[LoopUnroll] Fold add chains during unrolling

Loop unrolling tends to produce chains of
`%x1 = add %x0, 1; %x2 = add %x1, 1; ...` with one add per unrolled
iteration. This patch simplifies these adds to `%xN = add %x0, N`
directly during unrolling, rather than waiting for InstCombine to do so.

The motivation for this is that having a single add (rather than
an add chain) on the induction variable makes it a simple recurrence,
which we specially recognize in a number of places. This allows
InstCombine to directly perform folds with that knowledge, instead
of first folding the add chains, and then doing other folds in another
InstCombine iteration.

Due to the reduced number of InstCombine iterations, this also
results in a small compile-time improvement.

Differential Revision: https://reviews.llvm.org/D153540

diff --git a/llvm/lib/Transforms/Utils/LoopUnroll.cpp b/llvm/lib/Transforms/Utils/LoopUnroll.cpp
index 29661cd..511dd61 100644 (file)
--- a/llvm/lib/Transforms/Utils/LoopUnroll.cpp
+++ b/llvm/lib/Transforms/Utils/LoopUnroll.cpp
@@ -45,6 +45,7 @@
 #include "llvm/IR/IntrinsicInst.h"
 #include "llvm/IR/Metadata.h"
 #include "llvm/IR/Module.h"
+#include "llvm/IR/PatternMatch.h"
 #include "llvm/IR/Use.h"
 #include "llvm/IR/User.h"
 #include "llvm/IR/ValueHandle.h"
@@ -216,6 +217,8 @@ void llvm::simplifyLoopAfterUnroll(Loop *L, bool SimplifyIVs, LoopInfo *LI,
                                    ScalarEvolution *SE, DominatorTree *DT,
                                    AssumptionCache *AC,
                                    const TargetTransformInfo *TTI) {
+  using namespace llvm::PatternMatch;
+
   // Simplify any new induction variables in the partially unrolled loop.
   if (SE && SimplifyIVs) {
     SmallVector<WeakTrackingVH, 16> DeadInsts;
@@ -241,6 +244,30 @@ void llvm::simplifyLoopAfterUnroll(Loop *L, bool SimplifyIVs, LoopInfo *LI,
           Inst.replaceAllUsesWith(V);
       if (isInstructionTriviallyDead(&Inst))
         DeadInsts.emplace_back(&Inst);
+
+      // Fold ((add X, C1), C2) to (add X, C1+C2). This is very common in
+      // unrolled loops, and handling this early allows following code to
+      // identify the IV as a "simple recurrence" without first folding away
+      // a long chain of adds.
+      {
+        Value *X;
+        const APInt *C1, *C2;
+        if (match(&Inst, m_Add(m_Add(m_Value(X), m_APInt(C1)), m_APInt(C2)))) {
+          auto *InnerI = dyn_cast<Instruction>(Inst.getOperand(0));
+          auto *InnerOBO = cast<OverflowingBinaryOperator>(Inst.getOperand(0));
+          bool SignedOverflow;
+          APInt NewC = C1->sadd_ov(*C2, SignedOverflow);
+          Inst.setOperand(0, X);
+          Inst.setOperand(1, ConstantInt::get(Inst.getType(), NewC));
+          Inst.setHasNoUnsignedWrap(Inst.hasNoUnsignedWrap() &&
+                                    InnerOBO->hasNoUnsignedWrap());
+          Inst.setHasNoSignedWrap(Inst.hasNoSignedWrap() &&
+                                  InnerOBO->hasNoSignedWrap() &&
+                                  !SignedOverflow);
+          if (InnerI && isInstructionTriviallyDead(InnerI))
+            DeadInsts.emplace_back(InnerI);
+        }
+      }
     }
     // We can't do recursive deletion until we're done iterating, as we might
     // have a phi which (potentially indirectly) uses instructions later in

diff --git a/llvm/test/Analysis/ScalarEvolution/2012-05-29-MulAddRec.ll b/llvm/test/Analysis/ScalarEvolution/2012-05-29-MulAddRec.ll
index a41781a..1d735d9 100644 (file)
--- a/llvm/test/Analysis/ScalarEvolution/2012-05-29-MulAddRec.ll
+++ b/llvm/test/Analysis/ScalarEvolution/2012-05-29-MulAddRec.ll
@@ -18,7 +18,7 @@ declare void @use(i8 %x)
 ; CHECK: br label %for.body
 
 ; CHECK: for.body:
-; CHECK: %inc.9 = add i8 %inc.8, 1
+; CHECK: %inc.9 = add i8 %inc1, 10
 ; CHECK: %0 = add i8 %inc1, 10
 ; CHECK: br label %for.cond
 

diff --git a/llvm/test/DebugInfo/unrolled-loop-remainder.ll b/llvm/test/DebugInfo/unrolled-loop-remainder.ll
index c4e9796..6a7fe77 100644 (file)
--- a/llvm/test/DebugInfo/unrolled-loop-remainder.ll
+++ b/llvm/test/DebugInfo/unrolled-loop-remainder.ll
@@ -42,7 +42,7 @@ define i32 @func_c() local_unnamed_addr #0 !dbg !14 {
 ; CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[ARRAYIDX_PROL_1]], align 4, !dbg [[DBG28]], !tbaa [[TBAA20]]
 ; CHECK-NEXT:    [[CONV_PROL_1:%.*]] = sext i32 [[TMP7]] to i64, !dbg [[DBG28]]
 ; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[CONV_PROL_1]] to ptr, !dbg [[DBG28]]
-; CHECK-NEXT:    [[ADD_PROL_1:%.*]] = add nsw i32 [[ADD_PROL]], 2, !dbg [[DBG29]]
+; CHECK-NEXT:    [[ADD_PROL_1:%.*]] = add nsw i32 [[DOTPR]], 4, !dbg [[DBG29]]
 ; CHECK-NEXT:    [[PROL_ITER_CMP_1:%.*]] = icmp ne i32 2, [[XTRAITER]], !dbg [[DBG24]]
 ; CHECK-NEXT:    br i1 [[PROL_ITER_CMP_1]], label [[FOR_BODY_PROL_2:%.*]], label [[FOR_BODY_PROL_LOOPEXIT_UNR_LCSSA]], !dbg [[DBG24]]
 ; CHECK:       for.body.prol.2:
@@ -50,7 +50,7 @@ define i32 @func_c() local_unnamed_addr #0 !dbg !14 {
 ; CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[ARRAYIDX_PROL_2]], align 4, !dbg [[DBG28]], !tbaa [[TBAA20]]
 ; CHECK-NEXT:    [[CONV_PROL_2:%.*]] = sext i32 [[TMP9]] to i64, !dbg [[DBG28]]
 ; CHECK-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[CONV_PROL_2]] to ptr, !dbg [[DBG28]]
-; CHECK-NEXT:    [[ADD_PROL_2:%.*]] = add nsw i32 [[ADD_PROL_1]], 2, !dbg [[DBG29]]
+; CHECK-NEXT:    [[ADD_PROL_2:%.*]] = add nsw i32 [[DOTPR]], 6, !dbg [[DBG29]]
 ; CHECK-NEXT:    br label [[FOR_BODY_PROL_LOOPEXIT_UNR_LCSSA]]
 ; CHECK:       for.body.prol.loopexit.unr-lcssa:
 ; CHECK-NEXT:    [[DOTLCSSA_UNR_PH:%.*]] = phi ptr [ [[TMP6]], [[FOR_BODY_PROL]] ], [ [[TMP8]], [[FOR_BODY_PROL_1]] ], [ [[TMP10]], [[FOR_BODY_PROL_2]] ]
@@ -72,22 +72,19 @@ define i32 @func_c() local_unnamed_addr #0 !dbg !14 {
 ; CHECK-NEXT:    [[TMP14:%.*]] = load i32, ptr [[ARRAYIDX]], align 4, !dbg [[DBG28]], !tbaa [[TBAA20]]
 ; CHECK-NEXT:    [[CONV:%.*]] = sext i32 [[TMP14]] to i64, !dbg [[DBG28]]
 ; CHECK-NEXT:    [[TMP15:%.*]] = inttoptr i64 [[CONV]] to ptr, !dbg [[DBG28]]
-; CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP13]], 2, !dbg [[DBG29]]
 ; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, ptr [[TMP15]], i64 1, !dbg [[DBG28]]
 ; CHECK-NEXT:    [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX_1]], align 4, !dbg [[DBG28]], !tbaa [[TBAA20]]
 ; CHECK-NEXT:    [[CONV_1:%.*]] = sext i32 [[TMP16]] to i64, !dbg [[DBG28]]
 ; CHECK-NEXT:    [[TMP17:%.*]] = inttoptr i64 [[CONV_1]] to ptr, !dbg [[DBG28]]
-; CHECK-NEXT:    [[ADD_1:%.*]] = add nsw i32 [[ADD]], 2, !dbg [[DBG29]]
 ; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, ptr [[TMP17]], i64 1, !dbg [[DBG28]]
 ; CHECK-NEXT:    [[TMP18:%.*]] = load i32, ptr [[ARRAYIDX_2]], align 4, !dbg [[DBG28]], !tbaa [[TBAA20]]
 ; CHECK-NEXT:    [[CONV_2:%.*]] = sext i32 [[TMP18]] to i64, !dbg [[DBG28]]
 ; CHECK-NEXT:    [[TMP19:%.*]] = inttoptr i64 [[CONV_2]] to ptr, !dbg [[DBG28]]
-; CHECK-NEXT:    [[ADD_2:%.*]] = add nsw i32 [[ADD_1]], 2, !dbg [[DBG29]]
 ; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, ptr [[TMP19]], i64 1, !dbg [[DBG28]]
 ; CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[ARRAYIDX_3]], align 4, !dbg [[DBG28]], !tbaa [[TBAA20]]
 ; CHECK-NEXT:    [[CONV_3:%.*]] = sext i32 [[TMP20]] to i64, !dbg [[DBG28]]
 ; CHECK-NEXT:    [[TMP21]] = inttoptr i64 [[CONV_3]] to ptr, !dbg [[DBG28]]
-; CHECK-NEXT:    [[ADD_3]] = add nsw i32 [[ADD_2]], 2, !dbg [[DBG29]]
+; CHECK-NEXT:    [[ADD_3]] = add nsw i32 [[TMP13]], 8, !dbg [[DBG29]]
 ; CHECK-NEXT:    [[TOBOOL_3:%.*]] = icmp eq i32 [[ADD_3]], 0, !dbg [[DBG24]]
 ; CHECK-NEXT:    br i1 [[TOBOOL_3]], label [[FOR_COND_FOR_END_CRIT_EDGE_UNR_LCSSA:%.*]], label [[FOR_BODY]], !dbg [[DBG24]], !llvm.loop [[LOOP30:![0-9]+]]
 ; CHECK:       for.cond.for.end_crit_edge.unr-lcssa:

diff --git a/llvm/test/Transforms/LoopUnroll/2011-08-08-PhiUpdate.ll b/llvm/test/Transforms/LoopUnroll/2011-08-08-PhiUpdate.ll
index d7365d7..a965bc2 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/2011-08-08-PhiUpdate.ll
+++ b/llvm/test/Transforms/LoopUnroll/2011-08-08-PhiUpdate.ll
@@ -83,7 +83,7 @@ define i32 @test2(ptr nocapture %p, i32 %n) nounwind readonly {
 ; CHECK-NEXT:    [[TMP4:%.*]] = add nsw i32 [[TMP3]], [[TMP2]]
 ; CHECK-NEXT:    br label [[BB1_1:%.*]]
 ; CHECK:       bb1.1:
-; CHECK-NEXT:    [[INDVAR_NEXT_1:%.*]] = add nuw nsw i64 [[INDVAR_NEXT]], 1
+; CHECK-NEXT:    [[INDVAR_NEXT_1:%.*]] = add nuw nsw i64 [[INDVAR]], 2
 ; CHECK-NEXT:    [[EXITCOND_1:%.*]] = icmp ne i64 [[INDVAR_NEXT_1]], [[TMP]]
 ; CHECK-NEXT:    br i1 [[EXITCOND_1]], label [[BB_2:%.*]], label [[BB1_BB2_CRIT_EDGE]]
 ; CHECK:       bb.2:
@@ -92,7 +92,7 @@ define i32 @test2(ptr nocapture %p, i32 %n) nounwind readonly {
 ; CHECK-NEXT:    [[TMP6:%.*]] = add nsw i32 [[TMP5]], [[TMP4]]
 ; CHECK-NEXT:    br label [[BB1_2:%.*]]
 ; CHECK:       bb1.2:
-; CHECK-NEXT:    [[INDVAR_NEXT_2:%.*]] = add nuw nsw i64 [[INDVAR_NEXT_1]], 1
+; CHECK-NEXT:    [[INDVAR_NEXT_2:%.*]] = add nuw nsw i64 [[INDVAR]], 3
 ; CHECK-NEXT:    [[EXITCOND_2:%.*]] = icmp ne i64 [[INDVAR_NEXT_2]], [[TMP]]
 ; CHECK-NEXT:    br i1 [[EXITCOND_2]], label [[BB_3:%.*]], label [[BB1_BB2_CRIT_EDGE]]
 ; CHECK:       bb.3:
@@ -101,7 +101,7 @@ define i32 @test2(ptr nocapture %p, i32 %n) nounwind readonly {
 ; CHECK-NEXT:    [[TMP8]] = add nsw i32 [[TMP7]], [[TMP6]]
 ; CHECK-NEXT:    br label [[BB1_3]]
 ; CHECK:       bb1.3:
-; CHECK-NEXT:    [[INDVAR_NEXT_3]] = add i64 [[INDVAR_NEXT_2]], 1
+; CHECK-NEXT:    [[INDVAR_NEXT_3]] = add i64 [[INDVAR]], 4
 ; CHECK-NEXT:    [[EXITCOND_3:%.*]] = icmp ne i64 [[INDVAR_NEXT_3]], [[TMP]]
 ; CHECK-NEXT:    br i1 [[EXITCOND_3]], label [[BB]], label [[BB1_BB2_CRIT_EDGE]], !llvm.loop [[LOOP2:![0-9]+]]
 ; CHECK:       bb1.bb2_crit_edge:

diff --git a/llvm/test/Transforms/LoopUnroll/AArch64/partial.ll b/llvm/test/Transforms/LoopUnroll/AArch64/partial.ll
index b753b18..37206b0 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/AArch64/partial.ll
+++ b/llvm/test/Transforms/LoopUnroll/AArch64/partial.ll
@@ -15,14 +15,7 @@ exit:
   ret void
 }
 
-; CHECK:      add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
+; CHECK: add{{.*}}, 8
 ; CHECK-NEXT: icmp
 
 ; Partial unroll 16 times for this loop.
@@ -57,20 +50,5 @@ exit:
 
 
 
-; CHECK:      add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
+; CHECK: add{{.*}}, 16
 ; CHECK-NEXT: icmp

