[Scalarizer] ExtractElement handling w/ constant extract index

[lebedev.ri]

Summary

It appears to be better IR-wise to aggressively scalarize it,
rather than relying on gathering it, and leaving it as-is.

[lebedev.ri]

llvm/test/Transforms/Scalarizer/basic.ll

611 ↗

(On Diff #275249)

Previously we'd end up with https://godbolt.org/z/HDKdqh

%val0 = load <4 x i32>, <4 x i32>* %src, align 16
%val0.i0 = extractelement <4 x i32> %val0, i32 0
%val1.i0 = shl i32 1, %val0.i0
%val0.i1 = extractelement <4 x i32> %val0, i32 1
%val1.i1 = shl i32 2, %val0.i1
%val0.i2 = extractelement <4 x i32> %val0, i32 2
%val1.i2 = shl i32 3, %val0.i2
%val0.i3 = extractelement <4 x i32> %val0, i32 3
%val1.i3 = shl i32 4, %val0.i3
%val1.upto0 = insertelement <4 x i32> undef, i32 %val1.i0, i32 0
%val1.upto1 = insertelement <4 x i32> %val1.upto0, i32 %val1.i1, i32 1
%val1.upto2 = insertelement <4 x i32> %val1.upto1, i32 %val1.i2, i32 2
%val1 = insertelement <4 x i32> %val1.upto2, i32 %val1.i3, i32 3
%val2 = extractelement <4 x i32> %val1, i32 3
ret i32 %val2

[jdoerfert]

LGTM.

[lebedev.ri]

Reordered, precommitted tests.

[foad]

@lebedev.ri this is causing assertion failures and verification failures in some of our downstream tests. Here's a test case:

$ cat reduced.ll
define void @main(<3 x i32> inreg %w) {
entry:
  %a = extractelement <3 x i32> undef, i32 0
  %b = extractelement <3 x i32> undef, i32 1
  %x = extractelement <3 x i32> %w, i32 2
  %y = insertelement <4 x i32> undef, i32 %x, i32 2
  %z = insertelement <4 x i32> %y, i32 undef, i32 3
  store <4 x i32> %z, <4 x i32> addrspace(7)* undef, align 16
  ret void
}
$ ~/llvm-debug/bin/opt -scalarizer -o /dev/null reduced.ll
Instruction does not dominate all uses!
  <badref> = extractelement [145938144 x half] <badref>, i32 undef
  %z.upto2 = insertelement <4 x i32> undef, i32 <badref>, i32 2
in function main
LLVM ERROR: Broken function found, compilation aborted!

[lebedev.ri]

In D83101#2133062, @foad wrote:

@lebedev.ri this is causing assertion failures and verification failures in some of our downstream tests. Here's a test case:

$ cat reduced.ll
define void @main(<3 x i32> inreg %w) {
entry:
  %a = extractelement <3 x i32> undef, i32 0
  %b = extractelement <3 x i32> undef, i32 1
  %x = extractelement <3 x i32> %w, i32 2
  %y = insertelement <4 x i32> undef, i32 %x, i32 2
  %z = insertelement <4 x i32> %y, i32 undef, i32 3
  store <4 x i32> %z, <4 x i32> addrspace(7)* undef, align 16
  ret void
}
$ ~/llvm-debug/bin/opt -scalarizer -o /dev/null reduced.ll
Instruction does not dominate all uses!
  <badref> = extractelement [145938144 x half] <badref>, i32 undef
  %z.upto2 = insertelement <4 x i32> undef, i32 <badref>, i32 2
in function main
LLVM ERROR: Broken function found, compilation aborted!

Thanks for test case, looking.

