %1 = load i32, ptr %field22, align 4
+; CHECK:        - c: <unknown>
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   %1 = load i32, ptr %field22, align 4
+; CHECK:   %field22 = getelementptr i32, ptr %field2, i32 0
+; CHECK:   %field2 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 2
+; CHECK:   %f = alloca [10 x i32], align 4
+; CHECK: [4-5] : 1
+; CHECK:      - 9 -   store i8 10, ptr %field11, align 4
+; CHECK:        - c: i8 10
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   store i8 10, ptr %field11, align 4
+; CHECK:   %field11 = getelementptr i32, ptr %field1, i32 0
+; CHECK:   %field1 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 1
+; CHECK:   %f = alloca [10 x i32], align 4
+; CHECK: [32-36] : 1
+; CHECK:      - 9 -   store i32 %3, ptr %field8, align 4
+; CHECK:        - c:   %3 = load i32, ptr %val, align 4
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   store i32 %3, ptr %field8, align 4
+; CHECK:   %field8 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 8
+; CHECK:   %f = alloca [10 x i32], align 4
+; CHECK: [4-8] : 1
+; CHECK:      - 5 -   %0 = load i32, ptr %field11, align 4
+; CHECK:        - c: <unknown>
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   %0 = load i32, ptr %field11, align 4
+; CHECK:   %field11 = getelementptr i32, ptr %field1, i32 0
+; CHECK:   %field1 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 1
+; CHECK:   %f = alloca [10 x i32], align 4
+; CHECK: [8-9] : 1
+; CHECK:      - 9 -   store i8 12, ptr %field22, align 4
+; CHECK:        - c: i8 12
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   store i8 12, ptr %field22, align 4
+; CHECK:   %field22 = getelementptr i32, ptr %field2, i32 0
+; CHECK:   %field2 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 2
+; CHECK:   %f = alloca [10 x i32], align 4
+define dso_local i32 @track_chain(ptr nocapture %val) #0 {
+entry:
+  %f = alloca [10 x i32]
+  %field1 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 1
+  %field2 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 2
+  %field3 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 3
+  %field8 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 8
+
+  %field11 = getelementptr i32, ptr %field1, i32 0
+  %field22 = getelementptr i32, ptr %field2, i32 0
+  store i8 10, ptr %field11, align 4
+  store i8 12, ptr %field22, align 4
+
+  %1 = load i32, ptr %field11, align 4
+  %2 = load i32, ptr %field22, align 4
+  %3 = add i32 %1, %2
+
+  %4 = load i32, ptr %val, align 4
+  store i32 %4, ptr %field8, align 4
+
+  %5 = add i32 %4, %3
+
+  ret i32 %5
+}
+
+
+; CHECK: Accesses by bin after update:
+; CHECK: [12-16] : 1
+; CHECK:      - 5 -   %0 = load i32, ptr %field11, align 4
+; CHECK:        - c: <unknown>
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   %0 = load i32, ptr %field11, align 4
+; CHECK:   %field11 = getelementptr i32, ptr %field1, i32 2
+; CHECK:   %field1 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 1
+; CHECK:   %f = alloca [10 x i32], align 4
+; CHECK: [16-17] : 1
+; CHECK:      - 9 -   store i8 12, ptr %field22, align 4
+; CHECK:        - c: i8 12
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   store i8 12, ptr %field22, align 4
+; CHECK:   %field22 = getelementptr i32, ptr %field2, i32 2
+; CHECK:   %field2 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 2
+; CHECK:   %f = alloca [10 x i32], align 4
+; CHECK: [32-36] : 1
+; CHECK:      - 9 -   store i32 %3, ptr %field8, align 4
+; CHECK:        - c:   %3 = load i32, ptr %val, align 4
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   store i32 %3, ptr %field8, align 4
+; CHECK:   %field8 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 8
+; CHECK:   %f = alloca [10 x i32], align 4
+; CHECK: [16-20] : 1
+; CHECK:      - 5 -   %1 = load i32, ptr %field22, align 4
+; CHECK:        - c: <unknown>
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   %1 = load i32, ptr %field22, align 4
+; CHECK:   %field22 = getelementptr i32, ptr %field2, i32 2
+; CHECK:   %field2 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 2
+; CHECK:   %f = alloca [10 x i32], align 4
+; CHECK: [12-13] : 1
+; CHECK:      - 9 -   store i8 10, ptr %field11, align 4
+; CHECK:        - c: i8 10
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   store i8 10, ptr %field11, align 4
+; CHECK:   %field11 = getelementptr i32, ptr %field1, i32 2
+; CHECK:   %field1 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 1
+; CHECK:   %f = alloca [10 x i32], align 4
+define dso_local i32 @track_chain_2(ptr nocapture %val) #0 {
+entry:
+  %f = alloca [10 x i32]
+  %field1 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 1
+  %field2 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 2
+  %field3 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 3
+  %field8 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 8
+
+  %field11 = getelementptr i32, ptr %field1, i32 2
+  %field22 = getelementptr i32, ptr %field2, i32 2
+  store i8 10, ptr %field11, align 4
+  store i8 12, ptr %field22, align 4
+
+  %1 = load i32, ptr %field11, align 4
+  %2 = load i32, ptr %field22, align 4
+  %3 = add i32 %1, %2
+
+  %4 = load i32, ptr %val, align 4
+  store i32 %4, ptr %field8, align 4
+
+  %5 = add i32 %4, %3
+
+  ret i32 %5
+}
+
+
+; CHECK: Accesses by bin after update:
+; CHECK: [12-16] : 3
+; CHECK:      - 5 -   %0 = load i32, ptr %field11, align 4
+; CHECK:       - c: <unknown>
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   %0 = load i32, ptr %field11, align 4
+; CHECK:   %field11 = getelementptr i32, ptr %field1, i32 2
+; CHECK:   %field1 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 1
+; CHECK:   %f = alloca [10 x i32], align 4
+; CHECK:      - 5 -   %b = load i32, ptr %field3, align 4
+; CHECK:        - c: <unknown>
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   %b = load i32, ptr %field3, align 4
+; CHECK:   %field3 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 3
+; CHECK:   %f = alloca [10 x i32], align 4
+; CHECK:      - 10 -   store i32 1000, ptr %6, align 4
+; CHECK:        - c: i32 1000
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   store i32 1000, ptr %6, align 4
+; CHECK:   %6 = select i1 %cond, ptr %field3, ptr %field8
+; CHECK:   %field3 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 3
+; CHECK:   %f = alloca [10 x i32], align 4
+; CHECK: Backtrack a unique access path:
+; CHECK:   store i32 1000, ptr %6, align 4
+; CHECK:   %6 = select i1 %cond, ptr %field3, ptr %field8
+; CHECK:   %field8 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 8
+; CHECK:   %f = alloca [10 x i32], align 4
+; CHECK: [16-17] : 1
+; CHECK:      - 9 -   store i8 12, ptr %field22, align 4
+; CHECK:       - c: i8 12
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   store i8 12, ptr %field22, align 4
+; CHECK:   %field22 = getelementptr i32, ptr %field2, i32 2
+; CHECK:   %field2 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 2
+; CHECK:   %f = alloca [10 x i32], align 4
+; CHECK: [32-36] : 4
+; CHECK:      - 9 -   store i32 %3, ptr %field8, align 4
+; CHECK:        - c:   %3 = load i32, ptr %val, align 4
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   store i32 %3, ptr %field8, align 4
+; CHECK:   %field8 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 8
+; CHECK:   %f = alloca [10 x i32], align 4
+; CHECK:      - 5 -   %a1 = load i32, ptr %field8, align 4
+; CHECK:        - c: <unknown>
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   %a1 = load i32, ptr %field8, align 4
+; CHECK:   %field8 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 8
+; CHECK:   %f = alloca [10 x i32], align 4
+; CHECK:      - 10 -   store i32 1000, ptr %6, align 4
+; CHECK:        - c: i32 1000
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   store i32 1000, ptr %6, align 4
+; CHECK:   %6 = select i1 %cond, ptr %field3, ptr %field8
+; CHECK:   %field3 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 3
+; CHECK:   %f = alloca [10 x i32], align 4
+; CHECK: Backtrack a unique access path:
+; CHECK:   store i32 1000, ptr %6, align 4
+; CHECK:   %6 = select i1 %cond, ptr %field3, ptr %field8
+; CHECK:  %field8 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 8
+; CHECK:   %f = alloca [10 x i32], align 4
+; CHECK:      - 5 -   %8 = load i32, ptr %field8, align 4
+; CHECK:        - c: <unknown>
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:  %8 = load i32, ptr %field8, align 4
+; CHECK:  %field8 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 8
+; CHECK:  %f = alloca [10 x i32], align 4
+; CHECK: [16-20] : 1
+; CHECK:      - 5 -   %1 = load i32, ptr %field22, align 4
+; CHECK:        - c: <unknown>
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   %1 = load i32, ptr %field22, align 4
+; CHECK:   %field22 = getelementptr i32, ptr %field2, i32 2
+; CHECK:   %field2 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 2
+; CHECK:   %f = alloca [10 x i32], align 4
+; CHECK: [12-13] : 1
+; CHECK:      - 9 -   store i8 10, ptr %field11, align 4
+; CHECK:        - c: i8 10
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   store i8 10, ptr %field11, align 4
+; CHECK:   %field11 = getelementptr i32, ptr %field1, i32 2
+; CHECK:   %field1 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 1
+; CHECK:   %f = alloca [10 x i32], align 4
+define dso_local i32 @track_chain_3(ptr nocapture %val, i1 %cond) #0 {
+entry:
+  %f = alloca [10 x i32]
+  %field1 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 1
+  %field2 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 2
+  %field3 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 3
+  %field8 = getelementptr inbounds [10 x i32], ptr %f, i32 0, i32 8
+
+  %field11 = getelementptr i32, ptr %field1, i32 2
+  %field22 = getelementptr i32, ptr %field2, i32 2
+  store i8 10, ptr %field11, align 4
+  store i8 12, ptr %field22, align 4
+  %1 = load i32, ptr %field11, align 4
+  %2 = load i32, ptr %field22, align 4
+  %3 = add i32 %1, %2
+  %4 = load i32, ptr %val, align 4
+  store i32 %4, ptr %field8, align 4
+  %5 = add i32 %4, %3
+  %6 = load i32, ptr %val
+  %a1 = load i32, ptr %field8
+  %a = add i32 %a1, %6
+  %b = load i32, ptr %field3
+  ;%b  = sub i32 %b1, %6
+  %7 = select i1 %cond, ptr %field3, ptr %field8
+  store i32 1000, ptr %7
+  %8 = add i32 %5, %b
+  %9 = load i32, ptr %field8
+  %10 = add i32 %9, %8
+  ret i32 %10
+}
+
+; CHECK: Accesses by bin after update:
+; CHECK: [8-12] : 2
+; CHECK:      - 9 -   store i32 %0, ptr %field2, align 4
+; CHECK:        - c:   %0 = load i32, ptr %val, align 4
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   store i32 %0, ptr %field2, align 4
+; CHECK:   %field2 = getelementptr i32, ptr @globalBytes, i32 2
+; CHECK: @globalBytes = internal global [1024 x i8] zeroinitializer, align 16
+; CHECK:      - 6 -   %ret = load i32, ptr %x, align 4
+; CHECK:        - c: <unknown>
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   %ret = load i32, ptr %x, align 4
+; CHECK:   %x = phi ptr [ %field2, %then ], [ %field8, %else ]
+; CHECK:   %field2 = getelementptr i32, ptr @globalBytes, i32 2
+; CHECK: @globalBytes = internal global [1024 x i8] zeroinitializer, align 16
+; CHECK: Backtrack a unique access path:
+; CHECK:   %ret = load i32, ptr %x, align 4
+; CHECK:   %x = phi ptr [ %field2, %then ], [ %field8, %else ]
+; CHECK:   %field8 = getelementptr i32, ptr @globalBytes, i32 8
+; CHECK: @globalBytes = internal global [1024 x i8] zeroinitializer, align 16
+; CHECK: [32-36] : 5
+; CHECK:      - 6 -   %ret = load i32, ptr %x, align 4
+; CHECK:        - c: <unknown>
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   %ret = load i32, ptr %x, align 4
+; CHECK:   %x = phi ptr [ %field2, %then ], [ %field8, %else ]
+; CHECK:   %field2 = getelementptr i32, ptr @globalBytes, i32 2
+; CHECK: @globalBytes = internal global [1024 x i8] zeroinitializer, align 16
+; CHECK: Backtrack a unique access path:
+; CHECK:   %ret = load i32, ptr %x, align 4
+; CHECK:   %x = phi ptr [ %field2, %then ], [ %field8, %else ]
+; CHECK:   %field8 = getelementptr i32, ptr @globalBytes, i32 8
+; CHECK: @globalBytes = internal global [1024 x i8] zeroinitializer, align 16
+; CHECK:      - 9 -   store i32 %1, ptr %field8, align 4
+; CHECK:        - c:   %1 = load i32, ptr %val2, align 4
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   store i32 %1, ptr %field8, align 4
+; CHECK:   %field8 = getelementptr i32, ptr @globalBytes, i32 8
+; CHECK: @globalBytes = internal global [1024 x i8] zeroinitializer, align 16
+; CHECK:      - 9 -   store i32 %0, ptr %field2, align 4
+; CHECK:        - c:   %0 = load i32, ptr %val, align 4
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   store i32 %0, ptr %field2, align 4
+; CHECK:   %field2 = getelementptr i32, ptr @globalBytes, i32 8
+; CHECK: @globalBytes = internal global [1024 x i8] zeroinitializer, align 16
+; CHECK:      - 6 -   %ret = load i32, ptr %x, align 4
+; CHECK:        - c: <unknown>
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   %ret = load i32, ptr %x, align 4
+; CHECK:   %x = phi ptr [ %field2, %then ], [ %field8, %else ]
+; CHECK:   %field2 = getelementptr i32, ptr @globalBytes, i32 8
+; CHECK: @globalBytes = internal global [1024 x i8] zeroinitializer, align 16
+; CHECK: Backtrack a unique access path:
+; CHECK:   %ret = load i32, ptr %x, align 4
+; CHECK:   %x = phi ptr [ %field2, %then ], [ %field8, %else ]
+; CHECK:   %field8 = getelementptr i32, ptr @globalBytes, i32 8
+; CHECK: @globalBytes = internal global [1024 x i8] zeroinitializer, align 16
+; CHECK:      - 9 -   store i32 %1, ptr %field8, align 4
+; CHECK:        - c:   %1 = load i32, ptr %val2, align 4
+; CHECK: Print all access paths found:
+; CHECK: Backtrack a unique access path:
+; CHECK:   store i32 %1, ptr %field8, align 4
+; CHECK:   %field8 = getelementptr i32, ptr @globalBytes, i32 8
+; CHECK: @globalBytes = internal global [1024 x i8] zeroinitializer, align 16
+
+define dso_local i32 @phi_different_offsets(ptr nocapture %val, ptr nocapture %val2, i1 %cmp) {
+entry:
+  br i1 %cmp, label %then, label %else
+
+then:
+  %field2 = getelementptr i32, ptr @globalBytes, i32 2
+  %0 = load i32, ptr %val
+  store i32 %0, ptr %field2
+  br label %end
+
+else:
+  %field8 = getelementptr i32, ptr @globalBytes, i32 8
+  %2 = load i32, ptr %val2
+  store i32 %2, ptr %field8
+  br label %end
+
+end:
+  %x = phi ptr [ %field2, %then ], [ %field8, %else ]
+  %ret = load i32, ptr %x
+  ret i32 %ret
+
+}
+
+define dso_local i32 @phi_same_offsets(ptr nocapture %val, ptr nocapture %val2, i1 %cmp) {
+entry:
+  br i1 %cmp, label %then, label %else
+
+then:
+  %field2 = getelementptr i32, ptr @globalBytes, i32 8
+  %0 = load i32, ptr %val
+  store i32 %0, ptr %field2
+  br label %end
+
+else:
+  %field8 = getelementptr i32, ptr @globalBytes, i32 8
+  %2 = load i32, ptr %val2
+  store i32 %2, ptr %field8
+  br label %end
+
+end:
+  %x = phi ptr [ %field2, %then ], [ %field8, %else ]
+  %ret = load i32, ptr %x
+  ret i32 %ret
+}
\ No newline at end of file

