[NewGVN][LoadCoercion][2/3] Add support for load coercion between load instructions.

[kmitropoulou]

Summary

This patch checks all the live-in loads and tries to apply load coercion.

[kmitropoulou]

Updating D124230: [2/3][NewGVN][LoadCoercion] Add support for load coercion between load instructions.

[kmitropoulou]

ping

[kmitropoulou]

[2/3][NewGVN][LoadCoercion] Add support for load coercion between load instructions.

[kmitropoulou]

Updating D124230: [NewGVN][LoadCoercion][2/3] Add support for load coercion between load instructions.

[kmitropoulou]

ping

[kmitropoulou]

Updating D124230: [NewGVN][LoadCoercion][2/3] Add support for load coercion between load instructions.

[kmitropoulou]

Updating D124230: [NewGVN][LoadCoercion][2/3] Add support for load coercion between load instructions.

[kmitropoulou]

Updating D124230: [NewGVN][LoadCoercion][2/3] Add support for load coercion between load instructions.

llvm/lib/Transforms/Scalar/NewGVN.cpp

#endif
   SmallVector<Value *, 32> DFSToInstr;
 
   // Deletion info.
   SmallPtrSet<Instruction *, 8> InstructionsToErase;
 
   // Maps the loads to their depending instructions.
   std::map<Value *, SmallPtrSet<Instruction *, 2>> LoadCoercion;
 
-  // During load coercion, new loads might be generated. We do not want to apply
-  // load coercion for them for now.
+  // During load coercion, new loads might be generated. For now, we do not want
+  // to apply load coercion for them.
   SmallVector<Instruction *, 2> NewLoadsInLoadCoercion;
 
 public:
   NewGVN(Function &F, DominatorTree *DT, AssumptionCache *AC,
          TargetLibraryInfo *TLI, AliasAnalysis *AA, MemorySSA *MSSA,
          const DataLayout &DL)
       : F(F), DT(DT), TLI(TLI), AA(AA), MSSA(MSSA), AC(AC), DL(DL),
         PredInfo(std::make_unique<PredicateInfo>(F, *DT, *AC)),
const Expression *NewGVN::performSymbolicStoreEvaluation(Instruction *I) const {
   }
 
   // If the store is not equivalent to anything, value number it as a store that
   // produces a unique memory state (instead of using it's MemoryUse, we use
   // it's MemoryDef).
   return createStoreExpression(SI, StoreAccess);
 }
 
+// A load can have one or more dependencies as the following examples show:
+//
+// Example 1:
+//  BB1:
+//   ...
+//   store i32 %V1, i32* %P
+//   ...
+//   %V2 = load i32, i32* %P
+//   ...
+//
+// Example 2:
+//  BB1:                       BB2:
+//   store i32 %V1, i32* %P     %V2 = load i32, i32* %P
+//   br label %BB3              br label %BB3
+//                      \      /
+//                     BB3:
+//                      %V3 = load i32, i32* %P
+//
+// In the first example, the load (%V2) has only one dependency. In the second
+// example, the load (%V3) has two dependencies. Therefore, we add the load
+// along with its two dependencies in LoadCoercion map. However, this is not
+// always the case as it is shown below:
+//
+// Example 3:
+//                   BB1:
+//                    %P1 = bitcast i32* %P to <4 x i32>*
+//                    %V1 = load <4 x i32>, <4 x i32>* %P1
+//                    br i1 %cond, label %BB2, label %BB3
+//                   /                          \
+//   BB2:                                      BB3:
+//    %P2 = bitcast i32* %P to <2 x i32>*       %V3 = load i32, i32* %P
+//    %V2 = load <2 x i32>, <2 x i32>* %P2      br label %BB4
+//    br label %BB4                             /
+//         \                         /
+//                  BB4:
+//                   %V4 = load i32, i32* %P
+//
+// In the above example, the load (%V4) can be optimized out by any of the loads
+// (%V1, %V2, %V3). But, loads %V2 and %V3 can also be optimized out by %V1. For
+// this reason, we need to do an extra check before we add the load in the map.
+// We check if the load is already in the map and if the existing depending
+// instruction dominates the current depending instruction. If so, then we do
+// not add the new depending instruction in LoadCoercion map. If the current
+// depending instruction dominates the existing depending instruction, then we
+// remove the existing depending instruction from LoadCoercion map and we add
+// the current depending instruction. Therefore, in Example 3, the load
+// (%V4) has only one dependency (%V1) and we add only this one in LoadCoercion
+// map.
 void NewGVN::tryAddLoadDepInsnIntoLoadCoercionMap(
     LoadInst *LI, Instruction *CurrentDepI) const {
+  // Check if LI already exit in LoadCoercion map.
+  auto It = const_cast<NewGVN *>(this)->LoadCoercion.find(LI);
+  if (It != LoadCoercion.end()) {
+    auto &ExistingDepIs = It->second;
+    // Iterate over all the existing depending instructions of LI.
+    for (Instruction *ExistingDepI :
+         llvm::make_early_inc_range(ExistingDepIs)) {
+
+      if (MSSAWalker->getClobberingMemoryAccess(getMemoryAccess(CurrentDepI)) ==
+              MSSAWalker->getClobberingMemoryAccess(
+                  getMemoryAccess(ExistingDepI)) &&
+          isa<LoadInst>(ExistingDepI) && isa<LoadInst>(CurrentDepI)) {
+        // If the existing depending instruction dominates the current depending
+        // instruction, then we should not add the current depending instruction
+        // in LoadCoercion map (Example 3).
+        if (DT->dominates(ExistingDepI, CurrentDepI))
+          return;
+
+        // If the current depending instruction dominates the existing one, then
+        // we remove the existing depending instruction from the LoadCoercion
+        // map. Next, we add the current depending instruction in LoadCoercion
+        // map.
+        if (DT->dominates(CurrentDepI, ExistingDepI))
+          ExistingDepIs.erase(ExistingDepI);
+      }
+    }
+  }
   // Add the load and the corresponding depending instruction in LoadCoercion
   // map.
   const_cast<NewGVN *>(this)->LoadCoercion[LI].insert(CurrentDepI);
 }
 
