<div dir="ltr"><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Tue, May 25, 2021 at 1:31 PM Roman Lebedev via llvm-commits <<a href="mailto:llvm-commits@lists.llvm.org">llvm-commits@lists.llvm.org</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><br>
Author: Roman Lebedev<br>
Date: 2021-05-25T14:30:49+03:00<br>
New Revision: f1c5f78d3813584f7796f8b84b92fa0725964c17<br>
<br>
URL: <a href="https://github.com/llvm/llvm-project/commit/f1c5f78d3813584f7796f8b84b92fa0725964c17" rel="noreferrer" target="_blank">https://github.com/llvm/llvm-project/commit/f1c5f78d3813584f7796f8b84b92fa0725964c17</a><br>
DIFF: <a href="https://github.com/llvm/llvm-project/commit/f1c5f78d3813584f7796f8b84b92fa0725964c17.diff" rel="noreferrer" target="_blank">https://github.com/llvm/llvm-project/commit/f1c5f78d3813584f7796f8b84b92fa0725964c17.diff</a><br>
<br>
LOG: [LoopIdiom] Support 'arithmetic right-shift until zero' idiom<br>
<br>
This adds support for the "count active bits" pattern, i.e.:<br>
```<br>
int countActiveBits(signed val) {<br>
    int cnt = 0;<br>
    for( ; (val >> cnt) != 0; ++cnt)<br>
        ;<br>
    return cnt;<br>
}<br>
```<br>
but a somewhat more general one:<br>
```<br>
int countActiveBits(signed val, int start, int off) {<br>
    int cnt;<br>
    for (cnt = start; val >> (cnt + off); cnt++)<br>
        ;<br>
    return cnt;<br>
}<br>
```<br>
<br>
This directly matches the existing 'logical right-shift until zero' idiom.<br>
alive2 is happy with all the tests there.<br>
<br>
Note that, again, much like with the original unsigned case,<br>
we don't require the `val != 0` guard.<br>
<br>
The old `detectShiftUntilZeroIdiom()` already supports this pattern,<br>
the idea here is that the `val` must be positive (have at least one<br>
leading zero), because otherwise the loop is non-terminating,<br>
but since it is not `while(1)`, that would have been UB.<br></blockquote><div><br></div><div>Just to double check, are you making a "must progress" assumption here (which nowadays is illegal without the presence of the attribute/metadata)? I'm missing wider context and just going by this one comment, so I'm probably misunderstanding.<br></div><div><br></div><div>Regards,<br></div><div>Nikita<br></div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">

Added: <br>
<br>
<br>
Modified: <br>
    llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp<br>
    llvm/test/Transforms/LoopIdiom/X86/arithmetic-right-shift-until-zero.ll<br>
<br>
Removed: <br>
<br>
<br>
<br>
################################################################################<br>
diff  --git a/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp b/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp<br>
index ae8ef00913c4..d02db245cb88 100644<br>
--- a/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp<br>
+++ b/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp<br>
@@ -2416,7 +2416,7 @@ bool LoopIdiomRecognize::recognizeShiftUntilBitTest() {<br>
 ///   loop:<br>
 ///     %iv = phi i8 [ %start, %entry ], [ %iv.next, %for.cond ]<br>
 ///     %nbits = add nsw i8 %iv, %extraoffset<br>
-///     %val.shifted = lshr i8 %val, %nbits<br>
+///     %val.shifted = [la]shr i8 %val, %nbits<br>
 ///     %val.shifted.iszero = icmp eq i8 %val.shifted, 0<br>
 ///     %iv.next = add i8 %iv, 1<br>
 ///     <...><br>
@@ -2469,8 +2469,8 @@ static bool detectShiftUntilZeroIdiom(Loop *CurLoop, ScalarEvolution *SE,<br>
<br>
   // Step 2: Check if the comparison's operand is in desirable form.<br>
   // FIXME: Val could be a one-input PHI node, which we should look past.<br>
-  if (!match(ValShifted, m_LShr(m_LoopInvariant(m_Value(Val), CurLoop),<br>
-                                m_Instruction(NBits)))) {<br>
+  if (!match(ValShifted, m_Shr(m_LoopInvariant(m_Value(Val), CurLoop),<br>
+                               m_Instruction(NBits)))) {<br>
     LLVM_DEBUG(dbgs() << DEBUG_TYPE " Bad comparisons value computation.\n");<br>
     return false;<br>
   }<br>
@@ -2541,7 +2541,7 @@ static bool detectShiftUntilZeroIdiom(Loop *CurLoop, ScalarEvolution *SE,<br>
 ///   loop:<br>
 ///     %iv = phi i8 [ %start, %entry ], [ %iv.next, %for.cond ]<br>
 ///     %nbits = add nsw i8 %iv, %extraoffset<br>
-///     %val.shifted = lshr i8 %val, %nbits<br>
+///     %val.shifted = [la]shr i8 %val, %nbits<br>
 ///     %val.shifted.iszero = icmp eq i8 %val.shifted, 0<br>
 ///     %iv.next = add i8 %iv, 1<br>
 ///     <...><br>
<br>
diff  --git a/llvm/test/Transforms/LoopIdiom/X86/arithmetic-right-shift-until-zero.ll b/llvm/test/Transforms/LoopIdiom/X86/arithmetic-right-shift-until-zero.ll<br>
index 5f25dfd68614..b1e14f9f299e 100644<br>
--- a/llvm/test/Transforms/LoopIdiom/X86/arithmetic-right-shift-until-zero.ll<br>
+++ b/llvm/test/Transforms/LoopIdiom/X86/arithmetic-right-shift-until-zero.ll<br>
@@ -11,20 +11,29 @@ declare i8 @gen.i8()<br>
 define i8 @p0(i8 %val, i8 %start, i8 %extraoffset) {<br>
 ; CHECK-LABEL: @p0(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i8 @llvm.ctlz.i8(i8 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i8 8, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i8 0, [[EXTRAOFFSET:%.*]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i8 [[VAL_NUMACTIVEBITS]], [[TMP0]]<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i8 @llvm.smax.i8(i8 [[VAL_NUMACTIVEBITS_OFFSET]], i8 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i8 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i8 [[LOOP_BACKEDGETAKENCOUNT]], 1<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i8 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i8 [[IV]], [[EXTRAOFFSET:%.*]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i8 [[VAL:%.*]], [[NBITS]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i8 [[VAL_SHIFTED]], 0<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i8 [[IV]], 1<br>
-; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i8 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i8 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i8 [[LOOP_IV]], 1<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i8 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i8 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i8 [[IV]], [[EXTRAOFFSET]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i8 [[VAL]], [[NBITS]]<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i8 [[IV]], 1<br>
+; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i8 [[IV_NEXT]])<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i8 [ [[NBITS]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i8 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i8 [ [[IV_NEXT]], [[LOOP]] ]<br>
 ; 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]])<br>
 ; CHECK-NEXT:    ret i8 [[IV_RES]]<br>
@@ -59,20 +68,29 @@ end:<br>
 define i8 @p1(i8 %val, i8 %start, i8 %extraoffset) {<br>
 ; CHECK-LABEL: @p1(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i8 @llvm.ctlz.i8(i8 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i8 8, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i8 0, [[EXTRAOFFSET:%.*]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i8 [[VAL_NUMACTIVEBITS]], [[TMP0]]<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i8 @llvm.smax.i8(i8 [[VAL_NUMACTIVEBITS_OFFSET]], i8 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i8 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i8 [[LOOP_BACKEDGETAKENCOUNT]], 1<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i8 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[NBITS:%.*]] = add nuw i8 [[IV]], [[EXTRAOFFSET:%.*]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i8 [[VAL:%.*]], [[NBITS]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i8 [[VAL_SHIFTED]], 0<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i8 [[IV]], 1<br>
-; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i8 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i8 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i8 [[LOOP_IV]], 1<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i8 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i8 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[NBITS:%.*]] = add nuw i8 [[IV]], [[EXTRAOFFSET]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i8 [[VAL]], [[NBITS]]<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i8 [[IV]], 1<br>
+; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i8 [[IV_NEXT]])<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i8 [ [[NBITS]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i8 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i8 [ [[IV_NEXT]], [[LOOP]] ]<br>
 ; 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]])<br>
 ; CHECK-NEXT:    ret i8 [[IV_RES]]<br>
@@ -107,20 +125,28 @@ end:<br>
 define i8 @p2(i8 %val, i8 %start, i8 %extraoffset) {<br>
 ; CHECK-LABEL: @p2(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i8 @llvm.ctlz.i8(i8 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i8 8, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i8 [[VAL_NUMACTIVEBITS]], [[EXTRAOFFSET:%.*]]<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i8 @llvm.smax.i8(i8 [[VAL_NUMACTIVEBITS_OFFSET]], i8 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i8 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i8 [[LOOP_BACKEDGETAKENCOUNT]], 1<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i8 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[NBITS:%.*]] = sub nsw i8 [[IV]], [[EXTRAOFFSET:%.*]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i8 [[VAL:%.*]], [[NBITS]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i8 [[VAL_SHIFTED]], 0<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i8 [[IV]], 1<br>
-; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i8 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i8 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i8 [[LOOP_IV]], 1<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i8 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i8 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[NBITS:%.*]] = sub nsw i8 [[IV]], [[EXTRAOFFSET]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i8 [[VAL]], [[NBITS]]<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i8 [[IV]], 1<br>
+; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i8 [[IV_NEXT]])<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i8 [ [[NBITS]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i8 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i8 [ [[IV_NEXT]], [[LOOP]] ]<br>
 ; 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]])<br>
 ; CHECK-NEXT:    ret i8 [[IV_RES]]<br>
@@ -299,18 +325,26 @@ end:<br>
 define i8 @p6(i8 %val, i8 %start) {<br>
 ; CHECK-LABEL: @p6(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i8 @llvm.ctlz.i8(i8 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i8 8, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nuw nsw i8 [[VAL_NUMACTIVEBITS]], 0<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i8 @llvm.smax.i8(i8 [[VAL_NUMACTIVEBITS_OFFSET]], i8 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nuw nsw i8 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i8 [[LOOP_BACKEDGETAKENCOUNT]], 1<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i8 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i8 [[VAL:%.*]], [[IV]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i8 [[VAL_SHIFTED]], 0<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i8 [[IV]], 1<br>
-; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[IV]], i8 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i8 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i8 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i8 [[LOOP_IV]], 1<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i8 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i8 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i8 [[VAL]], [[IV]]<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i8 [[IV]], 1<br>
+; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[IV]], i8 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i8 [[IV_NEXT]])<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i8 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i8 [ [[IV_NEXT]], [[LOOP]] ]<br>
 ; 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]])<br>
 ; CHECK-NEXT:    ret i8 [[IV_RES]]<br>
@@ -346,20 +380,29 @@ declare void @escape_outer.i7(i7, i7, i7, i1, i7)<br>
 define i7 @p7(i7 %val, i7 %start, i7 %extraoffset) {<br>
 ; CHECK-LABEL: @p7(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i7 @llvm.ctlz.i7(i7 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i7 7, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i7 0, [[EXTRAOFFSET:%.*]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i7 [[VAL_NUMACTIVEBITS]], [[TMP0]]<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i7 @llvm.smax.i7(i7 [[VAL_NUMACTIVEBITS_OFFSET]], i7 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i7 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i7 [[LOOP_BACKEDGETAKENCOUNT]], 1<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i7 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i7 [[IV]], [[EXTRAOFFSET:%.*]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i7 [[VAL:%.*]], [[NBITS]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i7 [[VAL_SHIFTED]], 0<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i7 [[IV]], 1<br>
-; CHECK-NEXT:    call void @escape_inner.i7(i7 [[IV]], i7 [[NBITS]], i7 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i7 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i7 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i7 [[LOOP_IV]], 1<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i7 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i7 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i7 [[IV]], [[EXTRAOFFSET]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i7 [[VAL]], [[NBITS]]<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i7 [[IV]], 1<br>
+; CHECK-NEXT:    call void @escape_inner.i7(i7 [[IV]], i7 [[NBITS]], i7 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i7 [[IV_NEXT]])<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i7 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i7 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i7 [ [[NBITS]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i7 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i7 [ [[IV_NEXT]], [[LOOP]] ]<br>
 ; 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]])<br>
 ; CHECK-NEXT:    ret i7 [[IV_RES]]<br>
@@ -442,17 +485,27 @@ end:<br>
 define i8 @t9(i8 %val, i8 %start, i8 %extraoffset) {<br>
 ; CHECK-LABEL: @t9(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i8 @llvm.ctlz.i8(i8 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i8 8, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i8 0, [[EXTRAOFFSET:%.*]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i8 [[VAL_NUMACTIVEBITS]], [[TMP0]]<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i8 @llvm.smax.i8(i8 [[VAL_NUMACTIVEBITS_OFFSET]], i8 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i8 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i8 [[LOOP_BACKEDGETAKENCOUNT]], 1<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i8 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i8 [[IV]], [[EXTRAOFFSET:%.*]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i8 [[VAL:%.*]], [[NBITS]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISNOTZERO:%.*]] = icmp ne i8 [[VAL_SHIFTED]], 0<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i8 [[IV]], 1<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i8 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i8 [[LOOP_IV]], 1<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i8 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISNOTZERO:%.*]] = xor i1 [[LOOP_IVCHECK]], true<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i8 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i8 [[IV]], [[EXTRAOFFSET]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i8 [[VAL]], [[NBITS]]<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i8 [[IV]], 1<br>
 ; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISNOTZERO]], i8 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISNOTZERO]], label [[LOOP]], label [[END:%.*]]<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i8 [ [[NBITS]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i8 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_ISNOTZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISNOTZERO]], [[LOOP]] ]<br>
@@ -783,20 +836,29 @@ end:<br>
 define i8 @t16(i8 %val, i8 %start, i8 %extraoffset) {<br>
 ; CHECK-LABEL: @t16(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i8 @llvm.ctlz.i8(i8 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i8 8, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i8 0, [[EXTRAOFFSET:%.*]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i8 [[VAL_NUMACTIVEBITS]], [[TMP0]]<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i8 @llvm.smax.i8(i8 [[VAL_NUMACTIVEBITS_OFFSET]], i8 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i8 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i8 [[LOOP_BACKEDGETAKENCOUNT]], 1<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i8 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i8 [[EXTRAOFFSET:%.*]], [[IV]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i8 [[VAL:%.*]], [[NBITS]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i8 [[VAL_SHIFTED]], 0<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i8 [[IV]], 1<br>
-; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i8 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i8 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i8 [[LOOP_IV]], 1<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i8 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i8 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i8 [[EXTRAOFFSET]], [[IV]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i8 [[VAL]], [[NBITS]]<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i8 [[IV]], 1<br>
+; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i8 [[IV_NEXT]])<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i8 [ [[NBITS]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i8 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i8 [ [[IV_NEXT]], [[LOOP]] ]<br>
 ; 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]])<br>
 ; CHECK-NEXT:    ret i8 [[IV_RES]]<br>
@@ -1130,22 +1192,31 @@ end:<br>
 define i8 @n23(i8 %val, i8 %start, i8 %extraoffset) {<br>
 ; CHECK-LABEL: @n23(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i8 @llvm.ctlz.i8(i8 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i8 8, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i8 0, [[EXTRAOFFSET:%.*]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i8 [[VAL_NUMACTIVEBITS]], [[TMP0]]<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i8 @llvm.smax.i8(i8 [[VAL_NUMACTIVEBITS_OFFSET]], i8 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i8 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i8 [[LOOP_BACKEDGETAKENCOUNT]], 1<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i8 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i8 [[IV]], [[EXTRAOFFSET:%.*]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i8 [[VAL:%.*]], [[NBITS]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i8 [[VAL_SHIFTED]], 0<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i8 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i8 [[LOOP_IV]], 1<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i8 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i8 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i8 [[IV]], [[EXTRAOFFSET]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i8 [[VAL]], [[NBITS]]<br>
 ; CHECK-NEXT:    [[NOT_IV_NEXT:%.*]] = add i8 [[IV]], 1<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i8 [[IV]], 1<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i8 [[IV]], 1<br>
 ; CHECK-NEXT:    [[ALSO_IV_NEXT:%.*]] = add i8 [[IV]], 1<br>
-; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i8 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]<br>
+; CHECK-NEXT:    call void @escape_inner(i8 [[IV]], i8 [[NBITS]], i8 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i8 [[IV_NEXT]])<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i8 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i8 [ [[NBITS]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i8 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i8 [ [[IV_NEXT]], [[LOOP]] ]<br>
 ; 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]])<br>
 ; CHECK-NEXT:    ret i8 [[IV_RES]]<br>
@@ -1238,18 +1309,26 @@ declare void @escape_outer.i3(i3, i3, i3, i1, i3)<br>
 define i1 @t25_nooffset_i1(i1 %val, i1 %start) {<br>
 ; CHECK-LABEL: @t25_nooffset_i1(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i1 @llvm.ctlz.i1(i1 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i1 true, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nuw nsw i1 [[VAL_NUMACTIVEBITS]], false<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i1 @llvm.smax.i1(i1 [[VAL_NUMACTIVEBITS_OFFSET]], i1 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nuw nsw i1 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i1 [[LOOP_BACKEDGETAKENCOUNT]], true<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i1 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i1 [[VAL:%.*]], [[IV]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i1 [[VAL_SHIFTED]], false<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i1 [[IV]], true<br>
-; CHECK-NEXT:    call void @escape_inner.i1(i1 [[IV]], i1 [[IV]], i1 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i1 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i1 [ false, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i1 [[LOOP_IV]], true<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i1 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i1 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i1 [[VAL]], [[IV]]<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i1 [[IV]], true<br>
+; CHECK-NEXT:    call void @escape_inner.i1(i1 [[IV]], i1 [[IV]], i1 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i1 [[IV_NEXT]])<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i1 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i1 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i1 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i1 [ [[IV_NEXT]], [[LOOP]] ]<br>
 ; 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]])<br>
 ; CHECK-NEXT:    ret i1 [[IV_RES]]<br>
@@ -1280,18 +1359,26 @@ end:<br>
 define i2 @t26_nooffset_i2(i2 %val, i2 %start) {<br>
 ; CHECK-LABEL: @t26_nooffset_i2(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i2 @llvm.ctlz.i2(i2 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw i2 -2, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nuw nsw i2 [[VAL_NUMACTIVEBITS]], 0<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i2 @llvm.smax.i2(i2 [[VAL_NUMACTIVEBITS_OFFSET]], i2 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nuw nsw i2 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw i2 [[LOOP_BACKEDGETAKENCOUNT]], 1<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i2 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i2 [[VAL:%.*]], [[IV]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i2 [[VAL_SHIFTED]], 0<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i2 [[IV]], 1<br>