>From ea992a180392d4c35699f272393cdd1c417d1569 Mon Sep 17 00:00:00 2001
From: Vidush Singhal <singhal2 at ruby964.llnl.gov>
Date: Tue, 4 Jun 2024 16:51:06 -0700
Subject: [PATCH 2/2] [Attributor]: Change allocation size and load/store
 offsets using AAPointerInfo for Alloca instructions

---
 llvm/include/llvm/Transforms/IPO/Attributor.h |   8 +
 .../Transforms/IPO/AttributorAttributes.cpp   | 471 ++++++++++++++++--
 .../Attributor/ArgumentPromotion/crash.ll     |   6 +-
 .../live_called_from_dead.ll                  |   3 +-
 .../live_called_from_dead_2.ll                |   3 +-
 .../nonzero-address-spaces.ll                 |   3 +-
 .../Attributor/IPConstantProp/pthreads.ll     |   4 +-
 llvm/test/Transforms/Attributor/allocator.ll  | 193 +++++--
 .../Attributor/call-simplify-pointer-info.ll  |  28 +-
 .../Transforms/Attributor/heap_to_stack.ll    |   3 +-
 .../Attributor/heap_to_stack_gpu.ll           |   3 +-
 llvm/test/Transforms/Attributor/liveness.ll   |   8 +-
 llvm/test/Transforms/Attributor/nodelete.ll   |   5 +-
 .../Transforms/Attributor/pointer-info.ll     |   6 +-
 .../Attributor/value-simplify-pointer-info.ll |  18 +-
 15 files changed, 634 insertions(+), 128 deletions(-)

diff --git a/llvm/include/llvm/Transforms/IPO/Attributor.h b/llvm/include/llvm/Transforms/IPO/Attributor.h
index e2065b496e210..2845aede72abd 100644
--- a/llvm/include/llvm/Transforms/IPO/Attributor.h
+++ b/llvm/include/llvm/Transforms/IPO/Attributor.h
@@ -5758,8 +5758,10 @@ struct AANonConvergent : public StateWrapper<BooleanState, AbstractAttribute> {
 
 /// An abstract interface for struct information.
 struct AAPointerInfo : public AbstractAttribute {
+protected:
   AAPointerInfo(const IRPosition &IRP) : AbstractAttribute(IRP) {}
 
+public:
   /// See AbstractAttribute::isValidIRPositionForInit
   static bool isValidIRPositionForInit(Attributor &A, const IRPosition &IRP) {
     if (!IRP.getAssociatedType()->isPtrOrPtrVectorTy())
@@ -6309,6 +6311,9 @@ struct AAPointerInfo : public AbstractAttribute {
   virtual const_bin_iterator begin() const = 0;
   virtual const_bin_iterator end() const = 0;
   virtual int64_t numOffsetBins() const = 0;
+  virtual void dumpState(raw_ostream &O) const = 0;
+  virtual const Access &getBinAccess(unsigned Index) const = 0;
+  virtual const DenseMap<Value *, OffsetInfo> &getOffsetInfoMap() const = 0;
 
   /// Call \p CB on all accesses that might interfere with \p Range and return
   /// true if all such accesses were known and the callback returned true for
@@ -6480,6 +6485,9 @@ struct AAAllocationInfo : public StateWrapper<BooleanState, AbstractAttribute> {
 
   virtual std::optional<TypeSize> getAllocatedSize() const = 0;
 
+  using NewOffsetsTy = DenseMap<AA::RangeTy, AA::RangeTy>;
+  virtual const NewOffsetsTy &getNewOffsets() const = 0;
+
   /// See AbstractAttribute::getName()
   const std::string getName() const override { return "AAAllocationInfo"; }
 
diff --git a/llvm/lib/Transforms/IPO/AttributorAttributes.cpp b/llvm/lib/Transforms/IPO/AttributorAttributes.cpp
index ca7b6e555111c..a508d55967b64 100644
--- a/llvm/lib/Transforms/IPO/AttributorAttributes.cpp
+++ b/llvm/lib/Transforms/IPO/AttributorAttributes.cpp
@@ -1144,6 +1144,15 @@ struct AAPointerInfoImpl
     return State::numOffsetBins();
   }
 
+  virtual const Access &getBinAccess(unsigned Index) const override {
+    return getAccess(Index);
+  }
+
+  virtual const DenseMap<Value *, OffsetInfo> &
+  getOffsetInfoMap() const override {
+    return OffsetInfoMap;
+  }
+
   bool forallInterferingAccesses(
       AA::RangeTy Range,
       function_ref<bool(const AAPointerInfo::Access &, bool)> CB)
@@ -1487,12 +1496,15 @@ struct AAPointerInfoImpl
     return Changed;
   }
 
+  // /// Offsets Info Map
+  // DenseMap<Value *, OffsetInfo> OffsetInfoMap;
+
   /// Statistic tracking for all AAPointerInfo implementations.
   /// See AbstractAttribute::trackStatistics().
   void trackPointerInfoStatistics(const IRPosition &IRP) const {}
 
   /// Dump the state into \p O.
-  void dumpState(raw_ostream &O) {
+  virtual void dumpState(raw_ostream &O) const override {
     for (auto &It : OffsetBins) {
       O << "[" << It.first.Offset << "-" << It.first.Offset + It.first.Size
         << "] : " << It.getSecond().size() << "\n";
@@ -12765,6 +12777,11 @@ struct AAAllocationInfoImpl : public AAAllocationInfo {
     return AssumedAllocatedSize;
   }
 
+  const NewOffsetsTy &getNewOffsets() const override {
+    assert(isValidState() && "the AA is invalid");
+    return NewComputedOffsets;
+  }
+
   std::optional<TypeSize> findInitialAllocationSize(Instruction *I,
                                                     const DataLayout &DL) {
 
