From 895d1ce31b0528288a4c94eff2912586e3de87e3 Mon Sep 17 00:00:00 2001
From: Roman Lebedev <lebedev.ri@gmail.com>
Date: Tue, 25 May 2021 15:00:20 +0300
Subject: [PATCH] [LoopIdiom] Support 'left-shift until zero' idiom

This adds support for the "count active bits" pattern, i.e.:
```
int countBits(unsigned val) {
    int cnt = 0;
    for( ; (val << cnt) != 0; ++cnt)
        ;
    return cnt;
}
```
but a somewhat more general one:
```
int countBits(unsigned val, int start, int off) {
    int cnt;
    for (cnt = start; val << (cnt + off); cnt++)
        ;
    return cnt;
}
```

alive2 is happy with all the tests there.

Note that, again, much like with the right-shift cases,
we don't require the `val != 0` guard.

This is the last pattern that was supported by
`detectShiftUntilZeroIdiom()`, which now becomes obsolete.
---
 lib/Transforms/Scalar/LoopIdiomRecognize.cpp  |  26 +-
 .../LoopIdiom/X86/left-shift-until-zero.ll    | 519 ++++++++++++------
 2 files changed, 360 insertions(+), 185 deletions(-)

diff --git a/lib/Transforms/Scalar/LoopIdiomRecognize.cpp b/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
index d02db245cb8..089715a04f8 100644
--- a/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
+++ b/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
@@ -2416,7 +2416,7 @@ bool LoopIdiomRecognize::recognizeShiftUntilBitTest() {
 ///   loop:
 ///     %iv = phi i8 [ %start, %entry ], [ %iv.next, %for.cond ]
 ///     %nbits = add nsw i8 %iv, %extraoffset
-///     %val.shifted = [la]shr i8 %val, %nbits
+///     %val.shifted = {{l,a}shr,shl} i8 %val, %nbits
 ///     %val.shifted.iszero = icmp eq i8 %val.shifted, 0
 ///     %iv.next = add i8 %iv, 1
 ///     <...>
@@ -2432,8 +2432,9 @@ bool LoopIdiomRecognize::recognizeShiftUntilBitTest() {
 /// \endcode
 static bool detectShiftUntilZeroIdiom(Loop *CurLoop, ScalarEvolution *SE,
                                       Instruction *&ValShiftedIsZero,
-                                      Instruction *&IV, Value *&Start,
-                                      Value *&Val, const SCEV *&ExtraOffsetExpr,
+                                      Intrinsic::ID &IntrinID, Instruction *&IV,
+                                      Value *&Start, Value *&Val,
+                                      const SCEV *&ExtraOffsetExpr,
                                       bool &InvertedCond) {
   LLVM_DEBUG(dbgs() << DEBUG_TYPE
              " Performing shift-until-zero idiom detection.\n");
@@ -2469,11 +2470,13 @@ static bool detectShiftUntilZeroIdiom(Loop *CurLoop, ScalarEvolution *SE,
 
   // Step 2: Check if the comparison's operand is in desirable form.
   // FIXME: Val could be a one-input PHI node, which we should look past.
-  if (!match(ValShifted, m_Shr(m_LoopInvariant(m_Value(Val), CurLoop),
-                               m_Instruction(NBits)))) {
+  if (!match(ValShifted, m_Shift(m_LoopInvariant(m_Value(Val), CurLoop),
+                                 m_Instruction(NBits)))) {
     LLVM_DEBUG(dbgs() << DEBUG_TYPE " Bad comparisons value computation.\n");
     return false;
   }
+  IntrinID = ValShifted->getOpcode() == Instruction::Shl ? Intrinsic::cttz
+                                                         : Intrinsic::ctlz;
 
   // Step 3: Check if the shift amount is in desirable form.
 
@@ -2541,7 +2544,7 @@ static bool detectShiftUntilZeroIdiom(Loop *CurLoop, ScalarEvolution *SE,
 ///   loop:
 ///     %iv = phi i8 [ %start, %entry ], [ %iv.next, %for.cond ]
 ///     %nbits = add nsw i8 %iv, %extraoffset
-///     %val.shifted = [la]shr i8 %val, %nbits
+///     %val.shifted = {{l,a}shr,shl} i8 %val, %nbits
 ///     %val.shifted.iszero = icmp eq i8 %val.shifted, 0
 ///     %iv.next = add i8 %iv, 1
 ///     <...>
@@ -2563,7 +2566,7 @@ static bool detectShiftUntilZeroIdiom(Loop *CurLoop, ScalarEvolution *SE,
 ///     %start = <...>
 ///     %extraoffset = <...>
 ///     <...>
-///     %val.numleadingzeros = call i8 @llvm.ctlz.i8(i8 %val, i1 0)
+///     %val.numleadingzeros = call i8 @llvm.ct{l,t}z.i8(i8 %val, i1 0)
 ///     %val.numactivebits = sub i8 8, %val.numleadingzeros
 ///     %extraoffset.neg = sub i8 0, %extraoffset
 ///     %tmp = add i8 %val.numactivebits, %extraoffset.neg
@@ -2586,12 +2589,14 @@ static bool detectShiftUntilZeroIdiom(Loop *CurLoop, ScalarEvolution *SE,
 bool LoopIdiomRecognize::recognizeShiftUntilZero() {
   bool MadeChange = false;
 
-  Instruction *ValShiftedIsZero, *IV;
+  Instruction *ValShiftedIsZero;
+  Intrinsic::ID IntrID;
+  Instruction *IV;
   Value *Start, *Val;
   const SCEV *ExtraOffsetExpr;
   bool InvertedCond;
-  if (!detectShiftUntilZeroIdiom(CurLoop, SE, ValShiftedIsZero, IV, Start, Val,
-                                 ExtraOffsetExpr, InvertedCond)) {
+  if (!detectShiftUntilZeroIdiom(CurLoop, SE, ValShiftedIsZero, IntrID, IV,
+                                 Start, Val, ExtraOffsetExpr, InvertedCond)) {
     LLVM_DEBUG(dbgs() << DEBUG_TYPE
                " shift-until-zero idiom detection failed.\n");
     return MadeChange;
@@ -2611,7 +2616,6 @@ bool LoopIdiomRecognize::recognizeShiftUntilZero() {
   IRBuilder<> Builder(LoopPreheaderBB->getTerminator());
   Builder.SetCurrentDebugLocation(IV->getDebugLoc());
 
-  Intrinsic::ID IntrID = Intrinsic::ctlz;
   Type *Ty = Val->getType();
   unsigned Bitwidth = Ty->getScalarSizeInBits();
 
diff --git a/test/Transforms/LoopIdiom/X86/left-shift-until-zero.ll b/test/Transforms/LoopIdiom/X86/left-shift-until-zero.ll
index 86575aff6f9..d32fd50f299 100644
--- a/test/Transforms/LoopIdiom/X86/left-shift-until-zero.ll
+++ b/test/Transforms/LoopIdiom/X86/left-shift-until-zero.ll
@@ -11,20 +11,29 @@ declare i8 @gen.i8()
 define i8 @p0(i8 %val, i8 %start, i8 %extraoffset) {
 ; CHECK-LABEL: @p0(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i8 @llvm.cttz.i8(i8 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i8 8, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i8 0, [[EXTRAOFFSET:%.*]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i8 [[VAL_NUMACTIVEBITS]], [[TMP0]]