-; CHECK-NEXT:    call void @escape_inner.i2(i2 [[IV]], i2 [[IV]], i2 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i2 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i2 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw i2 [[LOOP_IV]], 1<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i2 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i2 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i2 [[VAL]], [[IV]]<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i2 [[IV]], 1<br>
+; CHECK-NEXT:    call void @escape_inner.i2(i2 [[IV]], i2 [[IV]], i2 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i2 [[IV_NEXT]])<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i2 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i2 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i2 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i2 [ [[IV_NEXT]], [[LOOP]] ]<br>
 ; 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]])<br>
 ; CHECK-NEXT:    ret i2 [[IV_RES]]<br>
@@ -1322,18 +1409,26 @@ end:<br>
 define i3 @t27_nooffset_i3(i3 %val, i3 %start) {<br>
 ; CHECK-LABEL: @t27_nooffset_i3(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i3 @llvm.ctlz.i3(i3 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i3 3, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nuw nsw i3 [[VAL_NUMACTIVEBITS]], 0<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i3 @llvm.smax.i3(i3 [[VAL_NUMACTIVEBITS_OFFSET]], i3 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nuw nsw i3 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i3 [[LOOP_BACKEDGETAKENCOUNT]], 1<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i3 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i3 [[VAL:%.*]], [[IV]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i3 [[VAL_SHIFTED]], 0<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i3 [[IV]], 1<br>
-; CHECK-NEXT:    call void @escape_inner.i3(i3 [[IV]], i3 [[IV]], i3 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i3 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i3 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i3 [[LOOP_IV]], 1<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i3 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i3 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i3 [[VAL]], [[IV]]<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i3 [[IV]], 1<br>
+; CHECK-NEXT:    call void @escape_inner.i3(i3 [[IV]], i3 [[IV]], i3 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i3 [[IV_NEXT]])<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i3 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i3 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i3 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i3 [ [[IV_NEXT]], [[LOOP]] ]<br>
 ; 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]])<br>
 ; CHECK-NEXT:    ret i3 [[IV_RES]]<br>
@@ -1365,20 +1460,28 @@ end:<br>
 define i1 @t28_addnsw_i1(i1 %val, i1 %start, i1 %extraoffset) {<br>
 ; CHECK-LABEL: @t28_addnsw_i1(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i1 @llvm.ctlz.i1(i1 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i1 true, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i1 [[VAL_NUMACTIVEBITS]], [[EXTRAOFFSET:%.*]]<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i1 @llvm.smax.i1(i1 [[VAL_NUMACTIVEBITS_OFFSET]], i1 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i1 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i1 [[LOOP_BACKEDGETAKENCOUNT]], true<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i1 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i1 [[IV]], [[EXTRAOFFSET:%.*]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i1 [[VAL:%.*]], [[NBITS]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i1 [[VAL_SHIFTED]], false<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i1 [[IV]], true<br>
-; CHECK-NEXT:    call void @escape_inner.i1(i1 [[IV]], i1 [[NBITS]], i1 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i1 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i1 [ false, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i1 [[LOOP_IV]], true<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i1 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i1 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i1 [[IV]], [[EXTRAOFFSET]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i1 [[VAL]], [[NBITS]]<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i1 [[IV]], true<br>
+; CHECK-NEXT:    call void @escape_inner.i1(i1 [[IV]], i1 [[NBITS]], i1 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i1 [[IV_NEXT]])<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i1 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i1 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i1 [ [[NBITS]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i1 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i1 [ [[IV_NEXT]], [[LOOP]] ]<br>
 ; 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]])<br>
 ; CHECK-NEXT:    ret i1 [[IV_RES]]<br>
@@ -1411,20 +1514,29 @@ end:<br>
 define i2 @t29_addnsw_i2(i2 %val, i2 %start, i2 %extraoffset) {<br>
 ; CHECK-LABEL: @t29_addnsw_i2(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i2 @llvm.ctlz.i2(i2 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw i2 -2, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i2 0, [[EXTRAOFFSET:%.*]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i2 [[VAL_NUMACTIVEBITS]], [[TMP0]]<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i2 @llvm.smax.i2(i2 [[VAL_NUMACTIVEBITS_OFFSET]], i2 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i2 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw i2 [[LOOP_BACKEDGETAKENCOUNT]], 1<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i2 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i2 [[IV]], [[EXTRAOFFSET:%.*]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i2 [[VAL:%.*]], [[NBITS]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i2 [[VAL_SHIFTED]], 0<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i2 [[IV]], 1<br>