@@ -12809,42 +12826,53 @@ struct AAAllocationInfoImpl : public AAAllocationInfo {
     if (!AllocationSize)
       return indicatePessimisticFixpoint();
 
-    // For zero sized allocations, we give up.
-    // Since we can't reduce further
+    // For zero sized allocations, we give up
+    // because we cannot reduce them any further.
     if (*AllocationSize == 0)
       return indicatePessimisticFixpoint();
 
-    int64_t BinSize = PI->numOffsetBins();
-
-    // TODO: implement for multiple bins
-    if (BinSize > 1)
-      return indicatePessimisticFixpoint();
-
-    if (BinSize == 0) {
+    int64_t NumBins = PI->numOffsetBins();
+    if (NumBins == 0) {
       auto NewAllocationSize = std::optional<TypeSize>(TypeSize(0, false));
       if (!changeAllocationSize(NewAllocationSize))
         return ChangeStatus::UNCHANGED;
       return ChangeStatus::CHANGED;
     }
 
-    // TODO: refactor this to be part of multiple bin case
-    const auto &It = PI->begin();
+    // For each access bin we compute its new start offset
+    // and store the results in a new map (NewOffsetBins).
+    // NewOffsetsBins is a Map from AA::RangeTy OldRange to AA::RangeTy
+    // NewRange.
+    unsigned long PrevBinEndOffset = 0;
+    bool ChangedOffsets = false;
+    for (AAPointerInfo::OffsetBinsTy::const_iterator It = PI->begin();
+         It != PI->end(); It++) {
+      const AA::RangeTy &OldRange = It->getFirst();
 
-    // TODO: handle if Offset is not zero
-    if (It->first.Offset != 0)
-      return indicatePessimisticFixpoint();
+      // If any byte range has an unknown offset or size, we should leave the
+      // original allocation unmodified.
+      if (OldRange.offsetOrSizeAreUnknown())
+        return indicatePessimisticFixpoint();
 
-    uint64_t SizeOfBin = It->first.Offset + It->first.Size;
+      unsigned long NewStartOffset = PrevBinEndOffset;
+      unsigned long NewEndOffset = NewStartOffset + OldRange.Size;
+      PrevBinEndOffset = NewEndOffset;
 
-    if (SizeOfBin >= *AllocationSize)
-      return indicatePessimisticFixpoint();
+      ChangedOffsets |= setNewOffsets(OldRange, OldRange.Offset, NewStartOffset,
+                                      OldRange.Size);
+    }
 
+    // Set the new size of the allocation. The new size of the Allocation should
+    // be the size of PrevBinEndOffset * 8 in bits.
     auto NewAllocationSize =
-        std::optional<TypeSize>(TypeSize(SizeOfBin * 8, false));
+        std::optional<TypeSize>(TypeSize(PrevBinEndOffset * 8, false));
 
     if (!changeAllocationSize(NewAllocationSize))
       return ChangeStatus::UNCHANGED;
 
+    if (!ChangedOffsets)
+      return ChangeStatus::UNCHANGED;
+
     return ChangeStatus::CHANGED;
   }
 
@@ -12854,39 +12882,314 @@ struct AAAllocationInfoImpl : public AAAllocationInfo {
     assert(isValidState() &&
            "Manifest should only be called if the state is valid.");
 
-    Instruction *I = getIRPosition().getCtxI();
+    bool Changed = false;
+    const IRPosition &IRP = getIRPosition();
+    Instruction *I = IRP.getCtxI();
+
+    // Check if simplified values exist.
+    if (checkIfSimplifiedValuesExists(A, I))
+      return ChangeStatus::UNCHANGED;
 
-    auto FixedAllocatedSizeInBits = getAllocatedSize()->getFixedValue();
+    if (getAllocatedSize() == HasNoAllocationSize)
+      return ChangeStatus::UNCHANGED;
 
-    unsigned long NumBytesToAllocate = (FixedAllocatedSizeInBits + 7) / 8;
+    const AAPointerInfo *PI =
+        A.getOrCreateAAFor<AAPointerInfo>(IRP, *this, DepClassTy::REQUIRED);
 
-    switch (I->getOpcode()) {
-    // TODO: add case for malloc like calls
-    case Instruction::Alloca: {
+    if (!PI)
+      return ChangeStatus::UNCHANGED;
 
-      AllocaInst *AI = cast<AllocaInst>(I);
+    assert(PI->getState().isValidState() &&
+           "[AAAllocationinfo]: AAPointerinfo was not in valid state!");
 
-      Type *CharType = Type::getInt8Ty(I->getContext());
+    // Store a map where each instruction is mapped to a map containing
+    // old bins accessed by that instruction to the corresponding new
+    // bins in the allocation.
+    DenseMap<Instruction *, DenseMap<AA::RangeTy, AA::RangeTy>>
+        AccessedInstructionsToBinsMap;
 
-      auto *NumBytesToValue =
-          ConstantInt::get(I->getContext(), APInt(32, NumBytesToAllocate));
+    auto AddBins =
+        [](DenseMap<Instruction *, DenseMap<AA::RangeTy, AA::RangeTy>> &Map,
+           Instruction *LocalInst, const AA::RangeTy &OldRange,
+           const AA::RangeTy &NewRange) {
+          DenseMap<AA::RangeTy, AA::RangeTy> &NewBinsForInstruction =
+              Map.getOrInsertDefault(LocalInst);
 
-      BasicBlock::iterator insertPt = AI->getIterator();
-      insertPt = std::next(insertPt);
-      AllocaInst *NewAllocaInst =
-          new AllocaInst(CharType, AI->getAddressSpace(), NumBytesToValue,
-                         AI->getAlign(), AI->getName(), insertPt);
+          NewBinsForInstruction.insert(std::make_pair(OldRange, NewRange));
+        };
 
-      if (A.changeAfterManifest(IRPosition::inst(*AI), *NewAllocaInst))
-        return ChangeStatus::CHANGED;
+    const auto &NewOffsetsMap = getNewOffsets();
+    const auto &OffsetInfoMap = PI->getOffsetInfoMap();
+
+    // Map access causing instructions to a tuple of (Old, New) bins.
+    // The access causing instruction contains the pointer operand
+    // which comes from the allocation we may want to backtrack that
+    // pointer operand, there are 2 cases that may arise.
+    // A) A GEP exists that calculates the pointer operand from the original
+    // allocation instruction: I
+    // B) A GEP does not exists in which case we need to insert a GEP just
+    // before the access causing instruction with the shift value from the
+    // original offset.
+    for (AAPointerInfo::OffsetBinsTy::const_iterator It = PI->begin();
+         It != PI->end(); It++) {
+      const auto &OldOffsetRange = It->getFirst();
+      // If the OldOffsetRange is not in the map, offsets for that bin did not
+      // change. We should just continue and skip changing the offsets in that
+      // case.
+      if (!NewOffsetsMap.contains(OldOffsetRange))
+        continue;
+
+      const auto &NewOffsetRange = NewOffsetsMap.lookup(OldOffsetRange);
+      for (const auto AccIndex : It->getSecond()) {
+        const auto &AccessInstruction = PI->getBinAccess(AccIndex);
+        Instruction *LocalInst = AccessInstruction.getLocalInst();
+
+        if (checkIfSimplifiedValuesExists(A, LocalInst))
+          return ChangeStatus::UNCHANGED;
+
+        if (checkIfAccessChainUsesMultipleBins(A, LocalInst, OffsetInfoMap))
+          return ChangeStatus::UNCHANGED;
+
+        // Check if we can backtrack the access causing instruction to a GEP
+        // from the original allocation, if yes, then we prefer to change the
+        // GEP rather than inserting a new one just before the access causing
+        // instruction.
+        switch (LocalInst->getOpcode()) {
+        case Instruction::Call: {
+          CallInst *CallInstruction = cast<CallInst>(LocalInst);
+          for (auto *It = CallInstruction->op_begin();
+               It != CallInstruction->op_end(); It++) {
+            if (Instruction *OperandInstruction = dyn_cast<Instruction>(It)) {
+              // Operand does not cause an access in the current byte range.
+              if (!OffsetInfoMap.contains(OperandInstruction))
+                continue;
+
+              // Find the old offset and the corresponding new offset for the
+              // call argument.
+              auto OffsetsVecArg =
+                  OffsetInfoMap.lookup(OperandInstruction).Ranges;
+              int64_t OldOffsetArg = OffsetsVecArg.front().Offset;
+              int NewOffsetArg = 0;
+              for (auto OldToNewRange : NewOffsetsMap) {
+                auto Old = OldToNewRange.getFirst();
+                if (Old.Offset == OldOffsetArg)
+                  NewOffsetArg = OldToNewRange.getSecond().Offset;
+              }
+
+              // If the offsets did not change, continue.
+              if (NewOffsetArg == OldOffsetArg)
+                continue;
+
+              // We don't have access to the size of the offset here but it is
+              // ok since we do not need it here.
+              AA::RangeTy &CallArgOldRange = OffsetsVecArg.front();
+              AA::RangeTy CallArgNewRange =
+                  AA::RangeTy(NewOffsetArg, CallArgOldRange.Size);
+
+              // Find the chain the call instruction is part of
+              const AAPointerInfo::AccessPathSetTy *AccessPaths =
+                  AccessInstruction.getAccessChain();
+
+              const AAPointerInfo::AccessPathTy *ChainWithArg = nullptr;
+              for (auto *Chain : *AccessPaths) {
+
+                if (std::find(Chain->begin(), Chain->end(),
+                              OperandInstruction) != Chain->end()) {
+                  ChainWithArg = Chain;
+                }
+              }
+
+              bool BackTrackInstructionToGEP = false;
+              if (ChainWithArg) {
+                bool Exists = false;
+                for (auto *V : *ChainWithArg) {
+
+                  GetElementPtrInst *GepI = dyn_cast<GetElementPtrInst>(V);
+
+                  if (!GepI)
+                    continue;
+
+                  if (AccessedInstructionsToBinsMap.contains(GepI)) {
+                    Exists = true;
+                    continue;
+                  }
+
+                  // check if its a GEP and weather the GEP accesses the
+                  // Allocation
+                  if (GepI->getPointerOperand() == I) {
+                    if (checkIfSimplifiedValuesExists(A, GepI))
+                      return ChangeStatus::UNCHANGED;
 