[lebedev.ri]

In D83101#2134056, @lebedev.ri wrote:

In D83101#2133062, @foad wrote:

@lebedev.ri this is causing assertion failures and verification failures in some of our downstream tests. Here's a test case:

$ cat reduced.ll
define void @main(<3 x i32> inreg %w) {
entry:
  %a = extractelement <3 x i32> undef, i32 0
  %b = extractelement <3 x i32> undef, i32 1
  %x = extractelement <3 x i32> %w, i32 2
  %y = insertelement <4 x i32> undef, i32 %x, i32 2
  %z = insertelement <4 x i32> %y, i32 undef, i32 3
  store <4 x i32> %z, <4 x i32> addrspace(7)* undef, align 16
  ret void
}
$ ~/llvm-debug/bin/opt -scalarizer -o /dev/null reduced.ll
Instruction does not dominate all uses!
  <badref> = extractelement [145938144 x half] <badref>, i32 undef
  %z.upto2 = insertelement <4 x i32> undef, i32 <badref>, i32 2
in function main
LLVM ERROR: Broken function found, compilation aborted!

Thanks for test case, looking.

Thank you for a great reproducer! Fixed in rGdb05f2e34a5e9380ddcc199d6687531108d795e4.

[foad]

In D83101#2134412, @lebedev.ri wrote:

Fixed in rGdb05f2e34a5e9380ddcc199d6687531108d795e4.

Thanks. I can confirm that it fixes all the failures I was seeing.

[bjope]

I unfortunately still see some problems (related to the Scalarizer changes, and probably this patch):

> cat scalarizer-bug.ll
; RUN: opt < %s -scalarizer -S -o -

define void @foo() {
vector.ph:
  br label %vector.body115

vector.body115:                                   ; preds = %vector.body115, %vector.ph
  %vector.recur = phi <4 x i16> [ undef, %vector.ph ], [ %wide.load125, %vector.body115 ]
  %wide.load125 = load <4 x i16>, <4 x i16>* undef, align 1
  br i1 undef, label %middle.block113, label %vector.body115

middle.block113:                                  ; preds = %vector.body115
  ret void
}


-----------------------------------------------------

> ~/opt.master -scalarizer scalarizer-bug.ll -S
opt.master: ../lib/IR/Value.cpp:458: void llvm::Value::doRAUW(llvm::Value *, llvm::Value::ReplaceMetadataUses): Assertion `!contains(New, this) && "this->replaceAllUsesWith(expr(this)) is NOT valid!"' failed.
PLEASE submit a bug report to https://bugs.llvm.org/ and include the crash backtrace.
Stack dump:
0.      Program arguments: /home/uabbpet/opt.master -scalarizer scalarizer-bug.ll -S 
1.      Running pass 'Function Pass Manager' on module 'scalarizer-bug.ll'.
2.      Running pass 'Scalarize vector operations' on function '@foo'
Abort

[lebedev.ri]

In D83101#2135945, @bjope wrote:

I unfortunately still see some problems (related to the Scalarizer changes, and probably this patch):

> cat scalarizer-bug.ll
; RUN: opt < %s -scalarizer -S -o -

define void @foo() {
vector.ph:
  br label %vector.body115

vector.body115:                                   ; preds = %vector.body115, %vector.ph
  %vector.recur = phi <4 x i16> [ undef, %vector.ph ], [ %wide.load125, %vector.body115 ]
  %wide.load125 = load <4 x i16>, <4 x i16>* undef, align 1
  br i1 undef, label %middle.block113, label %vector.body115

middle.block113:                                  ; preds = %vector.body115
  ret void
}


-----------------------------------------------------

> ~/opt.master -scalarizer scalarizer-bug.ll -S
opt.master: ../lib/IR/Value.cpp:458: void llvm::Value::doRAUW(llvm::Value *, llvm::Value::ReplaceMetadataUses): Assertion `!contains(New, this) && "this->replaceAllUsesWith(expr(this)) is NOT valid!"' failed.
PLEASE submit a bug report to https://bugs.llvm.org/ and include the crash backtrace.
Stack dump:
0.      Program arguments: /home/uabbpet/opt.master -scalarizer scalarizer-bug.ll -S 
1.      Running pass 'Function Pass Manager' on module 'scalarizer-bug.ll'.
2.      Running pass 'Scalarize vector operations' on function '@foo'
Abort

Acknowledged, looking.