-; CHECK-NEXT:    call void @escape_inner.i2(i2 [[IV]], i2 [[NBITS]], i2 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i2 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i2 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw i2 [[LOOP_IV]], 1<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i2 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i2 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i2 [[IV]], [[EXTRAOFFSET]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i2 [[VAL]], [[NBITS]]<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i2 [[IV]], 1<br>
+; CHECK-NEXT:    call void @escape_inner.i2(i2 [[IV]], i2 [[NBITS]], i2 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i2 [[IV_NEXT]])<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i2 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i2 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i2 [ [[NBITS]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i2 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i2 [ [[IV_NEXT]], [[LOOP]] ]<br>
 ; 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]])<br>
 ; CHECK-NEXT:    ret i2 [[IV_RES]]<br>
@@ -1457,20 +1569,29 @@ end:<br>
 define i3 @t30_addnsw_i3(i3 %val, i3 %start, i3 %extraoffset) {<br>
 ; CHECK-LABEL: @t30_addnsw_i3(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i3 @llvm.ctlz.i3(i3 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i3 3, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i3 0, [[EXTRAOFFSET:%.*]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i3 [[VAL_NUMACTIVEBITS]], [[TMP0]]<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i3 @llvm.smax.i3(i3 [[VAL_NUMACTIVEBITS_OFFSET]], i3 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i3 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i3 [[LOOP_BACKEDGETAKENCOUNT]], 1<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i3 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i3 [[IV]], [[EXTRAOFFSET:%.*]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i3 [[VAL:%.*]], [[NBITS]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i3 [[VAL_SHIFTED]], 0<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i3 [[IV]], 1<br>
-; CHECK-NEXT:    call void @escape_inner.i3(i3 [[IV]], i3 [[NBITS]], i3 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i3 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i3 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i3 [[LOOP_IV]], 1<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i3 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i3 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[NBITS:%.*]] = add nsw i3 [[IV]], [[EXTRAOFFSET]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i3 [[VAL]], [[NBITS]]<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i3 [[IV]], 1<br>
+; CHECK-NEXT:    call void @escape_inner.i3(i3 [[IV]], i3 [[NBITS]], i3 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i3 [[IV_NEXT]])<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i3 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i3 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i3 [ [[NBITS]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i3 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i3 [ [[IV_NEXT]], [[LOOP]] ]<br>
 ; 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]])<br>
 ; CHECK-NEXT:    ret i3 [[IV_RES]]<br>
@@ -1504,20 +1625,28 @@ end:<br>
 define i1 @t31_addnuw_i1(i1 %val, i1 %start, i1 %extraoffset) {<br>
 ; CHECK-LABEL: @t31_addnuw_i1(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i1 @llvm.ctlz.i1(i1 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i1 true, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i1 [[VAL_NUMACTIVEBITS]], [[EXTRAOFFSET:%.*]]<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i1 @llvm.smax.i1(i1 [[VAL_NUMACTIVEBITS_OFFSET]], i1 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i1 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i1 [[LOOP_BACKEDGETAKENCOUNT]], true<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i1 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[NBITS:%.*]] = add nuw i1 [[IV]], [[EXTRAOFFSET:%.*]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i1 [[VAL:%.*]], [[NBITS]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i1 [[VAL_SHIFTED]], false<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i1 [[IV]], true<br>
-; CHECK-NEXT:    call void @escape_inner.i1(i1 [[IV]], i1 [[NBITS]], i1 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i1 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i1 [ false, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i1 [[LOOP_IV]], true<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i1 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i1 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[NBITS:%.*]] = add nuw i1 [[IV]], [[EXTRAOFFSET]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i1 [[VAL]], [[NBITS]]<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i1 [[IV]], true<br>
+; CHECK-NEXT:    call void @escape_inner.i1(i1 [[IV]], i1 [[NBITS]], i1 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i1 [[IV_NEXT]])<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i1 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i1 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i1 [ [[NBITS]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i1 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i1 [ [[IV_NEXT]], [[LOOP]] ]<br>
 ; 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]])<br>
 ; CHECK-NEXT:    ret i1 [[IV_RES]]<br>
@@ -1550,20 +1679,29 @@ end:<br>
 define i2 @t32_addnuw_i2(i2 %val, i2 %start, i2 %extraoffset) {<br>