+                    AddBins(AccessedInstructionsToBinsMap, GepI,
+                            CallArgOldRange, CallArgNewRange);
+                    BackTrackInstructionToGEP = true;
+                  }
+                }
+
+                if (Exists)
+                  continue;
+              }
+
+              if (!BackTrackInstructionToGEP) {
+                AddBins(AccessedInstructionsToBinsMap, OperandInstruction,
+                        CallArgOldRange, CallArgNewRange);
+                continue;
+              }
+            }
+          }
+          break;
+        }
+        default: {
+
+          bool BackTrackInstructionToGEP = false;
+          bool Exists = false;
+          const AAPointerInfo::AccessPathSetTy *AccessPaths =
+              AccessInstruction.getAccessChain();
+          for (auto *Chain : *AccessPaths) {
+            for (auto *V : *Chain) {
+
+              GetElementPtrInst *GepI = dyn_cast<GetElementPtrInst>(V);
+
+              if (!GepI)
+                continue;
+
+              if (AccessedInstructionsToBinsMap.contains(GepI)) {
+                Exists = true;
+                continue;
+              }
+
+              // check if its a GEP and weather the GEP accesses the Allocation
+              if (GepI->getPointerOperand() == I) {
+                if (checkIfSimplifiedValuesExists(A, GepI))
+                  return ChangeStatus::UNCHANGED;
+
+                AddBins(AccessedInstructionsToBinsMap, GepI, OldOffsetRange,
+                        NewOffsetRange);
+                BackTrackInstructionToGEP = true;
+              }
+            }
+          }
+
+          if (Exists)
+            continue;
+
+          if (!BackTrackInstructionToGEP)
+            AddBins(AccessedInstructionsToBinsMap, LocalInst, OldOffsetRange,
+                    NewOffsetRange);
+
+          break;
+        }
+        }
+      }
+    }
+
+    unsigned long FixedAllocatedSizeInBits =
+        getAllocatedSize()->getFixedValue();
+    unsigned long NumBytesToAllocate = (FixedAllocatedSizeInBits + 7) / 8;
+    Type *NewAllocationType = nullptr;
+    switch (I->getOpcode()) {
+    // TODO: add case for malloc like calls
+    case Instruction::Alloca: {
+      AllocaInst *OldAllocaInst = cast<AllocaInst>(I);
+      const DataLayout &DL = A.getDataLayout();
+      auto OriginalAllocationSize = OldAllocaInst->getAllocationSizeInBits(DL);
+
+      if (*OriginalAllocationSize <= FixedAllocatedSizeInBits)
+        return ChangeStatus::UNCHANGED;
+
+      Type *CharType = Type::getInt8Ty(I->getContext());
+      Type *CharArrayType = ArrayType::get(CharType, NumBytesToAllocate);
+      NewAllocationType = CharArrayType;
+      BasicBlock::iterator InsertPt = OldAllocaInst->getIterator();
+      InsertPt = std::next(InsertPt);
+      Instruction *NewAllocationInstruction =
+          new AllocaInst(CharArrayType, OldAllocaInst->getAddressSpace(),
+                         OldAllocaInst->getName(), InsertPt);
+
+      Changed |= A.changeAfterManifest(IRPosition::inst(*I),
+                                       *NewAllocationInstruction);
+      A.deleteAfterManifest(*I);
       break;
     }
     default:
       break;
     }
 
-    return ChangeStatus::UNCHANGED;
+    for (auto &It : AccessedInstructionsToBinsMap) {
+      Instruction *LocalInst = It.first;
+      // Get a hold of a map, mapping old to new bins.
+      DenseMap<AA::RangeTy, AA::RangeTy> &OldToNewBins = It.second;
+      IntegerType *Int64TyInteger =
+          IntegerType::get(LocalInst->getContext(), 64);
+      switch (LocalInst->getOpcode()) {
+      case Instruction::Load: {
+        // The number of bytes to shift the load/store by.
+        int64_t OffsetOld = OldToNewBins.begin()->getFirst().Offset;
+        int64_t OffsetNew = OldToNewBins.begin()->getSecond().Offset;
+        LoadInst *OldLoadInst = cast<LoadInst>(LocalInst);
+        Instruction *PointerOperand =
+            cast<Instruction>(OldLoadInst->getPointerOperand());
+        Type *PointeeTy = OldLoadInst->getPointerOperandType();
+        int64_t ShiftValue = OffsetNew - OffsetOld;
+        Value *IndexList[1] = {ConstantInt::get(Int64TyInteger, ShiftValue)};
+        Value *GepToNewAddress = GetElementPtrInst::Create(
+            PointeeTy, PointerOperand, IndexList, "NewGep", OldLoadInst);
+
+        LoadInst *NewLoadInst = new LoadInst(
+            OldLoadInst->getType(), GepToNewAddress, OldLoadInst->getName(),
+            false, OldLoadInst->getAlign(), OldLoadInst);
+
+        Changed |=
+            A.changeAfterManifest(IRPosition::inst(*OldLoadInst), *NewLoadInst);
+
+        A.deleteAfterManifest(*OldLoadInst);
+        break;
+      }
+      case Instruction::Store: {
+        // The number of bytes to shift the load/store by.
+        int64_t OffsetOld = OldToNewBins.begin()->getFirst().Offset;
+        int64_t OffsetNew = OldToNewBins.begin()->getSecond().Offset;
+        int64_t ShiftValue = OffsetNew - OffsetOld;
+        StoreInst *OldStoreInst = cast<StoreInst>(LocalInst);
+        Instruction *PointerOperand =
+            cast<Instruction>(OldStoreInst->getPointerOperand());
+        Type *PointeeTy = OldStoreInst->getPointerOperandType();
+        Value *IndexList[1] = {ConstantInt::get(Int64TyInteger, ShiftValue)};
+        Value *GepToNewAddress = GetElementPtrInst::Create(
+            PointeeTy, PointerOperand, IndexList, "NewGep", OldStoreInst);
+
+        StoreInst *NewStoreInst =
+            new StoreInst(OldStoreInst->getValueOperand(), GepToNewAddress,
+                          false, OldStoreInst->getAlign(), OldStoreInst);
+
+        Changed |= A.changeAfterManifest(IRPosition::inst(*OldStoreInst),
+                                         *NewStoreInst);
+
+        A.deleteAfterManifest(*OldStoreInst);
+        break;
+      }
+      case Instruction::GetElementPtr: {
+        GetElementPtrInst *OldGEP = cast<GetElementPtrInst>(LocalInst);
+        int64_t OffsetNew = OldToNewBins.begin()->getSecond().Offset;
+        Value *IndexList[1] = {ConstantInt::get(Int64TyInteger, OffsetNew)};
+        Value *OldPointerOperand = OldGEP->getPointerOperand();
+        Value *GepToNewAddress = GetElementPtrInst::Create(
+            NewAllocationType, OldPointerOperand, IndexList, "NewGep", OldGEP);
+
+        Changed |=
+            A.changeAfterManifest(IRPosition::inst(*OldGEP), *GepToNewAddress);
+
+        A.deleteAfterManifest(*OldGEP);
+        break;
+      }
+      default:
+        break;
+      }
+    }
+
+    if (!Changed)
+      return ChangeStatus::UNCHANGED;
+    return ChangeStatus::CHANGED;
   }
 
   /// See AbstractAttribute::getAsStr().
@@ -12900,8 +13203,28 @@ struct AAAllocationInfoImpl : public AAAllocationInfo {
            ")";
   }
 
+  void dumpNewOffsetBins(raw_ostream &O) {
+
+    O << "Printing Map from [OldOffsetsRange] : [NewOffsetsRange] if the "
+         "offsets changed."
+      << "\n";
+    const auto &NewOffsetsMap = getNewOffsets();
+    for (auto It = NewOffsetsMap.begin(); It != NewOffsetsMap.end(); It++) {
+
+      const auto &OldRange = It->getFirst();
+      const auto &NewRange = It->getSecond();
+
+      O << "[" << OldRange.Offset << "," << OldRange.Offset + OldRange.Size
+        << "] : ";
+      O << "[" << NewRange.Offset << "," << NewRange.Offset + NewRange.Size
+        << "]";
+      O << "\n";
+    }
+  }
+
 private:
   std::optional<TypeSize> AssumedAllocatedSize = HasNoAllocationSize;
+  NewOffsetsTy NewComputedOffsets;
 
   // Maintain the computed allocation size of the object.
   // Returns (bool) weather the size of the allocation was modified or not.
@@ -12913,6 +13236,80 @@ struct AAAllocationInfoImpl : public AAAllocationInfo {
     }
     return false;
   }
+
+  // Maps an old byte range to its new offset range in the new allocation.
+  // Returns (bool) weather the old byte range's offsets changed or not.
+  bool setNewOffsets(const AA::RangeTy &OldRange, int64_t OldOffset,
+                     int64_t NewComputedOffset, int64_t Size) {
+
+    if (OldOffset == NewComputedOffset)
+      return false;
+
+    AA::RangeTy &NewRange = NewComputedOffsets.getOrInsertDefault(OldRange);
+    NewRange.Offset = NewComputedOffset;
+    NewRange.Size = Size;
+
+    return true;
+  }
+
+  // A helper function to check if simplified values exists for the current
+  // instruction.
+  // Right now we don't change the value and give up
+  // on modifying the size and offsets of the allocation
+  // but this may be sub-optimal.
+  // TODO: handle case for a similified value
+  bool checkIfSimplifiedValuesExists(Attributor &A, Instruction *LocalInst) {
+
+    // If there are potential values that replace the accessed instruction, we
+    // should use those values instead.
+    bool UsedAssumedInformation = false;
+    SmallVector<AA::ValueAndContext> Values;
+    if (A.getAssumedSimplifiedValues(IRPosition::inst(*LocalInst), *this,
+                                     Values, AA::AnyScope,
+                                     UsedAssumedInformation))
+
+      for (auto &ValAndContext : Values)
+        // Don't modify the instruction if any simplified value exists.
+        if (ValAndContext.getValue() && ValAndContext.getValue() != LocalInst)
+          return true;
+
+    return false;
+  }
+
+  bool checkIfAccessChainUsesMultipleBins(
+      Attributor &A, Instruction *LocalInst,
+      const DenseMap<Value *, AAPointerInfo::OffsetInfo> &OffsetInfoMap) {
+
+    // BackTrack and check if any instruction in the access causing chain
+    // accessed multiple byte ranges. If they do, we currently give up.
+    SmallVector<Instruction *> ReadyList;
+    DenseMap<Instruction *, bool> Visited;
+    ReadyList.push_back(LocalInst);
+    while (!ReadyList.empty()) {
+      Instruction *GetBack = ReadyList.back();
+      ReadyList.pop_back();
+
+      if (!Visited.insert(std::make_pair(GetBack, true)).second)
+        continue;
+
+      // Check if the Instruction has multiple bins, if so give up
+      // for calls it is okay to have multiple bins since they may
+      // come from different call arguments and we can address them
+      // seperately.
+      // TODO: handle when one instruction has multiple bins
+      auto OffsetsVecArg = OffsetInfoMap.lookup(GetBack).Ranges;
+      if (GetBack->getOpcode() != Instruction::Call && OffsetsVecArg.size() > 1)
+        return true;
+
+      for (auto *It = GetBack->op_begin(); It != GetBack->op_end(); It++) {
+        if (Instruction *Ins = dyn_cast<Instruction>(*It)) {
+          ReadyList.push_back(Ins);
+        }
+      }
+    }
+
+    return false;
+  }
 };
 