 // See if we can extract the value of a loaded pointer from a load, a store, or
 // a memory instruction. Load coercion has two phases. In the first phase (the
 // code below), we collect the load intructions and we add them to the
if (auto *DepSI = dyn_cast<StoreInst>(DepInst)) {
     }
   } else if (auto *DepLI = dyn_cast<LoadInst>(DepInst)) {
     // Can't forward from non-atomic to atomic without violating memory model.
     if (LI->isAtomic() > DepLI->isAtomic())
       return nullptr;
     int Offset = analyzeLoadFromClobberingLoad(LoadType, LoadPtr, DepLI, DL);
     if (Offset >= 0) {
       // We can coerce a constant load into a load.
-      if (auto *C = dyn_cast<Constant>(lookupOperandLeader(DepLI)))
+      if (auto *C = dyn_cast<Constant>(lookupOperandLeader(DepLI))) {
         if (auto *PossibleConstant =
                 getConstantLoadValueForLoad(C, Offset, LoadType, DL)) {
           LLVM_DEBUG(dbgs() << "Coercing load from load " << *LI
                             << " to constant " << *PossibleConstant << "\n");
           return createConstantExpression(PossibleConstant);
         }
+      } else if (EnableLoadCoercion) {
+        // Similarly, we do not create a load expression for the loads that are
+        // elimianted with load coercion.
+        tryAddLoadDepInsnIntoLoadCoercionMap(LI, DepInst);
+        return nullptr;
+      }
     }
   } else if (auto *DepMI = dyn_cast<MemIntrinsic>(DepInst)) {
     int Offset = analyzeLoadFromClobberingMemInst(LoadType, LoadPtr, DepMI, DL);
     if (Offset >= 0) {
       if (auto *PossibleConstant =
               getConstantMemInstValueForLoad(DepMI, Offset, LoadType, DL)) {
         LLVM_DEBUG(dbgs() << "Coercing load from meminst " << *DepMI
                           << " to constant " << *PossibleConstant << "\n");
if (auto *MD = dyn_cast<MemoryDef>(DefiningAccess)) {
       // defining access has a different type, or it is a pointer produced by
       // certain memory operations that cause the memory to have a fixed value
       // (IE things like calloc).
       if (const auto *CoercionResult =
               performSymbolicLoadCoercion(LI->getType(), LoadAddressLeader, LI,
                                           DefiningInst, DefiningAccess))
         return CoercionResult;
     }
+  } else if (EnableLoadCoercion) {
+    // Check if any of the live-in loads can be eliminated with load coercion.
+    for (const auto &U : DefiningAccess->uses())
+      if (auto *MemUse = dyn_cast<MemoryUse>(U.getUser())) {
+        LoadInst *DependingLoad = dyn_cast<LoadInst>(MemUse->getMemoryInst());
+
+        if (!DependingLoad || LI == DependingLoad)
+          continue;
+
+        // The DependingLoad should have bigger bit size than the load that we
+        // should optimize.
+        if (DL.getTypeSizeInBits(DependingLoad->getType()).getFixedSize() <
+            DL.getTypeSizeInBits(LI->getType()).getFixedSize())
+          continue;
+
+        // If the depending load does not have any uses, then we should not do
+        // load coercion because the depending load will be eliminated.
+        if (DependingLoad->getNumUses() == 0)
+          continue;
+
+        // If two load instructions have the same operands, then it is not
+        // load coercion.
+        bool DependingLoadDomintatesLI = DT->dominates(DependingLoad, LI);
+        if (DependingLoad->getPointerOperand() == LI->getPointerOperand() &&
+            DependingLoadDomintatesLI)
+          continue;
+
+        // The two loads should be executed in the right order.
+        if (DependingLoadDomintatesLI)
+          performSymbolicLoadCoercion(LI->getType(), LI->getPointerOperand(),
+                                      LI, DependingLoad,
+                                      getMemoryAccess(DependingLoad));
+      }
   }
 
-  // If we cannot apply load coercion, then we create a load expression.
+  // If the load is a candidate for load coercion, then we do not create a load
+  // expression.
+  if (EnableLoadCoercion && LoadCoercion.count(LI))
+    return nullptr;
+  // Otherwise, we create a load expression.
   return createLoadExpAndUpdateMemUses(LI, LoadAddressLeader, OriginalAccess,
                                        DefiningAccess);
 }
 
 NewGVN::ExprResult
 NewGVN::performSymbolicPredicateInfoEvaluation(IntrinsicInst *I) const {
   auto *PI = PredInfo->getPredicateInfoFor(I);
   if (!PI)

llvm/test/Transforms/NewGVN/load_coercion_between_loads.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -S -gvn < %s | FileCheck %s -check-prefixes=GVN,OLDGVN
 ; RUN: opt -S -newgvn < %s | FileCheck %s -check-prefixes=GVN,NEWGVN
 
 define i8 @test1(i32* %P1) {
 ; OLDGVN-LABEL: @test1(
 ; OLDGVN-NEXT:    [[V1:%.*]] = load i32, i32* [[P1:%.*]], align 4
 ; OLDGVN-NEXT:    [[P2:%.*]] = bitcast i32* [[P1]] to i8*
 ; OLDGVN-NEXT:    [[TMP1:%.*]] = trunc i32 [[V1]] to i8
 ; OLDGVN-NEXT:    [[V4:%.*]] = add i8 [[TMP1]], [[TMP1]]
 ; OLDGVN-NEXT:    ret i8 [[V4]]
 ;
 ; NEWGVN-LABEL: @test1(
 ; NEWGVN-NEXT:    [[V1:%.*]] = load i32, i32* [[P1:%.*]], align 4
+; NEWGVN-NEXT:    [[TMP1:%.*]] = trunc i32 [[V1]] to i8
 ; NEWGVN-NEXT:    [[P2:%.*]] = bitcast i32* [[P1]] to i8*
-; NEWGVN-NEXT:    [[V2:%.*]] = load i8, i8* [[P2]], align 1
-; NEWGVN-NEXT:    [[V3:%.*]] = trunc i32 [[V1]] to i8
-; NEWGVN-NEXT:    [[V4:%.*]] = add i8 [[V2]], [[V3]]
+; NEWGVN-NEXT:    [[V4:%.*]] = add i8 [[TMP1]], [[TMP1]]
 ; NEWGVN-NEXT:    ret i8 [[V4]]
 ;
   %V1 = load i32, i32* %P1
   %P2 = bitcast i32* %P1 to i8*
   %V2 = load i8, i8* %P2
   %V3 = trunc i32 %V1 to i8
   %V4 = add i8 %V2, %V3
   ret i8 %V4