[lebedev.ri]

In D83101#2135962, @lebedev.ri wrote:

In D83101#2135945, @bjope wrote:

I unfortunately still see some problems (related to the Scalarizer changes, and probably this patch):

> cat scalarizer-bug.ll
; RUN: opt < %s -scalarizer -S -o -

define void @foo() {
vector.ph:
  br label %vector.body115

vector.body115:                                   ; preds = %vector.body115, %vector.ph
  %vector.recur = phi <4 x i16> [ undef, %vector.ph ], [ %wide.load125, %vector.body115 ]
  %wide.load125 = load <4 x i16>, <4 x i16>* undef, align 1
  br i1 undef, label %middle.block113, label %vector.body115

middle.block113:                                  ; preds = %vector.body115
  ret void
}


-----------------------------------------------------

> ~/opt.master -scalarizer scalarizer-bug.ll -S
opt.master: ../lib/IR/Value.cpp:458: void llvm::Value::doRAUW(llvm::Value *, llvm::Value::ReplaceMetadataUses): Assertion `!contains(New, this) && "this->replaceAllUsesWith(expr(this)) is NOT valid!"' failed.
PLEASE submit a bug report to https://bugs.llvm.org/ and include the crash backtrace.
Stack dump:
0.      Program arguments: /home/uabbpet/opt.master -scalarizer scalarizer-bug.ll -S 
1.      Running pass 'Function Pass Manager' on module 'scalarizer-bug.ll'.
2.      Running pass 'Scalarize vector operations' on function '@foo'
Abort

Acknowledged, looking.

Hm, this is saddening. I've fixed it in rG16266e63963ad6ee27ad21983a9366ab313dfd03, but are there more?

[lebedev.ri]

Forgot to say, thank you for the reduced reproducer!

[bjope]

In D83101#2136062, @lebedev.ri wrote:

In D83101#2135962, @lebedev.ri wrote:

In D83101#2135945, @bjope wrote:

I unfortunately still see some problems (related to the Scalarizer changes, and probably this patch):

> cat scalarizer-bug.ll
; RUN: opt < %s -scalarizer -S -o -

define void @foo() {
vector.ph:
  br label %vector.body115

vector.body115:                                   ; preds = %vector.body115, %vector.ph
  %vector.recur = phi <4 x i16> [ undef, %vector.ph ], [ %wide.load125, %vector.body115 ]
  %wide.load125 = load <4 x i16>, <4 x i16>* undef, align 1
  br i1 undef, label %middle.block113, label %vector.body115

middle.block113:                                  ; preds = %vector.body115
  ret void
}


-----------------------------------------------------

> ~/opt.master -scalarizer scalarizer-bug.ll -S
opt.master: ../lib/IR/Value.cpp:458: void llvm::Value::doRAUW(llvm::Value *, llvm::Value::ReplaceMetadataUses): Assertion `!contains(New, this) && "this->replaceAllUsesWith(expr(this)) is NOT valid!"' failed.
PLEASE submit a bug report to https://bugs.llvm.org/ and include the crash backtrace.
Stack dump:
0.      Program arguments: /home/uabbpet/opt.master -scalarizer scalarizer-bug.ll -S 
1.      Running pass 'Function Pass Manager' on module 'scalarizer-bug.ll'.
2.      Running pass 'Scalarize vector operations' on function '@foo'
Abort

Acknowledged, looking.

Hm, this is saddening. I've fixed it in rG16266e63963ad6ee27ad21983a9366ab313dfd03, but are there more?

Ah, maybe I was so occupied reducing the fault so I missed that there was another fix. I actually did take a look in github just to avoid reporting a problem that had been fixed, but must have done it just before rG16266e63963ad6ee27ad21983a9366ab313dfd03 landed.

I'll fetch, rebuild, and test. Let you know if it didn't help.