 struct AAAllocationInfoFloating : AAAllocationInfoImpl {
diff --git a/llvm/test/Transforms/Attributor/ArgumentPromotion/crash.ll b/llvm/test/Transforms/Attributor/ArgumentPromotion/crash.ll
index 595cb37c6c93e..f0efa2a0ae3c1 100644
--- a/llvm/test/Transforms/Attributor/ArgumentPromotion/crash.ll
+++ b/llvm/test/Transforms/Attributor/ArgumentPromotion/crash.ll
@@ -106,10 +106,8 @@ define i32 @test_inf_promote_caller(i32 %arg) {
 ; CGSCC-LABEL: define {{[^@]+}}@test_inf_promote_caller
 ; CGSCC-SAME: (i32 [[ARG:%.*]]) #[[ATTR3:[0-9]+]] {
 ; CGSCC-NEXT:  bb:
-; CGSCC-NEXT:    [[TMP:%.*]] = alloca [[S:%.*]], align 8
-; CGSCC-NEXT:    [[TMP3:%.*]] = alloca i8, i32 0, align 8
-; CGSCC-NEXT:    [[TMP1:%.*]] = alloca [[S]], align 8
-; CGSCC-NEXT:    [[TMP14:%.*]] = alloca i8, i32 0, align 8
+; CGSCC-NEXT:    [[TMP3:%.*]] = alloca [0 x i8], align 1
+; CGSCC-NEXT:    [[TMP14:%.*]] = alloca [0 x i8], align 1
 ; CGSCC-NEXT:    ret i32 0
 ;
 bb:
diff --git a/llvm/test/Transforms/Attributor/ArgumentPromotion/live_called_from_dead.ll b/llvm/test/Transforms/Attributor/ArgumentPromotion/live_called_from_dead.ll
index 1c34fff8dd755..63dbc4da7da37 100644
--- a/llvm/test/Transforms/Attributor/ArgumentPromotion/live_called_from_dead.ll
+++ b/llvm/test/Transforms/Attributor/ArgumentPromotion/live_called_from_dead.ll
@@ -36,8 +36,7 @@ define internal i32 @caller(ptr %B) {
 ; CGSCC: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(none)
 ; CGSCC-LABEL: define {{[^@]+}}@caller
 ; CGSCC-SAME: () #[[ATTR0]] {
-; CGSCC-NEXT:    [[A:%.*]] = alloca i32, align 4
-; CGSCC-NEXT:    [[A1:%.*]] = alloca i8, i32 0, align 4
+; CGSCC-NEXT:    [[A1:%.*]] = alloca [0 x i8], align 1
 ; CGSCC-NEXT:    ret i32 0
 ;
   %A = alloca i32
diff --git a/llvm/test/Transforms/Attributor/ArgumentPromotion/live_called_from_dead_2.ll b/llvm/test/Transforms/Attributor/ArgumentPromotion/live_called_from_dead_2.ll
index b42647840f7cf..956fa0e88b028 100644
--- a/llvm/test/Transforms/Attributor/ArgumentPromotion/live_called_from_dead_2.ll
+++ b/llvm/test/Transforms/Attributor/ArgumentPromotion/live_called_from_dead_2.ll
@@ -53,8 +53,7 @@ define internal i32 @caller(ptr %B) {
 ; CGSCC: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(argmem: write)
 ; CGSCC-LABEL: define {{[^@]+}}@caller
 ; CGSCC-SAME: (ptr noalias nocapture nofree noundef nonnull writeonly align 4 dereferenceable(4) [[B:%.*]]) #[[ATTR0]] {
-; CGSCC-NEXT:    [[A:%.*]] = alloca i32, align 4
-; CGSCC-NEXT:    [[A1:%.*]] = alloca i8, i32 0, align 4
+; CGSCC-NEXT:    [[A1:%.*]] = alloca [0 x i8], align 1
 ; CGSCC-NEXT:    [[C:%.*]] = call i32 @test(ptr noalias nocapture nofree noundef nonnull writeonly align 4 dereferenceable(4) [[B]]) #[[ATTR2:[0-9]+]]
 ; CGSCC-NEXT:    ret i32 0
 ;
diff --git a/llvm/test/Transforms/Attributor/ArgumentPromotion/nonzero-address-spaces.ll b/llvm/test/Transforms/Attributor/ArgumentPromotion/nonzero-address-spaces.ll
index b588a399e5bd9..7b5e1276ac212 100644
--- a/llvm/test/Transforms/Attributor/ArgumentPromotion/nonzero-address-spaces.ll
+++ b/llvm/test/Transforms/Attributor/ArgumentPromotion/nonzero-address-spaces.ll
@@ -29,8 +29,7 @@ define internal i32 @foo(ptr) {
 ; CHECK-LABEL: define {{[^@]+}}@foo
 ; CHECK-SAME: () addrspace(1) #[[ATTR0:[0-9]+]] {
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[RETVAL1:%.*]] = alloca i8, i32 0, align 4
+; CHECK-NEXT:    [[RETVAL1:%.*]] = alloca [0 x i8], align 1
 ; CHECK-NEXT:    call addrspace(0) void asm sideeffect "ldr r0, [r0] \0Abx lr \0A", ""()
 ; CHECK-NEXT:    unreachable
 ;
diff --git a/llvm/test/Transforms/Attributor/IPConstantProp/pthreads.ll b/llvm/test/Transforms/Attributor/IPConstantProp/pthreads.ll
index 490894d129023..af2d1ef1eabba 100644
--- a/llvm/test/Transforms/Attributor/IPConstantProp/pthreads.ll
+++ b/llvm/test/Transforms/Attributor/IPConstantProp/pthreads.ll
@@ -34,8 +34,8 @@ target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 define dso_local i32 @main() {
 ; TUNIT-LABEL: define {{[^@]+}}@main() {
 ; TUNIT-NEXT:  entry:
-; TUNIT-NEXT:    [[ALLOC11:%.*]] = alloca i8, i32 0, align 8
-; TUNIT-NEXT:    [[ALLOC22:%.*]] = alloca i8, i32 0, align 8
+; TUNIT-NEXT:    [[ALLOC11:%.*]] = alloca [0 x i8], align 1
+; TUNIT-NEXT:    [[ALLOC22:%.*]] = alloca [0 x i8], align 1
 ; TUNIT-NEXT:    [[THREAD:%.*]] = alloca i64, align 8
 ; TUNIT-NEXT:    [[CALL:%.*]] = call i32 @pthread_create(ptr noundef nonnull align 8 dereferenceable(8) [[THREAD]], ptr noundef align 4294967296 null, ptr noundef nonnull @foo, ptr nofree readnone align 4294967296 undef)
 ; TUNIT-NEXT:    [[CALL1:%.*]] = call i32 @pthread_create(ptr noundef nonnull align 8 dereferenceable(8) [[THREAD]], ptr noundef align 4294967296 null, ptr noundef nonnull @bar, ptr noalias nocapture nofree nonnull readnone align 8 dereferenceable(8) undef)
diff --git a/llvm/test/Transforms/Attributor/allocator.ll b/llvm/test/Transforms/Attributor/allocator.ll
index f2d9ecd1d8fa4..693386f053615 100644
--- a/llvm/test/Transforms/Attributor/allocator.ll
+++ b/llvm/test/Transforms/Attributor/allocator.ll
@@ -13,8 +13,8 @@ define dso_local void @positive_alloca_1(i32 noundef %val) #0 {
 ; CHECK-LABEL: define dso_local void @positive_alloca_1
 ; CHECK-SAME: (i32 noundef [[VAL:%.*]]) {
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[VAL_ADDR1:%.*]] = alloca i8, i32 4, align 4
-; CHECK-NEXT:    [[F2:%.*]] = alloca i8, i32 4, align 4
+; CHECK-NEXT:    [[VAL_ADDR1:%.*]] = alloca [4 x i8], align 1
+; CHECK-NEXT:    [[F2:%.*]] = alloca [4 x i8], align 1
 ; CHECK-NEXT:    store i32 [[VAL]], ptr [[VAL_ADDR1]], align 4
 ; CHECK-NEXT:    store i32 10, ptr [[F2]], align 4
 ; CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr [[F2]], align 4
@@ -164,37 +164,52 @@ entry:
 ;TODO: The allocation can be reduced here.
 ;However, the offsets (load/store etc.) Need to be changed.
 ; Function Attrs: noinline nounwind uwtable
-define dso_local { i64, ptr } @positive_test_not_a_single_start_offset(i32 noundef %val) #0 {
-; CHECK: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(none)
-; CHECK-LABEL: define dso_local { i64, ptr } @positive_test_not_a_single_start_offset
-; CHECK-SAME: (i32 noundef [[VAL:%.*]]) #[[ATTR0:[0-9]+]] {
+define dso_local void @positive_test_not_a_single_start_offset(i32 noundef %val) #0 {
+; CHECK-LABEL: define dso_local void @positive_test_not_a_single_start_offset
+; CHECK-SAME: (i32 noundef [[VAL:%.*]]) {
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[RETVAL:%.*]] = alloca [[STRUCT_FOO:%.*]], align 8
 ; CHECK-NEXT:    [[VAL_ADDR:%.*]] = alloca i32, align 4
+; CHECK-NEXT:    [[F1:%.*]] = alloca [5 x i8], align 1
 ; CHECK-NEXT:    store i32 [[VAL]], ptr [[VAL_ADDR]], align 4
-; CHECK-NEXT:    store i32 2, ptr [[RETVAL]], align 8
-; CHECK-NEXT:    [[FIELD3:%.*]] = getelementptr inbounds [[STRUCT_FOO]], ptr [[RETVAL]], i32 0, i32 2
-; CHECK-NEXT:    store ptr [[VAL_ADDR]], ptr [[FIELD3]], align 8
-; CHECK-NEXT:    [[TMP0:%.*]] = load { i64, ptr }, ptr [[RETVAL]], align 8
-; CHECK-NEXT:    ret { i64, ptr } [[TMP0]]
+; CHECK-NEXT:    [[MUL:%.*]] = mul nsw i32 2, [[VAL]]
+; CHECK-NEXT:    store i32 [[MUL]], ptr [[F1]], align 4
+; CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr [[F1]], align 4
+; CHECK-NEXT:    [[CALL:%.*]] = call i32 (ptr, ...) @printf(ptr noundef nonnull dereferenceable(17) @.str, i32 noundef [[TMP0]])
+; CHECK-NEXT:    [[NEWGEP:%.*]] = getelementptr [5 x i8], ptr [[F1]], i64 4
+; CHECK-NEXT:    [[CONV1:%.*]] = trunc i32 [[TMP0]] to i8
+; CHECK-NEXT:    store i8 [[CONV1]], ptr [[NEWGEP]], align 4
+; CHECK-NEXT:    [[NEWGEP2:%.*]] = getelementptr [5 x i8], ptr [[F1]], i64 4
+; CHECK-NEXT:    [[TMP1:%.*]] = load i8, ptr [[NEWGEP2]], align 4
+; CHECK-NEXT:    [[CONV:%.*]] = sext i8 [[TMP1]] to i32
+; CHECK-NEXT:    [[CALL3:%.*]] = call i32 (ptr, ...) @printf(ptr noundef nonnull dereferenceable(17) @.str, i32 noundef [[CONV]])
+; CHECK-NEXT:    ret void
 ;
 entry:
-  %retval = alloca %struct.Foo, align 8
   %val.addr = alloca i32, align 4
+  %f = alloca %struct.Foo, align 4
   store i32 %val, ptr %val.addr, align 4
-  %field1 = getelementptr inbounds %struct.Foo, ptr %retval, i32 0, i32 0
-  store i32 2, ptr %field1, align 8
-  %field3 = getelementptr inbounds %struct.Foo, ptr %retval, i32 0, i32 2
-  store ptr %val.addr, ptr %field3, align 8
-  %0 = load { i64, ptr }, ptr %retval, align 8
-  ret { i64, ptr } %0
+  %0 = load i32, ptr %val.addr, align 4
+  %mul = mul nsw i32 2, %0
+  %a = getelementptr inbounds %struct.Foo, ptr %f, i32 0, i32 0
+  store i32 %mul, ptr %a, align 4
+  %a1 = getelementptr inbounds %struct.Foo, ptr %f, i32 0, i32 0
+  %1 = load i32, ptr %a1, align 4
+  %call = call i32 (ptr, ...) @printf(ptr noundef @.str, i32 noundef %1)
+  %c = getelementptr inbounds %struct.Foo, ptr %f, i32 0, i32 2
+  %conv1 = trunc i32 %1 to i8
+  store i8 %conv1, ptr %c, align 4
+  %c2 = getelementptr inbounds %struct.Foo, ptr %f, i32 0, i32 2
+  %2 = load i8, ptr %c2, align 4
+  %conv = sext i8 %2 to i32
+  %call3 = call i32 (ptr, ...) @printf(ptr noundef @.str, i32 noundef %conv)
+  ret void
 }
 