 ; OLDGVN-NEXT:    [[TMP1:%.*]] = trunc i64 [[TMP0]] to i32
 ; OLDGVN-NEXT:    [[V5:%.*]] = add i32 [[TMP1]], [[TMP1]]
 ; OLDGVN-NEXT:    ret i32 [[V5]]
 ;
 ; NEWGVN-LABEL: @test2(
 ; NEWGVN-NEXT:  Entry:
 ; NEWGVN-NEXT:    [[P1:%.*]] = bitcast i8* [[P:%.*]] to <2 x i32>*
 ; NEWGVN-NEXT:    [[V1:%.*]] = load <2 x i32>, <2 x i32>* [[P1]], align 8
+; NEWGVN-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V1]] to i64
+; NEWGVN-NEXT:    [[TMP1:%.*]] = trunc i64 [[TMP0]] to i32
 ; NEWGVN-NEXT:    [[P2:%.*]] = bitcast i8* [[P]] to i32*
-; NEWGVN-NEXT:    [[V2:%.*]] = load i32, i32* [[P2]], align 4
-; NEWGVN-NEXT:    [[V3:%.*]] = bitcast <2 x i32> [[V1]] to i64
-; NEWGVN-NEXT:    [[V4:%.*]] = trunc i64 [[V3]] to i32
-; NEWGVN-NEXT:    [[V5:%.*]] = add i32 [[V2]], [[V4]]
+; NEWGVN-NEXT:    [[V5:%.*]] = add i32 [[TMP1]], [[TMP1]]
 ; NEWGVN-NEXT:    ret i32 [[V5]]
 ;
 Entry:
   %P1 = bitcast i8* %P to <2 x i32>*
   %V1 = load <2 x i32>, <2 x i32>* %P1
   %P2 = bitcast i8* %P to i32*
   %V2 = load i32, i32* %P2
   %V3 = bitcast <2 x i32> %V1 to i64

 ; OLDGVN-NEXT:    [[I1:%.*]] = insertvalue <{ <2 x i32>, i32 }> poison, <2 x i32> [[V1]], 0
 ; OLDGVN-NEXT:    [[I2:%.*]] = insertvalue <{ <2 x i32>, i32 }> [[I1]], i32 [[TMP1]], 1
 ; OLDGVN-NEXT:    ret <{ <2 x i32>, i32 }> [[I2]]
 ;
 ; NEWGVN-LABEL: @test3(
 ; NEWGVN-NEXT:  Entry:
 ; NEWGVN-NEXT:    [[P1:%.*]] = bitcast i8* [[P:%.*]] to <2 x i32>*
 ; NEWGVN-NEXT:    [[V1:%.*]] = load <2 x i32>, <2 x i32>* [[P1]], align 8
+; NEWGVN-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V1]] to i64
+; NEWGVN-NEXT:    [[TMP1:%.*]] = trunc i64 [[TMP0]] to i32
 ; NEWGVN-NEXT:    [[P2:%.*]] = bitcast i8* [[P]] to i32*
-; NEWGVN-NEXT:    [[V2:%.*]] = load i32, i32* [[P2]], align 4
 ; NEWGVN-NEXT:    [[I1:%.*]] = insertvalue <{ <2 x i32>, i32 }> poison, <2 x i32> [[V1]], 0
-; NEWGVN-NEXT:    [[I2:%.*]] = insertvalue <{ <2 x i32>, i32 }> [[I1]], i32 [[V2]], 1
+; NEWGVN-NEXT:    [[I2:%.*]] = insertvalue <{ <2 x i32>, i32 }> [[I1]], i32 [[TMP1]], 1
 ; NEWGVN-NEXT:    ret <{ <2 x i32>, i32 }> [[I2]]
 ;
 Entry:
   %P1 = bitcast i8* %P to <2 x i32>*
   %V1 = load <2 x i32>, <2 x i32>* %P1
   %P2 = bitcast i8* %P to i32*
   %V2 = load i32, i32* %P2
   %I1 = insertvalue <{<2 x i32>, i32}> poison, <2 x i32> %V1, 0

 ; OLDGVN-NEXT:    [[I3:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I2]], i32 [[TMP3]], 2
 ; OLDGVN-NEXT:    [[I4:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I3]], <2 x i32> [[TMP2]], 3
 ; OLDGVN-NEXT:    ret <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I4]]
 ;
 ; NEWGVN-LABEL: @test6(
 ; NEWGVN-NEXT:  Entry:
 ; NEWGVN-NEXT:    [[P1:%.*]] = bitcast i8* [[P:%.*]] to <4 x i32>*
 ; NEWGVN-NEXT:    [[V1:%.*]] = load <4 x i32>, <4 x i32>* [[P1]], align 16
+; NEWGVN-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V1]] to i128
+; NEWGVN-NEXT:    [[TMP1:%.*]] = trunc i128 [[TMP0]] to i64
+; NEWGVN-NEXT:    [[TMP2:%.*]] = bitcast i64 [[TMP1]] to <2 x i32>
+; NEWGVN-NEXT:    [[TMP3:%.*]] = trunc i128 [[TMP0]] to i32
 ; NEWGVN-NEXT:    [[P2:%.*]] = bitcast i8* [[P]] to <2 x i32>*
-; NEWGVN-NEXT:    [[V2:%.*]] = load <2 x i32>, <2 x i32>* [[P2]], align 8
 ; NEWGVN-NEXT:    [[P3:%.*]] = bitcast i8* [[P]] to i32*
-; NEWGVN-NEXT:    [[V3:%.*]] = load i32, i32* [[P3]], align 4
 ; NEWGVN-NEXT:    [[I1:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> poison, <4 x i32> [[V1]], 0
-; NEWGVN-NEXT:    [[I2:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I1]], <2 x i32> [[V2]], 1
-; NEWGVN-NEXT:    [[I3:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I2]], i32 [[V3]], 2
-; NEWGVN-NEXT:    [[I4:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I3]], <2 x i32> [[V2]], 3
+; NEWGVN-NEXT:    [[I2:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I1]], <2 x i32> [[TMP2]], 1
+; NEWGVN-NEXT:    [[I3:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I2]], i32 [[TMP3]], 2
+; NEWGVN-NEXT:    [[I4:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I3]], <2 x i32> [[TMP2]], 3
 ; NEWGVN-NEXT:    ret <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I4]]
 ;
 Entry:
   %P1 = bitcast i8* %P to <4 x i32>*
   %V1 = load <4 x i32>, <4 x i32>* %P1
   %P2 = bitcast i8* %P to <2 x i32>*
   %V2 = load <2 x i32>, <2 x i32>* %P2
   %P3 = bitcast i8* %P to i32*