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i8 @llvm.smax.i8(i8 [[VAL_NUMACTIVEBITS_OFFSET]], i8 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i8 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i8 [[LOOP_BACKEDGETAKENCOUNT]], 1
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i8 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i8 [[IV]], [[EXTRAOFFSET:%.*]]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i8 [[VAL:%.*]], [[NBITS]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i8 [[VAL_SHIFTED]], 0
-; CHECK-NEXT:    [[IV_NEXT]] = add i8 [[IV]], 1
-; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i8 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i8 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i8 [[LOOP_IV]], 1
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i8 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i8 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i8 [[IV]], [[EXTRAOFFSET]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i8 [[VAL]], [[NBITS]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i8 [[IV]], 1
+; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i8 [[IV_NEXT]])
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i8 [ [[NBITS]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i8 [ [[VAL_SHIFTED]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i8 [ [[IV_NEXT]], [[LOOP]] ]
 ; CHECK-NEXT:    call void @escape_outer(i8 [[IV_RES]], i8 [[NBITS_RES]], i8 [[VAL_SHIFTED_RES]], i1 [[VAL_SHIFTED_ISZERO_RES]], i8 [[IV_NEXT_RES]])
 ; CHECK-NEXT:    ret i8 [[IV_RES]]
@@ -59,20 +68,29 @@ end:
 define i8 @p1(i8 %val, i8 %start, i8 %extraoffset) {
 ; CHECK-LABEL: @p1(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i8 @llvm.cttz.i8(i8 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i8 8, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i8 0, [[EXTRAOFFSET:%.*]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i8 [[VAL_NUMACTIVEBITS]], [[TMP0]]
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i8 @llvm.smax.i8(i8 [[VAL_NUMACTIVEBITS_OFFSET]], i8 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i8 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i8 [[LOOP_BACKEDGETAKENCOUNT]], 1
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i8 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[NBITS:%.*]] = add nuw i8 [[IV]], [[EXTRAOFFSET:%.*]]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i8 [[VAL:%.*]], [[NBITS]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i8 [[VAL_SHIFTED]], 0
-; CHECK-NEXT:    [[IV_NEXT]] = add i8 [[IV]], 1
-; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i8 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i8 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i8 [[LOOP_IV]], 1
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i8 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i8 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[NBITS:%.*]] = add nuw i8 [[IV]], [[EXTRAOFFSET]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i8 [[VAL]], [[NBITS]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i8 [[IV]], 1
+; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i8 [[IV_NEXT]])
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i8 [ [[NBITS]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i8 [ [[VAL_SHIFTED]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i8 [ [[IV_NEXT]], [[LOOP]] ]
 ; CHECK-NEXT:    call void @escape_outer(i8 [[IV_RES]], i8 [[NBITS_RES]], i8 [[VAL_SHIFTED_RES]], i1 [[VAL_SHIFTED_ISZERO_RES]], i8 [[IV_NEXT_RES]])
 ; CHECK-NEXT:    ret i8 [[IV_RES]]
@@ -107,20 +125,28 @@ end:
 define i8 @p2(i8 %val, i8 %start, i8 %extraoffset) {
 ; CHECK-LABEL: @p2(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i8 @llvm.cttz.i8(i8 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i8 8, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i8 [[VAL_NUMACTIVEBITS]], [[EXTRAOFFSET:%.*]]
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i8 @llvm.smax.i8(i8 [[VAL_NUMACTIVEBITS_OFFSET]], i8 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i8 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i8 [[LOOP_BACKEDGETAKENCOUNT]], 1
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i8 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[NBITS:%.*]] = sub nsw i8 [[IV]], [[EXTRAOFFSET:%.*]]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i8 [[VAL:%.*]], [[NBITS]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i8 [[VAL_SHIFTED]], 0
-; CHECK-NEXT:    [[IV_NEXT]] = add i8 [[IV]], 1
-; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i8 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i8 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i8 [[LOOP_IV]], 1
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i8 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i8 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[NBITS:%.*]] = sub nsw i8 [[IV]], [[EXTRAOFFSET]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i8 [[VAL]], [[NBITS]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i8 [[IV]], 1
+; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i8 [[IV_NEXT]])
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i8 [ [[NBITS]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i8 [ [[VAL_SHIFTED]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i8 [ [[IV_NEXT]], [[LOOP]] ]
 ; CHECK-NEXT:    call void @escape_outer(i8 [[IV_RES]], i8 [[NBITS_RES]], i8 [[VAL_SHIFTED_RES]], i1 [[VAL_SHIFTED_ISZERO_RES]], i8 [[IV_NEXT_RES]])
 ; CHECK-NEXT:    ret i8 [[IV_RES]]
@@ -299,18 +325,26 @@ end:
 define i8 @p6(i8 %val, i8 %start) {
 ; CHECK-LABEL: @p6(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i8 @llvm.cttz.i8(i8 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i8 8, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nuw nsw i8 [[VAL_NUMACTIVEBITS]], 0
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i8 @llvm.smax.i8(i8 [[VAL_NUMACTIVEBITS_OFFSET]], i8 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nuw nsw i8 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i8 [[LOOP_BACKEDGETAKENCOUNT]], 1