 ; Function Attrs: noinline nounwind uwtable
 define dso_local void @positive_test_reduce_array_allocation_1() {
 ; CHECK-LABEL: define dso_local void @positive_test_reduce_array_allocation_1() {
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[ARRAY1:%.*]] = alloca i8, i32 4, align 8
+; CHECK-NEXT:    [[ARRAY1:%.*]] = alloca [4 x i8], align 1
 ; CHECK-NEXT:    store i32 0, ptr [[ARRAY1]], align 8
 ; CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr [[ARRAY1]], align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[TMP0]], 2
@@ -275,37 +290,37 @@ entry:
 define dso_local void @positive_test_reduce_array_allocation_2() #0 {
 ; CHECK-LABEL: define dso_local void @positive_test_reduce_array_allocation_2() {
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[ARRAY:%.*]] = alloca ptr, align 8
-; CHECK-NEXT:    [[I:%.*]] = alloca i32, align 4
+; CHECK-NEXT:    [[ARRAY1:%.*]] = alloca ptr, align 8
+; CHECK-NEXT:    [[I2:%.*]] = alloca i32, align 4
 ; CHECK-NEXT:    [[CALL:%.*]] = call noalias ptr @malloc(i64 noundef 40000)
-; CHECK-NEXT:    store ptr [[CALL]], ptr [[ARRAY]], align 8
-; CHECK-NEXT:    store i32 0, ptr [[I]], align 4
+; CHECK-NEXT:    store ptr [[CALL]], ptr [[ARRAY1]], align 8
+; CHECK-NEXT:    store i32 0, ptr [[I2]], align 4
 ; CHECK-NEXT:    br label [[FOR_COND:%.*]]
 ; CHECK:       for.cond:
-; CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr [[I]], align 4
+; CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr [[I2]], align 4
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 10000
 ; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
 ; CHECK:       for.body:
-; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[I]], align 4
-; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[I]], align 4
+; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[I2]], align 4
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[I2]], align 4
 ; CHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP2]] to i64
 ; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[CALL]], i64 [[IDXPROM]]
 ; CHECK-NEXT:    store i32 [[TMP1]], ptr [[ARRAYIDX]], align 4
 ; CHECK-NEXT:    br label [[FOR_INC:%.*]]
 ; CHECK:       for.inc:
-; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[I]], align 4
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[I2]], align 4
 ; CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], 2
-; CHECK-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
+; CHECK-NEXT:    store i32 [[ADD]], ptr [[I2]], align 4
 ; CHECK-NEXT:    br label [[FOR_COND]]
 ; CHECK:       for.end:
-; CHECK-NEXT:    store i32 0, ptr [[I]], align 4
+; CHECK-NEXT:    store i32 0, ptr [[I2]], align 4
 ; CHECK-NEXT:    br label [[FOR_COND1:%.*]]
 ; CHECK:       for.cond1:
-; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I]], align 4
+; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I2]], align 4
 ; CHECK-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[TMP4]], 10000
 ; CHECK-NEXT:    br i1 [[CMP2]], label [[FOR_BODY3:%.*]], label [[FOR_END9:%.*]]
 ; CHECK:       for.body3:
-; CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[I]], align 4
+; CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[I2]], align 4
 ; CHECK-NEXT:    [[IDXPROM4:%.*]] = sext i32 [[TMP5]] to i64
 ; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds i32, ptr [[CALL]], i64 [[IDXPROM4]]
 ; CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[ARRAYIDX5]], align 4
@@ -313,28 +328,28 @@ define dso_local void @positive_test_reduce_array_allocation_2() #0 {
 ; CHECK-NEXT:    store i32 [[ADD6]], ptr [[ARRAYIDX5]], align 4
 ; CHECK-NEXT:    br label [[FOR_INC7:%.*]]
 ; CHECK:       for.inc7:
-; CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[I]], align 4
+; CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[I2]], align 4
 ; CHECK-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP7]], 2
-; CHECK-NEXT:    store i32 [[ADD8]], ptr [[I]], align 4
+; CHECK-NEXT:    store i32 [[ADD8]], ptr [[I2]], align 4
 ; CHECK-NEXT:    br label [[FOR_COND1]]
 ; CHECK:       for.end9:
-; CHECK-NEXT:    store i32 0, ptr [[I]], align 4
+; CHECK-NEXT:    store i32 0, ptr [[I2]], align 4
 ; CHECK-NEXT:    br label [[FOR_COND10:%.*]]
 ; CHECK:       for.cond10:
-; CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[I]], align 4
+; CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr [[I2]], align 4
 ; CHECK-NEXT:    [[CMP11:%.*]] = icmp slt i32 [[TMP8]], 10000
 ; CHECK-NEXT:    br i1 [[CMP11]], label [[FOR_BODY12:%.*]], label [[FOR_END18:%.*]]
 ; CHECK:       for.body12:
-; CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[I]], align 4
+; CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[I2]], align 4
 ; CHECK-NEXT:    [[IDXPROM13:%.*]] = sext i32 [[TMP9]] to i64
 ; CHECK-NEXT:    [[ARRAYIDX14:%.*]] = getelementptr inbounds i32, ptr [[CALL]], i64 [[IDXPROM13]]
 ; CHECK-NEXT:    [[TMP10:%.*]] = load i32, ptr [[ARRAYIDX14]], align 4
 ; CHECK-NEXT:    [[CALL15:%.*]] = call i32 (ptr, ...) @printf(ptr noundef nonnull dereferenceable(17) @.str, i32 noundef [[TMP10]])
 ; CHECK-NEXT:    br label [[FOR_INC16:%.*]]
 ; CHECK:       for.inc16:
-; CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[I]], align 4
+; CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[I2]], align 4
 ; CHECK-NEXT:    [[ADD17:%.*]] = add nsw i32 [[TMP11]], 2
-; CHECK-NEXT:    store i32 [[ADD17]], ptr [[I]], align 4
+; CHECK-NEXT:    store i32 [[ADD17]], ptr [[I2]], align 4
 ; CHECK-NEXT:    br label [[FOR_COND10]]
 ; CHECK:       for.end18:
 ; CHECK-NEXT:    ret void
@@ -426,7 +441,7 @@ define dso_local void @pthread_test(){
 ; TUNIT-NEXT:    [[ARG1:%.*]] = alloca i8, align 8
 ; TUNIT-NEXT:    [[THREAD:%.*]] = alloca i64, align 8
 ; TUNIT-NEXT:    [[CALL1:%.*]] = call i32 @pthread_create(ptr noundef nonnull align 8 dereferenceable(8) [[THREAD]], ptr noundef align 4294967296 null, ptr noundef nonnull @pthread_allocation_should_remain_same, ptr noundef nonnull align 8 dereferenceable(1) [[ARG1]])
-; TUNIT-NEXT:    [[F1:%.*]] = alloca i8, i32 4, align 4
+; TUNIT-NEXT:    [[F1:%.*]] = alloca [4 x i8], align 1
 ; TUNIT-NEXT:    [[CALL2:%.*]] = call i32 @pthread_create(ptr noundef nonnull align 8 dereferenceable(8) [[THREAD]], ptr noundef align 4294967296 null, ptr noundef nonnull @pthread_allocation_should_be_reduced, ptr noalias nocapture nofree nonnull readnone align 4 dereferenceable(12) undef)
 ; TUNIT-NEXT:    [[F2:%.*]] = alloca [[STRUCT_FOO:%.*]], align 4
 ; TUNIT-NEXT:    [[CALL3:%.*]] = call i32 @pthread_create(ptr noundef nonnull align 8 dereferenceable(8) [[THREAD]], ptr noundef align 4294967296 null, ptr noundef nonnull @pthread_check_captured_pointer, ptr noundef nonnull align 4 dereferenceable(12) [[F2]])
@@ -452,6 +467,46 @@ define dso_local void @pthread_test(){
   ret void
 }
 
+
+define dso_local void @select_case(i1 %cond){
+; CHECK: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(write)
+; CHECK-LABEL: define dso_local void @select_case
+; CHECK-SAME: (i1 [[COND:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:    [[A:%.*]] = alloca [100 x i8], align 1
+; CHECK-NEXT:    [[B:%.*]] = getelementptr inbounds [100 x i8], ptr [[A]], i64 0, i64 3
+; CHECK-NEXT:    [[C:%.*]] = getelementptr inbounds [100 x i8], ptr [[A]], i64 0, i64 1
+; CHECK-NEXT:    [[SEL:%.*]] = select i1 [[COND]], ptr [[B]], ptr [[C]]
+; CHECK-NEXT:    store i8 100, ptr [[SEL]], align 1
+; CHECK-NEXT:    ret void
+;
+  %a = alloca [100 x i8], align 1
+  %b = getelementptr inbounds [100 x i8], ptr %a, i64 0, i64 3
+  %c = getelementptr inbounds [100 x i8], ptr %a, i64 0, i64 1
+  %sel = select i1 %cond, ptr %b, ptr %c
+  store i8 100, ptr %sel, align 1
+  ret void
+}
+
+define dso_local void @select_case_2(i1 %cond){
+; CHECK: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(write)
+; CHECK-LABEL: define dso_local void @select_case_2
+; CHECK-SAME: (i1 [[COND:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[A:%.*]] = alloca [100 x i32], align 1
+; CHECK-NEXT:    [[B:%.*]] = getelementptr inbounds [100 x i32], ptr [[A]], i64 0, i64 3
+; CHECK-NEXT:    [[C:%.*]] = getelementptr inbounds [100 x i32], ptr [[A]], i64 0, i64 1
+; CHECK-NEXT:    [[SEL:%.*]] = select i1 [[COND]], ptr [[B]], ptr [[C]]
+; CHECK-NEXT:    store i8 100, ptr [[SEL]], align 1
+; CHECK-NEXT:    ret void
+;
+  %a = alloca [100 x i32], align 1
+  %b = getelementptr inbounds [100 x i32], ptr %a, i64 0, i64 3
+  %c = getelementptr inbounds [100 x i32], ptr %a, i64 0, i64 1
+  %sel = select i1 %cond, ptr %b, ptr %c
+  %sel2 = getelementptr inbounds i32, ptr %sel, i64 0
+  store i8 100, ptr %sel2, align 1
+  ret void
+}
+
 define internal ptr @pthread_allocation_should_remain_same(ptr %arg) {
 ; CHECK-LABEL: define internal noundef nonnull align 8 dereferenceable(1) ptr @pthread_allocation_should_remain_same
 ; CHECK-SAME: (ptr noundef nonnull returned align 8 dereferenceable(1) [[ARG:%.*]]) {
@@ -499,6 +554,58 @@ entry:
   ret void
 }
 