[lebedev.ri]

In D83101#2136101, @bjope wrote:

In D83101#2136062, @lebedev.ri wrote:

In D83101#2135962, @lebedev.ri wrote:

In D83101#2135945, @bjope wrote:

I unfortunately still see some problems (related to the Scalarizer changes, and probably this patch):

> cat scalarizer-bug.ll
; RUN: opt < %s -scalarizer -S -o -

define void @foo() {
vector.ph:
  br label %vector.body115

vector.body115:                                   ; preds = %vector.body115, %vector.ph
  %vector.recur = phi <4 x i16> [ undef, %vector.ph ], [ %wide.load125, %vector.body115 ]
  %wide.load125 = load <4 x i16>, <4 x i16>* undef, align 1
  br i1 undef, label %middle.block113, label %vector.body115

middle.block113:                                  ; preds = %vector.body115
  ret void
}


-----------------------------------------------------

> ~/opt.master -scalarizer scalarizer-bug.ll -S
opt.master: ../lib/IR/Value.cpp:458: void llvm::Value::doRAUW(llvm::Value *, llvm::Value::ReplaceMetadataUses): Assertion `!contains(New, this) && "this->replaceAllUsesWith(expr(this)) is NOT valid!"' failed.
PLEASE submit a bug report to https://bugs.llvm.org/ and include the crash backtrace.
Stack dump:
0.      Program arguments: /home/uabbpet/opt.master -scalarizer scalarizer-bug.ll -S 
1.      Running pass 'Function Pass Manager' on module 'scalarizer-bug.ll'.
2.      Running pass 'Scalarize vector operations' on function '@foo'
Abort

Acknowledged, looking.

Hm, this is saddening. I've fixed it in rG16266e63963ad6ee27ad21983a9366ab313dfd03, but are there more?

Ah, maybe I was so occupied reducing the fault so I missed that there was another fix. I actually did take a look in github just to avoid reporting a problem that had been fixed, but must have done it just before rG16266e63963ad6ee27ad21983a9366ab313dfd03 landed.

Nono, rG16266e63963ad6ee27ad21983a9366ab313dfd03 is the fix as a reaction to the bug you reported, your report wasn't a duplicate, sorry for confusion.

I'll fetch, rebuild, and test. Let you know if it didn't help.

Yes, please, thank you!

[bjope]

In D83101#2136116, @lebedev.ri wrote:

Ah, maybe I was so occupied reducing the fault so I missed that there was another fix. I actually did take a look in github just to avoid reporting a problem that had been fixed, but must have done it just before rG16266e63963ad6ee27ad21983a9366ab313dfd03 landed.

Nono, rG16266e63963ad6ee27ad21983a9366ab313dfd03 is the fix as a reaction to the bug you reported, your report wasn't a duplicate, sorry for confusion.

Ah, thanks! (That was such a quick fix/response so I thought someone else had discovered the same problem.)

I'll fetch, rebuild, and test. Let you know if it didn't help.

Yes, please, thank you!

The fix works fine. So I don't know about more problems right now.

[materi]

We have seen some issues with missing symbols during linking that I think are caused by this patch. The origin of this is downstream fuzz testing.

The global variable aglobal is renamed:

$ cat global.ll
@aglobal = dso_local global i16 0, align 1
@b = dso_local local_unnamed_addr global i16 0, align 1

define dso_local void @c() local_unnamed_addr {
entry:
  %d.sroa.0.1.vec.extract = extractelement <4 x i16*> <i16* @aglobal, i16* @aglobal, i16* @aglobal, i16* @aglobal>, i32 1
  %0 = ptrtoint i16* %d.sroa.0.1.vec.extract to i16
  store i16 %0, i16* @b, align 1
  ret void
}

$ opt -scalarizer global.ll -S
; ModuleID = 'global.ll'
source_filename = "global.ll"

@d.sroa.0.1.vec.extract = dso_local global i16 0, align 1
@b = dso_local local_unnamed_addr global i16 0, align 1

define dso_local void @c() local_unnamed_addr {
entry:
  %0 = ptrtoint i16* @d.sroa.0.1.vec.extract to i16
  store i16 %0, i16* @b, align 1
  ret void
}

[lebedev.ri]

In D83101#2223689, @materi wrote:

We have seen some issues with missing symbols during linking that I think are caused by this patch. The origin of this is downstream fuzz testing.