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i8 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i8 [[VAL:%.*]], [[IV]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i8 [[VAL_SHIFTED]], 0
-; CHECK-NEXT:    [[IV_NEXT]] = add i8 [[IV]], 1
-; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[IV]], i8 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i8 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i8 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i8 [[LOOP_IV]], 1
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i8 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i8 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i8 [[VAL]], [[IV]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i8 [[IV]], 1
+; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[IV]], i8 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i8 [[IV_NEXT]])
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i8 [ [[VAL_SHIFTED]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i8 [ [[IV_NEXT]], [[LOOP]] ]
 ; CHECK-NEXT:    call void @escape_outer(i8 [[IV_RES]], i8 [[IV_RES]], i8 [[VAL_SHIFTED_RES]], i1 [[VAL_SHIFTED_ISZERO_RES]], i8 [[IV_NEXT_RES]])
 ; CHECK-NEXT:    ret i8 [[IV_RES]]
@@ -346,20 +380,29 @@ declare void @escape_outer.i7(i7, i7, i7, i1, i7)
 define i7 @p7(i7 %val, i7 %start, i7 %extraoffset) {
 ; CHECK-LABEL: @p7(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i7 @llvm.cttz.i7(i7 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i7 7, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i7 0, [[EXTRAOFFSET:%.*]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i7 [[VAL_NUMACTIVEBITS]], [[TMP0]]
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i7 @llvm.smax.i7(i7 [[VAL_NUMACTIVEBITS_OFFSET]], i7 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i7 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i7 [[LOOP_BACKEDGETAKENCOUNT]], 1
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i7 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i7 [[IV]], [[EXTRAOFFSET:%.*]]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i7 [[VAL:%.*]], [[NBITS]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i7 [[VAL_SHIFTED]], 0
-; CHECK-NEXT:    [[IV_NEXT]] = add i7 [[IV]], 1
-; CHECK-NEXT:    call void @escape_inner.i7(i7 [[IV]], i7 [[NBITS]], i7 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i7 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i7 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i7 [[LOOP_IV]], 1
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i7 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i7 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i7 [[IV]], [[EXTRAOFFSET]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i7 [[VAL]], [[NBITS]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i7 [[IV]], 1
+; CHECK-NEXT:    call void @escape_inner.i7(i7 [[IV]], i7 [[NBITS]], i7 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i7 [[IV_NEXT]])
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i7 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i7 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i7 [ [[NBITS]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i7 [ [[VAL_SHIFTED]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i7 [ [[IV_NEXT]], [[LOOP]] ]
 ; CHECK-NEXT:    call void @escape_outer.i7(i7 [[IV_RES]], i7 [[NBITS_RES]], i7 [[VAL_SHIFTED_RES]], i1 [[VAL_SHIFTED_ISZERO_RES]], i7 [[IV_NEXT_RES]])
 ; CHECK-NEXT:    ret i7 [[IV_RES]]
@@ -442,17 +485,27 @@ end:
 define i8 @t9(i8 %val, i8 %start, i8 %extraoffset) {
 ; CHECK-LABEL: @t9(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i8 @llvm.cttz.i8(i8 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i8 8, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i8 0, [[EXTRAOFFSET:%.*]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i8 [[VAL_NUMACTIVEBITS]], [[TMP0]]
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i8 @llvm.smax.i8(i8 [[VAL_NUMACTIVEBITS_OFFSET]], i8 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i8 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i8 [[LOOP_BACKEDGETAKENCOUNT]], 1
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i8 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i8 [[IV]], [[EXTRAOFFSET:%.*]]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i8 [[VAL:%.*]], [[NBITS]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISNOTZERO:%.*]] = icmp ne i8 [[VAL_SHIFTED]], 0
-; CHECK-NEXT:    [[IV_NEXT]] = add i8 [[IV]], 1
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i8 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i8 [[LOOP_IV]], 1
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i8 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISNOTZERO:%.*]] = xor i1 [[LOOP_IVCHECK]], true
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i8 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i8 [[IV]], [[EXTRAOFFSET]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i8 [[VAL]], [[NBITS]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i8 [[IV]], 1
 ; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISNOTZERO]], i8 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISNOTZERO]], label [[LOOP]], label [[END:%.*]]
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i8 [ [[NBITS]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i8 [ [[VAL_SHIFTED]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_ISNOTZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISNOTZERO]], [[LOOP]] ]
@@ -735,20 +788,29 @@ end:
 define i8 @t15(i8 %val, i8 %start, i8 %extraoffset) {
 ; CHECK-LABEL: @t15(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i8 @llvm.cttz.i8(i8 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i8 8, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i8 0, [[EXTRAOFFSET:%.*]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i8 [[VAL_NUMACTIVEBITS]], [[TMP0]]