diff --git a/llvm/test/Transforms/LoopUnroll/ARM/instr-size-costs.ll b/llvm/test/Transforms/LoopUnroll/ARM/instr-size-costs.ll
index ad934e0..e1701ee 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/ARM/instr-size-costs.ll
+++ b/llvm/test/Transforms/LoopUnroll/ARM/instr-size-costs.ll
@@ -22,7 +22,7 @@ define void @test_i32_add_optsize(ptr %a, ptr %b, ptr %c) #0 {
 ; CHECK-V8-NEXT:    [[RES_1:%.*]] = add i32 [[DATA_A_1]], [[DATA_B_1]]
 ; CHECK-V8-NEXT:    [[ADDR_C_1:%.*]] = getelementptr i32, ptr [[C]], i32 [[COUNT]]
 ; CHECK-V8-NEXT:    store i32 [[RES_1]], ptr [[ADDR_C_1]], align 4
-; CHECK-V8-NEXT:    [[COUNT_1]] = add nuw nsw i32 [[COUNT]], 1
+; CHECK-V8-NEXT:    [[COUNT_1]] = add nuw nsw i32 [[IV]], 2
 ; CHECK-V8-NEXT:    [[END_1:%.*]] = icmp ne i32 [[COUNT_1]], 100
 ; CHECK-V8-NEXT:    br i1 [[END_1]], label [[LOOP]], label [[EXIT:%.*]]
 ; CHECK-V8:       exit:
@@ -69,7 +69,7 @@ define void @test_i32_add_minsize(ptr %a, ptr %b, ptr %c) #1 {
 ; CHECK-V8-NEXT:    [[RES_1:%.*]] = add i32 [[DATA_A_1]], [[DATA_B_1]]
 ; CHECK-V8-NEXT:    [[ADDR_C_1:%.*]] = getelementptr i32, ptr [[C]], i32 [[COUNT]]
 ; CHECK-V8-NEXT:    store i32 [[RES_1]], ptr [[ADDR_C_1]], align 4
-; CHECK-V8-NEXT:    [[COUNT_1]] = add nuw nsw i32 [[COUNT]], 1
+; CHECK-V8-NEXT:    [[COUNT_1]] = add nuw nsw i32 [[IV]], 2
 ; CHECK-V8-NEXT:    [[END_1:%.*]] = icmp ne i32 [[COUNT_1]], 100
 ; CHECK-V8-NEXT:    br i1 [[END_1]], label [[LOOP]], label [[EXIT:%.*]]
 ; CHECK-V8:       exit:
@@ -116,7 +116,7 @@ define void @test_i64_add_optsize(ptr %a, ptr %b, ptr %c) #0 {
 ; CHECK-V8-NEXT:    [[RES_1:%.*]] = add i64 [[DATA_A_1]], [[DATA_B_1]]
 ; CHECK-V8-NEXT:    [[ADDR_C_1:%.*]] = getelementptr i64, ptr [[C]], i32 [[COUNT]]
 ; CHECK-V8-NEXT:    store i64 [[RES_1]], ptr [[ADDR_C_1]], align 4
-; CHECK-V8-NEXT:    [[COUNT_1]] = add nuw nsw i32 [[COUNT]], 1
+; CHECK-V8-NEXT:    [[COUNT_1]] = add nuw nsw i32 [[IV]], 2
 ; CHECK-V8-NEXT:    [[END_1:%.*]] = icmp ne i32 [[COUNT_1]], 100
 ; CHECK-V8-NEXT:    br i1 [[END_1]], label [[LOOP]], label [[EXIT:%.*]]
 ; CHECK-V8:       exit:
@@ -163,7 +163,7 @@ define void @test_i64_add_minsize(ptr %a, ptr %b, ptr %c) #1 {
 ; CHECK-V8-NEXT:    [[RES_1:%.*]] = add i64 [[DATA_A_1]], [[DATA_B_1]]
 ; CHECK-V8-NEXT:    [[ADDR_C_1:%.*]] = getelementptr i64, ptr [[C]], i32 [[COUNT]]
 ; CHECK-V8-NEXT:    store i64 [[RES_1]], ptr [[ADDR_C_1]], align 4
-; CHECK-V8-NEXT:    [[COUNT_1]] = add nuw nsw i32 [[COUNT]], 1
+; CHECK-V8-NEXT:    [[COUNT_1]] = add nuw nsw i32 [[IV]], 2
 ; CHECK-V8-NEXT:    [[END_1:%.*]] = icmp ne i32 [[COUNT_1]], 100
 ; CHECK-V8-NEXT:    br i1 [[END_1]], label [[LOOP]], label [[EXIT:%.*]]
 ; CHECK-V8:       exit:
@@ -215,7 +215,7 @@ define i32 @test_i32_select_optsize(ptr %a, ptr %b, ptr %c) #0 {
 ; CHECK-V8-NEXT:    [[ACC_NEXT_1]] = add i32 [[UMAX_1]], [[ACC_NEXT]]
 ; CHECK-V8-NEXT:    [[ADDR_C_1:%.*]] = getelementptr i32, ptr [[C]], i32 [[COUNT]]
 ; CHECK-V8-NEXT:    store i32 [[UMAX_1]], ptr [[ADDR_C_1]], align 4
-; CHECK-V8-NEXT:    [[COUNT_1]] = add nuw nsw i32 [[COUNT]], 1
+; CHECK-V8-NEXT:    [[COUNT_1]] = add nuw nsw i32 [[IV]], 2
 ; CHECK-V8-NEXT:    [[END_1:%.*]] = icmp ne i32 [[COUNT_1]], 100
 ; CHECK-V8-NEXT:    br i1 [[END_1]], label [[LOOP]], label [[EXIT:%.*]]
 ; CHECK-V8:       exit:
@@ -271,7 +271,7 @@ define i32 @test_i32_select_minsize(ptr %a, ptr %b, ptr %c) #1 {
 ; CHECK-V8-NEXT:    [[ACC_NEXT_1]] = add i32 [[UMAX_1]], [[ACC_NEXT]]
 ; CHECK-V8-NEXT:    [[ADDR_C_1:%.*]] = getelementptr i32, ptr [[C]], i32 [[COUNT]]
 ; CHECK-V8-NEXT:    store i32 [[UMAX_1]], ptr [[ADDR_C_1]], align 4
-; CHECK-V8-NEXT:    [[COUNT_1]] = add nuw nsw i32 [[COUNT]], 1
+; CHECK-V8-NEXT:    [[COUNT_1]] = add nuw nsw i32 [[IV]], 2
 ; CHECK-V8-NEXT:    [[END_1:%.*]] = icmp ne i32 [[COUNT_1]], 100
 ; CHECK-V8-NEXT:    br i1 [[END_1]], label [[LOOP]], label [[EXIT:%.*]]
 ; CHECK-V8:       exit:

diff --git a/llvm/test/Transforms/LoopUnroll/ARM/loop-unrolling.ll b/llvm/test/Transforms/LoopUnroll/ARM/loop-unrolling.ll
index dab06b9..c00f744 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/ARM/loop-unrolling.ll
+++ b/llvm/test/Transforms/LoopUnroll/ARM/loop-unrolling.ll
@@ -15,27 +15,27 @@ for.body:
 
 ; CHECK-NOUNROLL: [[IV0:%[a-z.0-9]+]] = phi i32 [ 0, %entry ], [ [[IV2:%[a-z.0-9]+]], %for.body ]
 ; CHECK-NOUNROLL: [[IV1:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 1
-; CHECK-NOUNROLL: [[IV2]] = add nuw nsw i32 [[IV1]], 1
+; CHECK-NOUNROLL: [[IV2]] = add nuw nsw i32 [[IV0]], 2
 ; CHECK-NOUNROLL: [[CMP:%[a-z.0-9]+]] = icmp eq i32 [[IV2]], 1024
 ; CHECK-NOUNROLL: br i1 [[CMP]], label [[END:%[a-z.]+]], label %for.body
 
 ; CHECK-UNROLL: [[IV0:%[a-z.0-9]+]] = phi i32 [ 0, %entry ], [ [[IV16:%[a-z.0-9]+]], %for.body ]
 ; CHECK-UNROLL: [[IV1:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 1
-; CHECK-UNROLL: [[IV2:%[a-z.0-9]+]] = add nuw nsw i32 [[IV1]], 1
-; CHECK-UNROLL: [[IV3:%[a-z.0-9]+]] = add nuw nsw i32 [[IV2]], 1
-; CHECK-UNROLL: [[IV4:%[a-z.0-9]+]] = add nuw nsw i32 [[IV3]], 1
-; CHECK-UNROLL: [[IV5:%[a-z.0-9]+]] = add nuw nsw i32 [[IV4]], 1
-; CHECK-UNROLL: [[IV6:%[a-z.0-9]+]] = add nuw nsw i32 [[IV5]], 1
-; CHECK-UNROLL: [[IV7:%[a-z.0-9]+]] = add nuw nsw i32 [[IV6]], 1
-; CHECK-UNROLL: [[IV8:%[a-z.0-9]+]] = add nuw nsw i32 [[IV7]], 1
-; CHECK-UNROLL: [[IV9:%[a-z.0-9]+]] = add nuw nsw i32 [[IV8]], 1
-; CHECK-UNROLL: [[IV10:%[a-z.0-9]+]] = add nuw nsw i32 [[IV9]], 1
-; CHECK-UNROLL: [[IV11:%[a-z.0-9]+]] = add nuw nsw i32 [[IV10]], 1
-; CHECK-UNROLL: [[IV12:%[a-z.0-9]+]] = add nuw nsw i32 [[IV11]], 1
-; CHECK-UNROLL: [[IV13:%[a-z.0-9]+]] = add nuw nsw i32 [[IV12]], 1
-; CHECK-UNROLL: [[IV14:%[a-z.0-9]+]] = add nuw nsw i32 [[IV13]], 1
-; CHECK-UNROLL: [[IV15:%[a-z.0-9]+]] = add nuw nsw i32 [[IV14]], 1
-; CHECK-UNROLL: [[IV16]] = add nuw nsw i32 [[IV15]], 1
+; CHECK-UNROLL: [[IV2:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 2
+; CHECK-UNROLL: [[IV3:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 3
+; CHECK-UNROLL: [[IV4:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 4
+; CHECK-UNROLL: [[IV5:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 5
+; CHECK-UNROLL: [[IV6:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 6
+; CHECK-UNROLL: [[IV7:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 7
+; CHECK-UNROLL: [[IV8:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 8
+; CHECK-UNROLL: [[IV9:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 9
+; CHECK-UNROLL: [[IV10:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 10
+; CHECK-UNROLL: [[IV11:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 11
+; CHECK-UNROLL: [[IV12:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 12
+; CHECK-UNROLL: [[IV13:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 13
+; CHECK-UNROLL: [[IV14:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 14
+; CHECK-UNROLL: [[IV15:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 15
+; CHECK-UNROLL: [[IV16]] = add nuw nsw i32 [[IV0]], 16
 ; CHECK-UNROLL: [[CMP:%[a-z.0-9]+]] = icmp eq i32 [[IV16]], 1024
 ; CHECK-UNROLL: br i1 [[CMP]], label [[END:%[a-z.]+]], label %for.body
 
@@ -65,14 +65,14 @@ entry:
 for.body:
 ; CHECK-NOUNROLL: [[IV0:%[a-z.0-9]+]] = phi i32 [ 0, [[PRE:%[a-z.0-9]+]] ], [ [[IV2:%[a-z.0-9]+]], %for.body ]
 ; CHECK-NOUNROLL: [[IV1:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 1
-; CHECK-NOUNROLL: [[IV2]] = add nuw i32 [[IV1]], 1
+; CHECK-NOUNROLL: [[IV2]] = add nuw i32 [[IV0]], 2
 ; CHECK-NOUNROLL: br
 
 ; CHECK-UNROLL: [[IV0:%[a-z.0-9]+]] = phi i32 [ 0, [[PRE:%[a-z.0-9]+]] ], [ [[IV4:%[a-z.0-9]+]], %for.body ]
 ; CHECK-UNROLL: [[IV1:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 1
-; CHECK-UNROLL: [[IV2:%[a-z.0-9]+]] = add nuw nsw i32 [[IV1]], 1
-; CHECK-UNROLL: [[IV3:%[a-z.0-9]+]] = add nuw nsw i32 [[IV2]], 1
-; CHECK-UNROLL: [[IV4]] = add nuw i32 [[IV3]], 1
+; CHECK-UNROLL: [[IV2:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 2
+; CHECK-UNROLL: [[IV3:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 3
+; CHECK-UNROLL: [[IV4]] = add nuw i32 [[IV0]], 4
 ; CHECK-UNROLL: br
 
 ; CHECK-UNROLL: for.body.epil:
@@ -125,9 +125,9 @@ for.body4:
 ; CHECK-UNROLL: for.body4.epil.2:
 ; CHECK-UNROLL: [[IV0:%[a-z.0-9]+]] = phi i32 [ 0, [[PRE:%[a-z0-9.]+]] ], [ [[IV4:%[a-z.0-9]+]], %for.body4 ]
 ; CHECK-UNROLL: [[IV1:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 1
-; CHECK-UNROLL: [[IV2:%[a-z.0-9]+]] = add nuw nsw i32 [[IV1]], 1
-; CHECK-UNROLL: [[IV3:%[a-z.0-9]+]] = add nuw nsw i32 [[IV2]], 1
-; CHECK-UNROLL: [[IV4]] = add nuw i32 [[IV3]], 1
+; CHECK-UNROLL: [[IV2:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 2
+; CHECK-UNROLL: [[IV3:%[a-z.0-9]+]] = add nuw nsw i32 [[IV0]], 3
+; CHECK-UNROLL: [[IV4]] = add nuw i32 [[IV0]], 4
 ; CHECK-UNROLL: br
 
   %w.024 = phi i32 [ 0, %for.body4.lr.ph ], [ %inc, %for.body4 ]

