[llvm] f1c5f78 - [LoopIdiom] Support 'arithmetic right-shift until zero' idiom
Roman Lebedev via llvm-commits
llvm-commits at lists.llvm.org
Tue May 25 04:31:12 PDT 2021
Author: Roman Lebedev
Date: 2021-05-25T14:30:49+03:00
New Revision: f1c5f78d3813584f7796f8b84b92fa0725964c17
URL: https://github.com/llvm/llvm-project/commit/f1c5f78d3813584f7796f8b84b92fa0725964c17
DIFF: https://github.com/llvm/llvm-project/commit/f1c5f78d3813584f7796f8b84b92fa0725964c17.diff
LOG: [LoopIdiom] Support 'arithmetic right-shift until zero' idiom
This adds support for the "count active bits" pattern, i.e.:
```
int countActiveBits(signed val) {
int cnt = 0;
for( ; (val >> cnt) != 0; ++cnt)
;
return cnt;
}
```
but a somewhat more general one:
```
int countActiveBits(signed val, int start, int off) {
int cnt;
for (cnt = start; val >> (cnt + off); cnt++)
;
return cnt;
}
```
This directly matches the existing 'logical right-shift until zero' idiom.
alive2 is happy with all the tests there.
Note that, again, much like with the original unsigned case,
we don't require the `val != 0` guard.
The old `detectShiftUntilZeroIdiom()` already supports this pattern,
the idea here is that the `val` must be positive (have at least one
leading zero), because otherwise the loop is non-terminating,
but since it is not `while(1)`, that would have been UB.
Added:
Modified:
llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
llvm/test/Transforms/LoopIdiom/X86/arithmetic-right-shift-until-zero.ll
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp b/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
index ae8ef00913c4..d02db245cb88 100644
--- a/llvm/lib/Transforms/Scalar/LoopIdiomRecognize.cpp
+++ b/llvm/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 = lshr i8 %val, %nbits
+/// %val.shifted = [la]shr i8 %val, %nbits
/// %val.shifted.iszero = icmp eq i8 %val.shifted, 0
/// %iv.next = add i8 %iv, 1
/// <...>
@@ -2469,8 +2469,8 @@ 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_LShr(m_LoopInvariant(m_Value(Val), CurLoop),
- m_Instruction(NBits)))) {
+ if (!match(ValShifted, m_Shr(m_LoopInvariant(m_Value(Val), CurLoop),
+ m_Instruction(NBits)))) {
LLVM_DEBUG(dbgs() << DEBUG_TYPE " Bad comparisons value computation.\n");
return false;
}
@@ -2541,7 +2541,7 @@ 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 = lshr i8 %val, %nbits
+/// %val.shifted = [la]shr i8 %val, %nbits
/// %val.shifted.iszero = icmp eq i8 %val.shifted, 0
/// %iv.next = add i8 %iv, 1
/// <...>
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
index 5f25dfd68614..b1e14f9f299e 100644
--- a/llvm/test/Transforms/LoopIdiom/X86/arithmetic-right-shift-until-zero.ll
+++ b/llvm/test/Transforms/LoopIdiom/X86/arithmetic-right-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.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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]] ]
@@ -783,20 +836,29 @@ end:
define i8 @t16(i8 %val, i8 %start, i8 %extraoffset) {
; CHECK-LABEL: @t16(
; CHECK-NEXT: entry:
+; CHECK-NEXT: [[VAL_NUMLEADINGZEROS:%.*]] = call i8 @llvm.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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]]
@@ -1130,22 +1192,31 @@ end:
define i8 @n23(i8 %val, i8 %start, i8 %extraoffset) {
; CHECK-LABEL: @n23(
; CHECK-NEXT: entry:
+; CHECK-NEXT: [[VAL_NUMLEADINGZEROS:%.*]] = call i8 @llvm.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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]]
@@ -1238,18 +1309,26 @@ declare void @escape_outer.i3(i3, i3, i3, i1, i3)
define i1 @t25_nooffset_i1(i1 %val, i1 %start) {
; CHECK-LABEL: @t25_nooffset_i1(
; CHECK-NEXT: entry:
+; CHECK-NEXT: [[VAL_NUMLEADINGZEROS:%.*]] = call i1 @llvm.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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]]
@@ -1280,18 +1359,26 @@ end:
define i2 @t26_nooffset_i2(i2 %val, i2 %start) {
; CHECK-LABEL: @t26_nooffset_i2(
; CHECK-NEXT: entry:
+; CHECK-NEXT: [[VAL_NUMLEADINGZEROS:%.*]] = call i2 @llvm.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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]]
@@ -1322,18 +1409,26 @@ end:
define i3 @t27_nooffset_i3(i3 %val, i3 %start) {
; CHECK-LABEL: @t27_nooffset_i3(
; CHECK-NEXT: entry:
+; CHECK-NEXT: [[VAL_NUMLEADINGZEROS:%.*]] = call i3 @llvm.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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]]
@@ -1365,20 +1460,28 @@ end:
define i1 @t28_addnsw_i1(i1 %val, i1 %start, i1 %extraoffset) {
; CHECK-LABEL: @t28_addnsw_i1(
; CHECK-NEXT: entry:
+; CHECK-NEXT: [[VAL_NUMLEADINGZEROS:%.*]] = call i1 @llvm.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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]]
@@ -1411,20 +1514,29 @@ end:
define i2 @t29_addnsw_i2(i2 %val, i2 %start, i2 %extraoffset) {
; CHECK-LABEL: @t29_addnsw_i2(
; CHECK-NEXT: entry:
+; CHECK-NEXT: [[VAL_NUMLEADINGZEROS:%.*]] = call i2 @llvm.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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]]
@@ -1457,20 +1569,29 @@ end:
define i3 @t30_addnsw_i3(i3 %val, i3 %start, i3 %extraoffset) {
; CHECK-LABEL: @t30_addnsw_i3(
; CHECK-NEXT: entry:
+; CHECK-NEXT: [[VAL_NUMLEADINGZEROS:%.*]] = call i3 @llvm.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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]]
@@ -1504,20 +1625,28 @@ end:
define i1 @t31_addnuw_i1(i1 %val, i1 %start, i1 %extraoffset) {
; CHECK-LABEL: @t31_addnuw_i1(
; CHECK-NEXT: entry:
+; CHECK-NEXT: [[VAL_NUMLEADINGZEROS:%.*]] = call i1 @llvm.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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]]
@@ -1550,20 +1679,29 @@ end:
define i2 @t32_addnuw_i2(i2 %val, i2 %start, i2 %extraoffset) {
; CHECK-LABEL: @t32_addnuw_i2(
; CHECK-NEXT: entry:
+; CHECK-NEXT: [[VAL_NUMLEADINGZEROS:%.*]] = call i2 @llvm.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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]]
@@ -1596,20 +1734,29 @@ end:
define i3 @t33_addnuw_i3(i3 %val, i3 %start, i3 %extraoffset) {
; CHECK-LABEL: @t33_addnuw_i3(
; CHECK-NEXT: entry:
+; CHECK-NEXT: [[VAL_NUMLEADINGZEROS:%.*]] = call i3 @llvm.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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]]
@@ -1644,20 +1791,28 @@ end:
define i1 @t34_subnsw_i1(i1 %val, i1 %start, i1 %extraoffset) {
; CHECK-LABEL: @t34_subnsw_i1(
; CHECK-NEXT: entry:
+; CHECK-NEXT: [[VAL_NUMLEADINGZEROS:%.*]] = call i1 @llvm.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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]]
@@ -1690,20 +1845,28 @@ end:
define i2 @t35_addnuw_i2(i2 %val, i2 %start, i2 %extraoffset) {
; CHECK-LABEL: @t35_addnuw_i2(
; CHECK-NEXT: entry:
+; CHECK-NEXT: [[VAL_NUMLEADINGZEROS:%.*]] = call i2 @llvm.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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]]
@@ -1736,20 +1899,28 @@ end:
define i3 @t36_addnuw_i3(i3 %val, i3 %start, i3 %extraoffset) {
; CHECK-LABEL: @t36_addnuw_i3(
; CHECK-NEXT: entry:
+; CHECK-NEXT: [[VAL_NUMLEADINGZEROS:%.*]] = call i3 @llvm.ctlz.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:%.*]] = ashr 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:%.*]] = ashr 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]]
More information about the llvm-commits
mailing list