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i8 @llvm.smax.i8(i8 [[VAL_NUMACTIVEBITS_OFFSET]], i8 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i8 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i8 [[LOOP_BACKEDGETAKENCOUNT]], 1
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i8 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i8 [[EXTRAOFFSET:%.*]], [[IV]]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i8 [[VAL:%.*]], [[NBITS]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i8 [[VAL_SHIFTED]], 0
-; CHECK-NEXT:    [[IV_NEXT]] = add i8 [[IV]], 1
-; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i8 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i8 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i8 [[LOOP_IV]], 1
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i8 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i8 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i8 [[EXTRAOFFSET]], [[IV]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i8 [[VAL]], [[NBITS]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i8 [[IV]], 1
+; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i8 [[IV_NEXT]])
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i8 [ [[NBITS]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i8 [ [[VAL_SHIFTED]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i8 [ [[IV_NEXT]], [[LOOP]] ]
 ; CHECK-NEXT:    call void @escape_outer(i8 [[IV_RES]], i8 [[NBITS_RES]], i8 [[VAL_SHIFTED_RES]], i1 [[VAL_SHIFTED_ISZERO_RES]], i8 [[IV_NEXT_RES]])
 ; CHECK-NEXT:    ret i8 [[IV_RES]]
@@ -1082,22 +1144,31 @@ end:
 define i8 @n22(i8 %val, i8 %start, i8 %extraoffset) {
 ; CHECK-LABEL: @n22(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i8 @llvm.cttz.i8(i8 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i8 8, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i8 0, [[EXTRAOFFSET:%.*]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i8 [[VAL_NUMACTIVEBITS]], [[TMP0]]
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i8 @llvm.smax.i8(i8 [[VAL_NUMACTIVEBITS_OFFSET]], i8 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i8 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i8 [[LOOP_BACKEDGETAKENCOUNT]], 1
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i8 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i8 [[IV]], [[EXTRAOFFSET:%.*]]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i8 [[VAL:%.*]], [[NBITS]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i8 [[VAL_SHIFTED]], 0
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i8 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i8 [[LOOP_IV]], 1
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i8 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i8 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i8 [[IV]], [[EXTRAOFFSET]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i8 [[VAL]], [[NBITS]]
 ; CHECK-NEXT:    [[NOT_IV_NEXT:%.*]] = add i8 [[IV]], 1
-; CHECK-NEXT:    [[IV_NEXT]] = add i8 [[IV]], 1
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i8 [[IV]], 1
 ; CHECK-NEXT:    [[ALSO_IV_NEXT:%.*]] = add i8 [[IV]], 1
-; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i8 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]
+; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i8 [[IV_NEXT]])
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i8 [ [[NBITS]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i8 [ [[VAL_SHIFTED]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i8 [ [[IV_NEXT]], [[LOOP]] ]
 ; CHECK-NEXT:    call void @escape_outer(i8 [[IV_RES]], i8 [[NBITS_RES]], i8 [[VAL_SHIFTED_RES]], i1 [[VAL_SHIFTED_ISZERO_RES]], i8 [[IV_NEXT_RES]])
 ; CHECK-NEXT:    ret i8 [[IV_RES]]
@@ -1190,18 +1261,26 @@ declare void @escape_outer.i3(i3, i3, i3, i1, i3)
 define i1 @t24_nooffset_i1(i1 %val, i1 %start) {
 ; CHECK-LABEL: @t24_nooffset_i1(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i1 @llvm.cttz.i1(i1 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i1 true, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nuw nsw i1 [[VAL_NUMACTIVEBITS]], false
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i1 @llvm.smax.i1(i1 [[VAL_NUMACTIVEBITS_OFFSET]], i1 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nuw nsw i1 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i1 [[LOOP_BACKEDGETAKENCOUNT]], true
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i1 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i1 [[VAL:%.*]], [[IV]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i1 [[VAL_SHIFTED]], false
-; CHECK-NEXT:    [[IV_NEXT]] = add i1 [[IV]], true
-; CHECK-NEXT:    call void @escape_inner.i1(i1 [[IV]], i1 [[IV]], i1 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i1 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i1 [ false, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i1 [[LOOP_IV]], true
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i1 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i1 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i1 [[VAL]], [[IV]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i1 [[IV]], true
+; CHECK-NEXT:    call void @escape_inner.i1(i1 [[IV]], i1 [[IV]], i1 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i1 [[IV_NEXT]])
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i1 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i1 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i1 [ [[VAL_SHIFTED]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i1 [ [[IV_NEXT]], [[LOOP]] ]
 ; CHECK-NEXT:    call void @escape_outer.i1(i1 [[IV_RES]], i1 [[IV_RES]], i1 [[VAL_SHIFTED_RES]], i1 [[VAL_SHIFTED_ISZERO_RES]], i1 [[IV_NEXT_RES]])
 ; CHECK-NEXT:    ret i1 [[IV_RES]]
@@ -1232,18 +1311,26 @@ end:
 define i2 @t25_nooffset_i2(i2 %val, i2 %start) {
 ; CHECK-LABEL: @t25_nooffset_i2(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i2 @llvm.cttz.i2(i2 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw i2 -2, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nuw nsw i2 [[VAL_NUMACTIVEBITS]], 0