+define dso_local void @alloca_array_multi_offset(){
+; CHECK: Function Attrs: nofree norecurse nosync nounwind memory(none)
+; CHECK-LABEL: define dso_local void @alloca_array_multi_offset
+; CHECK-SAME: () #[[ATTR1:[0-9]+]] {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[I:%.*]] = alloca i32, align 4
+; CHECK-NEXT:    store i32 0, ptr [[I]], align 4
+; CHECK-NEXT:    br label [[FOR_COND:%.*]]
+; CHECK:       for.cond:
+; CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr [[I]], align 4
+; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP0]], 10
+; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
+; CHECK:       for.body:
+; CHECK-NEXT:    br label [[FOR_INC:%.*]]
+; CHECK:       for.inc:
+; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[I]], align 4
+; CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP1]], 2
+; CHECK-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
+; CHECK-NEXT:    br label [[FOR_COND]]
+; CHECK:       for.end:
+; CHECK-NEXT:    ret void
+;
+entry:
+  %arr = alloca i8, i32 10, align 4
+  %i = alloca i32, align 4
+  store i32 0, ptr %i, align 4
+  br label %for.cond
+
+for.cond:
+  %0 = load i32, ptr %i, align 4
+  %cmp = icmp slt i32 %0, 10
+  br i1 %cmp, label %for.body, label %for.end
+
+for.body:
+  %1 = load i32, ptr %i, align 4
+  %2 = load ptr, ptr %arr, align 8
+  %3 = load i32, ptr %i, align 4
+  %arrayidx = getelementptr inbounds i32, ptr %2, i32 %3
+  store i32 %1, ptr %arrayidx, align 4
+  br label %for.inc
+
+for.inc:
+  %4 = load i32, ptr %i, align 4
+  %add = add nsw i32 %4, 2
+  store i32 %add, ptr %i, align 4
+  br label %for.cond
+
+for.end:
+  ret void
+
+}
+
 
 declare external void @external_call(ptr)
 