 ; CHECK-LABEL: @t32_addnuw_i2(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i2 @llvm.ctlz.i2(i2 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw i2 -2, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i2 0, [[EXTRAOFFSET:%.*]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i2 [[VAL_NUMACTIVEBITS]], [[TMP0]]<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i2 @llvm.smax.i2(i2 [[VAL_NUMACTIVEBITS_OFFSET]], i2 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i2 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw i2 [[LOOP_BACKEDGETAKENCOUNT]], 1<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i2 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[NBITS:%.*]] = add nuw i2 [[IV]], [[EXTRAOFFSET:%.*]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i2 [[VAL:%.*]], [[NBITS]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i2 [[VAL_SHIFTED]], 0<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i2 [[IV]], 1<br>
-; CHECK-NEXT:    call void @escape_inner.i2(i2 [[IV]], i2 [[NBITS]], i2 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i2 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i2 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw i2 [[LOOP_IV]], 1<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i2 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i2 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[NBITS:%.*]] = add nuw i2 [[IV]], [[EXTRAOFFSET]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i2 [[VAL]], [[NBITS]]<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i2 [[IV]], 1<br>
+; CHECK-NEXT:    call void @escape_inner.i2(i2 [[IV]], i2 [[NBITS]], i2 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i2 [[IV_NEXT]])<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i2 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i2 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i2 [ [[NBITS]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i2 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i2 [ [[IV_NEXT]], [[LOOP]] ]<br>
 ; 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]])<br>
 ; CHECK-NEXT:    ret i2 [[IV_RES]]<br>
@@ -1596,20 +1734,29 @@ end:<br>
 define i3 @t33_addnuw_i3(i3 %val, i3 %start, i3 %extraoffset) {<br>
 ; CHECK-LABEL: @t33_addnuw_i3(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i3 @llvm.ctlz.i3(i3 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i3 3, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i3 0, [[EXTRAOFFSET:%.*]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i3 [[VAL_NUMACTIVEBITS]], [[TMP0]]<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i3 @llvm.smax.i3(i3 [[VAL_NUMACTIVEBITS_OFFSET]], i3 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i3 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i3 [[LOOP_BACKEDGETAKENCOUNT]], 1<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i3 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[NBITS:%.*]] = add nuw i3 [[IV]], [[EXTRAOFFSET:%.*]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i3 [[VAL:%.*]], [[NBITS]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i3 [[VAL_SHIFTED]], 0<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i3 [[IV]], 1<br>
-; CHECK-NEXT:    call void @escape_inner.i3(i3 [[IV]], i3 [[NBITS]], i3 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i3 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i3 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i3 [[LOOP_IV]], 1<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i3 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i3 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[NBITS:%.*]] = add nuw i3 [[IV]], [[EXTRAOFFSET]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i3 [[VAL]], [[NBITS]]<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i3 [[IV]], 1<br>
+; CHECK-NEXT:    call void @escape_inner.i3(i3 [[IV]], i3 [[NBITS]], i3 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i3 [[IV_NEXT]])<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i3 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i3 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i3 [ [[NBITS]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i3 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i3 [ [[IV_NEXT]], [[LOOP]] ]<br>
 ; 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]])<br>
 ; CHECK-NEXT:    ret i3 [[IV_RES]]<br>
@@ -1644,20 +1791,28 @@ end:<br>
 define i1 @t34_subnsw_i1(i1 %val, i1 %start, i1 %extraoffset) {<br>
 ; CHECK-LABEL: @t34_subnsw_i1(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i1 @llvm.ctlz.i1(i1 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i1 true, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i1 [[VAL_NUMACTIVEBITS]], [[EXTRAOFFSET:%.*]]<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i1 @llvm.smax.i1(i1 [[VAL_NUMACTIVEBITS_OFFSET]], i1 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i1 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i1 [[LOOP_BACKEDGETAKENCOUNT]], true<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i1 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[NBITS:%.*]] = sub nsw i1 [[IV]], [[EXTRAOFFSET:%.*]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i1 [[VAL:%.*]], [[NBITS]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i1 [[VAL_SHIFTED]], false<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i1 [[IV]], true<br>
-; CHECK-NEXT:    call void @escape_inner.i1(i1 [[IV]], i1 [[NBITS]], i1 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i1 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i1 [ false, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i1 [[LOOP_IV]], true<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i1 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i1 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[NBITS:%.*]] = sub nsw i1 [[IV]], [[EXTRAOFFSET]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i1 [[VAL]], [[NBITS]]<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i1 [[IV]], true<br>