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i2 @llvm.smax.i2(i2 [[VAL_NUMACTIVEBITS_OFFSET]], i2 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nuw nsw i2 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw i2 [[LOOP_BACKEDGETAKENCOUNT]], 1
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i2 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i2 [[VAL:%.*]], [[IV]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i2 [[VAL_SHIFTED]], 0
-; CHECK-NEXT:    [[IV_NEXT]] = add i2 [[IV]], 1
-; CHECK-NEXT:    call void @escape_inner.i2(i2 [[IV]], i2 [[IV]], i2 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i2 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i2 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw i2 [[LOOP_IV]], 1
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i2 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i2 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i2 [[VAL]], [[IV]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i2 [[IV]], 1
+; CHECK-NEXT:    call void @escape_inner.i2(i2 [[IV]], i2 [[IV]], i2 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i2 [[IV_NEXT]])
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i2 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i2 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i2 [ [[VAL_SHIFTED]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i2 [ [[IV_NEXT]], [[LOOP]] ]
 ; CHECK-NEXT:    call void @escape_outer.i2(i2 [[IV_RES]], i2 [[IV_RES]], i2 [[VAL_SHIFTED_RES]], i1 [[VAL_SHIFTED_ISZERO_RES]], i2 [[IV_NEXT_RES]])
 ; CHECK-NEXT:    ret i2 [[IV_RES]]
@@ -1274,18 +1361,26 @@ end:
 define i3 @t26_nooffset_i3(i3 %val, i3 %start) {
 ; CHECK-LABEL: @t26_nooffset_i3(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i3 @llvm.cttz.i3(i3 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i3 3, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nuw nsw i3 [[VAL_NUMACTIVEBITS]], 0
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i3 @llvm.smax.i3(i3 [[VAL_NUMACTIVEBITS_OFFSET]], i3 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nuw nsw i3 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i3 [[LOOP_BACKEDGETAKENCOUNT]], 1
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i3 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i3 [[VAL:%.*]], [[IV]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i3 [[VAL_SHIFTED]], 0
-; CHECK-NEXT:    [[IV_NEXT]] = add i3 [[IV]], 1
-; CHECK-NEXT:    call void @escape_inner.i3(i3 [[IV]], i3 [[IV]], i3 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i3 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i3 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i3 [[LOOP_IV]], 1
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i3 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i3 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i3 [[VAL]], [[IV]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i3 [[IV]], 1
+; CHECK-NEXT:    call void @escape_inner.i3(i3 [[IV]], i3 [[IV]], i3 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i3 [[IV_NEXT]])
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i3 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i3 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i3 [ [[VAL_SHIFTED]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i3 [ [[IV_NEXT]], [[LOOP]] ]
 ; CHECK-NEXT:    call void @escape_outer.i3(i3 [[IV_RES]], i3 [[IV_RES]], i3 [[VAL_SHIFTED_RES]], i1 [[VAL_SHIFTED_ISZERO_RES]], i3 [[IV_NEXT_RES]])
 ; CHECK-NEXT:    ret i3 [[IV_RES]]
@@ -1317,20 +1412,28 @@ end:
 define i1 @t27_addnsw_i1(i1 %val, i1 %start, i1 %extraoffset) {
 ; CHECK-LABEL: @t27_addnsw_i1(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i1 @llvm.cttz.i1(i1 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i1 true, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i1 [[VAL_NUMACTIVEBITS]], [[EXTRAOFFSET:%.*]]
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i1 @llvm.smax.i1(i1 [[VAL_NUMACTIVEBITS_OFFSET]], i1 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i1 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i1 [[LOOP_BACKEDGETAKENCOUNT]], true
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i1 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i1 [[IV]], [[EXTRAOFFSET:%.*]]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i1 [[VAL:%.*]], [[NBITS]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i1 [[VAL_SHIFTED]], false
-; CHECK-NEXT:    [[IV_NEXT]] = add i1 [[IV]], true
-; CHECK-NEXT:    call void @escape_inner.i1(i1 [[IV]], i1 [[NBITS]], i1 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i1 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i1 [ false, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i1 [[LOOP_IV]], true
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i1 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i1 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i1 [[IV]], [[EXTRAOFFSET]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i1 [[VAL]], [[NBITS]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i1 [[IV]], true
+; CHECK-NEXT:    call void @escape_inner.i1(i1 [[IV]], i1 [[NBITS]], i1 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i1 [[IV_NEXT]])
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i1 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i1 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i1 [ [[NBITS]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i1 [ [[VAL_SHIFTED]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i1 [ [[IV_NEXT]], [[LOOP]] ]
 ; CHECK-NEXT:    call void @escape_outer.i1(i1 [[IV_RES]], i1 [[NBITS_RES]], i1 [[VAL_SHIFTED_RES]], i1 [[VAL_SHIFTED_ISZERO_RES]], i1 [[IV_NEXT_RES]])
 ; CHECK-NEXT:    ret i1 [[IV_RES]]
@@ -1363,20 +1466,29 @@ end:
 define i2 @t28_addnsw_i2(i2 %val, i2 %start, i2 %extraoffset) {
 ; CHECK-LABEL: @t28_addnsw_i2(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i2 @llvm.cttz.i2(i2 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw i2 -2, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i2 0, [[EXTRAOFFSET:%.*]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i2 [[VAL_NUMACTIVEBITS]], [[TMP0]]