diff --git a/llvm/test/Transforms/LoopUnroll/ARM/multi-blocks.ll b/llvm/test/Transforms/LoopUnroll/ARM/multi-blocks.ll
index 757a3d9..f988c15 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/ARM/multi-blocks.ll
+++ b/llvm/test/Transforms/LoopUnroll/ARM/multi-blocks.ll
@@ -55,7 +55,7 @@ define void @test_three_blocks(ptr nocapture %Output, ptr nocapture readonly %Co
 ; CHECK-NEXT:    br label [[FOR_INC_EPIL_1]]
 ; CHECK:       for.inc.epil.1:
 ; CHECK-NEXT:    [[TEMP_1_EPIL_1:%.*]] = phi i32 [ [[ADD_EPIL_1]], [[IF_THEN_EPIL_1]] ], [ [[TEMP_1_EPIL]], [[FOR_BODY_EPIL_1]] ]
-; CHECK-NEXT:    [[INC_EPIL_1:%.*]] = add nuw i32 [[INC_EPIL]], 1
+; CHECK-NEXT:    [[INC_EPIL_1:%.*]] = add nuw i32 [[J_010_UNR]], 2
 ; CHECK-NEXT:    [[EPIL_ITER_CMP_1:%.*]] = icmp ne i32 2, [[XTRAITER]]
 ; CHECK-NEXT:    br i1 [[EPIL_ITER_CMP_1]], label [[FOR_BODY_EPIL_2:%.*]], label [[FOR_COND_CLEANUP_LOOPEXIT_EPILOG_LCSSA]]
 ; CHECK:       for.body.epil.2:
@@ -97,7 +97,6 @@ define void @test_three_blocks(ptr nocapture %Output, ptr nocapture readonly %Co
 ; CHECK:       for.inc:
 ; CHECK-NEXT:    [[TEMP_1:%.*]] = phi i32 [ [[ADD]], [[IF_THEN]] ], [ [[TEMP_09]], [[FOR_BODY]] ]
 ; CHECK-NEXT:    [[INC:%.*]] = add nuw nsw i32 [[J_010]], 1
-; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i32 [[NITER]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, ptr [[CONDITION]], i32 [[INC]]
 ; CHECK-NEXT:    [[I_1:%.*]] = load i32, ptr [[ARRAYIDX_1]], align 4
 ; CHECK-NEXT:    [[TOBOOL_1:%.*]] = icmp eq i32 [[I_1]], 0
@@ -109,8 +108,7 @@ define void @test_three_blocks(ptr nocapture %Output, ptr nocapture readonly %Co
 ; CHECK-NEXT:    br label [[FOR_INC_1]]
 ; CHECK:       for.inc.1:
 ; CHECK-NEXT:    [[TEMP_1_1:%.*]] = phi i32 [ [[ADD_1]], [[IF_THEN_1]] ], [ [[TEMP_1]], [[FOR_INC]] ]
-; CHECK-NEXT:    [[INC_1:%.*]] = add nuw nsw i32 [[INC]], 1
-; CHECK-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i32 [[NITER_NEXT]], 1
+; CHECK-NEXT:    [[INC_1:%.*]] = add nuw nsw i32 [[J_010]], 2
 ; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, ptr [[CONDITION]], i32 [[INC_1]]
 ; CHECK-NEXT:    [[I_2:%.*]] = load i32, ptr [[ARRAYIDX_2]], align 4
 ; CHECK-NEXT:    [[TOBOOL_2:%.*]] = icmp eq i32 [[I_2]], 0
@@ -122,8 +120,7 @@ define void @test_three_blocks(ptr nocapture %Output, ptr nocapture readonly %Co
 ; CHECK-NEXT:    br label [[FOR_INC_2]]
 ; CHECK:       for.inc.2:
 ; CHECK-NEXT:    [[TEMP_1_2:%.*]] = phi i32 [ [[ADD_2]], [[IF_THEN_2]] ], [ [[TEMP_1_1]], [[FOR_INC_1]] ]
-; CHECK-NEXT:    [[INC_2:%.*]] = add nuw nsw i32 [[INC_1]], 1
-; CHECK-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i32 [[NITER_NEXT_1]], 1
+; CHECK-NEXT:    [[INC_2:%.*]] = add nuw nsw i32 [[J_010]], 3
 ; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, ptr [[CONDITION]], i32 [[INC_2]]
 ; CHECK-NEXT:    [[I_3:%.*]] = load i32, ptr [[ARRAYIDX_3]], align 4
 ; CHECK-NEXT:    [[TOBOOL_3:%.*]] = icmp eq i32 [[I_3]], 0
@@ -135,8 +132,8 @@ define void @test_three_blocks(ptr nocapture %Output, ptr nocapture readonly %Co
 ; CHECK-NEXT:    br label [[FOR_INC_3]]
 ; CHECK:       for.inc.3:
 ; CHECK-NEXT:    [[TEMP_1_3]] = phi i32 [ [[ADD_3]], [[IF_THEN_3]] ], [ [[TEMP_1_2]], [[FOR_INC_2]] ]
-; CHECK-NEXT:    [[INC_3]] = add nuw i32 [[INC_2]], 1
-; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER_NEXT_2]], 1
+; CHECK-NEXT:    [[INC_3]] = add nuw i32 [[J_010]], 4
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER]], 4
 ; CHECK-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i32 [[NITER_NEXT_3]], [[UNROLL_ITER]]
 ; CHECK-NEXT:    br i1 [[NITER_NCMP_3]], label [[FOR_COND_CLEANUP_LOOPEXIT_UNR_LCSSA_LOOPEXIT]], label [[FOR_BODY]]
 ;
@@ -207,7 +204,7 @@ define void @test_two_exits(ptr nocapture %Output, ptr nocapture readonly %Condi
 ; CHECK-NEXT:    [[TOBOOL_1:%.*]] = icmp eq i32 [[I1_1]], 0
 ; CHECK-NEXT:    [[ADD_1:%.*]] = select i1 [[TOBOOL_1]], i32 0, i32 [[I_1]]
 ; CHECK-NEXT:    [[TEMP_0_ADD_1:%.*]] = add i32 [[ADD_1]], [[TEMP_0_ADD]]
-; CHECK-NEXT:    [[INC_1:%.*]] = add nuw nsw i32 [[INC]], 1
+; CHECK-NEXT:    [[INC_1:%.*]] = add nuw nsw i32 [[J_016]], 2
 ; CHECK-NEXT:    [[CMP_1:%.*]] = icmp ult i32 [[INC_1]], [[MAXJ]]
 ; CHECK-NEXT:    br i1 [[CMP_1]], label [[FOR_BODY_2:%.*]], label [[CLEANUP_LOOPEXIT]]
 ; CHECK:       for.body.2:
@@ -221,7 +218,7 @@ define void @test_two_exits(ptr nocapture %Output, ptr nocapture readonly %Condi
 ; CHECK-NEXT:    [[TOBOOL_2:%.*]] = icmp eq i32 [[I1_2]], 0
 ; CHECK-NEXT:    [[ADD_2:%.*]] = select i1 [[TOBOOL_2]], i32 0, i32 [[I_2]]
 ; CHECK-NEXT:    [[TEMP_0_ADD_2:%.*]] = add i32 [[ADD_2]], [[TEMP_0_ADD_1]]
-; CHECK-NEXT:    [[INC_2:%.*]] = add nuw nsw i32 [[INC_1]], 1
+; CHECK-NEXT:    [[INC_2:%.*]] = add nuw nsw i32 [[J_016]], 3
 ; CHECK-NEXT:    [[CMP_2:%.*]] = icmp ult i32 [[INC_2]], [[MAXJ]]
 ; CHECK-NEXT:    br i1 [[CMP_2]], label [[FOR_BODY_3:%.*]], label [[CLEANUP_LOOPEXIT]]
 ; CHECK:       for.body.3:
@@ -235,7 +232,7 @@ define void @test_two_exits(ptr nocapture %Output, ptr nocapture readonly %Condi
 ; CHECK-NEXT:    [[TOBOOL_3:%.*]] = icmp eq i32 [[I1_3]], 0
 ; CHECK-NEXT:    [[ADD_3:%.*]] = select i1 [[TOBOOL_3]], i32 0, i32 [[I_3]]
 ; CHECK-NEXT:    [[TEMP_0_ADD_3]] = add i32 [[ADD_3]], [[TEMP_0_ADD_2]]
-; CHECK-NEXT:    [[INC_3]] = add nuw i32 [[INC_2]], 1
+; CHECK-NEXT:    [[INC_3]] = add nuw i32 [[J_016]], 4
 ; CHECK-NEXT:    [[CMP_3:%.*]] = icmp ult i32 [[INC_3]], [[MAXJ]]
 ; CHECK-NEXT:    br i1 [[CMP_3]], label [[FOR_BODY]], label [[CLEANUP_LOOPEXIT]]
 ; CHECK:       cleanup.loopexit:
@@ -423,7 +420,7 @@ define void @test_four_blocks(ptr nocapture %Output, ptr nocapture readonly %Con
 ; CHECK-NEXT:    br label [[FOR_INC_EPIL_1]]
 ; CHECK:       for.inc.epil.1:
 ; CHECK-NEXT:    [[TEMP_1_EPIL_1:%.*]] = phi i32 [ [[ADD_EPIL_1]], [[IF_THEN_EPIL_1]] ], [ [[SUB13_EPIL_1]], [[IF_ELSE_EPIL_1]] ]
-; CHECK-NEXT:    [[INC_EPIL_1:%.*]] = add nuw i32 [[INC_EPIL]], 1
+; CHECK-NEXT:    [[INC_EPIL_1:%.*]] = add nuw i32 [[J_027_UNR]], 2
 ; CHECK-NEXT:    [[EPIL_ITER_CMP_1:%.*]] = icmp ne i32 2, [[XTRAITER]]
 ; CHECK-NEXT:    br i1 [[EPIL_ITER_CMP_1]], label [[FOR_BODY_EPIL_2:%.*]], label [[FOR_COND_CLEANUP_LOOPEXIT_EPILOG_LCSSA]]
 ; CHECK:       for.body.epil.2:
@@ -486,7 +483,6 @@ define void @test_four_blocks(ptr nocapture %Output, ptr nocapture readonly %Con
 ; CHECK:       for.inc:
 ; CHECK-NEXT:    [[TEMP_1:%.*]] = phi i32 [ [[ADD]], [[IF_THEN]] ], [ [[SUB13]], [[IF_ELSE]] ]
 ; CHECK-NEXT:    [[INC:%.*]] = add nuw nsw i32 [[J_027]], 1
-; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i32 [[NITER]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, ptr [[CONDITION]], i32 [[INC]]
 ; CHECK-NEXT:    [[I1_1:%.*]] = load i32, ptr [[ARRAYIDX_1]], align 4
 ; CHECK-NEXT:    [[CMP1_1:%.*]] = icmp ugt i32 [[I1_1]], 65535
@@ -508,8 +504,7 @@ define void @test_four_blocks(ptr nocapture %Output, ptr nocapture readonly %Con
 ; CHECK-NEXT:    br label [[FOR_INC_1]]
 ; CHECK:       for.inc.1:
 ; CHECK-NEXT:    [[TEMP_1_1:%.*]] = phi i32 [ [[ADD_1]], [[IF_THEN_1]] ], [ [[SUB13_1]], [[IF_ELSE_1]] ]
-; CHECK-NEXT:    [[INC_1:%.*]] = add nuw nsw i32 [[INC]], 1
-; CHECK-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i32 [[NITER_NEXT]], 1
+; CHECK-NEXT:    [[INC_1:%.*]] = add nuw nsw i32 [[J_027]], 2
 ; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, ptr [[CONDITION]], i32 [[INC_1]]
 ; CHECK-NEXT:    [[I1_2:%.*]] = load i32, ptr [[ARRAYIDX_2]], align 4
 ; CHECK-NEXT:    [[CMP1_2:%.*]] = icmp ugt i32 [[I1_2]], 65535
@@ -531,8 +526,7 @@ define void @test_four_blocks(ptr nocapture %Output, ptr nocapture readonly %Con
 ; CHECK-NEXT:    br label [[FOR_INC_2]]
 ; CHECK:       for.inc.2:
 ; CHECK-NEXT:    [[TEMP_1_2:%.*]] = phi i32 [ [[ADD_2]], [[IF_THEN_2]] ], [ [[SUB13_2]], [[IF_ELSE_2]] ]
-; CHECK-NEXT:    [[INC_2:%.*]] = add nuw i32 [[INC_1]], 1
-; CHECK-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i32 [[NITER_NEXT_1]], 1
+; CHECK-NEXT:    [[INC_2:%.*]] = add nuw i32 [[J_027]], 3
 ; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, ptr [[CONDITION]], i32 [[INC_2]]
 ; CHECK-NEXT:    [[I1_3:%.*]] = load i32, ptr [[ARRAYIDX_3]], align 4
 ; CHECK-NEXT:    [[CMP1_3:%.*]] = icmp ugt i32 [[I1_3]], 65535
@@ -554,8 +548,8 @@ define void @test_four_blocks(ptr nocapture %Output, ptr nocapture readonly %Con
 ; CHECK-NEXT:    br label [[FOR_INC_3]]
 ; CHECK:       for.inc.3:
 ; CHECK-NEXT:    [[TEMP_1_3]] = phi i32 [ [[ADD_3]], [[IF_THEN_3]] ], [ [[SUB13_3]], [[IF_ELSE_3]] ]
-; CHECK-NEXT:    [[INC_3]] = add nuw nsw i32 [[INC_2]], 1
-; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER_NEXT_2]], 1
+; CHECK-NEXT:    [[INC_3]] = add nuw i32 [[J_027]], 4
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER]], 4
 ; CHECK-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i32 [[NITER_NEXT_3]], [[UNROLL_ITER]]
 ; CHECK-NEXT:    br i1 [[NITER_NCMP_3]], label [[FOR_COND_CLEANUP_LOOPEXIT_UNR_LCSSA_LOOPEXIT]], label [[FOR_BODY]]
 ;

diff --git a/llvm/test/Transforms/LoopUnroll/PowerPC/p7-unrolling.ll b/llvm/test/Transforms/LoopUnroll/PowerPC/p7-unrolling.ll
index e9385ee..f1e5d0f 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/PowerPC/p7-unrolling.ll
+++ b/llvm/test/Transforms/LoopUnroll/PowerPC/p7-unrolling.ll
@@ -14,37 +14,6 @@ exit:
 }
 
 ; CHECK-LABEL: @unroll_default
-; CHECK:      add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
-; CHECK-NEXT: add
+; CHECK: add{{.*}}, 32
 ; CHECK-NEXT: icmp
 

diff --git a/llvm/test/Transforms/LoopUnroll/PowerPC/p8-unrolling-legalize-vectors-inseltpoison.ll b/llvm/test/Transforms/LoopUnroll/PowerPC/p8-unrolling-legalize-vectors-inseltpoison.ll
index 97f31b6..d004166 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/PowerPC/p8-unrolling-legalize-vectors-inseltpoison.ll
+++ b/llvm/test/Transforms/LoopUnroll/PowerPC/p8-unrolling-legalize-vectors-inseltpoison.ll
@@ -40,16 +40,15 @@ define ptr @f(ptr returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    store <16 x i8> [[TMP7]], ptr [[TMP8]], align 1
 ; CHECK-NEXT:    [[INDEX_NEXT:%.*]] = add nuw nsw i64 [[INDEX]], 16
 ; CHECK-NEXT:    [[VEC_IND_NEXT13:%.*]] = add <16 x i32> [[VEC_IND12]], <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
-; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i64 [[NITER]], 1
 ; CHECK-NEXT:    [[TMP9:%.*]] = shl <16 x i32> <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>, [[VEC_IND_NEXT13]]
 ; CHECK-NEXT:    [[TMP10:%.*]] = and <16 x i32> [[TMP9]], [[BROADCAST_SPLAT]]
 ; CHECK-NEXT:    [[TMP11:%.*]] = icmp eq <16 x i32> [[TMP10]], zeroinitializer
 ; CHECK-NEXT:    [[TMP12:%.*]] = select <16 x i1> [[TMP11]], <16 x i8> <i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48>, <16 x i8> <i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49>
 ; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i8, ptr [[S]], i64 [[INDEX_NEXT]]
 ; CHECK-NEXT:    store <16 x i8> [[TMP12]], ptr [[TMP13]], align 1
-; CHECK-NEXT:    [[INDEX_NEXT_1]] = add i64 [[INDEX_NEXT]], 16
-; CHECK-NEXT:    [[VEC_IND_NEXT13_1]] = add <16 x i32> [[VEC_IND_NEXT13]], <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
-; CHECK-NEXT:    [[NITER_NEXT_1]] = add i64 [[NITER_NEXT]], 1
+; CHECK-NEXT:    [[INDEX_NEXT_1]] = add i64 [[INDEX]], 32
+; CHECK-NEXT:    [[VEC_IND_NEXT13_1]] = add <16 x i32> [[VEC_IND12]], <i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32>
+; CHECK-NEXT:    [[NITER_NEXT_1]] = add i64 [[NITER]], 2
 ; CHECK-NEXT:    [[NITER_NCMP_1:%.*]] = icmp eq i64 [[NITER_NEXT_1]], [[UNROLL_ITER]]
 ; CHECK-NEXT:    br i1 [[NITER_NCMP_1]], label [[MIDDLE_BLOCK_UNR_LCSSA_LOOPEXIT:%.*]], label [[VECTOR_BODY]]
 ; CHECK:       middle.block.unr-lcssa.loopexit:
@@ -125,7 +124,7 @@ define ptr @f(ptr returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    [[CONV_1:%.*]] = select i1 [[TOBOOL_1]], i8 48, i8 49
 ; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i8, ptr [[S]], i64 [[INDVARS_IV_NEXT]]
 ; CHECK-NEXT:    store i8 [[CONV_1]], ptr [[ARRAYIDX_1]], align 1
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 2
 ; CHECK-NEXT:    [[TMP26:%.*]] = trunc i64 [[INDVARS_IV_NEXT_1]] to i32
 ; CHECK-NEXT:    [[SHL_2:%.*]] = shl i32 1, [[TMP26]]
 ; CHECK-NEXT:    [[AND_2:%.*]] = and i32 [[SHL_2]], [[X]]
@@ -133,7 +132,7 @@ define ptr @f(ptr returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    [[CONV_2:%.*]] = select i1 [[TOBOOL_2]], i8 48, i8 49
 ; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i8, ptr [[S]], i64 [[INDVARS_IV_NEXT_1]]
 ; CHECK-NEXT:    store i8 [[CONV_2]], ptr [[ARRAYIDX_2]], align 1
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_1]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 3
 ; CHECK-NEXT:    [[TMP27:%.*]] = trunc i64 [[INDVARS_IV_NEXT_2]] to i32
 ; CHECK-NEXT:    [[SHL_3:%.*]] = shl i32 1, [[TMP27]]
 ; CHECK-NEXT:    [[AND_3:%.*]] = and i32 [[SHL_3]], [[X]]
@@ -141,7 +140,7 @@ define ptr @f(ptr returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    [[CONV_3:%.*]] = select i1 [[TOBOOL_3]], i8 48, i8 49
 ; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i8, ptr [[S]], i64 [[INDVARS_IV_NEXT_2]]
 ; CHECK-NEXT:    store i8 [[CONV_3]], ptr [[ARRAYIDX_3]], align 1
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_3:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_2]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_3:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 4
 ; CHECK-NEXT:    [[TMP28:%.*]] = trunc i64 [[INDVARS_IV_NEXT_3]] to i32
 ; CHECK-NEXT:    [[SHL_4:%.*]] = shl i32 1, [[TMP28]]
 ; CHECK-NEXT:    [[AND_4:%.*]] = and i32 [[SHL_4]], [[X]]
@@ -149,7 +148,7 @@ define ptr @f(ptr returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    [[CONV_4:%.*]] = select i1 [[TOBOOL_4]], i8 48, i8 49
 ; CHECK-NEXT:    [[ARRAYIDX_4:%.*]] = getelementptr inbounds i8, ptr [[S]], i64 [[INDVARS_IV_NEXT_3]]
 ; CHECK-NEXT:    store i8 [[CONV_4]], ptr [[ARRAYIDX_4]], align 1
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_4:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_3]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_4:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 5
 ; CHECK-NEXT:    [[TMP29:%.*]] = trunc i64 [[INDVARS_IV_NEXT_4]] to i32
 ; CHECK-NEXT:    [[SHL_5:%.*]] = shl i32 1, [[TMP29]]
 ; CHECK-NEXT:    [[AND_5:%.*]] = and i32 [[SHL_5]], [[X]]
@@ -157,7 +156,7 @@ define ptr @f(ptr returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    [[CONV_5:%.*]] = select i1 [[TOBOOL_5]], i8 48, i8 49
 ; CHECK-NEXT:    [[ARRAYIDX_5:%.*]] = getelementptr inbounds i8, ptr [[S]], i64 [[INDVARS_IV_NEXT_4]]
 ; CHECK-NEXT:    store i8 [[CONV_5]], ptr [[ARRAYIDX_5]], align 1
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_5:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_4]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_5:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 6
 ; CHECK-NEXT:    [[TMP30:%.*]] = trunc i64 [[INDVARS_IV_NEXT_5]] to i32
 ; CHECK-NEXT:    [[SHL_6:%.*]] = shl i32 1, [[TMP30]]
 ; CHECK-NEXT:    [[AND_6:%.*]] = and i32 [[SHL_6]], [[X]]
@@ -165,7 +164,7 @@ define ptr @f(ptr returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    [[CONV_6:%.*]] = select i1 [[TOBOOL_6]], i8 48, i8 49
 ; CHECK-NEXT:    [[ARRAYIDX_6:%.*]] = getelementptr inbounds i8, ptr [[S]], i64 [[INDVARS_IV_NEXT_5]]
 ; CHECK-NEXT:    store i8 [[CONV_6]], ptr [[ARRAYIDX_6]], align 1
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_6:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_5]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_6:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 7
 ; CHECK-NEXT:    [[TMP31:%.*]] = trunc i64 [[INDVARS_IV_NEXT_6]] to i32
 ; CHECK-NEXT:    [[SHL_7:%.*]] = shl i32 1, [[TMP31]]
 ; CHECK-NEXT:    [[AND_7:%.*]] = and i32 [[SHL_7]], [[X]]
@@ -173,7 +172,7 @@ define ptr @f(ptr returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    [[CONV_7:%.*]] = select i1 [[TOBOOL_7]], i8 48, i8 49
 ; CHECK-NEXT:    [[ARRAYIDX_7:%.*]] = getelementptr inbounds i8, ptr [[S]], i64 [[INDVARS_IV_NEXT_6]]
 ; CHECK-NEXT:    store i8 [[CONV_7]], ptr [[ARRAYIDX_7]], align 1
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_7]] = add nuw nsw i64 [[INDVARS_IV_NEXT_6]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_7]] = add nuw nsw i64 [[INDVARS_IV]], 8
 ; CHECK-NEXT:    [[EXITCOND_7:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT_7]], [[WIDE_TRIP_COUNT]]
 ; CHECK-NEXT:    br i1 [[EXITCOND_7]], label [[FOR_END_LOOPEXIT_UNR_LCSSA:%.*]], label [[FOR_BODY]]
 ; CHECK:       for.end.loopexit.unr-lcssa:

diff --git a/llvm/test/Transforms/LoopUnroll/PowerPC/p8-unrolling-legalize-vectors.ll b/llvm/test/Transforms/LoopUnroll/PowerPC/p8-unrolling-legalize-vectors.ll
index 8e74063..c2a8675 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/PowerPC/p8-unrolling-legalize-vectors.ll
+++ b/llvm/test/Transforms/LoopUnroll/PowerPC/p8-unrolling-legalize-vectors.ll
@@ -40,16 +40,15 @@ define ptr @f(ptr returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    store <16 x i8> [[TMP7]], ptr [[TMP8]], align 1
 ; CHECK-NEXT:    [[INDEX_NEXT:%.*]] = add nuw nsw i64 [[INDEX]], 16
 ; CHECK-NEXT:    [[VEC_IND_NEXT13:%.*]] = add <16 x i32> [[VEC_IND12]], <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
-; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i64 [[NITER]], 1
 ; CHECK-NEXT:    [[TMP9:%.*]] = shl <16 x i32> <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>, [[VEC_IND_NEXT13]]
 ; CHECK-NEXT:    [[TMP10:%.*]] = and <16 x i32> [[TMP9]], [[BROADCAST_SPLAT]]
 ; CHECK-NEXT:    [[TMP11:%.*]] = icmp eq <16 x i32> [[TMP10]], zeroinitializer
 ; CHECK-NEXT:    [[TMP12:%.*]] = select <16 x i1> [[TMP11]], <16 x i8> <i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48, i8 48>, <16 x i8> <i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49, i8 49>
 ; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i8, ptr [[S]], i64 [[INDEX_NEXT]]
 ; CHECK-NEXT:    store <16 x i8> [[TMP12]], ptr [[TMP13]], align 1
-; CHECK-NEXT:    [[INDEX_NEXT_1]] = add i64 [[INDEX_NEXT]], 16
-; CHECK-NEXT:    [[VEC_IND_NEXT13_1]] = add <16 x i32> [[VEC_IND_NEXT13]], <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
-; CHECK-NEXT:    [[NITER_NEXT_1]] = add i64 [[NITER_NEXT]], 1
+; CHECK-NEXT:    [[INDEX_NEXT_1]] = add i64 [[INDEX]], 32
+; CHECK-NEXT:    [[VEC_IND_NEXT13_1]] = add <16 x i32> [[VEC_IND12]], <i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32>
+; CHECK-NEXT:    [[NITER_NEXT_1]] = add i64 [[NITER]], 2
 ; CHECK-NEXT:    [[NITER_NCMP_1:%.*]] = icmp eq i64 [[NITER_NEXT_1]], [[UNROLL_ITER]]
 ; CHECK-NEXT:    br i1 [[NITER_NCMP_1]], label [[MIDDLE_BLOCK_UNR_LCSSA_LOOPEXIT:%.*]], label [[VECTOR_BODY]]
 ; CHECK:       middle.block.unr-lcssa.loopexit:
@@ -125,7 +124,7 @@ define ptr @f(ptr returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    [[CONV_1:%.*]] = select i1 [[TOBOOL_1]], i8 48, i8 49
 ; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i8, ptr [[S]], i64 [[INDVARS_IV_NEXT]]
 ; CHECK-NEXT:    store i8 [[CONV_1]], ptr [[ARRAYIDX_1]], align 1
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 2
 ; CHECK-NEXT:    [[TMP26:%.*]] = trunc i64 [[INDVARS_IV_NEXT_1]] to i32
 ; CHECK-NEXT:    [[SHL_2:%.*]] = shl i32 1, [[TMP26]]
 ; CHECK-NEXT:    [[AND_2:%.*]] = and i32 [[SHL_2]], [[X]]
@@ -133,7 +132,7 @@ define ptr @f(ptr returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    [[CONV_2:%.*]] = select i1 [[TOBOOL_2]], i8 48, i8 49
 ; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i8, ptr [[S]], i64 [[INDVARS_IV_NEXT_1]]
 ; CHECK-NEXT:    store i8 [[CONV_2]], ptr [[ARRAYIDX_2]], align 1
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_1]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 3
 ; CHECK-NEXT:    [[TMP27:%.*]] = trunc i64 [[INDVARS_IV_NEXT_2]] to i32
 ; CHECK-NEXT:    [[SHL_3:%.*]] = shl i32 1, [[TMP27]]
 ; CHECK-NEXT:    [[AND_3:%.*]] = and i32 [[SHL_3]], [[X]]
@@ -141,7 +140,7 @@ define ptr @f(ptr returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    [[CONV_3:%.*]] = select i1 [[TOBOOL_3]], i8 48, i8 49
 ; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i8, ptr [[S]], i64 [[INDVARS_IV_NEXT_2]]
 ; CHECK-NEXT:    store i8 [[CONV_3]], ptr [[ARRAYIDX_3]], align 1
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_3:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_2]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_3:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 4
 ; CHECK-NEXT:    [[TMP28:%.*]] = trunc i64 [[INDVARS_IV_NEXT_3]] to i32
 ; CHECK-NEXT:    [[SHL_4:%.*]] = shl i32 1, [[TMP28]]
 ; CHECK-NEXT:    [[AND_4:%.*]] = and i32 [[SHL_4]], [[X]]
@@ -149,7 +148,7 @@ define ptr @f(ptr returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    [[CONV_4:%.*]] = select i1 [[TOBOOL_4]], i8 48, i8 49
 ; CHECK-NEXT:    [[ARRAYIDX_4:%.*]] = getelementptr inbounds i8, ptr [[S]], i64 [[INDVARS_IV_NEXT_3]]
 ; CHECK-NEXT:    store i8 [[CONV_4]], ptr [[ARRAYIDX_4]], align 1
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_4:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_3]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_4:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 5
 ; CHECK-NEXT:    [[TMP29:%.*]] = trunc i64 [[INDVARS_IV_NEXT_4]] to i32
 ; CHECK-NEXT:    [[SHL_5:%.*]] = shl i32 1, [[TMP29]]
 ; CHECK-NEXT:    [[AND_5:%.*]] = and i32 [[SHL_5]], [[X]]
@@ -157,7 +156,7 @@ define ptr @f(ptr returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    [[CONV_5:%.*]] = select i1 [[TOBOOL_5]], i8 48, i8 49
 ; CHECK-NEXT:    [[ARRAYIDX_5:%.*]] = getelementptr inbounds i8, ptr [[S]], i64 [[INDVARS_IV_NEXT_4]]
 ; CHECK-NEXT:    store i8 [[CONV_5]], ptr [[ARRAYIDX_5]], align 1
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_5:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_4]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_5:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 6
 ; CHECK-NEXT:    [[TMP30:%.*]] = trunc i64 [[INDVARS_IV_NEXT_5]] to i32
 ; CHECK-NEXT:    [[SHL_6:%.*]] = shl i32 1, [[TMP30]]
 ; CHECK-NEXT:    [[AND_6:%.*]] = and i32 [[SHL_6]], [[X]]
@@ -165,7 +164,7 @@ define ptr @f(ptr returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    [[CONV_6:%.*]] = select i1 [[TOBOOL_6]], i8 48, i8 49
 ; CHECK-NEXT:    [[ARRAYIDX_6:%.*]] = getelementptr inbounds i8, ptr [[S]], i64 [[INDVARS_IV_NEXT_5]]
 ; CHECK-NEXT:    store i8 [[CONV_6]], ptr [[ARRAYIDX_6]], align 1
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_6:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_5]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_6:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 7
 ; CHECK-NEXT:    [[TMP31:%.*]] = trunc i64 [[INDVARS_IV_NEXT_6]] to i32
 ; CHECK-NEXT:    [[SHL_7:%.*]] = shl i32 1, [[TMP31]]
 ; CHECK-NEXT:    [[AND_7:%.*]] = and i32 [[SHL_7]], [[X]]
@@ -173,7 +172,7 @@ define ptr @f(ptr returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    [[CONV_7:%.*]] = select i1 [[TOBOOL_7]], i8 48, i8 49
 ; CHECK-NEXT:    [[ARRAYIDX_7:%.*]] = getelementptr inbounds i8, ptr [[S]], i64 [[INDVARS_IV_NEXT_6]]
 ; CHECK-NEXT:    store i8 [[CONV_7]], ptr [[ARRAYIDX_7]], align 1
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_7]] = add nuw nsw i64 [[INDVARS_IV_NEXT_6]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_7]] = add nuw nsw i64 [[INDVARS_IV]], 8
 ; CHECK-NEXT:    [[EXITCOND_7:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT_7]], [[WIDE_TRIP_COUNT]]
 ; CHECK-NEXT:    br i1 [[EXITCOND_7]], label [[FOR_END_LOOPEXIT_UNR_LCSSA:%.*]], label [[FOR_BODY]]
 ; CHECK:       for.end.loopexit.unr-lcssa:

diff --git a/llvm/test/Transforms/LoopUnroll/RISCV/unroll.ll b/llvm/test/Transforms/LoopUnroll/RISCV/unroll.ll
index 015c4e4..18d22f5 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/RISCV/unroll.ll
+++ b/llvm/test/Transforms/LoopUnroll/RISCV/unroll.ll
@@ -22,7 +22,7 @@ define dso_local void @saxpy(float %a, ptr %x, ptr %y) {
 ; CHECK-NEXT:    [[TMP3:%.*]] = load float, ptr [[ARRAYIDX2_1]], align 4
 ; CHECK-NEXT:    [[ADD_1:%.*]] = fadd fast float [[MUL_1]], [[TMP3]]
 ; CHECK-NEXT:    store float [[ADD_1]], ptr [[ARRAYIDX2_1]], align 4
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 2
 ; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds float, ptr [[X]], i64 [[INDVARS_IV_NEXT_1]]
 ; CHECK-NEXT:    [[TMP4:%.*]] = load float, ptr [[ARRAYIDX_2]], align 4
 ; CHECK-NEXT:    [[MUL_2:%.*]] = fmul fast float [[TMP4]], [[A]]
@@ -30,7 +30,7 @@ define dso_local void @saxpy(float %a, ptr %x, ptr %y) {
 ; CHECK-NEXT:    [[TMP5:%.*]] = load float, ptr [[ARRAYIDX2_2]], align 4
 ; CHECK-NEXT:    [[ADD_2:%.*]] = fadd fast float [[MUL_2]], [[TMP5]]
 ; CHECK-NEXT:    store float [[ADD_2]], ptr [[ARRAYIDX2_2]], align 4
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_1]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 3
 ; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds float, ptr [[X]], i64 [[INDVARS_IV_NEXT_2]]
 ; CHECK-NEXT:    [[TMP6:%.*]] = load float, ptr [[ARRAYIDX_3]], align 4
 ; CHECK-NEXT:    [[MUL_3:%.*]] = fmul fast float [[TMP6]], [[A]]
@@ -38,7 +38,7 @@ define dso_local void @saxpy(float %a, ptr %x, ptr %y) {
 ; CHECK-NEXT:    [[TMP7:%.*]] = load float, ptr [[ARRAYIDX2_3]], align 4
 ; CHECK-NEXT:    [[ADD_3:%.*]] = fadd fast float [[MUL_3]], [[TMP7]]
 ; CHECK-NEXT:    store float [[ADD_3]], ptr [[ARRAYIDX2_3]], align 4
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_3:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_2]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_3:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 4
 ; CHECK-NEXT:    [[ARRAYIDX_4:%.*]] = getelementptr inbounds float, ptr [[X]], i64 [[INDVARS_IV_NEXT_3]]
 ; CHECK-NEXT:    [[TMP8:%.*]] = load float, ptr [[ARRAYIDX_4]], align 4
 ; CHECK-NEXT:    [[MUL_4:%.*]] = fmul fast float [[TMP8]], [[A]]
@@ -46,7 +46,7 @@ define dso_local void @saxpy(float %a, ptr %x, ptr %y) {
 ; CHECK-NEXT:    [[TMP9:%.*]] = load float, ptr [[ARRAYIDX2_4]], align 4
 ; CHECK-NEXT:    [[ADD_4:%.*]] = fadd fast float [[MUL_4]], [[TMP9]]
 ; CHECK-NEXT:    store float [[ADD_4]], ptr [[ARRAYIDX2_4]], align 4
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_4:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_3]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_4:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 5
 ; CHECK-NEXT:    [[ARRAYIDX_5:%.*]] = getelementptr inbounds float, ptr [[X]], i64 [[INDVARS_IV_NEXT_4]]
 ; CHECK-NEXT:    [[TMP10:%.*]] = load float, ptr [[ARRAYIDX_5]], align 4
 ; CHECK-NEXT:    [[MUL_5:%.*]] = fmul fast float [[TMP10]], [[A]]
@@ -54,7 +54,7 @@ define dso_local void @saxpy(float %a, ptr %x, ptr %y) {
 ; CHECK-NEXT:    [[TMP11:%.*]] = load float, ptr [[ARRAYIDX2_5]], align 4
 ; CHECK-NEXT:    [[ADD_5:%.*]] = fadd fast float [[MUL_5]], [[TMP11]]
 ; CHECK-NEXT:    store float [[ADD_5]], ptr [[ARRAYIDX2_5]], align 4
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_5:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_4]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_5:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 6
 ; CHECK-NEXT:    [[ARRAYIDX_6:%.*]] = getelementptr inbounds float, ptr [[X]], i64 [[INDVARS_IV_NEXT_5]]
 ; CHECK-NEXT:    [[TMP12:%.*]] = load float, ptr [[ARRAYIDX_6]], align 4
 ; CHECK-NEXT:    [[MUL_6:%.*]] = fmul fast float [[TMP12]], [[A]]
@@ -62,7 +62,7 @@ define dso_local void @saxpy(float %a, ptr %x, ptr %y) {
 ; CHECK-NEXT:    [[TMP13:%.*]] = load float, ptr [[ARRAYIDX2_6]], align 4
 ; CHECK-NEXT:    [[ADD_6:%.*]] = fadd fast float [[MUL_6]], [[TMP13]]
 ; CHECK-NEXT:    store float [[ADD_6]], ptr [[ARRAYIDX2_6]], align 4
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_6:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_5]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_6:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 7
 ; CHECK-NEXT:    [[ARRAYIDX_7:%.*]] = getelementptr inbounds float, ptr [[X]], i64 [[INDVARS_IV_NEXT_6]]
 ; CHECK-NEXT:    [[TMP14:%.*]] = load float, ptr [[ARRAYIDX_7]], align 4
 ; CHECK-NEXT:    [[MUL_7:%.*]] = fmul fast float [[TMP14]], [[A]]
@@ -70,7 +70,7 @@ define dso_local void @saxpy(float %a, ptr %x, ptr %y) {
 ; CHECK-NEXT:    [[TMP15:%.*]] = load float, ptr [[ARRAYIDX2_7]], align 4
 ; CHECK-NEXT:    [[ADD_7:%.*]] = fadd fast float [[MUL_7]], [[TMP15]]
 ; CHECK-NEXT:    store float [[ADD_7]], ptr [[ARRAYIDX2_7]], align 4
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_7:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_6]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_7:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 8
 ; CHECK-NEXT:    [[ARRAYIDX_8:%.*]] = getelementptr inbounds float, ptr [[X]], i64 [[INDVARS_IV_NEXT_7]]
 ; CHECK-NEXT:    [[TMP16:%.*]] = load float, ptr [[ARRAYIDX_8]], align 4
 ; CHECK-NEXT:    [[MUL_8:%.*]] = fmul fast float [[TMP16]], [[A]]
@@ -78,7 +78,7 @@ define dso_local void @saxpy(float %a, ptr %x, ptr %y) {
 ; CHECK-NEXT:    [[TMP17:%.*]] = load float, ptr [[ARRAYIDX2_8]], align 4
 ; CHECK-NEXT:    [[ADD_8:%.*]] = fadd fast float [[MUL_8]], [[TMP17]]
 ; CHECK-NEXT:    store float [[ADD_8]], ptr [[ARRAYIDX2_8]], align 4
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_8:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_7]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_8:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 9
 ; CHECK-NEXT:    [[ARRAYIDX_9:%.*]] = getelementptr inbounds float, ptr [[X]], i64 [[INDVARS_IV_NEXT_8]]
 ; CHECK-NEXT:    [[TMP18:%.*]] = load float, ptr [[ARRAYIDX_9]], align 4
 ; CHECK-NEXT:    [[MUL_9:%.*]] = fmul fast float [[TMP18]], [[A]]
@@ -86,7 +86,7 @@ define dso_local void @saxpy(float %a, ptr %x, ptr %y) {
 ; CHECK-NEXT:    [[TMP19:%.*]] = load float, ptr [[ARRAYIDX2_9]], align 4
 ; CHECK-NEXT:    [[ADD_9:%.*]] = fadd fast float [[MUL_9]], [[TMP19]]
 ; CHECK-NEXT:    store float [[ADD_9]], ptr [[ARRAYIDX2_9]], align 4
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_9:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_8]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_9:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 10
 ; CHECK-NEXT:    [[ARRAYIDX_10:%.*]] = getelementptr inbounds float, ptr [[X]], i64 [[INDVARS_IV_NEXT_9]]
 ; CHECK-NEXT:    [[TMP20:%.*]] = load float, ptr [[ARRAYIDX_10]], align 4
 ; CHECK-NEXT:    [[MUL_10:%.*]] = fmul fast float [[TMP20]], [[A]]
@@ -94,7 +94,7 @@ define dso_local void @saxpy(float %a, ptr %x, ptr %y) {
 ; CHECK-NEXT:    [[TMP21:%.*]] = load float, ptr [[ARRAYIDX2_10]], align 4
 ; CHECK-NEXT:    [[ADD_10:%.*]] = fadd fast float [[MUL_10]], [[TMP21]]
 ; CHECK-NEXT:    store float [[ADD_10]], ptr [[ARRAYIDX2_10]], align 4
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_10:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_9]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_10:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 11
 ; CHECK-NEXT:    [[ARRAYIDX_11:%.*]] = getelementptr inbounds float, ptr [[X]], i64 [[INDVARS_IV_NEXT_10]]
 ; CHECK-NEXT:    [[TMP22:%.*]] = load float, ptr [[ARRAYIDX_11]], align 4
 ; CHECK-NEXT:    [[MUL_11:%.*]] = fmul fast float [[TMP22]], [[A]]
@@ -102,7 +102,7 @@ define dso_local void @saxpy(float %a, ptr %x, ptr %y) {
 ; CHECK-NEXT:    [[TMP23:%.*]] = load float, ptr [[ARRAYIDX2_11]], align 4
 ; CHECK-NEXT:    [[ADD_11:%.*]] = fadd fast float [[MUL_11]], [[TMP23]]
 ; CHECK-NEXT:    store float [[ADD_11]], ptr [[ARRAYIDX2_11]], align 4
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_11:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_10]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_11:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 12
 ; CHECK-NEXT:    [[ARRAYIDX_12:%.*]] = getelementptr inbounds float, ptr [[X]], i64 [[INDVARS_IV_NEXT_11]]
 ; CHECK-NEXT:    [[TMP24:%.*]] = load float, ptr [[ARRAYIDX_12]], align 4
 ; CHECK-NEXT:    [[MUL_12:%.*]] = fmul fast float [[TMP24]], [[A]]
@@ -110,7 +110,7 @@ define dso_local void @saxpy(float %a, ptr %x, ptr %y) {
 ; CHECK-NEXT:    [[TMP25:%.*]] = load float, ptr [[ARRAYIDX2_12]], align 4
 ; CHECK-NEXT:    [[ADD_12:%.*]] = fadd fast float [[MUL_12]], [[TMP25]]
 ; CHECK-NEXT:    store float [[ADD_12]], ptr [[ARRAYIDX2_12]], align 4
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_12:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_11]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_12:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 13
 ; CHECK-NEXT:    [[ARRAYIDX_13:%.*]] = getelementptr inbounds float, ptr [[X]], i64 [[INDVARS_IV_NEXT_12]]
 ; CHECK-NEXT:    [[TMP26:%.*]] = load float, ptr [[ARRAYIDX_13]], align 4
 ; CHECK-NEXT:    [[MUL_13:%.*]] = fmul fast float [[TMP26]], [[A]]
@@ -118,7 +118,7 @@ define dso_local void @saxpy(float %a, ptr %x, ptr %y) {
 ; CHECK-NEXT:    [[TMP27:%.*]] = load float, ptr [[ARRAYIDX2_13]], align 4
 ; CHECK-NEXT:    [[ADD_13:%.*]] = fadd fast float [[MUL_13]], [[TMP27]]
 ; CHECK-NEXT:    store float [[ADD_13]], ptr [[ARRAYIDX2_13]], align 4
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_13:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_12]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_13:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 14
 ; CHECK-NEXT:    [[ARRAYIDX_14:%.*]] = getelementptr inbounds float, ptr [[X]], i64 [[INDVARS_IV_NEXT_13]]
 ; CHECK-NEXT:    [[TMP28:%.*]] = load float, ptr [[ARRAYIDX_14]], align 4
 ; CHECK-NEXT:    [[MUL_14:%.*]] = fmul fast float [[TMP28]], [[A]]
@@ -126,7 +126,7 @@ define dso_local void @saxpy(float %a, ptr %x, ptr %y) {
 ; CHECK-NEXT:    [[TMP29:%.*]] = load float, ptr [[ARRAYIDX2_14]], align 4
 ; CHECK-NEXT:    [[ADD_14:%.*]] = fadd fast float [[MUL_14]], [[TMP29]]
 ; CHECK-NEXT:    store float [[ADD_14]], ptr [[ARRAYIDX2_14]], align 4
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_14:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_13]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_14:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 15
 ; CHECK-NEXT:    [[ARRAYIDX_15:%.*]] = getelementptr inbounds float, ptr [[X]], i64 [[INDVARS_IV_NEXT_14]]
 ; CHECK-NEXT:    [[TMP30:%.*]] = load float, ptr [[ARRAYIDX_15]], align 4
 ; CHECK-NEXT:    [[MUL_15:%.*]] = fmul fast float [[TMP30]], [[A]]
@@ -134,7 +134,7 @@ define dso_local void @saxpy(float %a, ptr %x, ptr %y) {
 ; CHECK-NEXT:    [[TMP31:%.*]] = load float, ptr [[ARRAYIDX2_15]], align 4
 ; CHECK-NEXT:    [[ADD_15:%.*]] = fadd fast float [[MUL_15]], [[TMP31]]
 ; CHECK-NEXT:    store float [[ADD_15]], ptr [[ARRAYIDX2_15]], align 4
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_15]] = add nuw nsw i64 [[INDVARS_IV_NEXT_14]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_15]] = add nuw nsw i64 [[INDVARS_IV]], 16
 ; CHECK-NEXT:    [[EXITCOND_NOT_15:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT_15]], 64
 ; CHECK-NEXT:    br i1 [[EXITCOND_NOT_15]], label [[EXIT_LOOP:%.*]], label [[FOR_BODY]]
 ; CHECK:       exit_loop:

diff --git a/llvm/test/Transforms/LoopUnroll/X86/mmx.ll b/llvm/test/Transforms/LoopUnroll/X86/mmx.ll
index 979ce63..b460b79 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/X86/mmx.ll
+++ b/llvm/test/Transforms/LoopUnroll/X86/mmx.ll
@@ -10,14 +10,8 @@ define x86_mmx @f() #0 {
 ; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
 ; CHECK:       for.body:
 ; CHECK-NEXT:    [[PHI:%.*]] = phi i32 [ 1, [[ENTRY:%.*]] ], [ [[ADD_7:%.*]], [[FOR_BODY]] ]
-; CHECK-NEXT:    [[ADD:%.*]] = add nuw nsw i32 [[PHI]], 1
-; CHECK-NEXT:    [[ADD_1:%.*]] = add nuw nsw i32 [[ADD]], 1
-; CHECK-NEXT:    [[ADD_2:%.*]] = add nuw nsw i32 [[ADD_1]], 1
-; CHECK-NEXT:    [[ADD_3:%.*]] = add nuw nsw i32 [[ADD_2]], 1
-; CHECK-NEXT:    [[ADD_4:%.*]] = add nuw nsw i32 [[ADD_3]], 1
-; CHECK-NEXT:    [[ADD_5:%.*]] = add nuw nsw i32 [[ADD_4]], 1
-; CHECK-NEXT:    [[ADD_6:%.*]] = add i32 [[ADD_5]], 1
-; CHECK-NEXT:    [[ADD_7]] = add nuw nsw i32 [[ADD_6]], 1
+; CHECK-NEXT:    [[ADD_6:%.*]] = add i32 [[PHI]], 7
+; CHECK-NEXT:    [[ADD_7]] = add i32 [[PHI]], 8
 ; CHECK-NEXT:    [[CMP_7:%.*]] = icmp eq i32 [[ADD_6]], 0
 ; CHECK-NEXT:    br i1 [[CMP_7]], label [[EXIT:%.*]], label [[FOR_BODY]]
 ; CHECK:       exit:

diff --git a/llvm/test/Transforms/LoopUnroll/add-folding-overflow.ll b/llvm/test/Transforms/LoopUnroll/add-folding-overflow.ll
index 07d2dd8..b265e16 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/add-folding-overflow.ll
+++ b/llvm/test/Transforms/LoopUnroll/add-folding-overflow.ll
@@ -22,43 +22,40 @@ define void @test(i8 %start1, i8 %start2, i8 %start3, i8 %start4) {
 ; CHECK-NEXT:    call void @use(i8 [[IV3_NEXT]])
 ; CHECK-NEXT:    [[IV4_NEXT:%.*]] = add nsw i8 [[IV4]], 127
 ; CHECK-NEXT:    call void @use(i8 [[IV4_NEXT]])
-; CHECK-NEXT:    [[IV_NEXT:%.*]] = add nuw nsw i32 [[IV]], 1
-; CHECK-NEXT:    [[IV1_NEXT_1:%.*]] = add i8 [[IV1_NEXT]], 1
+; CHECK-NEXT:    [[IV1_NEXT_1:%.*]] = add i8 [[IV1]], 2
 ; CHECK-NEXT:    call void @use(i8 [[IV1_NEXT_1]])
-; CHECK-NEXT:    [[IV2_NEXT_1:%.*]] = add nuw i8 [[IV2_NEXT]], 1
+; CHECK-NEXT:    [[IV2_NEXT_1:%.*]] = add nuw i8 [[IV2]], 2
 ; CHECK-NEXT:    call void @use(i8 [[IV2_NEXT_1]])
-; CHECK-NEXT:    [[IV3_NEXT_1:%.*]] = add nsw i8 [[IV3_NEXT]], 1
+; CHECK-NEXT:    [[IV3_NEXT_1:%.*]] = add nsw i8 [[IV3]], 2
 ; CHECK-NEXT:    call void @use(i8 [[IV3_NEXT_1]])
-; CHECK-NEXT:    [[IV4_NEXT_1:%.*]] = add nsw i8 [[IV4_NEXT]], 127
+; CHECK-NEXT:    [[IV4_NEXT_1:%.*]] = add i8 [[IV4]], -2
 ; CHECK-NEXT:    call void @use(i8 [[IV4_NEXT_1]])
-; CHECK-NEXT:    [[IV_NEXT_1:%.*]] = add nuw nsw i32 [[IV_NEXT]], 1
-; CHECK-NEXT:    [[IV1_NEXT_2:%.*]] = add i8 [[IV1_NEXT_1]], 1
+; CHECK-NEXT:    [[IV1_NEXT_2:%.*]] = add i8 [[IV1]], 3
 ; CHECK-NEXT:    call void @use(i8 [[IV1_NEXT_2]])
-; CHECK-NEXT:    [[IV2_NEXT_2:%.*]] = add nuw i8 [[IV2_NEXT_1]], 1
+; CHECK-NEXT:    [[IV2_NEXT_2:%.*]] = add nuw i8 [[IV2]], 3
 ; CHECK-NEXT:    call void @use(i8 [[IV2_NEXT_2]])
-; CHECK-NEXT:    [[IV3_NEXT_2:%.*]] = add nsw i8 [[IV3_NEXT_1]], 1
+; CHECK-NEXT:    [[IV3_NEXT_2:%.*]] = add nsw i8 [[IV3]], 3
 ; CHECK-NEXT:    call void @use(i8 [[IV3_NEXT_2]])
-; CHECK-NEXT:    [[IV4_NEXT_2:%.*]] = add nsw i8 [[IV4_NEXT_1]], 127
+; CHECK-NEXT:    [[IV4_NEXT_2:%.*]] = add i8 [[IV4]], 125
 ; CHECK-NEXT:    call void @use(i8 [[IV4_NEXT_2]])
-; CHECK-NEXT:    [[IV_NEXT_2:%.*]] = add nuw nsw i32 [[IV_NEXT_1]], 1
-; CHECK-NEXT:    [[IV1_NEXT_3:%.*]] = add i8 [[IV1_NEXT_2]], 1
+; CHECK-NEXT:    [[IV1_NEXT_3:%.*]] = add i8 [[IV1]], 4
 ; CHECK-NEXT:    call void @use(i8 [[IV1_NEXT_3]])
-; CHECK-NEXT:    [[IV2_NEXT_3:%.*]] = add nuw i8 [[IV2_NEXT_2]], 1
+; CHECK-NEXT:    [[IV2_NEXT_3:%.*]] = add nuw i8 [[IV2]], 4
 ; CHECK-NEXT:    call void @use(i8 [[IV2_NEXT_3]])
-; CHECK-NEXT:    [[IV3_NEXT_3:%.*]] = add nsw i8 [[IV3_NEXT_2]], 1
+; CHECK-NEXT:    [[IV3_NEXT_3:%.*]] = add nsw i8 [[IV3]], 4
 ; CHECK-NEXT:    call void @use(i8 [[IV3_NEXT_3]])
-; CHECK-NEXT:    [[IV4_NEXT_3:%.*]] = add nsw i8 [[IV4_NEXT_2]], 127
+; CHECK-NEXT:    [[IV4_NEXT_3:%.*]] = add i8 [[IV4]], -4
 ; CHECK-NEXT:    call void @use(i8 [[IV4_NEXT_3]])
-; CHECK-NEXT:    [[IV_NEXT_3:%.*]] = add nuw nsw i32 [[IV_NEXT_2]], 1
-; CHECK-NEXT:    [[IV1_NEXT_4]] = add i8 [[IV1_NEXT_3]], 1
+; CHECK-NEXT:    [[IV_NEXT_3:%.*]] = add nuw nsw i32 [[IV]], 4
+; CHECK-NEXT:    [[IV1_NEXT_4]] = add i8 [[IV1]], 5
 ; CHECK-NEXT:    call void @use(i8 [[IV1_NEXT_4]])
-; CHECK-NEXT:    [[IV2_NEXT_4]] = add nuw i8 [[IV2_NEXT_3]], 1
+; CHECK-NEXT:    [[IV2_NEXT_4]] = add nuw i8 [[IV2]], 5
 ; CHECK-NEXT:    call void @use(i8 [[IV2_NEXT_4]])
-; CHECK-NEXT:    [[IV3_NEXT_4]] = add nsw i8 [[IV3_NEXT_3]], 1
+; CHECK-NEXT:    [[IV3_NEXT_4]] = add nsw i8 [[IV3]], 5
 ; CHECK-NEXT:    call void @use(i8 [[IV3_NEXT_4]])
-; CHECK-NEXT:    [[IV4_NEXT_4]] = add nsw i8 [[IV4_NEXT_3]], 127
+; CHECK-NEXT:    [[IV4_NEXT_4]] = add i8 [[IV4]], 123
 ; CHECK-NEXT:    call void @use(i8 [[IV4_NEXT_4]])
-; CHECK-NEXT:    [[IV_NEXT_4]] = add nuw nsw i32 [[IV_NEXT_3]], 1
+; CHECK-NEXT:    [[IV_NEXT_4]] = add nuw nsw i32 [[IV]], 5
 ; CHECK-NEXT:    [[CMP_4:%.*]] = icmp eq i32 [[IV_NEXT_3]], 1024
 ; CHECK-NEXT:    br i1 [[CMP_4]], label [[EXIT:%.*]], label [[LOOP]]
 ; CHECK:       exit:

diff --git a/llvm/test/Transforms/LoopUnroll/full-unroll-keep-first-exit.ll b/llvm/test/Transforms/LoopUnroll/full-unroll-keep-first-exit.ll
index 18cd7cb..a8752db 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/full-unroll-keep-first-exit.ll
+++ b/llvm/test/Transforms/LoopUnroll/full-unroll-keep-first-exit.ll
@@ -54,7 +54,7 @@ define void @s32_max2(i32 %n, ptr %p) {
 ; CHECK:       do.body.1:
 ; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr i32, ptr [[P]], i32 [[INC]]
 ; CHECK-NEXT:    store i32 [[INC]], ptr [[ARRAYIDX_1]], align 4
-; CHECK-NEXT:    [[INC_1:%.*]] = add i32 [[INC]], 1
+; CHECK-NEXT:    [[INC_1:%.*]] = add i32 [[N]], 2
 ; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr i32, ptr [[P]], i32 [[INC_1]]
 ; CHECK-NEXT:    store i32 [[INC_1]], ptr [[ARRAYIDX_2]], align 4
 ; CHECK-NEXT:    br label [[DO_END]]
@@ -202,7 +202,7 @@ define void @u32_max2(i32 %n, ptr %p) {
 ; CHECK:       do.body.1:
 ; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr i32, ptr [[P]], i32 [[INC]]
 ; CHECK-NEXT:    store i32 [[INC]], ptr [[ARRAYIDX_1]], align 4
-; CHECK-NEXT:    [[INC_1:%.*]] = add i32 [[INC]], 1
+; CHECK-NEXT:    [[INC_1:%.*]] = add i32 [[N]], 2
 ; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr i32, ptr [[P]], i32 [[INC_1]]
 ; CHECK-NEXT:    store i32 [[INC_1]], ptr [[ARRAYIDX_2]], align 4
 ; CHECK-NEXT:    br label [[DO_END]]

diff --git a/llvm/test/Transforms/LoopUnroll/nonlatchcondbr.ll b/llvm/test/Transforms/LoopUnroll/nonlatchcondbr.ll
index ec55d52..b42d538 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/nonlatchcondbr.ll
+++ b/llvm/test/Transforms/LoopUnroll/nonlatchcondbr.ll
@@ -93,7 +93,7 @@ define void @test2(ptr noalias %A) {
 ; CHECK-NEXT:    [[ARRAYIDX_PHI_TRANS_INSERT:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INC]]
 ; CHECK-NEXT:    [[DOTPRE:%.*]] = load i32, ptr [[ARRAYIDX_PHI_TRANS_INSERT]], align 4
 ; CHECK-NEXT:    call void @bar(i32 [[DOTPRE]])
-; CHECK-NEXT:    [[INC_1:%.*]] = add nuw nsw i64 [[INC]], 1
+; CHECK-NEXT:    [[INC_1:%.*]] = add nuw nsw i64 [[I]], 2
 ; CHECK-NEXT:    br i1 true, label [[FOR_BODY_1:%.*]], label [[FOR_BODY_FOR_BODY_CRIT_EDGE_1:%.*]]
 ; CHECK:       for.body.1:
 ; CHECK-NEXT:    [[CMP_1:%.*]] = call i1 @foo(i64 [[INC]])
@@ -102,7 +102,7 @@ define void @test2(ptr noalias %A) {
 ; CHECK-NEXT:    [[ARRAYIDX_PHI_TRANS_INSERT_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INC_1]]
 ; CHECK-NEXT:    [[DOTPRE_1:%.*]] = load i32, ptr [[ARRAYIDX_PHI_TRANS_INSERT_1]], align 4
 ; CHECK-NEXT:    call void @bar(i32 [[DOTPRE_1]])
-; CHECK-NEXT:    [[INC_2:%.*]] = add nuw nsw i64 [[INC_1]], 1
+; CHECK-NEXT:    [[INC_2:%.*]] = add nuw nsw i64 [[I]], 3
 ; CHECK-NEXT:    br i1 true, label [[FOR_BODY_2:%.*]], label [[FOR_BODY_FOR_BODY_CRIT_EDGE_2:%.*]]
 ; CHECK:       for.body.2:
 ; CHECK-NEXT:    [[CMP_2:%.*]] = call i1 @foo(i64 [[INC_1]])
@@ -111,7 +111,7 @@ define void @test2(ptr noalias %A) {
 ; CHECK-NEXT:    [[ARRAYIDX_PHI_TRANS_INSERT_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INC_2]]
 ; CHECK-NEXT:    [[DOTPRE_2:%.*]] = load i32, ptr [[ARRAYIDX_PHI_TRANS_INSERT_2]], align 4
 ; CHECK-NEXT:    call void @bar(i32 [[DOTPRE_2]])
-; CHECK-NEXT:    [[INC_3]] = add nsw i64 [[INC_2]], 1
+; CHECK-NEXT:    [[INC_3]] = add nsw i64 [[I]], 4
 ; CHECK-NEXT:    br i1 true, label [[FOR_BODY_3:%.*]], label [[FOR_BODY_FOR_BODY_CRIT_EDGE_3]]
 ; CHECK:       for.body.3:
 ; CHECK-NEXT:    [[CMP_3:%.*]] = call i1 @foo(i64 [[INC_2]])