+; CHECK-NEXT:    call void @escape_inner.i1(i1 [[IV]], i1 [[NBITS]], i1 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i1 [[IV_NEXT]])<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i1 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i1 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i1 [ [[NBITS]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i1 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i1 [ [[IV_NEXT]], [[LOOP]] ]<br>
 ; 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]])<br>
 ; CHECK-NEXT:    ret i1 [[IV_RES]]<br>
@@ -1690,20 +1845,28 @@ end:<br>
 define i2 @t35_addnuw_i2(i2 %val, i2 %start, i2 %extraoffset) {<br>
 ; CHECK-LABEL: @t35_addnuw_i2(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i2 @llvm.ctlz.i2(i2 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw i2 -2, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i2 [[VAL_NUMACTIVEBITS]], [[EXTRAOFFSET:%.*]]<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i2 @llvm.smax.i2(i2 [[VAL_NUMACTIVEBITS_OFFSET]], i2 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i2 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw i2 [[LOOP_BACKEDGETAKENCOUNT]], 1<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i2 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[NBITS:%.*]] = sub nsw i2 [[IV]], [[EXTRAOFFSET:%.*]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i2 [[VAL:%.*]], [[NBITS]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i2 [[VAL_SHIFTED]], 0<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i2 [[IV]], 1<br>
-; CHECK-NEXT:    call void @escape_inner.i2(i2 [[IV]], i2 [[NBITS]], i2 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i2 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i2 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw i2 [[LOOP_IV]], 1<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i2 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i2 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[NBITS:%.*]] = sub nsw i2 [[IV]], [[EXTRAOFFSET]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i2 [[VAL]], [[NBITS]]<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i2 [[IV]], 1<br>
+; CHECK-NEXT:    call void @escape_inner.i2(i2 [[IV]], i2 [[NBITS]], i2 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i2 [[IV_NEXT]])<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i2 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i2 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i2 [ [[NBITS]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i2 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i2 [ [[IV_NEXT]], [[LOOP]] ]<br>
 ; 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]])<br>
 ; CHECK-NEXT:    ret i2 [[IV_RES]]<br>
@@ -1736,20 +1899,28 @@ end:<br>
 define i3 @t36_addnuw_i3(i3 %val, i3 %start, i3 %extraoffset) {<br>
 ; CHECK-LABEL: @t36_addnuw_i3(<br>
 ; CHECK-NEXT:  entry:<br>
+; CHECK-NEXT:    [[VAL_NUMLEADINGZEROS:%.*]] = call i3 @llvm.ctlz.i3(i3 [[VAL:%.*]], i1 false)<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS:%.*]] = sub nuw nsw i3 3, [[VAL_NUMLEADINGZEROS]]<br>
+; CHECK-NEXT:    [[VAL_NUMACTIVEBITS_OFFSET:%.*]] = add nsw i3 [[VAL_NUMACTIVEBITS]], [[EXTRAOFFSET:%.*]]<br>
+; CHECK-NEXT:    [[IV_FINAL:%.*]] = call i3 @llvm.smax.i3(i3 [[VAL_NUMACTIVEBITS_OFFSET]], i3 [[START:%.*]])<br>
+; CHECK-NEXT:    [[LOOP_BACKEDGETAKENCOUNT:%.*]] = sub nsw i3 [[IV_FINAL]], [[START]]<br>
+; CHECK-NEXT:    [[LOOP_TRIPCOUNT:%.*]] = add nuw nsw i3 [[LOOP_BACKEDGETAKENCOUNT]], 1<br>
 ; CHECK-NEXT:    br label [[LOOP:%.*]]<br>
 ; CHECK:       loop:<br>
-; CHECK-NEXT:    [[IV:%.*]] = phi i3 [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[NBITS:%.*]] = sub nsw i3 [[IV]], [[EXTRAOFFSET:%.*]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i3 [[VAL:%.*]], [[NBITS]]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO:%.*]] = icmp eq i3 [[VAL_SHIFTED]], 0<br>
-; CHECK-NEXT:    [[IV_NEXT]] = add i3 [[IV]], 1<br>
-; CHECK-NEXT:    call void @escape_inner.i3(i3 [[IV]], i3 [[NBITS]], i3 [[VAL_SHIFTED]], i1 [[VAL_SHIFTED_ISZERO]], i3 [[IV_NEXT]])<br>
-; CHECK-NEXT:    br i1 [[VAL_SHIFTED_ISZERO]], label [[END:%.*]], label [[LOOP]]<br>
+; CHECK-NEXT:    [[LOOP_IV:%.*]] = phi i3 [ 0, [[ENTRY:%.*]] ], [ [[LOOP_IV_NEXT:%.*]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[LOOP_IV_NEXT]] = add nuw nsw i3 [[LOOP_IV]], 1<br>
+; CHECK-NEXT:    [[LOOP_IVCHECK:%.*]] = icmp eq i3 [[LOOP_IV_NEXT]], [[LOOP_TRIPCOUNT]]<br>
+; CHECK-NEXT:    [[IV:%.*]] = add nsw i3 [[LOOP_IV]], [[START]]<br>
+; CHECK-NEXT:    [[NBITS:%.*]] = sub nsw i3 [[IV]], [[EXTRAOFFSET]]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED:%.*]] = ashr i3 [[VAL]], [[NBITS]]<br>
+; CHECK-NEXT:    [[IV_NEXT:%.*]] = add i3 [[IV]], 1<br>
+; CHECK-NEXT:    call void @escape_inner.i3(i3 [[IV]], i3 [[NBITS]], i3 [[VAL_SHIFTED]], i1 [[LOOP_IVCHECK]], i3 [[IV_NEXT]])<br>
+; CHECK-NEXT:    br i1 [[LOOP_IVCHECK]], label [[END:%.*]], label [[LOOP]]<br>
 ; CHECK:       end:<br>
-; CHECK-NEXT:    [[IV_RES:%.*]] = phi i3 [ [[IV]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[IV_RES:%.*]] = phi i3 [ [[IV_FINAL]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[NBITS_RES:%.*]] = phi i3 [ [[NBITS]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[VAL_SHIFTED_RES:%.*]] = phi i3 [ [[VAL_SHIFTED]], [[LOOP]] ]<br>
-; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[VAL_SHIFTED_ISZERO]], [[LOOP]] ]<br>
+; CHECK-NEXT:    [[VAL_SHIFTED_ISZERO_RES:%.*]] = phi i1 [ [[LOOP_IVCHECK]], [[LOOP]] ]<br>
 ; CHECK-NEXT:    [[IV_NEXT_RES:%.*]] = phi i3 [ [[IV_NEXT]], [[LOOP]] ]<br>
 ; 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]])<br>
 ; CHECK-NEXT:    ret i3 [[IV_RES]]<br>
<br>
<br>
<br>
_______________________________________________<br>
llvm-commits mailing list<br>
<a href="mailto:llvm-commits@lists.llvm.org" target="_blank">llvm-commits@lists.llvm.org</a><br>
<a href="https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits" rel="noreferrer" target="_blank">https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits</a><br>
</blockquote></div></div>