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i2 @llvm.smax.i2(i2 [[VAL_NUMACTIVEBITS_OFFSET]], i2 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i2 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw i2 [[LOOP_BACKEDGETAKENCOUNT]], 1
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i2 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i2 [[IV]], [[EXTRAOFFSET:%.*]]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i2 [[VAL:%.*]], [[NBITS]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i2 [[VAL_SHIFTED]], 0
-; CHECK-NEXT:    [[IV_NEXT]] = add i2 [[IV]], 1
-; CHECK-NEXT:    call void @escape_inner.i2(i2 [[IV]], i2 [[NBITS]], i2 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i2 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i2 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw i2 [[LOOP_IV]], 1
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i2 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i2 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i2 [[IV]], [[EXTRAOFFSET]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i2 [[VAL]], [[NBITS]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i2 [[IV]], 1
+; CHECK-NEXT:    call void @escape_inner.i2(i2 [[IV]], i2 [[NBITS]], i2 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i2 [[IV_NEXT]])
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i2 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i2 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i2 [ [[NBITS]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i2 [ [[VAL_SHIFTED]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i2 [ [[IV_NEXT]], [[LOOP]] ]
 ; CHECK-NEXT:    call void @escape_outer.i2(i2 [[IV_RES]], i2 [[NBITS_RES]], i2 [[VAL_SHIFTED_RES]], i1 [[VAL_SHIFTED_ISZERO_RES]], i2 [[IV_NEXT_RES]])
 ; CHECK-NEXT:    ret i2 [[IV_RES]]
@@ -1409,20 +1521,29 @@ end:
 define i3 @t29_addnsw_i3(i3 %val, i3 %start, i3 %extraoffset) {
 ; CHECK-LABEL: @t29_addnsw_i3(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i3 @llvm.cttz.i3(i3 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i3 3, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i3 0, [[EXTRAOFFSET:%.*]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i3 [[VAL_NUMACTIVEBITS]], [[TMP0]]
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i3 @llvm.smax.i3(i3 [[VAL_NUMACTIVEBITS_OFFSET]], i3 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i3 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i3 [[LOOP_BACKEDGETAKENCOUNT]], 1
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i3 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i3 [[IV]], [[EXTRAOFFSET:%.*]]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i3 [[VAL:%.*]], [[NBITS]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i3 [[VAL_SHIFTED]], 0
-; CHECK-NEXT:    [[IV_NEXT]] = add i3 [[IV]], 1
-; CHECK-NEXT:    call void @escape_inner.i3(i3 [[IV]], i3 [[NBITS]], i3 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i3 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i3 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i3 [[LOOP_IV]], 1
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i3 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i3 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i3 [[IV]], [[EXTRAOFFSET]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i3 [[VAL]], [[NBITS]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i3 [[IV]], 1
+; CHECK-NEXT:    call void @escape_inner.i3(i3 [[IV]], i3 [[NBITS]], i3 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i3 [[IV_NEXT]])
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i3 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i3 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i3 [ [[NBITS]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i3 [ [[VAL_SHIFTED]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i3 [ [[IV_NEXT]], [[LOOP]] ]
 ; CHECK-NEXT:    call void @escape_outer.i3(i3 [[IV_RES]], i3 [[NBITS_RES]], i3 [[VAL_SHIFTED_RES]], i1 [[VAL_SHIFTED_ISZERO_RES]], i3 [[IV_NEXT_RES]])
 ; CHECK-NEXT:    ret i3 [[IV_RES]]
@@ -1456,20 +1577,28 @@ end:
 define i1 @t30_addnuw_i1(i1 %val, i1 %start, i1 %extraoffset) {
 ; CHECK-LABEL: @t30_addnuw_i1(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i1 @llvm.cttz.i1(i1 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i1 true, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i1 [[VAL_NUMACTIVEBITS]], [[EXTRAOFFSET:%.*]]
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i1 @llvm.smax.i1(i1 [[VAL_NUMACTIVEBITS_OFFSET]], i1 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i1 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i1 [[LOOP_BACKEDGETAKENCOUNT]], true
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i1 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[NBITS:%.*]] = add nuw i1 [[IV]], [[EXTRAOFFSET:%.*]]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i1 [[VAL:%.*]], [[NBITS]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i1 [[VAL_SHIFTED]], false
-; CHECK-NEXT:    [[IV_NEXT]] = add i1 [[IV]], true
-; CHECK-NEXT:    call void @escape_inner.i1(i1 [[IV]], i1 [[NBITS]], i1 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i1 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i1 [ false, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i1 [[LOOP_IV]], true
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i1 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i1 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[NBITS:%.*]] = add nuw i1 [[IV]], [[EXTRAOFFSET]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i1 [[VAL]], [[NBITS]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i1 [[IV]], true
+; CHECK-NEXT:    call void @escape_inner.i1(i1 [[IV]], i1 [[NBITS]], i1 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i1 [[IV_NEXT]])