 ; OLDGVN:       T:
 ; OLDGVN-NEXT:    ret float [[TMP0]]
 ; OLDGVN:       F:
 ; OLDGVN-NEXT:    ret float [[TMP0]]
 ;
 ; NEWGVN-LABEL: @test7(
 ; NEWGVN-NEXT:  Entry:
 ; NEWGVN-NEXT:    [[V1:%.*]] = load i32, i32* [[P1:%.*]], align 4
+; NEWGVN-NEXT:    [[TMP0:%.*]] = bitcast i32 [[V1]] to float
 ; NEWGVN-NEXT:    [[P2:%.*]] = bitcast i32* [[P1]] to float*
-; NEWGVN-NEXT:    [[V2:%.*]] = load float, float* [[P2]], align 4
 ; NEWGVN-NEXT:    br i1 [[COND:%.*]], label [[T:%.*]], label [[F:%.*]]
 ; NEWGVN:       T:
-; NEWGVN-NEXT:    ret float [[V2]]
+; NEWGVN-NEXT:    ret float [[TMP0]]
 ; NEWGVN:       F:
-; NEWGVN-NEXT:    [[V3:%.*]] = bitcast i32 [[V1]] to float
-; NEWGVN-NEXT:    ret float [[V3]]
+; NEWGVN-NEXT:    ret float [[TMP0]]
 ;
 Entry:
   %V1 = load i32, i32* %P1
   %P2 = bitcast i32* %P1 to float*
   %V2 = load float, float* %P2
   br i1 %cond, label %T, label %F
 T:
   ret float %V2

 ; OLDGVN-NEXT:    [[I3:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I2]], i32 [[TMP3]], 2
 ; OLDGVN-NEXT:    [[I4:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I3]], <2 x i32> [[TMP2]], 3
 ; OLDGVN-NEXT:    ret <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I4]]
 ;
 ; NEWGVN-LABEL: @test9(
 ; NEWGVN-NEXT:  Entry:
 ; NEWGVN-NEXT:    [[P1:%.*]] = bitcast i8* [[P:%.*]] to <4 x i32>*
 ; NEWGVN-NEXT:    [[V1:%.*]] = load <4 x i32>, <4 x i32>* [[P1]], align 16
+; NEWGVN-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V1]] to i128
+; NEWGVN-NEXT:    [[TMP1:%.*]] = trunc i128 [[TMP0]] to i64
+; NEWGVN-NEXT:    [[TMP2:%.*]] = bitcast i64 [[TMP1]] to <2 x i32>
+; NEWGVN-NEXT:    [[TMP3:%.*]] = trunc i128 [[TMP0]] to i32
 ; NEWGVN-NEXT:    [[P2:%.*]] = bitcast i8* [[P]] to <2 x i32>*
-; NEWGVN-NEXT:    [[V2:%.*]] = load <2 x i32>, <2 x i32>* [[P2]], align 8
 ; NEWGVN-NEXT:    [[P3:%.*]] = bitcast i8* [[P]] to i32*
-; NEWGVN-NEXT:    [[V3:%.*]] = load i32, i32* [[P3]], align 4
 ; NEWGVN-NEXT:    [[I1:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> poison, <4 x i32> [[V1]], 0
-; NEWGVN-NEXT:    [[I2:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I1]], <2 x i32> [[V2]], 1
-; NEWGVN-NEXT:    [[I3:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I2]], i32 [[V3]], 2
-; NEWGVN-NEXT:    [[I4:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I3]], <2 x i32> [[V2]], 3
+; NEWGVN-NEXT:    [[I2:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I1]], <2 x i32> [[TMP2]], 1
+; NEWGVN-NEXT:    [[I3:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I2]], i32 [[TMP3]], 2
+; NEWGVN-NEXT:    [[I4:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I3]], <2 x i32> [[TMP2]], 3
 ; NEWGVN-NEXT:    ret <{ <4 x i32>, <2 x i32>, i32, <2 x i32> }> [[I4]]
 ;
 Entry:
   %P1 = bitcast i8* %P to <4 x i32>*
   %V1 = load <4 x i32>, <4 x i32>* %P1
   %P2 = bitcast i8* %P to <2 x i32>*
   %V2 = load <2 x i32>, <2 x i32>* %P2
   %P3 = bitcast i8* %P to i32*

 ; OLDGVN-NEXT:    [[P3:%.*]] = bitcast i8* [[P]] to i32*
 ; OLDGVN-NEXT:    [[I3:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32 }> [[I1]], i32 [[TMP1]], 2
 ; OLDGVN-NEXT:    ret <{ <4 x i32>, <2 x i32>, i32 }> [[I3]]
 ;
 ; NEWGVN-LABEL: @test10(
 ; NEWGVN-NEXT:  Entry:
 ; NEWGVN-NEXT:    [[P1:%.*]] = bitcast i8* [[P:%.*]] to <4 x i32>*
 ; NEWGVN-NEXT:    [[V1:%.*]] = load <4 x i32>, <4 x i32>* [[P1]], align 16
+; NEWGVN-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V1]] to i128
+; NEWGVN-NEXT:    [[TMP1:%.*]] = trunc i128 [[TMP0]] to i32
+; NEWGVN-NEXT:    [[TMP2:%.*]] = trunc i128 [[TMP0]] to i64
+; NEWGVN-NEXT:    [[TMP3:%.*]] = bitcast i64 [[TMP2]] to <2 x i32>
 ; NEWGVN-NEXT:    [[I1:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32 }> poison, <4 x i32> [[V1]], 0
 ; NEWGVN-NEXT:    br i1 [[COND:%.*]], label [[T:%.*]], label [[F:%.*]]
 ; NEWGVN:       T:
 ; NEWGVN-NEXT:    [[P2:%.*]] = bitcast i8* [[P]] to <2 x i32>*
-; NEWGVN-NEXT:    [[V2:%.*]] = load <2 x i32>, <2 x i32>* [[P2]], align 8
-; NEWGVN-NEXT:    [[I2:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32 }> [[I1]], <2 x i32> [[V2]], 1
+; NEWGVN-NEXT:    [[I2:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32 }> [[I1]], <2 x i32> [[TMP3]], 1
 ; NEWGVN-NEXT:    ret <{ <4 x i32>, <2 x i32>, i32 }> [[I2]]
 ; NEWGVN:       F:
 ; NEWGVN-NEXT:    [[P3:%.*]] = bitcast i8* [[P]] to i32*
-; NEWGVN-NEXT:    [[V3:%.*]] = load i32, i32* [[P3]], align 4
-; NEWGVN-NEXT:    [[I3:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32 }> [[I1]], i32 [[V3]], 2
+; NEWGVN-NEXT:    [[I3:%.*]] = insertvalue <{ <4 x i32>, <2 x i32>, i32 }> [[I1]], i32 [[TMP1]], 2
 ; NEWGVN-NEXT:    ret <{ <4 x i32>, <2 x i32>, i32 }> [[I3]]
 ;
 Entry:
   %P1 = bitcast i8* %P to <4 x i32>*
   %V1 = load <4 x i32>, <4 x i32>* %P1
   %I1 = insertvalue <{<4 x i32>, <2 x i32>, i32}> poison, <4 x i32> %V1, 0
   br i1 %cond, label %T, label %F
 T:

 ; OLDGVN-NEXT:    [[I2:%.*]] = insertvalue <{ <2 x i32>, i32, float }> [[I1]], i32 [[TMP1]], 1
 ; OLDGVN-NEXT:    [[I3:%.*]] = insertvalue <{ <2 x i32>, i32, float }> [[I2]], float [[TMP2]], 2
 ; OLDGVN-NEXT:    ret <{ <2 x i32>, i32, float }> [[I3]]
 ;
 ; NEWGVN-LABEL: @test11(
 ; NEWGVN-NEXT:  Entry:
 ; NEWGVN-NEXT:    [[P1:%.*]] = bitcast i8* [[P:%.*]] to <2 x i32>*
 ; NEWGVN-NEXT:    [[V1:%.*]] = load <2 x i32>, <2 x i32>* [[P1]], align 8
+; NEWGVN-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V1]] to i64
+; NEWGVN-NEXT:    [[TMP1:%.*]] = trunc i64 [[TMP0]] to i32
+; NEWGVN-NEXT:    [[TMP2:%.*]] = bitcast i32 [[TMP1]] to float
 ; NEWGVN-NEXT:    [[P2:%.*]] = bitcast i8* [[P]] to i32*
-; NEWGVN-NEXT:    [[V2:%.*]] = load i32, i32* [[P2]], align 4
 ; NEWGVN-NEXT:    [[P3:%.*]] = bitcast i8* [[P]] to float*
-; NEWGVN-NEXT:    [[V3:%.*]] = load float, float* [[P3]], align 4
 ; NEWGVN-NEXT:    [[I1:%.*]] = insertvalue <{ <2 x i32>, i32, float }> poison, <2 x i32> [[V1]], 0
-; NEWGVN-NEXT:    [[I2:%.*]] = insertvalue <{ <2 x i32>, i32, float }> [[I1]], i32 [[V2]], 1
-; NEWGVN-NEXT:    [[I3:%.*]] = insertvalue <{ <2 x i32>, i32, float }> [[I2]], float [[V3]], 2
+; NEWGVN-NEXT:    [[I2:%.*]] = insertvalue <{ <2 x i32>, i32, float }> [[I1]], i32 [[TMP1]], 1
+; NEWGVN-NEXT:    [[I3:%.*]] = insertvalue <{ <2 x i32>, i32, float }> [[I2]], float [[TMP2]], 2
 ; NEWGVN-NEXT:    ret <{ <2 x i32>, i32, float }> [[I3]]
 ;
 Entry:
   %P1 = bitcast i8* %P to <2 x i32>*
   %V1 = load <2 x i32>, <2 x i32>* %P1
   %P2 = bitcast i8* %P to i32*
   %V2 = load i32, i32* %P2
   %P3 = bitcast i8* %P to float*

 ; OLDGVN-NEXT:    store i32 [[V:%.*]], i32* [[P1]], align 4
 ; OLDGVN-NEXT:    [[TMP2:%.*]] = trunc i32 [[V]] to i8
 ; OLDGVN-NEXT:    [[V5:%.*]] = add i8 [[TMP1]], [[TMP1]]
 ; OLDGVN-NEXT:    [[V6:%.*]] = add i8 [[TMP2]], [[V5]]
 ; OLDGVN-NEXT:    ret i8 [[V6]]
 ;
 ; NEWGVN-LABEL: @test12(
 ; NEWGVN-NEXT:    [[V1:%.*]] = load i32, i32* [[P1:%.*]], align 4
-; NEWGVN-NEXT:    [[P2:%.*]] = bitcast i32* [[P1]] to i8*
-; NEWGVN-NEXT:    [[V2:%.*]] = load i8, i8* [[P2]], align 1
-; NEWGVN-NEXT:    [[V3:%.*]] = trunc i32 [[V1]] to i8
+; NEWGVN-NEXT:    [[TMP1:%.*]] = trunc i32 [[V1]] to i8
 ; NEWGVN-NEXT:    store i32 [[V:%.*]], i32* [[P1]], align 4
-; NEWGVN-NEXT:    [[TMP1:%.*]] = trunc i32 [[V]] to i8
-; NEWGVN-NEXT:    [[V5:%.*]] = add i8 [[V2]], [[V3]]
-; NEWGVN-NEXT:    [[V6:%.*]] = add i8 [[TMP1]], [[V5]]
+; NEWGVN-NEXT:    [[TMP2:%.*]] = trunc i32 [[V]] to i8
+; NEWGVN-NEXT:    [[V5:%.*]] = add i8 [[TMP1]], [[TMP1]]
+; NEWGVN-NEXT:    [[V6:%.*]] = add i8 [[TMP2]], [[V5]]
 ; NEWGVN-NEXT:    ret i8 [[V6]]
 ;
   %V1 = load i32, i32* %P1
   %P2 = bitcast i32* %P1 to i8*
   %V2 = load i8, i8* %P2
   %V3 = trunc i32 %V1 to i8
   store i32 %V, i32* %P1
   %P4 = bitcast i32* %P1 to i8*