The global variable aglobal is renamed:

$ cat global.ll
@aglobal = dso_local global i16 0, align 1
@b = dso_local local_unnamed_addr global i16 0, align 1

define dso_local void @c() local_unnamed_addr {
entry:
  %d.sroa.0.1.vec.extract = extractelement <4 x i16*> <i16* @aglobal, i16* @aglobal, i16* @aglobal, i16* @aglobal>, i32 1
  %0 = ptrtoint i16* %d.sroa.0.1.vec.extract to i16
  store i16 %0, i16* @b, align 1
  ret void
}

$ opt -scalarizer global.ll -S
; ModuleID = 'global.ll'
source_filename = "global.ll"

@d.sroa.0.1.vec.extract = dso_local global i16 0, align 1
@b = dso_local local_unnamed_addr global i16 0, align 1

define dso_local void @c() local_unnamed_addr {
entry:
  %0 = ptrtoint i16* @d.sroa.0.1.vec.extract to i16
  store i16 %0, i16* @b, align 1
  ret void
}

I see that the global is renamed, https://godbolt.org/z/PM95se, it's not really intentional.
But i think something else is missing in this test - what's the failure? -verify passes

[materi]

In D83101#2223698, @lebedev.ri wrote:

In D83101#2223689, @materi wrote:

We have seen some issues with missing symbols during linking that I think are caused by this patch. The origin of this is downstream fuzz testing.

The global variable aglobal is renamed:

$ cat global.ll
@aglobal = dso_local global i16 0, align 1
@b = dso_local local_unnamed_addr global i16 0, align 1

define dso_local void @c() local_unnamed_addr {
entry:
  %d.sroa.0.1.vec.extract = extractelement <4 x i16*> <i16* @aglobal, i16* @aglobal, i16* @aglobal, i16* @aglobal>, i32 1
  %0 = ptrtoint i16* %d.sroa.0.1.vec.extract to i16
  store i16 %0, i16* @b, align 1
  ret void
}

$ opt -scalarizer global.ll -S
; ModuleID = 'global.ll'
source_filename = "global.ll"

@d.sroa.0.1.vec.extract = dso_local global i16 0, align 1
@b = dso_local local_unnamed_addr global i16 0, align 1

define dso_local void @c() local_unnamed_addr {
entry:
  %0 = ptrtoint i16* @d.sroa.0.1.vec.extract to i16
  store i16 %0, i16* @b, align 1
  ret void
}

I see that the global is renamed, https://godbolt.org/z/PM95se, it's not really intentional.
But i think something else is missing in this test - what's the failure? -verify passes

I don't think there is a verifier that points out this issue. But consider if there is another .ll file which has an external reference to the global:

@aglobal = external dso_local local_unnamed_addr global i16, align 1
define dso_local i16 @main() local_unnamed_addr #0 {
entry:
  %0 = load i16, i16* @aglobal, align 1
  ret i16 %0
}

In this case you get link error when @aglobal has been renamed in global.ll.

[bjope]

Proposed fix for the problem described by @materi : https://reviews.llvm.org/D86472

llvm/lib/Transforms/Scalar/Scalarizer.cpp

public:
   bool visitICmpInst(ICmpInst &ICI);
   bool visitFCmpInst(FCmpInst &FCI);
   bool visitUnaryOperator(UnaryOperator &UO);
   bool visitBinaryOperator(BinaryOperator &BO);
   bool visitGetElementPtrInst(GetElementPtrInst &GEPI);
   bool visitCastInst(CastInst &CI);
   bool visitBitCastInst(BitCastInst &BCI);
   bool visitInsertElementInst(InsertElementInst &IEI);
+  bool visitExtractElementInst(ExtractElementInst &EEI);
   bool visitShuffleVectorInst(ShuffleVectorInst &SVI);
   bool visitPHINode(PHINode &PHI);
   bool visitLoadInst(LoadInst &LI);
   bool visitStoreInst(StoreInst &SI);
   bool visitCallInst(CallInst &ICI);
 