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i1 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i1 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i1 [ [[NBITS]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i1 [ [[VAL_SHIFTED]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i1 [ [[IV_NEXT]], [[LOOP]] ]
 ; CHECK-NEXT:    call void @escape_outer.i1(i1 [[IV_RES]], i1 [[NBITS_RES]], i1 [[VAL_SHIFTED_RES]], i1 [[VAL_SHIFTED_ISZERO_RES]], i1 [[IV_NEXT_RES]])
 ; CHECK-NEXT:    ret i1 [[IV_RES]]
@@ -1502,20 +1631,29 @@ end:
 define i2 @t31_addnuw_i2(i2 %val, i2 %start, i2 %extraoffset) {
 ; CHECK-LABEL: @t31_addnuw_i2(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i2 @llvm.cttz.i2(i2 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw i2 -2, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i2 0, [[EXTRAOFFSET:%.*]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i2 [[VAL_NUMACTIVEBITS]], [[TMP0]]
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i2 @llvm.smax.i2(i2 [[VAL_NUMACTIVEBITS_OFFSET]], i2 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i2 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw i2 [[LOOP_BACKEDGETAKENCOUNT]], 1
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i2 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[NBITS:%.*]] = add nuw i2 [[IV]], [[EXTRAOFFSET:%.*]]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i2 [[VAL:%.*]], [[NBITS]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i2 [[VAL_SHIFTED]], 0
-; CHECK-NEXT:    [[IV_NEXT]] = add i2 [[IV]], 1
-; CHECK-NEXT:    call void @escape_inner.i2(i2 [[IV]], i2 [[NBITS]], i2 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i2 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i2 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw i2 [[LOOP_IV]], 1
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i2 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i2 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[NBITS:%.*]] = add nuw i2 [[IV]], [[EXTRAOFFSET]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i2 [[VAL]], [[NBITS]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i2 [[IV]], 1
+; CHECK-NEXT:    call void @escape_inner.i2(i2 [[IV]], i2 [[NBITS]], i2 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i2 [[IV_NEXT]])
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i2 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i2 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i2 [ [[NBITS]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i2 [ [[VAL_SHIFTED]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i2 [ [[IV_NEXT]], [[LOOP]] ]
 ; CHECK-NEXT:    call void @escape_outer.i2(i2 [[IV_RES]], i2 [[NBITS_RES]], i2 [[VAL_SHIFTED_RES]], i1 [[VAL_SHIFTED_ISZERO_RES]], i2 [[IV_NEXT_RES]])
 ; CHECK-NEXT:    ret i2 [[IV_RES]]
@@ -1548,20 +1686,29 @@ end:
 define i3 @t32_addnuw_i3(i3 %val, i3 %start, i3 %extraoffset) {
 ; CHECK-LABEL: @t32_addnuw_i3(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i3 @llvm.cttz.i3(i3 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i3 3, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i3 0, [[EXTRAOFFSET:%.*]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i3 [[VAL_NUMACTIVEBITS]], [[TMP0]]
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i3 @llvm.smax.i3(i3 [[VAL_NUMACTIVEBITS_OFFSET]], i3 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i3 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i3 [[LOOP_BACKEDGETAKENCOUNT]], 1
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i3 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[NBITS:%.*]] = add nuw i3 [[IV]], [[EXTRAOFFSET:%.*]]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i3 [[VAL:%.*]], [[NBITS]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i3 [[VAL_SHIFTED]], 0
-; CHECK-NEXT:    [[IV_NEXT]] = add i3 [[IV]], 1
-; CHECK-NEXT:    call void @escape_inner.i3(i3 [[IV]], i3 [[NBITS]], i3 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i3 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i3 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i3 [[LOOP_IV]], 1
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i3 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i3 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[NBITS:%.*]] = add nuw i3 [[IV]], [[EXTRAOFFSET]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i3 [[VAL]], [[NBITS]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i3 [[IV]], 1
+; CHECK-NEXT:    call void @escape_inner.i3(i3 [[IV]], i3 [[NBITS]], i3 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i3 [[IV_NEXT]])
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i3 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i3 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i3 [ [[NBITS]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i3 [ [[VAL_SHIFTED]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i3 [ [[IV_NEXT]], [[LOOP]] ]
 ; CHECK-NEXT:    call void @escape_outer.i3(i3 [[IV_RES]], i3 [[NBITS_RES]], i3 [[VAL_SHIFTED_RES]], i1 [[VAL_SHIFTED_ISZERO_RES]], i3 [[IV_NEXT_RES]])
 ; CHECK-NEXT:    ret i3 [[IV_RES]]
@@ -1596,20 +1743,28 @@ end:
 define i1 @t33_subnsw_i1(i1 %val, i1 %start, i1 %extraoffset) {
 ; CHECK-LABEL: @t33_subnsw_i1(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i1 @llvm.cttz.i1(i1 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i1 true, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i1 [[VAL_NUMACTIVEBITS]], [[EXTRAOFFSET:%.*]]
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i1 @llvm.smax.i1(i1 [[VAL_NUMACTIVEBITS_OFFSET]], i1 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i1 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i1 [[LOOP_BACKEDGETAKENCOUNT]], true
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i1 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[NBITS:%.*]] = sub nsw i1 [[IV]], [[EXTRAOFFSET:%.*]]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i1 [[VAL:%.*]], [[NBITS]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i1 [[VAL_SHIFTED]], false
-; CHECK-NEXT:    [[IV_NEXT]] = add i1 [[IV]], true