 define i32 @test14(i32* %P, i1 %Cond) {
 ;   Entry
 ;    /  \
 ;   T    F
 ;    \  /
 ;     vv
 ;    Exit
 ;
-; OLDGVN-LABEL: @test14(
-; OLDGVN-NEXT:  Entry:
-; OLDGVN-NEXT:    [[P1:%.*]] = bitcast i32* [[P:%.*]] to <2 x i32>*
-; OLDGVN-NEXT:    [[V1:%.*]] = load <2 x i32>, <2 x i32>* [[P1]], align 8
-; OLDGVN-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V1]] to i64
-; OLDGVN-NEXT:    [[TMP1:%.*]] = trunc i64 [[TMP0]] to i32
-; OLDGVN-NEXT:    br i1 [[COND:%.*]], label [[T:%.*]], label [[F:%.*]]
-; OLDGVN:       T:
-; OLDGVN-NEXT:    br label [[EXIT:%.*]]
-; OLDGVN:       F:
-; OLDGVN-NEXT:    br label [[EXIT]]
-; OLDGVN:       Exit:
-; OLDGVN-NEXT:    [[PHI:%.*]] = phi i32 [ 100, [[T]] ], [ 200, [[F]] ]
-; OLDGVN-NEXT:    [[V2:%.*]] = extractelement <2 x i32> [[V1]], i64 1
-; OLDGVN-NEXT:    [[V4:%.*]] = add i32 [[TMP1]], [[V2]]
-; OLDGVN-NEXT:    [[V5:%.*]] = add i32 [[V4]], [[PHI]]
-; OLDGVN-NEXT:    ret i32 [[V5]]
-;
-; NEWGVN-LABEL: @test14(
-; NEWGVN-NEXT:  Entry:
-; NEWGVN-NEXT:    [[P1:%.*]] = bitcast i32* [[P:%.*]] to <2 x i32>*
-; NEWGVN-NEXT:    [[V1:%.*]] = load <2 x i32>, <2 x i32>* [[P1]], align 8
-; NEWGVN-NEXT:    br i1 [[COND:%.*]], label [[T:%.*]], label [[F:%.*]]
-; NEWGVN:       T:
-; NEWGVN-NEXT:    br label [[EXIT:%.*]]
-; NEWGVN:       F:
-; NEWGVN-NEXT:    br label [[EXIT]]
-; NEWGVN:       Exit:
-; NEWGVN-NEXT:    [[PHI:%.*]] = phi i32 [ 100, [[T]] ], [ 200, [[F]] ]
-; NEWGVN-NEXT:    [[V2:%.*]] = extractelement <2 x i32> [[V1]], i64 1
-; NEWGVN-NEXT:    [[V3:%.*]] = load i32, i32* [[P]], align 4
-; NEWGVN-NEXT:    [[V4:%.*]] = add i32 [[V3]], [[V2]]
-; NEWGVN-NEXT:    [[V5:%.*]] = add i32 [[V4]], [[PHI]]
-; NEWGVN-NEXT:    ret i32 [[V5]]
+; GVN-LABEL: @test14(
+; GVN-NEXT:  Entry:
+; GVN-NEXT:    [[P1:%.*]] = bitcast i32* [[P:%.*]] to <2 x i32>*
+; GVN-NEXT:    [[V1:%.*]] = load <2 x i32>, <2 x i32>* [[P1]], align 8
+; GVN-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V1]] to i64
+; GVN-NEXT:    [[TMP1:%.*]] = trunc i64 [[TMP0]] to i32
+; GVN-NEXT:    br i1 [[COND:%.*]], label [[T:%.*]], label [[F:%.*]]
+; GVN:       T:
+; GVN-NEXT:    br label [[EXIT:%.*]]
+; GVN:       F:
+; GVN-NEXT:    br label [[EXIT]]
+; GVN:       Exit:
+; GVN-NEXT:    [[PHI:%.*]] = phi i32 [ 100, [[T]] ], [ 200, [[F]] ]
+; GVN-NEXT:    [[V2:%.*]] = extractelement <2 x i32> [[V1]], i64 1
+; GVN-NEXT:    [[V4:%.*]] = add i32 [[TMP1]], [[V2]]
+; GVN-NEXT:    [[V5:%.*]] = add i32 [[V4]], [[PHI]]
+; GVN-NEXT:    ret i32 [[V5]]
 ;
 Entry:
   %P1 = bitcast i32* %P to <2 x i32>*
   %V1 = load <2 x i32>, <2 x i32>* %P1
   br i1 %Cond, label %T, label %F
 
 T:
   br label %Exit

llvm/test/Transforms/NewGVN/load_coercion_replace_load_with_phi.ll

 ; OLDGVN:       Exit:
 ; OLDGVN-NEXT:    [[PHI:%.*]] = phi i32 [ [[V2]], [[ENTRY:%.*]] ], [ [[TMP1]], [[LOOP]] ]
 ; OLDGVN-NEXT:    ret i32 [[PHI]]
 ;
 ; NEWGVN-LABEL: @test12(
 ; NEWGVN-NEXT:  Entry:
 ; NEWGVN-NEXT:    [[P1:%.*]] = bitcast i32* [[P:%.*]] to <2 x i32>*
 ; NEWGVN-NEXT:    [[V1:%.*]] = load <2 x i32>, <2 x i32>* [[P1]], align 8
+; NEWGVN-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V1]] to i64
+; NEWGVN-NEXT:    [[TMP1:%.*]] = trunc i64 [[TMP0]] to i32
 ; NEWGVN-NEXT:    [[V2:%.*]] = extractelement <2 x i32> [[V1]], i64 1
 ; NEWGVN-NEXT:    br i1 [[COND1:%.*]], label [[LOOP:%.*]], label [[EXIT:%.*]]
 ; NEWGVN:       Loop:
-; NEWGVN-NEXT:    [[V3:%.*]] = load i32, i32* [[P]], align 4
 ; NEWGVN-NEXT:    br i1 [[COND2:%.*]], label [[LOOP]], label [[EXIT]]
 ; NEWGVN:       Exit:
-; NEWGVN-NEXT:    [[PHI:%.*]] = phi i32 [ [[V2]], [[ENTRY:%.*]] ], [ [[V3]], [[LOOP]] ]
+; NEWGVN-NEXT:    [[PHI:%.*]] = phi i32 [ [[V2]], [[ENTRY:%.*]] ], [ [[TMP1]], [[LOOP]] ]
 ; NEWGVN-NEXT:    ret i32 [[PHI]]
 ;
 Entry:
   %P1 = bitcast i32* %P to <2 x i32>*
   %V1 = load <2 x i32>, <2 x i32>* %P1
   %V2 = extractelement <2 x i32> %V1, i64 1
   br i1 %Cond1, label %Loop, label %Exit
 

 ;
 ; NEWGVN-LABEL: @test40(
 ; NEWGVN-NEXT:  Entry:
 ; NEWGVN-NEXT:    br label [[LOOP:%.*]]
 ; NEWGVN:       Loop:
 ; NEWGVN-NEXT:    [[INDEX:%.*]] = phi i64 [ 1, [[ENTRY:%.*]] ], [ [[INDEX_INC:%.*]], [[LOOP]] ]
 ; NEWGVN-NEXT:    [[P1:%.*]] = bitcast i64* [[P:%.*]] to <4 x i64>*
 ; NEWGVN-NEXT:    [[V1:%.*]] = load <4 x i64>, <4 x i64>* [[P1]], align 32
+; NEWGVN-NEXT:    [[TMP0:%.*]] = bitcast <4 x i64> [[V1]] to i256
+; NEWGVN-NEXT:    [[TMP1:%.*]] = trunc i256 [[TMP0]] to i64
 ; NEWGVN-NEXT:    [[V2:%.*]] = extractelement <4 x i64> [[V1]], i64 2
 ; NEWGVN-NEXT:    [[V3:%.*]] = add i64 [[V2]], [[INDEX]]
 ; NEWGVN-NEXT:    [[INDEX_INC]] = add i64 [[INDEX]], 1
 ; NEWGVN-NEXT:    [[COND1:%.*]] = icmp ne i64 [[INDEX_INC]], [[TC:%.*]]
 ; NEWGVN-NEXT:    br i1 [[COND1]], label [[LOOP]], label [[BB:%.*]]
 ; NEWGVN:       BB:
-; NEWGVN-NEXT:    [[V4:%.*]] = load i64, i64* [[P]], align 4
 ; NEWGVN-NEXT:    br label [[EXIT:%.*]]
 ; NEWGVN:       Exit:
-; NEWGVN-NEXT:    [[V5:%.*]] = add i64 [[V4]], [[V3]]
+; NEWGVN-NEXT:    [[V5:%.*]] = add i64 [[TMP1]], [[V3]]
 ; NEWGVN-NEXT:    ret i64 [[V5]]
 ;
 Entry:
   br label %Loop
 