 private:
   Scatterer scatter(Instruction *Point, Value *V);
bool ScalarizerVisitor::visitInsertElementInst(InsertElementInst &IEI) {
   } else {
     return false;
   }
 
   gather(&IEI, Res);
   return true;
 }
 
+bool ScalarizerVisitor::visitExtractElementInst(ExtractElementInst &EEI) {
+  VectorType *VT = dyn_cast<VectorType>(EEI.getOperand(0)->getType());
+  if (!VT)
+    return false;
+
+  IRBuilder<> Builder(&EEI);
+  Scatterer Op0 = scatter(&EEI, EEI.getOperand(0));
+  Value *ExtIdx = EEI.getOperand(1);
+
+  if (auto *CI = dyn_cast<ConstantInt>(ExtIdx)) {
+    Value *Res = Op0[CI->getValue().getZExtValue()];
+    gather(&EEI, {Res});
+    return true;
+  }
+
+  return false;
+}
+
 bool ScalarizerVisitor::visitShuffleVectorInst(ShuffleVectorInst &SVI) {
   VectorType *VT = dyn_cast<VectorType>(SVI.getType());
   if (!VT)
     return false;
 
   unsigned NumElems = VT->getNumElements();
   Scatterer Op0 = scatter(&SVI, SVI.getOperand(0));
   Scatterer Op1 = scatter(&SVI, SVI.getOperand(1));
bool ScalarizerVisitor::finish() {
   if (Gathered.empty() && Scattered.empty())
     return false;
   for (const auto &GMI : Gathered) {
     Instruction *Op = GMI.first;
     ValueVector &CV = *GMI.second;
     if (!Op->use_empty()) {
       // The value is still needed, so recreate it using a series of
       // InsertElements.
-      auto *Ty = cast<VectorType>(Op->getType());
-      Value *Res = UndefValue::get(Ty);
+      Value *Res = UndefValue::get(Op->getType());
+      if (auto *Ty = dyn_cast<VectorType>(Op->getType())) {
         BasicBlock *BB = Op->getParent();
         unsigned Count = Ty->getNumElements();
         IRBuilder<> Builder(Op);
         if (isa<PHINode>(Op))
           Builder.SetInsertPoint(BB, BB->getFirstInsertionPt());
         for (unsigned I = 0; I < Count; ++I)
           Res = Builder.CreateInsertElement(Res, CV[I], Builder.getInt32(I),
                                             Op->getName() + ".upto" + Twine(I));
+      } else {
+        assert(CV.size() == 1 && Op->getType() == CV[0]->getType());
+        Res = CV[0];
+      }
       Res->takeName(Op);
       Op->replaceAllUsesWith(Res);
     }
     Op->eraseFromParent();
   }
   Gathered.clear();
   Scattered.clear();
   return true;

llvm/test/Transforms/Scalarizer/constant-extractelement.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt %s -scalarizer -scalarize-load-store -dce -S | FileCheck --check-prefixes=ALL %s
 