diff --git a/llvm/test/Transforms/LoopUnroll/partial-unroll-non-latch-exit.ll b/llvm/test/Transforms/LoopUnroll/partial-unroll-non-latch-exit.ll
index 7738ffb..28e14c4 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/partial-unroll-non-latch-exit.ll
+++ b/llvm/test/Transforms/LoopUnroll/partial-unroll-non-latch-exit.ll
@@ -23,7 +23,7 @@ define i1 @test(ptr %a1, ptr %a2) {
 ; CHECK:       loop.1:
 ; CHECK-NEXT:    br label [[LATCH_1:%.*]]
 ; CHECK:       latch.1:
-; CHECK-NEXT:    [[IV_NEXT_1:%.*]] = add nuw nsw i64 [[IV_NEXT]], 1
+; CHECK-NEXT:    [[IV_NEXT_1:%.*]] = add nuw nsw i64 [[IV]], 2
 ; CHECK-NEXT:    [[GEP1_1:%.*]] = getelementptr inbounds i64, ptr [[A1]], i64 [[IV_NEXT]]
 ; CHECK-NEXT:    [[GEP2_1:%.*]] = getelementptr inbounds i64, ptr [[A2]], i64 [[IV_NEXT]]
 ; CHECK-NEXT:    [[LOAD1_1:%.*]] = load i64, ptr [[GEP1_1]], align 8
@@ -33,7 +33,7 @@ define i1 @test(ptr %a1, ptr %a2) {
 ; CHECK:       loop.2:
 ; CHECK-NEXT:    br label [[LATCH_2:%.*]]
 ; CHECK:       latch.2:
-; CHECK-NEXT:    [[IV_NEXT_2:%.*]] = add nuw nsw i64 [[IV_NEXT_1]], 1
+; CHECK-NEXT:    [[IV_NEXT_2:%.*]] = add nuw nsw i64 [[IV]], 3
 ; CHECK-NEXT:    [[GEP1_2:%.*]] = getelementptr inbounds i64, ptr [[A1]], i64 [[IV_NEXT_1]]
 ; CHECK-NEXT:    [[GEP2_2:%.*]] = getelementptr inbounds i64, ptr [[A2]], i64 [[IV_NEXT_1]]
 ; CHECK-NEXT:    [[LOAD1_2:%.*]] = load i64, ptr [[GEP1_2]], align 8
@@ -43,7 +43,7 @@ define i1 @test(ptr %a1, ptr %a2) {
 ; CHECK:       loop.3:
 ; CHECK-NEXT:    br label [[LATCH_3:%.*]]
 ; CHECK:       latch.3:
-; CHECK-NEXT:    [[IV_NEXT_3:%.*]] = add nuw nsw i64 [[IV_NEXT_2]], 1
+; CHECK-NEXT:    [[IV_NEXT_3:%.*]] = add nuw nsw i64 [[IV]], 4
 ; CHECK-NEXT:    [[GEP1_3:%.*]] = getelementptr inbounds i64, ptr [[A1]], i64 [[IV_NEXT_2]]
 ; CHECK-NEXT:    [[GEP2_3:%.*]] = getelementptr inbounds i64, ptr [[A2]], i64 [[IV_NEXT_2]]
 ; CHECK-NEXT:    [[LOAD1_3:%.*]] = load i64, ptr [[GEP1_3]], align 8
@@ -54,7 +54,7 @@ define i1 @test(ptr %a1, ptr %a2) {
 ; CHECK-NEXT:    [[EXITCOND_4:%.*]] = icmp eq i64 [[IV_NEXT_3]], 24
 ; CHECK-NEXT:    br i1 [[EXITCOND_4]], label [[EXIT]], label [[LATCH_4]]
 ; CHECK:       latch.4:
-; CHECK-NEXT:    [[IV_NEXT_4]] = add nuw nsw i64 [[IV_NEXT_3]], 1
+; CHECK-NEXT:    [[IV_NEXT_4]] = add nuw nsw i64 [[IV]], 5
 ; CHECK-NEXT:    [[GEP1_4:%.*]] = getelementptr inbounds i64, ptr [[A1]], i64 [[IV_NEXT_3]]
 ; CHECK-NEXT:    [[GEP2_4:%.*]] = getelementptr inbounds i64, ptr [[A2]], i64 [[IV_NEXT_3]]
 ; CHECK-NEXT:    [[LOAD1_4:%.*]] = load i64, ptr [[GEP1_4]], align 8

diff --git a/llvm/test/Transforms/LoopUnroll/partially-unroll-unconditional-latch.ll b/llvm/test/Transforms/LoopUnroll/partially-unroll-unconditional-latch.ll
index 0d7c4d9..8d50437 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/partially-unroll-unconditional-latch.ll
+++ b/llvm/test/Transforms/LoopUnroll/partially-unroll-unconditional-latch.ll
@@ -19,19 +19,19 @@ define i32 @test_partial_unroll_with_breakout_at_iter0() {
 ; CHECK-NEXT:    br label [[FOR_LATCH:%.*]]
 ; CHECK:       for.latch:
 ; CHECK-NEXT:    [[RED_NEXT_1:%.*]] = add nuw nsw i32 10, [[RED_NEXT]]
-; CHECK-NEXT:    [[IV_NEXT_1:%.*]] = add nuw nsw i64 [[IV_NEXT]], 2
+; CHECK-NEXT:    [[IV_NEXT_1:%.*]] = add nuw nsw i64 [[IV]], 4
 ; CHECK-NEXT:    [[PTR_1:%.*]] = getelementptr inbounds [344 x i32], ptr @table, i64 0, i64 [[IV_NEXT_1]]
 ; CHECK-NEXT:    store i32 [[RED_NEXT_1]], ptr [[PTR_1]], align 4
 ; CHECK-NEXT:    br label [[FOR_LATCH_1:%.*]]
 ; CHECK:       for.latch.1:
 ; CHECK-NEXT:    [[RED_NEXT_2:%.*]] = add nuw nsw i32 10, [[RED_NEXT_1]]
-; CHECK-NEXT:    [[IV_NEXT_2:%.*]] = add nuw nsw i64 [[IV_NEXT_1]], 2
+; CHECK-NEXT:    [[IV_NEXT_2:%.*]] = add nuw nsw i64 [[IV]], 6
 ; CHECK-NEXT:    [[PTR_2:%.*]] = getelementptr inbounds [344 x i32], ptr @table, i64 0, i64 [[IV_NEXT_2]]
 ; CHECK-NEXT:    store i32 [[RED_NEXT_2]], ptr [[PTR_2]], align 4
 ; CHECK-NEXT:    br label [[FOR_LATCH_2:%.*]]
 ; CHECK:       for.latch.2:
 ; CHECK-NEXT:    [[RED_NEXT_3]] = add nuw nsw i32 10, [[RED_NEXT_2]]
-; CHECK-NEXT:    [[IV_NEXT_3]] = add nuw nsw i64 [[IV_NEXT_2]], 2
+; CHECK-NEXT:    [[IV_NEXT_3]] = add nuw nsw i64 [[IV]], 8
 ; CHECK-NEXT:    [[PTR_3:%.*]] = getelementptr inbounds [344 x i32], ptr @table, i64 0, i64 [[IV_NEXT_3]]
 ; CHECK-NEXT:    store i32 [[RED_NEXT_3]], ptr [[PTR_3]], align 4
 ; CHECK-NEXT:    [[EXITCOND_1_I_3:%.*]] = icmp eq i64 [[IV_NEXT_3]], 344

diff --git a/llvm/test/Transforms/LoopUnroll/runtime-exit-phi-scev-invalidation.ll b/llvm/test/Transforms/LoopUnroll/runtime-exit-phi-scev-invalidation.ll
index 3f0e2b3..abf8df4 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/runtime-exit-phi-scev-invalidation.ll
+++ b/llvm/test/Transforms/LoopUnroll/runtime-exit-phi-scev-invalidation.ll
@@ -27,54 +27,40 @@ define void @pr56282() {
 ; CHECK:       inner.1.header:
 ; CHECK-NEXT:    [[INNER_1_IV:%.*]] = phi i64 [ 0, [[OUTER_HEADER_NEW]] ], [ [[INNER_1_IV_NEXT_7:%.*]], [[INNER_1_LATCH_7:%.*]] ]
 ; CHECK-NEXT:    [[NITER:%.*]] = phi i64 [ 0, [[OUTER_HEADER_NEW]] ], [ [[NITER_NEXT_7:%.*]], [[INNER_1_LATCH_7]] ]
-; CHECK-NEXT:    [[INNER_1_IV_NEXT:%.*]] = add nuw nsw i64 [[INNER_1_IV]], 1
 ; CHECK-NEXT:    [[V:%.*]] = call i32 @get()
 ; CHECK-NEXT:    [[C_1:%.*]] = icmp ugt i32 [[V]], 0
 ; CHECK-NEXT:    br i1 [[C_1]], label [[INNER_1_LATCH:%.*]], label [[EXIT_DEOPT_LOOPEXIT:%.*]]
 ; CHECK:       inner.1.latch:
-; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i64 [[NITER]], 1
-; CHECK-NEXT:    [[INNER_1_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INNER_1_IV_NEXT]], 1
 ; CHECK-NEXT:    [[V_1:%.*]] = call i32 @get()
 ; CHECK-NEXT:    [[C_1_1:%.*]] = icmp ugt i32 [[V_1]], 0
 ; CHECK-NEXT:    br i1 [[C_1_1]], label [[INNER_1_LATCH_1:%.*]], label [[EXIT_DEOPT_LOOPEXIT]]
 ; CHECK:       inner.1.latch.1:
-; CHECK-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i64 [[NITER_NEXT]], 1
-; CHECK-NEXT:    [[INNER_1_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INNER_1_IV_NEXT_1]], 1
 ; CHECK-NEXT:    [[V_2:%.*]] = call i32 @get()
 ; CHECK-NEXT:    [[C_1_2:%.*]] = icmp ugt i32 [[V_2]], 0
 ; CHECK-NEXT:    br i1 [[C_1_2]], label [[INNER_1_LATCH_2:%.*]], label [[EXIT_DEOPT_LOOPEXIT]]
 ; CHECK:       inner.1.latch.2:
-; CHECK-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i64 [[NITER_NEXT_1]], 1
-; CHECK-NEXT:    [[INNER_1_IV_NEXT_3:%.*]] = add nuw nsw i64 [[INNER_1_IV_NEXT_2]], 1
 ; CHECK-NEXT:    [[V_3:%.*]] = call i32 @get()
 ; CHECK-NEXT:    [[C_1_3:%.*]] = icmp ugt i32 [[V_3]], 0
 ; CHECK-NEXT:    br i1 [[C_1_3]], label [[INNER_1_LATCH_3:%.*]], label [[EXIT_DEOPT_LOOPEXIT]]
 ; CHECK:       inner.1.latch.3:
-; CHECK-NEXT:    [[NITER_NEXT_3:%.*]] = add nuw nsw i64 [[NITER_NEXT_2]], 1
-; CHECK-NEXT:    [[INNER_1_IV_NEXT_4:%.*]] = add nuw nsw i64 [[INNER_1_IV_NEXT_3]], 1
 ; CHECK-NEXT:    [[V_4:%.*]] = call i32 @get()
 ; CHECK-NEXT:    [[C_1_4:%.*]] = icmp ugt i32 [[V_4]], 0
 ; CHECK-NEXT:    br i1 [[C_1_4]], label [[INNER_1_LATCH_4:%.*]], label [[EXIT_DEOPT_LOOPEXIT]]
 ; CHECK:       inner.1.latch.4:
-; CHECK-NEXT:    [[NITER_NEXT_4:%.*]] = add nuw nsw i64 [[NITER_NEXT_3]], 1
-; CHECK-NEXT:    [[INNER_1_IV_NEXT_5:%.*]] = add nuw nsw i64 [[INNER_1_IV_NEXT_4]], 1
 ; CHECK-NEXT:    [[V_5:%.*]] = call i32 @get()
 ; CHECK-NEXT:    [[C_1_5:%.*]] = icmp ugt i32 [[V_5]], 0
 ; CHECK-NEXT:    br i1 [[C_1_5]], label [[INNER_1_LATCH_5:%.*]], label [[EXIT_DEOPT_LOOPEXIT]]
 ; CHECK:       inner.1.latch.5:
-; CHECK-NEXT:    [[NITER_NEXT_5:%.*]] = add nuw nsw i64 [[NITER_NEXT_4]], 1
-; CHECK-NEXT:    [[INNER_1_IV_NEXT_6:%.*]] = add nuw nsw i64 [[INNER_1_IV_NEXT_5]], 1
 ; CHECK-NEXT:    [[V_6:%.*]] = call i32 @get()
 ; CHECK-NEXT:    [[C_1_6:%.*]] = icmp ugt i32 [[V_6]], 0
 ; CHECK-NEXT:    br i1 [[C_1_6]], label [[INNER_1_LATCH_6:%.*]], label [[EXIT_DEOPT_LOOPEXIT]]
 ; CHECK:       inner.1.latch.6:
-; CHECK-NEXT:    [[NITER_NEXT_6:%.*]] = add nuw nsw i64 [[NITER_NEXT_5]], 1
-; CHECK-NEXT:    [[INNER_1_IV_NEXT_7]] = add nuw nsw i64 [[INNER_1_IV_NEXT_6]], 1
+; CHECK-NEXT:    [[INNER_1_IV_NEXT_7]] = add nuw nsw i64 [[INNER_1_IV]], 8
 ; CHECK-NEXT:    [[V_7:%.*]] = call i32 @get()
 ; CHECK-NEXT:    [[C_1_7:%.*]] = icmp ugt i32 [[V_7]], 0
 ; CHECK-NEXT:    br i1 [[C_1_7]], label [[INNER_1_LATCH_7]], label [[EXIT_DEOPT_LOOPEXIT]]
 ; CHECK:       inner.1.latch.7:
-; CHECK-NEXT:    [[NITER_NEXT_7]] = add i64 [[NITER_NEXT_6]], 1
+; CHECK-NEXT:    [[NITER_NEXT_7]] = add i64 [[NITER]], 8
 ; CHECK-NEXT:    [[NITER_NCMP_7:%.*]] = icmp ne i64 [[NITER_NEXT_7]], [[UNROLL_ITER]]
 ; CHECK-NEXT:    br i1 [[NITER_NCMP_7]], label [[INNER_1_HEADER]], label [[OUTER_MIDDLE_UNR_LCSSA_LOOPEXIT:%.*]]
 ; CHECK:       outer.middle.unr-lcssa.loopexit:
@@ -213,45 +199,39 @@ define void @pr56286(i64 %x, ptr %src, ptr %dst, ptr %ptr.src) !prof !0 {
 ; CHECK:       inner.1.latch:
 ; CHECK-NEXT:    [[L_1:%.*]] = load i32, ptr [[SRC]], align 4
 ; CHECK-NEXT:    store i32 [[L_1]], ptr [[DST]], align 8
-; CHECK-NEXT:    [[INNER_1_IV_NEXT:%.*]] = add i64 [[INNER_1_IV]], 1
 ; CHECK-NEXT:    br i1 false, label [[EXIT_DEOPT_LOOPEXIT]], label [[INNER_1_LATCH_1:%.*]]
 ; CHECK:       inner.1.latch.1:
 ; CHECK-NEXT:    [[L_1_1:%.*]] = load i32, ptr [[SRC]], align 4
 ; CHECK-NEXT:    store i32 [[L_1_1]], ptr [[DST]], align 8
-; CHECK-NEXT:    [[INNER_1_IV_NEXT_1:%.*]] = add i64 [[INNER_1_IV_NEXT]], 1
 ; CHECK-NEXT:    [[CMP_1_2:%.*]] = icmp sgt i32 [[OUTER_P]], 0
 ; CHECK-NEXT:    br i1 [[CMP_1_2]], label [[EXIT_DEOPT_LOOPEXIT]], label [[INNER_1_LATCH_2:%.*]]
 ; CHECK:       inner.1.latch.2:
 ; CHECK-NEXT:    [[L_1_2:%.*]] = load i32, ptr [[SRC]], align 4
 ; CHECK-NEXT:    store i32 [[L_1_2]], ptr [[DST]], align 8
-; CHECK-NEXT:    [[INNER_1_IV_NEXT_2:%.*]] = add i64 [[INNER_1_IV_NEXT_1]], 1
 ; CHECK-NEXT:    br i1 false, label [[EXIT_DEOPT_LOOPEXIT]], label [[INNER_1_LATCH_3:%.*]]
 ; CHECK:       inner.1.latch.3:
 ; CHECK-NEXT:    [[L_1_3:%.*]] = load i32, ptr [[SRC]], align 4
 ; CHECK-NEXT:    store i32 [[L_1_3]], ptr [[DST]], align 8
-; CHECK-NEXT:    [[INNER_1_IV_NEXT_3:%.*]] = add i64 [[INNER_1_IV_NEXT_2]], 1
 ; CHECK-NEXT:    [[CMP_1_4:%.*]] = icmp sgt i32 [[OUTER_P]], 0
 ; CHECK-NEXT:    br i1 [[CMP_1_4]], label [[EXIT_DEOPT_LOOPEXIT]], label [[INNER_1_LATCH_4:%.*]]
 ; CHECK:       inner.1.latch.4:
 ; CHECK-NEXT:    [[L_1_4:%.*]] = load i32, ptr [[SRC]], align 4
 ; CHECK-NEXT:    store i32 [[L_1_4]], ptr [[DST]], align 8
-; CHECK-NEXT:    [[INNER_1_IV_NEXT_4:%.*]] = add i64 [[INNER_1_IV_NEXT_3]], 1
 ; CHECK-NEXT:    br i1 false, label [[EXIT_DEOPT_LOOPEXIT]], label [[INNER_1_LATCH_5:%.*]]
 ; CHECK:       inner.1.latch.5:
 ; CHECK-NEXT:    [[L_1_5:%.*]] = load i32, ptr [[SRC]], align 4
 ; CHECK-NEXT:    store i32 [[L_1_5]], ptr [[DST]], align 8
-; CHECK-NEXT:    [[INNER_1_IV_NEXT_5:%.*]] = add i64 [[INNER_1_IV_NEXT_4]], 1
 ; CHECK-NEXT:    [[CMP_1_6:%.*]] = icmp sgt i32 [[OUTER_P]], 0
 ; CHECK-NEXT:    br i1 [[CMP_1_6]], label [[EXIT_DEOPT_LOOPEXIT]], label [[INNER_1_LATCH_6:%.*]]
 ; CHECK:       inner.1.latch.6:
 ; CHECK-NEXT:    [[L_1_6:%.*]] = load i32, ptr [[SRC]], align 4
 ; CHECK-NEXT:    store i32 [[L_1_6]], ptr [[DST]], align 8
-; CHECK-NEXT:    [[INNER_1_IV_NEXT_6:%.*]] = add i64 [[INNER_1_IV_NEXT_5]], 1
+; CHECK-NEXT:    [[INNER_1_IV_NEXT_6:%.*]] = add i64 [[INNER_1_IV]], 7
 ; CHECK-NEXT:    br i1 false, label [[EXIT_DEOPT_LOOPEXIT]], label [[INNER_1_LATCH_7]]
 ; CHECK:       inner.1.latch.7:
 ; CHECK-NEXT:    [[L_1_7:%.*]] = load i32, ptr [[SRC]], align 4
 ; CHECK-NEXT:    store i32 [[L_1_7]], ptr [[DST]], align 8
-; CHECK-NEXT:    [[INNER_1_IV_NEXT_7]] = add i64 [[INNER_1_IV_NEXT_6]], 1
+; CHECK-NEXT:    [[INNER_1_IV_NEXT_7]] = add i64 [[INNER_1_IV]], 8
 ; CHECK-NEXT:    [[CMP_2_7:%.*]] = icmp sgt i64 [[INNER_1_IV_NEXT_6]], 0
 ; CHECK-NEXT:    br i1 [[CMP_2_7]], label [[OUTER_MIDDLE_UNR_LCSSA:%.*]], label [[INNER_1_HEADER]], !prof [[PROF5:![0-9]+]]
 ; CHECK:       outer.middle.unr-lcssa:
@@ -269,7 +249,6 @@ define void @pr56286(i64 %x, ptr %src, ptr %dst, ptr %ptr.src) !prof !0 {
 ; CHECK-NEXT:    [[ADD_1:%.*]] = add i32 [[INNER_2_IV]], [[L_2]]
 ; CHECK-NEXT:    [[TMP281:%.*]] = call i32 @use.2(ptr [[TMP27]], i32 [[ADD_1]])
 ; CHECK-NEXT:    [[TMP31:%.*]] = shl nuw nsw i32 [[TMP15]], 16
-; CHECK-NEXT:    [[TMP32:%.*]] = add nuw i32 [[TMP31]], 262144
 ; CHECK-NEXT:    call void @bar()
 ; CHECK-NEXT:    call void @bar()
 ; CHECK-NEXT:    call void @bar()
@@ -279,11 +258,11 @@ define void @pr56286(i64 %x, ptr %src, ptr %dst, ptr %ptr.src) !prof !0 {
 ; CHECK-NEXT:    call void @bar()
 ; CHECK-NEXT:    call void @bar()
 ; CHECK-NEXT:    [[L_2_1:%.*]] = load i32, ptr [[SRC]], align 8
-; CHECK-NEXT:    [[INNER_2_IV_NEXT_1:%.*]] = add i32 [[INNER_2_IV_NEXT]], 1
+; CHECK-NEXT:    [[INNER_2_IV_NEXT_1:%.*]] = add i32 [[INNER_2_IV]], 2
 ; CHECK-NEXT:    [[TMP27_1:%.*]] = load ptr, ptr [[PTR_SRC]], align 8
 ; CHECK-NEXT:    [[ADD_1_1:%.*]] = add i32 [[INNER_2_IV_NEXT]], [[L_2_1]]
 ; CHECK-NEXT:    [[TMP281_1:%.*]] = call i32 @use.2(ptr [[TMP27_1]], i32 [[ADD_1_1]])
-; CHECK-NEXT:    [[TMP32_1:%.*]] = add nuw i32 [[TMP32]], 262144
+; CHECK-NEXT:    [[TMP32_1:%.*]] = add nuw i32 [[TMP31]], 524288
 ; CHECK-NEXT:    call void @bar()
 ; CHECK-NEXT:    call void @bar()
 ; CHECK-NEXT:    call void @bar()
@@ -293,11 +272,11 @@ define void @pr56286(i64 %x, ptr %src, ptr %dst, ptr %ptr.src) !prof !0 {
 ; CHECK-NEXT:    call void @bar()
 ; CHECK-NEXT:    call void @bar()
 ; CHECK-NEXT:    [[L_2_2:%.*]] = load i32, ptr [[SRC]], align 8
-; CHECK-NEXT:    [[INNER_2_IV_NEXT_2]] = add i32 [[INNER_2_IV_NEXT_1]], 1
+; CHECK-NEXT:    [[INNER_2_IV_NEXT_2]] = add i32 [[INNER_2_IV]], 3
 ; CHECK-NEXT:    [[TMP27_2:%.*]] = load ptr, ptr [[PTR_SRC]], align 8
 ; CHECK-NEXT:    [[ADD_1_2:%.*]] = add i32 [[INNER_2_IV_NEXT_1]], [[L_2_2]]
 ; CHECK-NEXT:    [[TMP281_2:%.*]] = call i32 @use.2(ptr [[TMP27_2]], i32 [[ADD_1_2]])
-; CHECK-NEXT:    [[TMP32_2:%.*]] = add nuw i32 [[TMP32_1]], 262144
+; CHECK-NEXT:    [[TMP32_2:%.*]] = add nuw i32 [[TMP31]], 786432
 ; CHECK-NEXT:    [[TMP33_2]] = ashr exact i32 [[TMP32_2]], 16
 ; CHECK-NEXT:    call void @bar()
 ; CHECK-NEXT:    call void @bar()