 Loop:
   %Index = phi i64 [ 1, %Entry ], [ %Index.inc, %Loop ]
   %P1 = bitcast i64* %P to <4 x i64>*

 ;
 ; NEWGVN-LABEL: @test45(
 ; NEWGVN-NEXT:  Entry:
 ; NEWGVN-NEXT:    br label [[LOOP1:%.*]]
 ; NEWGVN:       Loop1:
 ; NEWGVN-NEXT:    [[INDEX1:%.*]] = phi i64 [ 1, [[ENTRY:%.*]] ], [ [[INDEX1_INC:%.*]], [[LOOP1]] ]
 ; NEWGVN-NEXT:    [[P1:%.*]] = bitcast i64* [[P:%.*]] to <4 x i64>*
 ; NEWGVN-NEXT:    [[V1:%.*]] = load <4 x i64>, <4 x i64>* [[P1]], align 32
+; NEWGVN-NEXT:    [[TMP0:%.*]] = bitcast <4 x i64> [[V1]] to i256
+; NEWGVN-NEXT:    [[TMP1:%.*]] = trunc i256 [[TMP0]] to i64
+; NEWGVN-NEXT:    [[TMP2:%.*]] = trunc i256 [[TMP0]] to i128
+; NEWGVN-NEXT:    [[TMP3:%.*]] = bitcast i128 [[TMP2]] to <2 x i64>
 ; NEWGVN-NEXT:    [[V2:%.*]] = extractelement <4 x i64> [[V1]], i64 2
 ; NEWGVN-NEXT:    [[INDEX1_INC]] = add i64 [[INDEX1]], 1
 ; NEWGVN-NEXT:    [[COND2:%.*]] = icmp ne i64 [[INDEX1_INC]], [[TC1:%.*]]
 ; NEWGVN-NEXT:    br i1 [[COND2]], label [[LOOP1]], label [[BB1:%.*]]
 ; NEWGVN:       BB1:
 ; NEWGVN-NEXT:    br i1 [[COND1:%.*]], label [[BB2:%.*]], label [[BB3:%.*]]
 ; NEWGVN:       BB2:
-; NEWGVN-NEXT:    [[V4:%.*]] = load i64, i64* [[P]], align 4
 ; NEWGVN-NEXT:    br label [[LOOP2:%.*]]
 ; NEWGVN:       BB3:
 ; NEWGVN-NEXT:    br label [[LOOP2]]
 ; NEWGVN:       Loop2:
-; NEWGVN-NEXT:    [[INDEX2:%.*]] = phi i64 [ [[V4]], [[BB2]] ], [ [[V2]], [[BB3]] ], [ [[INDEX2_INC:%.*]], [[LOOP2]] ]
+; NEWGVN-NEXT:    [[INDEX2:%.*]] = phi i64 [ [[TMP1]], [[BB2]] ], [ [[V2]], [[BB3]] ], [ [[INDEX2_INC:%.*]], [[LOOP2]] ]
 ; NEWGVN-NEXT:    [[P2:%.*]] = bitcast i64* [[P]] to <2 x i64>*
-; NEWGVN-NEXT:    [[V5:%.*]] = load <2 x i64>, <2 x i64>* [[P2]], align 16
-; NEWGVN-NEXT:    [[V6:%.*]] = extractelement <2 x i64> [[V5]], i64 1
-; NEWGVN-NEXT:    [[V7:%.*]] = load i64, i64* [[P]], align 4
+; NEWGVN-NEXT:    [[V6:%.*]] = extractelement <2 x i64> [[TMP3]], i64 1
 ; NEWGVN-NEXT:    [[V8:%.*]] = add i64 [[V6]], [[INDEX2]]
-; NEWGVN-NEXT:    [[V9:%.*]] = add i64 [[V8]], [[V7]]
+; NEWGVN-NEXT:    [[V9:%.*]] = add i64 [[V8]], [[TMP1]]
 ; NEWGVN-NEXT:    [[INDEX2_INC]] = add i64 [[INDEX2]], 1
 ; NEWGVN-NEXT:    [[COND3:%.*]] = icmp ne i64 [[INDEX2_INC]], [[TC2:%.*]]
 ; NEWGVN-NEXT:    br i1 [[COND3]], label [[LOOP2]], label [[EXIT:%.*]]
 ; NEWGVN:       Exit:
 ; NEWGVN-NEXT:    ret i64 [[V9]]
 ;
 Entry:
   br label %Loop1

 ;
 ; NEWGVN-LABEL: @test46(
 ; NEWGVN-NEXT:  Entry:
 ; NEWGVN-NEXT:    br label [[LOOP:%.*]]
 ; NEWGVN:       Loop:
 ; NEWGVN-NEXT:    [[INDEX:%.*]] = phi i64 [ 1, [[ENTRY:%.*]] ], [ [[INDEX_INC:%.*]], [[LOOP]] ]
 ; NEWGVN-NEXT:    [[P2:%.*]] = bitcast i64* [[P:%.*]] to <2 x i64>*
 ; NEWGVN-NEXT:    [[V1:%.*]] = load <2 x i64>, <2 x i64>* [[P2]], align 16
+; NEWGVN-NEXT:    [[TMP0:%.*]] = bitcast <2 x i64> [[V1]] to i128
+; NEWGVN-NEXT:    [[TMP1:%.*]] = trunc i128 [[TMP0]] to i64
 ; NEWGVN-NEXT:    [[V2:%.*]] = extractelement <2 x i64> [[V1]], i64 1
 ; NEWGVN-NEXT:    [[V3:%.*]] = add i64 [[V2]], [[INDEX]]
-; NEWGVN-NEXT:    [[V4:%.*]] = load i64, i64* [[P]], align 4
-; NEWGVN-NEXT:    [[V5:%.*]] = add i64 [[V3]], [[V4]]
+; NEWGVN-NEXT:    [[V5:%.*]] = add i64 [[V3]], [[TMP1]]
 ; NEWGVN-NEXT:    [[INDEX_INC]] = add i64 [[INDEX]], 1
 ; NEWGVN-NEXT:    [[COND:%.*]] = icmp ne i64 [[INDEX_INC]], [[TC:%.*]]
 ; NEWGVN-NEXT:    br i1 [[COND]], label [[LOOP]], label [[EXIT:%.*]]
 ; NEWGVN:       Exit:
 ; NEWGVN-NEXT:    ret i64 [[V5]]
 ;
 Entry:
   br label %Loop