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
 
 ; Test that constant extracts are nicely scalarized
 define i32 @f1(<4 x i32> *%src, i32 %index) {
 ; ALL-LABEL: @f1(
 ; ALL-NEXT:    [[SRC_I0:%.*]] = bitcast <4 x i32>* [[SRC:%.*]] to i32*
-; ALL-NEXT:    [[VAL0_I0:%.*]] = load i32, i32* [[SRC_I0]], align 16
-; ALL-NEXT:    [[SRC_I1:%.*]] = getelementptr i32, i32* [[SRC_I0]], i32 1
-; ALL-NEXT:    [[VAL0_I1:%.*]] = load i32, i32* [[SRC_I1]], align 4
-; ALL-NEXT:    [[SRC_I2:%.*]] = getelementptr i32, i32* [[SRC_I0]], i32 2
-; ALL-NEXT:    [[VAL0_I2:%.*]] = load i32, i32* [[SRC_I2]], align 8
 ; ALL-NEXT:    [[SRC_I3:%.*]] = getelementptr i32, i32* [[SRC_I0]], i32 3
 ; ALL-NEXT:    [[VAL0_I3:%.*]] = load i32, i32* [[SRC_I3]], align 4
-; ALL-NEXT:    [[VAL1_I0:%.*]] = shl i32 1, [[VAL0_I0]]
-; ALL-NEXT:    [[VAL1_I1:%.*]] = shl i32 2, [[VAL0_I1]]
-; ALL-NEXT:    [[VAL1_I2:%.*]] = shl i32 3, [[VAL0_I2]]
-; ALL-NEXT:    [[VAL1_I3:%.*]] = shl i32 4, [[VAL0_I3]]
-; ALL-NEXT:    [[VAL1_UPTO0:%.*]] = insertelement <4 x i32> undef, i32 [[VAL1_I0]], i32 0
-; ALL-NEXT:    [[VAL1_UPTO1:%.*]] = insertelement <4 x i32> [[VAL1_UPTO0]], i32 [[VAL1_I1]], i32 1
-; ALL-NEXT:    [[VAL1_UPTO2:%.*]] = insertelement <4 x i32> [[VAL1_UPTO1]], i32 [[VAL1_I2]], i32 2
-; ALL-NEXT:    [[VAL1:%.*]] = insertelement <4 x i32> [[VAL1_UPTO2]], i32 [[VAL1_I3]], i32 3
-; ALL-NEXT:    [[VAL2:%.*]] = extractelement <4 x i32> [[VAL1]], i32 3
+; ALL-NEXT:    [[VAL2:%.*]] = shl i32 4, [[VAL0_I3]]
 ; ALL-NEXT:    ret i32 [[VAL2]]
 ;
   %val0 = load <4 x i32> , <4 x i32> *%src
   %val1 = shl <4 x i32> <i32 1, i32 2, i32 3, i32 4>, %val0
   %val2 = extractelement <4 x i32> %val1, i32 3
   ret i32 %val2
 }

llvm/test/Transforms/Scalarizer/phi-unreachable-pred.ll

 ; CHECK-NEXT:    br label [[FOR_COND:%.*]]
 ; CHECK:       for.cond:
 ; CHECK-NEXT:    br i1 undef, label [[FOR_BODY:%.*]], label [[FOR_END]]
 ; CHECK:       for.end:
 ; CHECK-NEXT:    [[PHI_I0:%.*]] = phi i16 [ 1, [[ENTRY:%.*]] ], [ undef, [[FOR_COND]] ]
 ; CHECK-NEXT:    [[PHI_I1:%.*]] = phi i16 [ 1, [[ENTRY]] ], [ undef, [[FOR_COND]] ]
 ; CHECK-NEXT:    [[PHI_I2:%.*]] = phi i16 [ 1, [[ENTRY]] ], [ undef, [[FOR_COND]] ]
 ; CHECK-NEXT:    [[PHI_I3:%.*]] = phi i16 [ 1, [[ENTRY]] ], [ undef, [[FOR_COND]] ]
-; CHECK-NEXT:    [[PHI_UPTO0:%.*]] = insertelement <4 x i16> undef, i16 [[PHI_I0]], i32 0
-; CHECK-NEXT:    [[PHI_UPTO1:%.*]] = insertelement <4 x i16> [[PHI_UPTO0]], i16 [[PHI_I1]], i32 1
-; CHECK-NEXT:    [[PHI_UPTO2:%.*]] = insertelement <4 x i16> [[PHI_UPTO1]], i16 [[PHI_I2]], i32 2
-; CHECK-NEXT:    [[PHI:%.*]] = insertelement <4 x i16> [[PHI_UPTO2]], i16 [[PHI_I3]], i32 3
-; CHECK-NEXT:    [[EXTRACT:%.*]] = extractelement <4 x i16> [[PHI]], i32 0
-; CHECK-NEXT:    ret i16 [[EXTRACT]]
+; CHECK-NEXT:    ret i16 [[PHI_I0]]
 ;
 entry:
   br label %for.end
 
 for.body:
   %insert = insertelement <4 x i16> %insert, i16 ptrtoint (i16 () * @f1 to i16), i32 0
   br label %for.cond