-; CHECK-NEXT:    call void @escape_inner.i1(i1 [[IV]], i1 [[NBITS]], i1 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i1 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i1 [ false, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i1 [[LOOP_IV]], true
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i1 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i1 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[NBITS:%.*]] = sub nsw i1 [[IV]], [[EXTRAOFFSET]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i1 [[VAL]], [[NBITS]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i1 [[IV]], true
+; CHECK-NEXT:    call void @escape_inner.i1(i1 [[IV]], i1 [[NBITS]], i1 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i1 [[IV_NEXT]])
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i1 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i1 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i1 [ [[NBITS]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i1 [ [[VAL_SHIFTED]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i1 [ [[IV_NEXT]], [[LOOP]] ]
 ; CHECK-NEXT:    call void @escape_outer.i1(i1 [[IV_RES]], i1 [[NBITS_RES]], i1 [[VAL_SHIFTED_RES]], i1 [[VAL_SHIFTED_ISZERO_RES]], i1 [[IV_NEXT_RES]])
 ; CHECK-NEXT:    ret i1 [[IV_RES]]
@@ -1642,20 +1797,28 @@ end:
 define i2 @t34_addnuw_i2(i2 %val, i2 %start, i2 %extraoffset) {
 ; CHECK-LABEL: @t34_addnuw_i2(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i2 @llvm.cttz.i2(i2 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw i2 -2, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i2 [[VAL_NUMACTIVEBITS]], [[EXTRAOFFSET:%.*]]
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i2 @llvm.smax.i2(i2 [[VAL_NUMACTIVEBITS_OFFSET]], i2 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i2 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw i2 [[LOOP_BACKEDGETAKENCOUNT]], 1
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i2 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[NBITS:%.*]] = sub nsw i2 [[IV]], [[EXTRAOFFSET:%.*]]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i2 [[VAL:%.*]], [[NBITS]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i2 [[VAL_SHIFTED]], 0
-; CHECK-NEXT:    [[IV_NEXT]] = add i2 [[IV]], 1
-; CHECK-NEXT:    call void @escape_inner.i2(i2 [[IV]], i2 [[NBITS]], i2 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i2 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i2 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw i2 [[LOOP_IV]], 1
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i2 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i2 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[NBITS:%.*]] = sub nsw i2 [[IV]], [[EXTRAOFFSET]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i2 [[VAL]], [[NBITS]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i2 [[IV]], 1
+; CHECK-NEXT:    call void @escape_inner.i2(i2 [[IV]], i2 [[NBITS]], i2 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i2 [[IV_NEXT]])
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i2 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i2 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i2 [ [[NBITS]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i2 [ [[VAL_SHIFTED]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i2 [ [[IV_NEXT]], [[LOOP]] ]
 ; CHECK-NEXT:    call void @escape_outer.i2(i2 [[IV_RES]], i2 [[NBITS_RES]], i2 [[VAL_SHIFTED_RES]], i1 [[VAL_SHIFTED_ISZERO_RES]], i2 [[IV_NEXT_RES]])
 ; CHECK-NEXT:    ret i2 [[IV_RES]]
@@ -1688,20 +1851,28 @@ end:
 define i3 @t35_addnuw_i3(i3 %val, i3 %start, i3 %extraoffset) {
 ; CHECK-LABEL: @t35_addnuw_i3(
 ; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i3 @llvm.cttz.i3(i3 [[VAL:%.*]], i1 false)
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i3 3, [[VAL_NUMLEADINGZEROS]]
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i3 [[VAL_NUMACTIVEBITS]], [[EXTRAOFFSET:%.*]]
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i3 @llvm.smax.i3(i3 [[VAL_NUMACTIVEBITS_OFFSET]], i3 [[START:%.*]])
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i3 [[IV_FINAL]], [[START]]
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i3 [[LOOP_BACKEDGETAKENCOUNT]], 1
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i3 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT:    [[NBITS:%.*]] = sub nsw i3 [[IV]], [[EXTRAOFFSET:%.*]]
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i3 [[VAL:%.*]], [[NBITS]]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i3 [[VAL_SHIFTED]], 0
-; CHECK-NEXT:    [[IV_NEXT]] = add i3 [[IV]], 1
-; CHECK-NEXT:    call void @escape_inner.i3(i3 [[IV]], i3 [[NBITS]], i3 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i3 [[IV_NEXT]])
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i3 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i3 [[LOOP_IV]], 1
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i3 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i3 [[LOOP_IV]], [[START]]
+; CHECK-NEXT:    [[NBITS:%.*]] = sub nsw i3 [[IV]], [[EXTRAOFFSET]]
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = shl i3 [[VAL]], [[NBITS]]
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i3 [[IV]], 1
+; CHECK-NEXT:    call void @escape_inner.i3(i3 [[IV]], i3 [[NBITS]], i3 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i3 [[IV_NEXT]])
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]
 ; CHECK:       end:
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i3 [ [[IV]], [[LOOP]] ]
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i3 [ [[IV_FINAL]], [[LOOP]] ]
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i3 [ [[NBITS]], [[LOOP]] ]
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i3 [ [[VAL_SHIFTED]], [[LOOP]] ]
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i3 [ [[IV_NEXT]], [[LOOP]] ]
 ; CHECK-NEXT:    call void @escape_outer.i3(i3 [[IV_RES]], i3 [[NBITS_RES]], i3 [[VAL_SHIFTED_RES]], i1 [[VAL_SHIFTED_ISZERO_RES]], i3 [[IV_NEXT_RES]])
 ; CHECK-NEXT:    ret i3 [[IV_RES]]