 ;
 ; NEWGVN-LABEL: @test50(
 ; NEWGVN-NEXT:  Entry:
 ; NEWGVN-NEXT:    br label [[LOOP1:%.*]]
 ; NEWGVN:       Loop1:
 ; NEWGVN-NEXT:    [[INDEX1:%.*]] = phi i64 [ 1, [[ENTRY:%.*]] ], [ [[INDEX1_INC:%.*]], [[LOOP1]] ]
 ; NEWGVN-NEXT:    [[P1:%.*]] = bitcast i64* [[P:%.*]] to <4 x i64>*
 ; NEWGVN-NEXT:    [[V1:%.*]] = load <4 x i64>, <4 x i64>* [[P1]], align 32
+; NEWGVN-NEXT:    [[TMP0:%.*]] = bitcast <4 x i64> [[V1]] to i256
+; NEWGVN-NEXT:    [[TMP1:%.*]] = trunc i256 [[TMP0]] to i64
 ; NEWGVN-NEXT:    [[V2:%.*]] = extractelement <4 x i64> [[V1]], i64 2
 ; NEWGVN-NEXT:    [[INDEX1_INC]] = add i64 [[INDEX1]], 1
 ; NEWGVN-NEXT:    [[COND2:%.*]] = icmp ne i64 [[INDEX1_INC]], [[TC1:%.*]]
 ; NEWGVN-NEXT:    br i1 [[COND2]], label [[LOOP1]], label [[BB1:%.*]]
 ; NEWGVN:       BB1:
 ; NEWGVN-NEXT:    br i1 [[COND1:%.*]], label [[BB2:%.*]], label [[BB3:%.*]]
 ; NEWGVN:       BB2:
-; NEWGVN-NEXT:    [[V4:%.*]] = load i64, i64* [[P]], align 4
 ; NEWGVN-NEXT:    br label [[LOOP2:%.*]]
 ; NEWGVN:       BB3:
 ; NEWGVN-NEXT:    br label [[LOOP2]]
 ; NEWGVN:       Loop2:
-; NEWGVN-NEXT:    [[INDEX2:%.*]] = phi i64 [ [[V4]], [[BB2]] ], [ [[V2]], [[BB3]] ], [ [[INDEX2_INC:%.*]], [[LOOP2]] ]
+; NEWGVN-NEXT:    [[INDEX2:%.*]] = phi i64 [ [[TMP1]], [[BB2]] ], [ [[V2]], [[BB3]] ], [ [[INDEX2_INC:%.*]], [[LOOP2]] ]
 ; NEWGVN-NEXT:    [[P2:%.*]] = bitcast i64* [[P]] to <2 x i64>*
 ; NEWGVN-NEXT:    [[V5:%.*]] = load <2 x i64>, <2 x i64>* [[P2]], align 16
 ; NEWGVN-NEXT:    [[V6:%.*]] = extractelement <2 x i64> [[V5]], i64 1
 ; NEWGVN-NEXT:    [[V7:%.*]] = load i64, i64* [[P]], align 4
 ; NEWGVN-NEXT:    [[V8:%.*]] = add i64 [[V6]], [[INDEX2]]
 ; NEWGVN-NEXT:    [[V9:%.*]] = add i64 [[V8]], [[V7]]
 ; NEWGVN-NEXT:    store i64 [[V9]], i64* [[P]], align 4
 ; NEWGVN-NEXT:    [[INDEX2_INC]] = add i64 [[INDEX2]], 1

 ;
 ; NEWGVN-LABEL: @test51(
 ; NEWGVN-NEXT:  Entry:
 ; NEWGVN-NEXT:    br label [[LOOP:%.*]]
 ; NEWGVN:       Loop:
 ; NEWGVN-NEXT:    [[INDEX:%.*]] = phi i64 [ 1, [[ENTRY:%.*]] ], [ [[INDEX_INC:%.*]], [[LOOP_LATCH:%.*]] ]
 ; NEWGVN-NEXT:    [[P1:%.*]] = bitcast i64* [[P:%.*]] to <4 x i64>*
 ; NEWGVN-NEXT:    [[V1:%.*]] = load <4 x i64>, <4 x i64>* [[P1]], align 32
+; NEWGVN-NEXT:    [[TMP0:%.*]] = bitcast <4 x i64> [[V1]] to i256
+; NEWGVN-NEXT:    [[TMP1:%.*]] = trunc i256 [[TMP0]] to i64
 ; NEWGVN-NEXT:    [[V2:%.*]] = extractelement <4 x i64> [[V1]], i64 2
-; NEWGVN-NEXT:    [[V3:%.*]] = load i64, i64* [[P]], align 4
-; NEWGVN-NEXT:    [[V4:%.*]] = add i64 [[V3]], [[V2]]
+; NEWGVN-NEXT:    [[V4:%.*]] = add i64 [[TMP1]], [[V2]]
 ; NEWGVN-NEXT:    br i1 [[COND:%.*]], label [[BB1:%.*]], label [[BB2:%.*]]
 ; NEWGVN:       BB1:
 ; NEWGVN-NEXT:    br label [[BB3:%.*]]
 ; NEWGVN:       BB2:
 ; NEWGVN-NEXT:    br label [[BB3]]
 ; NEWGVN:       BB3:
-; NEWGVN-NEXT:    [[PHI:%.*]] = phi i64 [ [[V4]], [[BB1]] ], [ [[V3]], [[BB2]] ]
+; NEWGVN-NEXT:    [[PHI:%.*]] = phi i64 [ [[V4]], [[BB1]] ], [ [[TMP1]], [[BB2]] ]
 ; NEWGVN-NEXT:    [[V6:%.*]] = add i64 [[PHI]], [[INDEX]]
 ; NEWGVN-NEXT:    br label [[LOOP_LATCH]]
 ; NEWGVN:       Loop.Latch:
 ; NEWGVN-NEXT:    [[INDEX_INC]] = add i64 [[INDEX]], 1
 ; NEWGVN-NEXT:    [[COND1:%.*]] = icmp ne i64 [[INDEX_INC]], [[TC:%.*]]
 ; NEWGVN-NEXT:    br i1 [[COND1]], label [[LOOP]], label [[EXIT:%.*]]
 ; NEWGVN:       Exit:
 ; NEWGVN-NEXT:    ret i64 [[V6]]