@@ -511,9 +618,11 @@ declare i32 @printf(ptr noundef, ...) #1
 ; Function Attrs: nounwind allocsize(0)
 declare noalias ptr @malloc(i64 noundef) #1
 ;.
-; TUNIT: attributes #[[ATTR0]] = { mustprogress nofree norecurse nosync nounwind willreturn memory(none) }
+; TUNIT: attributes #[[ATTR0]] = { mustprogress nofree norecurse nosync nounwind willreturn memory(write) }
+; TUNIT: attributes #[[ATTR1]] = { nofree norecurse nosync nounwind memory(none) }
 ;.
-; CGSCC: attributes #[[ATTR0]] = { mustprogress nofree norecurse nosync nounwind willreturn memory(none) }
+; CGSCC: attributes #[[ATTR0]] = { mustprogress nofree norecurse nosync nounwind willreturn memory(write) }
+; CGSCC: attributes #[[ATTR1]] = { nofree norecurse nosync nounwind memory(none) }
 ;.
 ; TUNIT: [[META0:![0-9]+]] = !{[[META1:![0-9]+]]}
 ; TUNIT: [[META1]] = !{i64 2, i64 3, i1 false}
diff --git a/llvm/test/Transforms/Attributor/call-simplify-pointer-info.ll b/llvm/test/Transforms/Attributor/call-simplify-pointer-info.ll
index 5bb795911ce40..ebbfd44017aa5 100644
--- a/llvm/test/Transforms/Attributor/call-simplify-pointer-info.ll
+++ b/llvm/test/Transforms/Attributor/call-simplify-pointer-info.ll
@@ -36,8 +36,8 @@ define i8 @call_simplifiable_1() {
 ; TUNIT-LABEL: define {{[^@]+}}@call_simplifiable_1
 ; TUNIT-SAME: () #[[ATTR0:[0-9]+]] {
 ; TUNIT-NEXT:  entry:
-; TUNIT-NEXT:    [[BYTES:%.*]] = alloca [1024 x i8], align 16
-; TUNIT-NEXT:    [[I0:%.*]] = getelementptr inbounds [1024 x i8], ptr [[BYTES]], i64 0, i64 2
+; TUNIT-NEXT:    [[BYTES1:%.*]] = alloca [1 x i8], align 1
+; TUNIT-NEXT:    [[NEWGEP:%.*]] = getelementptr [1 x i8], ptr [[BYTES1]], i64 0
 ; TUNIT-NEXT:    ret i8 2
 ;
 ; CGSCC: Function Attrs: mustprogress nofree nosync nounwind willreturn memory(none)
@@ -93,9 +93,9 @@ define i8 @call_simplifiable_2() {
 ; TUNIT-LABEL: define {{[^@]+}}@call_simplifiable_2
 ; TUNIT-SAME: () #[[ATTR0]] {
 ; TUNIT-NEXT:  entry:
-; TUNIT-NEXT:    [[BYTES:%.*]] = alloca [1024 x i8], align 16
-; TUNIT-NEXT:    [[I0:%.*]] = getelementptr inbounds [1024 x i8], ptr [[BYTES]], i64 0, i64 2
-; TUNIT-NEXT:    [[I1:%.*]] = getelementptr inbounds [1024 x i8], ptr [[BYTES]], i64 0, i64 3
+; TUNIT-NEXT:    [[BYTES1:%.*]] = alloca [2 x i8], align 1
+; TUNIT-NEXT:    [[NEWGEP:%.*]] = getelementptr [2 x i8], ptr [[BYTES1]], i64 0
+; TUNIT-NEXT:    [[NEWGEP2:%.*]] = getelementptr [2 x i8], ptr [[BYTES1]], i64 1
 ; TUNIT-NEXT:    ret i8 4
 ;
 ; CGSCC: Function Attrs: mustprogress nofree nosync nounwind willreturn memory(none)
@@ -125,8 +125,8 @@ define i8 @call_simplifiable_3() {
 ; TUNIT-LABEL: define {{[^@]+}}@call_simplifiable_3
 ; TUNIT-SAME: () #[[ATTR0]] {
 ; TUNIT-NEXT:  entry:
-; TUNIT-NEXT:    [[BYTES:%.*]] = alloca [1024 x i8], align 16
-; TUNIT-NEXT:    [[I2:%.*]] = getelementptr inbounds [1024 x i8], ptr [[BYTES]], i64 0, i64 2
+; TUNIT-NEXT:    [[BYTES1:%.*]] = alloca [1 x i8], align 1
+; TUNIT-NEXT:    [[NEWGEP:%.*]] = getelementptr [1 x i8], ptr [[BYTES1]], i64 0
 ; TUNIT-NEXT:    ret i8 2
 ;
 ; CGSCC: Function Attrs: mustprogress nofree nosync nounwind willreturn memory(none)
@@ -198,13 +198,13 @@ define i8 @call_partially_simplifiable_1() {
 ; TUNIT-LABEL: define {{[^@]+}}@call_partially_simplifiable_1
 ; TUNIT-SAME: () #[[ATTR0]] {
 ; TUNIT-NEXT:  entry:
-; TUNIT-NEXT:    [[BYTES:%.*]] = alloca [1024 x i8], align 16
-; TUNIT-NEXT:    [[I2:%.*]] = getelementptr inbounds [1024 x i8], ptr [[BYTES]], i64 0, i64 2
-; TUNIT-NEXT:    store i8 2, ptr [[I2]], align 2
-; TUNIT-NEXT:    [[I3:%.*]] = getelementptr inbounds [1024 x i8], ptr [[BYTES]], i64 0, i64 3
-; TUNIT-NEXT:    store i8 3, ptr [[I3]], align 1
-; TUNIT-NEXT:    [[I4:%.*]] = getelementptr inbounds [1024 x i8], ptr [[BYTES]], i64 0, i64 4
-; TUNIT-NEXT:    [[R:%.*]] = call i8 @sum_two_different_loads(ptr nocapture nofree noundef nonnull readonly align 2 dereferenceable(1022) [[I2]], ptr nocapture nofree noundef nonnull readonly dereferenceable(1021) [[I3]]) #[[ATTR3]]
+; TUNIT-NEXT:    [[BYTES1:%.*]] = alloca [3 x i8], align 1
+; TUNIT-NEXT:    [[NEWGEP:%.*]] = getelementptr [3 x i8], ptr [[BYTES1]], i64 0
+; TUNIT-NEXT:    store i8 2, ptr [[NEWGEP]], align 2
+; TUNIT-NEXT:    [[NEWGEP3:%.*]] = getelementptr [3 x i8], ptr [[BYTES1]], i64 1
+; TUNIT-NEXT:    store i8 3, ptr [[NEWGEP3]], align 1
+; TUNIT-NEXT:    [[NEWGEP2:%.*]] = getelementptr [3 x i8], ptr [[BYTES1]], i64 2
+; TUNIT-NEXT:    [[R:%.*]] = call i8 @sum_two_different_loads(ptr nocapture nofree noundef nonnull readonly align 2 dereferenceable(1022) [[NEWGEP]], ptr nocapture nofree noundef nonnull readonly dereferenceable(1021) [[NEWGEP3]]) #[[ATTR3]]
 ; TUNIT-NEXT:    ret i8 [[R]]
 ;
 ; CGSCC: Function Attrs: mustprogress nofree nosync nounwind willreturn memory(none)
diff --git a/llvm/test/Transforms/Attributor/heap_to_stack.ll b/llvm/test/Transforms/Attributor/heap_to_stack.ll
index 33ac066e43d09..846373e05be1a 100644
--- a/llvm/test/Transforms/Attributor/heap_to_stack.ll
+++ b/llvm/test/Transforms/Attributor/heap_to_stack.ll
@@ -502,8 +502,7 @@ define i32 @malloc_in_loop(i32 %arg) {
 ; CHECK-SAME: (i32 [[ARG:%.*]]) {
 ; CHECK-NEXT:  bb:
 ; CHECK-NEXT:    [[I:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[I1:%.*]] = alloca ptr, align 8
-; CHECK-NEXT:    [[I11:%.*]] = alloca i8, i32 0, align 8
+; CHECK-NEXT:    [[I11:%.*]] = alloca [0 x i8], align 1
 ; CHECK-NEXT:    store i32 [[ARG]], ptr [[I]], align 4
 ; CHECK-NEXT:    br label [[BB2:%.*]]
 ; CHECK:       bb2:
diff --git a/llvm/test/Transforms/Attributor/heap_to_stack_gpu.ll b/llvm/test/Transforms/Attributor/heap_to_stack_gpu.ll
index 2a5b3e94291a2..70aace8100abd 100644
--- a/llvm/test/Transforms/Attributor/heap_to_stack_gpu.ll
+++ b/llvm/test/Transforms/Attributor/heap_to_stack_gpu.ll
@@ -452,8 +452,7 @@ define i32 @malloc_in_loop(i32 %arg) {
 ; CHECK-SAME: (i32 [[ARG:%.*]]) {
 ; CHECK-NEXT:  bb:
 ; CHECK-NEXT:    [[I:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    [[I1:%.*]] = alloca ptr, align 8
-; CHECK-NEXT:    [[I11:%.*]] = alloca i8, i32 0, align 8
+; CHECK-NEXT:    [[I11:%.*]] = alloca [0 x i8], align 1
 ; CHECK-NEXT:    store i32 [[ARG]], ptr [[I]], align 4
 ; CHECK-NEXT:    br label [[BB2:%.*]]
 ; CHECK:       bb2:
diff --git a/llvm/test/Transforms/Attributor/liveness.ll b/llvm/test/Transforms/Attributor/liveness.ll
index f17bd5795a174..9eb79f8a46723 100644
--- a/llvm/test/Transforms/Attributor/liveness.ll
+++ b/llvm/test/Transforms/Attributor/liveness.ll
@@ -2587,8 +2587,8 @@ define void @bad_gep() {
 ; TUNIT-LABEL: define {{[^@]+}}@bad_gep
 ; TUNIT-SAME: () #[[ATTR13]] {
 ; TUNIT-NEXT:  entry:
-; TUNIT-NEXT:    [[N1:%.*]] = alloca i8, i32 0, align 1
-; TUNIT-NEXT:    [[M2:%.*]] = alloca i8, i32 0, align 1
+; TUNIT-NEXT:    [[N1:%.*]] = alloca [0 x i8], align 1
+; TUNIT-NEXT:    [[M2:%.*]] = alloca [0 x i8], align 1
 ; TUNIT-NEXT:    call void @llvm.lifetime.start.p0(i64 noundef 1, ptr noalias nocapture nofree noundef nonnull dereferenceable(1) [[N1]]) #[[ATTR18:[0-9]+]]
 ; TUNIT-NEXT:    br label [[EXIT:%.*]]
 ; TUNIT:       while.body:
@@ -2605,8 +2605,8 @@ define void @bad_gep() {
 ; CGSCC-LABEL: define {{[^@]+}}@bad_gep
 ; CGSCC-SAME: () #[[ATTR6]] {
 ; CGSCC-NEXT:  entry:
-; CGSCC-NEXT:    [[N1:%.*]] = alloca i8, i32 0, align 1
-; CGSCC-NEXT:    [[M2:%.*]] = alloca i8, i32 0, align 1
+; CGSCC-NEXT:    [[N1:%.*]] = alloca [0 x i8], align 1
+; CGSCC-NEXT:    [[M2:%.*]] = alloca [0 x i8], align 1
 ; CGSCC-NEXT:    call void @llvm.lifetime.start.p0(i64 noundef 1, ptr noalias nocapture nofree noundef nonnull dereferenceable(1) [[N1]]) #[[ATTR21:[0-9]+]]
 ; CGSCC-NEXT:    br label [[EXIT:%.*]]
 ; CGSCC:       while.body:
diff --git a/llvm/test/Transforms/Attributor/nodelete.ll b/llvm/test/Transforms/Attributor/nodelete.ll
index c28cb28379348..6357bf742bbf1 100644
--- a/llvm/test/Transforms/Attributor/nodelete.ll
+++ b/llvm/test/Transforms/Attributor/nodelete.ll
@@ -10,15 +10,14 @@ define hidden i64 @f1() align 2 {
 ; TUNIT-LABEL: define {{[^@]+}}@f1
 ; TUNIT-SAME: () #[[ATTR0:[0-9]+]] align 2 {
 ; TUNIT-NEXT:  entry:
-; TUNIT-NEXT:    [[REF_TMP1:%.*]] = alloca i8, i32 0, align 8
+; TUNIT-NEXT:    [[REF_TMP1:%.*]] = alloca [0 x i8], align 1
 ; TUNIT-NEXT:    ret i64 undef
 ;
 ; CGSCC: Function Attrs: mustprogress nofree nosync nounwind willreturn memory(none)
 ; CGSCC-LABEL: define {{[^@]+}}@f1
 ; CGSCC-SAME: () #[[ATTR0:[0-9]+]] align 2 {
 ; CGSCC-NEXT:  entry:
-; CGSCC-NEXT:    [[REF_TMP:%.*]] = alloca [[A:%.*]], align 8
-; CGSCC-NEXT:    [[REF_TMP1:%.*]] = alloca i8, i32 0, align 8
+; CGSCC-NEXT:    [[REF_TMP1:%.*]] = alloca [0 x i8], align 1
 ; CGSCC-NEXT:    [[CALL2:%.*]] = call i64 @f2() #[[ATTR2:[0-9]+]]
 ; CGSCC-NEXT:    ret i64 [[CALL2]]
 ;
diff --git a/llvm/test/Transforms/Attributor/pointer-info.ll b/llvm/test/Transforms/Attributor/pointer-info.ll
index 6afdbdaee317c..66dc0160a4e99 100644
--- a/llvm/test/Transforms/Attributor/pointer-info.ll
+++ b/llvm/test/Transforms/Attributor/pointer-info.ll
@@ -10,11 +10,11 @@ define void @foo(ptr %ptr) {
 ; TUNIT-LABEL: define {{[^@]+}}@foo
 ; TUNIT-SAME: (ptr nocapture nofree readnone [[PTR:%.*]]) #[[ATTR0:[0-9]+]] {
 ; TUNIT-NEXT:  entry:
-; TUNIT-NEXT:    [[TMP0:%.*]] = alloca [[STRUCT_TEST_A:%.*]], align 8
+; TUNIT-NEXT:    [[TMP0:%.*]] = alloca [8 x i8], align 1
 ; TUNIT-NEXT:    br label [[CALL_BR:%.*]]
 ; TUNIT:       call.br:
-; TUNIT-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_TEST_A]], ptr [[TMP0]], i64 0, i32 2
-; TUNIT-NEXT:    tail call void @bar(ptr noalias nocapture nofree noundef nonnull readonly byval([[STRUCT_TEST_A]]) align 8 dereferenceable(24) [[TMP0]]) #[[ATTR2:[0-9]+]]
+; TUNIT-NEXT:    [[NEWGEP:%.*]] = getelementptr [8 x i8], ptr [[TMP0]], i64 0
+; TUNIT-NEXT:    tail call void @bar(ptr noalias nocapture nofree noundef nonnull readonly byval([[STRUCT_TEST_A:%.*]]) align 8 dereferenceable(24) [[TMP0]]) #[[ATTR2:[0-9]+]]
 ; TUNIT-NEXT:    ret void
 ;
 ; CGSCC: Function Attrs: mustprogress nofree nosync nounwind willreturn memory(none)
diff --git a/llvm/test/Transforms/Attributor/value-simplify-pointer-info.ll b/llvm/test/Transforms/Attributor/value-simplify-pointer-info.ll
index 7a35b5c856097..07b25f6232436 100644
--- a/llvm/test/Transforms/Attributor/value-simplify-pointer-info.ll
+++ b/llvm/test/Transforms/Attributor/value-simplify-pointer-info.ll
@@ -2666,18 +2666,18 @@ define dso_local void @test_nested_memory(ptr %dst, ptr %src) {
 ; TUNIT-SAME: (ptr nocapture nofree writeonly [[DST:%.*]], ptr nocapture nofree readonly [[SRC:%.*]]) {
 ; TUNIT-NEXT:  entry:
 ; TUNIT-NEXT:    [[CALL_H2S:%.*]] = alloca i8, i64 24, align 1
-; TUNIT-NEXT:    [[LOCAL:%.*]] = alloca [[STRUCT_STY:%.*]], align 8
-; TUNIT-NEXT:    [[INNER:%.*]] = getelementptr inbounds [[STRUCT_STY]], ptr [[LOCAL]], i64 0, i32 2
-; TUNIT-NEXT:    store ptr @global, ptr [[INNER]], align 8
+; TUNIT-NEXT:    [[LOCAL1:%.*]] = alloca [8 x i8], align 1
+; TUNIT-NEXT:    [[NEWGEP:%.*]] = getelementptr [8 x i8], ptr [[LOCAL1]], i64 0
+; TUNIT-NEXT:    store ptr @global, ptr [[NEWGEP]], align 8
 ; TUNIT-NEXT:    store ptr [[DST]], ptr [[CALL_H2S]], align 8
 ; TUNIT-NEXT:    [[SRC2:%.*]] = getelementptr inbounds i8, ptr [[CALL_H2S]], i64 8
 ; TUNIT-NEXT:    store ptr [[SRC]], ptr [[SRC2]], align 8
-; TUNIT-NEXT:    store ptr [[CALL_H2S]], ptr getelementptr inbounds ([[STRUCT_STY]], ptr @global, i64 0, i32 2), align 8
-; TUNIT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[LOCAL]], align 8
-; TUNIT-NEXT:    [[LOCAL_B8:%.*]] = getelementptr i8, ptr [[LOCAL]], i64 8
-; TUNIT-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[LOCAL_B8]], align 8
-; TUNIT-NEXT:    [[LOCAL_B16:%.*]] = getelementptr i8, ptr [[LOCAL]], i64 16
-; TUNIT-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[LOCAL_B16]], align 8
+; TUNIT-NEXT:    store ptr [[CALL_H2S]], ptr getelementptr inbounds ([[STRUCT_STY:%.*]], ptr @global, i64 0, i32 2), align 8
+; TUNIT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[LOCAL1]], align 8
+; TUNIT-NEXT:    [[LOCAL1_B8:%.*]] = getelementptr i8, ptr [[LOCAL1]], i64 8
+; TUNIT-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[LOCAL1_B8]], align 8
+; TUNIT-NEXT:    [[LOCAL1_B16:%.*]] = getelementptr i8, ptr [[LOCAL1]], i64 16
+; TUNIT-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[LOCAL1_B16]], align 8
 ; TUNIT-NEXT:    call fastcc void @nested_memory_callee(ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]]) #[[ATTR21:[0-9]+]]
 ; TUNIT-NEXT:    ret void
 ;



More information about the llvm-commits mailing list