diff --git a/llvm/test/Transforms/LoopUnroll/runtime-loop-at-most-two-exits.ll b/llvm/test/Transforms/LoopUnroll/runtime-loop-at-most-two-exits.ll
index 9ccee95..d5864e9 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/runtime-loop-at-most-two-exits.ll
+++ b/llvm/test/Transforms/LoopUnroll/runtime-loop-at-most-two-exits.ll
@@ -24,59 +24,52 @@ define i32 @test(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 ; ENABLED-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], [[SUM_02]]
 ; ENABLED-NEXT:    [[INDVARS_IV_NEXT:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 1
-; ENABLED-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i64 [[NITER]], 1
 ; ENABLED-NEXT:    br i1 false, label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_1:%.*]]
 ; ENABLED:       for.body.1:
 ; ENABLED-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT]]
 ; ENABLED-NEXT:    [[TMP4:%.*]] = load i32, ptr [[ARRAYIDX_1]], align 4
 ; ENABLED-NEXT:    [[ADD_1:%.*]] = add nsw i32 [[TMP4]], [[ADD]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i64 [[NITER_NEXT]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 2
 ; ENABLED-NEXT:    [[CMP_2:%.*]] = icmp eq i64 [[N]], 42
 ; ENABLED-NEXT:    br i1 [[CMP_2]], label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_2:%.*]]
 ; ENABLED:       for.body.2:
 ; ENABLED-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_1]]
 ; ENABLED-NEXT:    [[TMP5:%.*]] = load i32, ptr [[ARRAYIDX_2]], align 4
 ; ENABLED-NEXT:    [[ADD_2:%.*]] = add nsw i32 [[TMP5]], [[ADD_1]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_1]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i64 [[NITER_NEXT_1]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 3
 ; ENABLED-NEXT:    br i1 false, label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_3:%.*]]
 ; ENABLED:       for.body.3:
 ; ENABLED-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_2]]
 ; ENABLED-NEXT:    [[TMP6:%.*]] = load i32, ptr [[ARRAYIDX_3]], align 4
 ; ENABLED-NEXT:    [[ADD_3:%.*]] = add nsw i32 [[TMP6]], [[ADD_2]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_3:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_2]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_3:%.*]] = add nuw nsw i64 [[NITER_NEXT_2]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_3:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 4
 ; ENABLED-NEXT:    [[CMP_4:%.*]] = icmp eq i64 [[N]], 42
 ; ENABLED-NEXT:    br i1 [[CMP_4]], label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_4:%.*]]
 ; ENABLED:       for.body.4:
 ; ENABLED-NEXT:    [[ARRAYIDX_4:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_3]]
 ; ENABLED-NEXT:    [[TMP7:%.*]] = load i32, ptr [[ARRAYIDX_4]], align 4
 ; ENABLED-NEXT:    [[ADD_4:%.*]] = add nsw i32 [[TMP7]], [[ADD_3]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_4:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_3]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_4:%.*]] = add nuw nsw i64 [[NITER_NEXT_3]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_4:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 5
 ; ENABLED-NEXT:    br i1 false, label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_5:%.*]]
 ; ENABLED:       for.body.5:
 ; ENABLED-NEXT:    [[ARRAYIDX_5:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_4]]
 ; ENABLED-NEXT:    [[TMP8:%.*]] = load i32, ptr [[ARRAYIDX_5]], align 4
 ; ENABLED-NEXT:    [[ADD_5:%.*]] = add nsw i32 [[TMP8]], [[ADD_4]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_5:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_4]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_5:%.*]] = add nuw nsw i64 [[NITER_NEXT_4]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_5:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 6
 ; ENABLED-NEXT:    [[CMP_6:%.*]] = icmp eq i64 [[N]], 42
 ; ENABLED-NEXT:    br i1 [[CMP_6]], label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_6:%.*]]
 ; ENABLED:       for.body.6:
 ; ENABLED-NEXT:    [[ARRAYIDX_6:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_5]]
 ; ENABLED-NEXT:    [[TMP9:%.*]] = load i32, ptr [[ARRAYIDX_6]], align 4
 ; ENABLED-NEXT:    [[ADD_6:%.*]] = add nsw i32 [[TMP9]], [[ADD_5]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_6:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_5]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_6:%.*]] = add nuw nsw i64 [[NITER_NEXT_5]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_6:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 7
 ; ENABLED-NEXT:    br i1 false, label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_7]]
 ; ENABLED:       for.body.7:
 ; ENABLED-NEXT:    [[ARRAYIDX_7:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_6]]
 ; ENABLED-NEXT:    [[TMP10:%.*]] = load i32, ptr [[ARRAYIDX_7]], align 4
 ; ENABLED-NEXT:    [[ADD_7]] = add nsw i32 [[TMP10]], [[ADD_6]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_7]] = add i64 [[INDVARS_IV_NEXT_6]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_7]] = add i64 [[NITER_NEXT_6]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_7]] = add i64 [[INDVARS_IV]], 8
+; ENABLED-NEXT:    [[NITER_NEXT_7]] = add i64 [[NITER]], 8
 ; ENABLED-NEXT:    [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NEXT_7]], [[UNROLL_ITER]]
 ; ENABLED-NEXT:    br i1 [[NITER_NCMP_7]], label [[FOR_END_UNR_LCSSA_LOOPEXIT:%.*]], label [[HEADER]]
 ; ENABLED:       for.end.unr-lcssa.loopexit:

diff --git a/llvm/test/Transforms/LoopUnroll/runtime-loop-multiexit-dom-verify.ll b/llvm/test/Transforms/LoopUnroll/runtime-loop-multiexit-dom-verify.ll
index 2c923c1..de54852 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/runtime-loop-multiexit-dom-verify.ll
+++ b/llvm/test/Transforms/LoopUnroll/runtime-loop-multiexit-dom-verify.ll
@@ -27,7 +27,7 @@ define i64 @test1() {
 ; CHECK-NEXT:    [[CMP2:%.*]] = icmp ult i64 [[SHFT]], [[TRIP]]
 ; CHECK-NEXT:    br i1 [[CMP2]], label [[HEADER_1:%.*]], label [[LATCHEXIT:%.*]]
 ; CHECK:       header.1:
-; CHECK-NEXT:    [[ADD_IV_1:%.*]] = add nuw nsw i64 [[ADD_IV]], 2
+; CHECK-NEXT:    [[ADD_IV_1:%.*]] = add nuw nsw i64 [[IV]], 4
 ; CHECK-NEXT:    [[CMP1_1:%.*]] = icmp ult i64 [[ADD_IV_1]], [[TRIP]]
 ; CHECK-NEXT:    br i1 [[CMP1_1]], label [[LATCH_1:%.*]], label [[HEADEREXIT]]
 ; CHECK:       latch.1:
@@ -35,7 +35,7 @@ define i64 @test1() {
 ; CHECK-NEXT:    [[CMP2_1:%.*]] = icmp ult i64 [[SHFT_1]], [[TRIP]]
 ; CHECK-NEXT:    br i1 [[CMP2_1]], label [[HEADER_2:%.*]], label [[LATCHEXIT]]
 ; CHECK:       header.2:
-; CHECK-NEXT:    [[ADD_IV_2:%.*]] = add nuw nsw i64 [[ADD_IV_1]], 2
+; CHECK-NEXT:    [[ADD_IV_2:%.*]] = add nuw nsw i64 [[IV]], 6
 ; CHECK-NEXT:    [[CMP1_2:%.*]] = icmp ult i64 [[ADD_IV_2]], [[TRIP]]
 ; CHECK-NEXT:    br i1 [[CMP1_2]], label [[LATCH_2:%.*]], label [[HEADEREXIT]]
 ; CHECK:       latch.2:
@@ -43,7 +43,7 @@ define i64 @test1() {
 ; CHECK-NEXT:    [[CMP2_2:%.*]] = icmp ult i64 [[SHFT_2]], [[TRIP]]
 ; CHECK-NEXT:    br i1 [[CMP2_2]], label [[HEADER_3:%.*]], label [[LATCHEXIT]]
 ; CHECK:       header.3:
-; CHECK-NEXT:    [[ADD_IV_3]] = add nuw nsw i64 [[ADD_IV_2]], 2
+; CHECK-NEXT:    [[ADD_IV_3]] = add nuw nsw i64 [[IV]], 8
 ; CHECK-NEXT:    [[CMP1_3:%.*]] = icmp ult i64 [[ADD_IV_3]], [[TRIP]]
 ; CHECK-NEXT:    br i1 [[CMP1_3]], label [[LATCH_3]], label [[HEADEREXIT]]
 ; CHECK:       latch.3:
@@ -109,7 +109,7 @@ define  void @test2(i1 %cond, i32 %n) {
 ; CHECK-NEXT:    [[CMP2:%.*]] = icmp ult i64 [[SHFT]], [[TRIP]]
 ; CHECK-NEXT:    br i1 [[CMP2]], label [[HEADER_1:%.*]], label [[LATCHEXIT:%.*]]
 ; CHECK:       header.1:
-; CHECK-NEXT:    [[ADD_IV_1:%.*]] = add nuw nsw i64 [[ADD_IV]], 2
+; CHECK-NEXT:    [[ADD_IV_1:%.*]] = add nuw nsw i64 [[IV]], 4
 ; CHECK-NEXT:    [[CMP1_1:%.*]] = icmp ult i64 [[ADD_IV_1]], [[TRIP]]
 ; CHECK-NEXT:    br i1 [[CMP1_1]], label [[LATCH_1:%.*]], label [[HEADEREXIT]]
 ; CHECK:       latch.1:
@@ -117,7 +117,7 @@ define  void @test2(i1 %cond, i32 %n) {
 ; CHECK-NEXT:    [[CMP2_1:%.*]] = icmp ult i64 [[SHFT_1]], [[TRIP]]
 ; CHECK-NEXT:    br i1 [[CMP2_1]], label [[HEADER_2:%.*]], label [[LATCHEXIT]]
 ; CHECK:       header.2:
-; CHECK-NEXT:    [[ADD_IV_2:%.*]] = add nuw nsw i64 [[ADD_IV_1]], 2
+; CHECK-NEXT:    [[ADD_IV_2:%.*]] = add nuw nsw i64 [[IV]], 6
 ; CHECK-NEXT:    [[CMP1_2:%.*]] = icmp ult i64 [[ADD_IV_2]], [[TRIP]]
 ; CHECK-NEXT:    br i1 [[CMP1_2]], label [[LATCH_2:%.*]], label [[HEADEREXIT]]
 ; CHECK:       latch.2:
@@ -125,7 +125,7 @@ define  void @test2(i1 %cond, i32 %n) {
 ; CHECK-NEXT:    [[CMP2_2:%.*]] = icmp ult i64 [[SHFT_2]], [[TRIP]]
 ; CHECK-NEXT:    br i1 [[CMP2_2]], label [[HEADER_3:%.*]], label [[LATCHEXIT]]
 ; CHECK:       header.3:
-; CHECK-NEXT:    [[ADD_IV_3]] = add nuw nsw i64 [[ADD_IV_2]], 2
+; CHECK-NEXT:    [[ADD_IV_3]] = add nuw nsw i64 [[IV]], 8
 ; CHECK-NEXT:    [[CMP1_3:%.*]] = icmp ult i64 [[ADD_IV_3]], [[TRIP]]
 ; CHECK-NEXT:    br i1 [[CMP1_3]], label [[LATCH_3]], label [[HEADEREXIT]]
 ; CHECK:       latch.3:
@@ -186,7 +186,7 @@ define i64 @test3(i32 %n) {
 ; CHECK-NEXT:    [[CMP2:%.*]] = icmp ult i64 [[SHFT]], [[TRIP]]
 ; CHECK-NEXT:    br i1 [[CMP2]], label [[HEADER_1:%.*]], label [[LATCHEXIT:%.*]]
 ; CHECK:       header.1:
-; CHECK-NEXT:    [[ADD_IV_1:%.*]] = add nuw nsw i64 [[ADD_IV]], 2
+; CHECK-NEXT:    [[ADD_IV_1:%.*]] = add nuw nsw i64 [[IV]], 4
 ; CHECK-NEXT:    [[CMP1_1:%.*]] = icmp ult i64 [[ADD_IV_1]], [[TRIP]]
 ; CHECK-NEXT:    br i1 [[CMP1_1]], label [[LATCH_1:%.*]], label [[HEADEREXIT]]
 ; CHECK:       latch.1:
@@ -194,7 +194,7 @@ define i64 @test3(i32 %n) {
 ; CHECK-NEXT:    [[CMP2_1:%.*]] = icmp ult i64 [[SHFT_1]], [[TRIP]]
 ; CHECK-NEXT:    br i1 [[CMP2_1]], label [[HEADER_2:%.*]], label [[LATCHEXIT]]
 ; CHECK:       header.2:
-; CHECK-NEXT:    [[ADD_IV_2:%.*]] = add nuw nsw i64 [[ADD_IV_1]], 2
+; CHECK-NEXT:    [[ADD_IV_2:%.*]] = add nuw nsw i64 [[IV]], 6
 ; CHECK-NEXT:    [[CMP1_2:%.*]] = icmp ult i64 [[ADD_IV_2]], [[TRIP]]
 ; CHECK-NEXT:    br i1 [[CMP1_2]], label [[LATCH_2:%.*]], label [[HEADEREXIT]]
 ; CHECK:       latch.2:
@@ -202,7 +202,7 @@ define i64 @test3(i32 %n) {
 ; CHECK-NEXT:    [[CMP2_2:%.*]] = icmp ult i64 [[SHFT_2]], [[TRIP]]
 ; CHECK-NEXT:    br i1 [[CMP2_2]], label [[HEADER_3:%.*]], label [[LATCHEXIT]]
 ; CHECK:       header.3:
-; CHECK-NEXT:    [[ADD_IV_3]] = add nuw nsw i64 [[ADD_IV_2]], 2
+; CHECK-NEXT:    [[ADD_IV_3]] = add nuw nsw i64 [[IV]], 8
 ; CHECK-NEXT:    [[CMP1_3:%.*]] = icmp ult i64 [[ADD_IV_3]], [[TRIP]]
 ; CHECK-NEXT:    br i1 [[CMP1_3]], label [[LATCH_3]], label [[HEADEREXIT]]
 ; CHECK:       latch.3:
@@ -292,7 +292,6 @@ define void @test4(i16 %c3) {
 ; CHECK-NEXT:    i16 95, label [[LATCH:%.*]]
 ; CHECK-NEXT:    ]
 ; CHECK:       latch:
-; CHECK-NEXT:    [[INDVARS_IV_NEXT:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 1
 ; CHECK-NEXT:    br label [[EXITING_1:%.*]]
 ; CHECK:       exiting.1:
 ; CHECK-NEXT:    switch i16 [[C3]], label [[DEFAULT_LOOPEXIT_LOOPEXIT]] [
@@ -300,7 +299,6 @@ define void @test4(i16 %c3) {
 ; CHECK-NEXT:    i16 95, label [[LATCH_1:%.*]]
 ; CHECK-NEXT:    ]
 ; CHECK:       latch.1:
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT]], 1
 ; CHECK-NEXT:    br label [[EXITING_2:%.*]]
 ; CHECK:       exiting.2:
 ; CHECK-NEXT:    switch i16 [[C3]], label [[DEFAULT_LOOPEXIT_LOOPEXIT]] [
@@ -308,7 +306,6 @@ define void @test4(i16 %c3) {
 ; CHECK-NEXT:    i16 95, label [[LATCH_2:%.*]]
 ; CHECK-NEXT:    ]
 ; CHECK:       latch.2:
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_1]], 1
 ; CHECK-NEXT:    br label [[EXITING_3:%.*]]
 ; CHECK:       exiting.3:
 ; CHECK-NEXT:    switch i16 [[C3]], label [[DEFAULT_LOOPEXIT_LOOPEXIT]] [
@@ -316,7 +313,7 @@ define void @test4(i16 %c3) {
 ; CHECK-NEXT:    i16 95, label [[LATCH_3]]
 ; CHECK-NEXT:    ]
 ; CHECK:       latch.3:
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_3]] = add nuw nsw i64 [[INDVARS_IV_NEXT_2]], 1
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_3]] = add nuw nsw i64 [[INDVARS_IV]], 4
 ; CHECK-NEXT:    [[C2_3:%.*]] = icmp ult i64 [[INDVARS_IV_NEXT_3]], [[C1]]
 ; CHECK-NEXT:    br i1 [[C2_3]], label [[HEADER]], label [[LATCHEXIT_UNR_LCSSA:%.*]], !llvm.loop [[LOOP5:![0-9]+]]
 ; CHECK:       latchexit.unr-lcssa:
@@ -469,7 +466,6 @@ define void @test5() {
 ; CHECK:       innerLatch.3:
 ; CHECK-NEXT:    br i1 false, label [[INNERH]], label [[OUTERLATCH]], !llvm.loop [[LOOP6]]
 ; CHECK:       outerLatch:
-; CHECK-NEXT:    [[TMP6:%.*]] = add i32 [[TMP4]], 1
 ; CHECK-NEXT:    br label [[INNERH_14:%.*]]
 ; CHECK:       innerH.14:
 ; CHECK-NEXT:    [[C1_13:%.*]] = call i1 @unknown(i32 0)
@@ -504,7 +500,6 @@ define void @test5() {
 ; CHECK:       innerLatch.3.1:
 ; CHECK-NEXT:    br i1 false, label [[INNERH_14]], label [[OUTERLATCH_1]], !llvm.loop [[LOOP6]]
 ; CHECK:       outerLatch.1:
-; CHECK-NEXT:    [[TMP6_1:%.*]] = add i32 [[TMP6]], 1
 ; CHECK-NEXT:    br label [[INNERH_29:%.*]]
 ; CHECK:       innerH.29:
 ; CHECK-NEXT:    [[C1_28:%.*]] = call i1 @unknown(i32 0)
@@ -539,7 +534,6 @@ define void @test5() {
 ; CHECK:       innerLatch.3.2:
 ; CHECK-NEXT:    br i1 false, label [[INNERH_29]], label [[OUTERLATCH_2]], !llvm.loop [[LOOP6]]
 ; CHECK:       outerLatch.2:
-; CHECK-NEXT:    [[TMP6_2:%.*]] = add i32 [[TMP6_1]], 1
 ; CHECK-NEXT:    br label [[INNERH_314:%.*]]
 ; CHECK:       innerH.314:
 ; CHECK-NEXT:    [[C1_313:%.*]] = call i1 @unknown(i32 0)
@@ -574,7 +568,7 @@ define void @test5() {
 ; CHECK:       innerLatch.3.3:
 ; CHECK-NEXT:    br i1 false, label [[INNERH_314]], label [[OUTERLATCH_3]], !llvm.loop [[LOOP6]]
 ; CHECK:       outerLatch.3:
-; CHECK-NEXT:    [[TMP6_3]] = add i32 [[TMP6_2]], 1
+; CHECK-NEXT:    [[TMP6_3]] = add i32 [[TMP4]], 4
 ; CHECK-NEXT:    [[TMP7_3:%.*]] = icmp sgt i32 [[TMP6_3]], 79
 ; CHECK-NEXT:    br i1 [[TMP7_3]], label [[OUTERLATCHEXIT_LOOPEXIT_UNR_LCSSA:%.*]], label [[OUTERH]], !llvm.loop [[LOOP8:![0-9]+]]
 ; CHECK:       outerLatchExit.loopexit.unr-lcssa:
@@ -703,17 +697,17 @@ define void @test6(i64 %start) {
 ; CHECK-NEXT:    [[C1_1:%.*]] = call i1 @unknown(i32 [[IV_I32_1]])
 ; CHECK-NEXT:    br i1 [[C1_1]], label [[LATCH_1:%.*]], label [[OTHEREXIT_LOOPEXIT]]
 ; CHECK:       latch.1:
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nsw i64 [[INDVARS_IV_NEXT]], 2
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nsw i64 [[INDVARS_IV]], 4
 ; CHECK-NEXT:    [[IV_I32_2:%.*]] = trunc i64 [[INDVARS_IV_NEXT_1]] to i32
 ; CHECK-NEXT:    [[C1_2:%.*]] = call i1 @unknown(i32 [[IV_I32_2]])
 ; CHECK-NEXT:    br i1 [[C1_2]], label [[LATCH_2:%.*]], label [[OTHEREXIT_LOOPEXIT]]
 ; CHECK:       latch.2:
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nsw i64 [[INDVARS_IV_NEXT_1]], 2
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nsw i64 [[INDVARS_IV]], 6
 ; CHECK-NEXT:    [[IV_I32_3:%.*]] = trunc i64 [[INDVARS_IV_NEXT_2]] to i32
 ; CHECK-NEXT:    [[C1_3:%.*]] = call i1 @unknown(i32 [[IV_I32_3]])
 ; CHECK-NEXT:    br i1 [[C1_3]], label [[LATCH_3]], label [[OTHEREXIT_LOOPEXIT]]
 ; CHECK:       latch.3:
-; CHECK-NEXT:    [[INDVARS_IV_NEXT_3]] = add nsw i64 [[INDVARS_IV_NEXT_2]], 2
+; CHECK-NEXT:    [[INDVARS_IV_NEXT_3]] = add nsw i64 [[INDVARS_IV]], 8
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp slt i64 [[INDVARS_IV_NEXT_3]], 616
 ; CHECK-NEXT:    br i1 [[TMP9]], label [[HEADER]], label [[LATCHEXIT_UNR_LCSSA:%.*]], !llvm.loop [[LOOP10:![0-9]+]]
 ; CHECK:       latchexit.unr-lcssa:

diff --git a/llvm/test/Transforms/LoopUnroll/runtime-loop-multiple-exits.ll b/llvm/test/Transforms/LoopUnroll/runtime-loop-multiple-exits.ll
index e0a916f..6cad2e1 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/runtime-loop-multiple-exits.ll
+++ b/llvm/test/Transforms/LoopUnroll/runtime-loop-multiple-exits.ll
@@ -34,64 +34,50 @@ define void @test1(i64 %trip, i1 %cond) {
 ; EPILOG:       exit3:
 ; EPILOG-NEXT:    ret void
 ; EPILOG:       loop_latch:
-; EPILOG-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
-; EPILOG-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.1, label %loop_exiting_bb1.1
 ; EPILOG:       loop_exiting_bb1.1:
 ; EPILOG-NEXT:    br i1 false, label %loop_exiting_bb2.1, label %exit1.loopexit
 ; EPILOG:       loop_exiting_bb2.1:
 ; EPILOG-NEXT:    br i1 false, label %loop_latch.1, label %exit3.loopexit
 ; EPILOG:       loop_latch.1:
-; EPILOG-NEXT:    %iv_next.1 = add nuw nsw i64 %iv_next, 1
-; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.2, label %loop_exiting_bb1.2
 ; EPILOG:       loop_exiting_bb1.2:
 ; EPILOG-NEXT:    br i1 false, label %loop_exiting_bb2.2, label %exit1.loopexit
 ; EPILOG:       loop_exiting_bb2.2:
 ; EPILOG-NEXT:    br i1 false, label %loop_latch.2, label %exit3.loopexit
 ; EPILOG:       loop_latch.2:
-; EPILOG-NEXT:    %iv_next.2 = add nuw nsw i64 %iv_next.1, 1
-; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.3, label %loop_exiting_bb1.3
 ; EPILOG:       loop_exiting_bb1.3:
 ; EPILOG-NEXT:    br i1 false, label %loop_exiting_bb2.3, label %exit1.loopexit
 ; EPILOG:       loop_exiting_bb2.3:
 ; EPILOG-NEXT:    br i1 false, label %loop_latch.3, label %exit3.loopexit
 ; EPILOG:       loop_latch.3:
-; EPILOG-NEXT:    %iv_next.3 = add nuw nsw i64 %iv_next.2, 1
-; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.4, label %loop_exiting_bb1.4
 ; EPILOG:       loop_exiting_bb1.4:
 ; EPILOG-NEXT:    br i1 false, label %loop_exiting_bb2.4, label %exit1.loopexit
 ; EPILOG:       loop_exiting_bb2.4:
 ; EPILOG-NEXT:    br i1 false, label %loop_latch.4, label %exit3.loopexit
 ; EPILOG:       loop_latch.4:
-; EPILOG-NEXT:    %iv_next.4 = add nuw nsw i64 %iv_next.3, 1
-; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.5, label %loop_exiting_bb1.5
 ; EPILOG:       loop_exiting_bb1.5:
 ; EPILOG-NEXT:    br i1 false, label %loop_exiting_bb2.5, label %exit1.loopexit
 ; EPILOG:       loop_exiting_bb2.5:
 ; EPILOG-NEXT:    br i1 false, label %loop_latch.5, label %exit3.loopexit
 ; EPILOG:       loop_latch.5:
-; EPILOG-NEXT:    %iv_next.5 = add nuw nsw i64 %iv_next.4, 1
-; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.6, label %loop_exiting_bb1.6
 ; EPILOG:       loop_exiting_bb1.6:
 ; EPILOG-NEXT:    br i1 false, label %loop_exiting_bb2.6, label %exit1.loopexit
 ; EPILOG:       loop_exiting_bb2.6:
 ; EPILOG-NEXT:    br i1 false, label %loop_latch.6, label %exit3.loopexit
 ; EPILOG:       loop_latch.6:
-; EPILOG-NEXT:    %iv_next.6 = add nuw nsw i64 %iv_next.5, 1
-; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.7, label %loop_exiting_bb1.7
 ; EPILOG:       loop_exiting_bb1.7:
 ; EPILOG-NEXT:    br i1 false, label %loop_exiting_bb2.7, label %exit1.loopexit
 ; EPILOG:       loop_exiting_bb2.7:
 ; EPILOG-NEXT:    br i1 false, label %loop_latch.7, label %exit3.loopexit
 ; EPILOG:       loop_latch.7:
-; EPILOG-NEXT:    %iv_next.7 = add i64 %iv_next.6, 1
-; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; EPILOG-NEXT:    %iv_next.7 = add i64 %iv, 8
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter, 8
 ; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.next.7, %unroll_iter
 ; EPILOG-NEXT:    br i1 %niter.ncmp.7, label %loop_header, label %exit2.loopexit.unr-lcssa.loopexit
 ; EPILOG:       exit1.loopexit:
@@ -151,16 +137,14 @@ define void @test1(i64 %trip, i1 %cond) {
 ; EPILOG-BLOCK:       exit3:
 ; EPILOG-BLOCK-NEXT:    ret void
 ; EPILOG-BLOCK:       loop_latch:
-; EPILOG-BLOCK-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
-; EPILOG-BLOCK-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-BLOCK-NEXT:    br i1 %cond, label %loop_latch.1, label %loop_exiting_bb1.1
 ; EPILOG-BLOCK:       loop_exiting_bb1.1:
 ; EPILOG-BLOCK-NEXT:    br i1 false, label %loop_exiting_bb2.1, label %exit1.loopexit
 ; EPILOG-BLOCK:       loop_exiting_bb2.1:
 ; EPILOG-BLOCK-NEXT:    br i1 false, label %loop_latch.1, label %exit3.loopexit
 ; EPILOG-BLOCK:       loop_latch.1:
-; EPILOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv_next, 1
-; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; EPILOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv, 2
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.next.1, %unroll_iter
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1, label %loop_header, label %exit2.loopexit.unr-lcssa.loopexit, !llvm.loop !0
 ; EPILOG-BLOCK:       exit1.loopexit:
@@ -231,56 +215,49 @@ define void @test1(i64 %trip, i1 %cond) {
 ; PROLOG:       exit3:
 ; PROLOG-NEXT:    ret void
 ; PROLOG:       loop_latch:
-; PROLOG-NEXT:    %iv_next = add i64 %iv, 1
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.1, label %loop_exiting_bb1.1
 ; PROLOG:       loop_exiting_bb1.1:
 ; PROLOG-NEXT:    br i1 false, label %loop_exiting_bb2.1, label %exit1.loopexit
 ; PROLOG:       loop_exiting_bb2.1:
 ; PROLOG-NEXT:    br i1 false, label %loop_latch.1, label %exit3.loopexit
 ; PROLOG:       loop_latch.1:
-; PROLOG-NEXT:    %iv_next.1 = add i64 %iv_next, 1
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.2, label %loop_exiting_bb1.2
 ; PROLOG:       loop_exiting_bb1.2:
 ; PROLOG-NEXT:    br i1 false, label %loop_exiting_bb2.2, label %exit1.loopexit
 ; PROLOG:       loop_exiting_bb2.2:
 ; PROLOG-NEXT:    br i1 false, label %loop_latch.2, label %exit3.loopexit
 ; PROLOG:       loop_latch.2:
-; PROLOG-NEXT:    %iv_next.2 = add i64 %iv_next.1, 1
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.3, label %loop_exiting_bb1.3
 ; PROLOG:       loop_exiting_bb1.3:
 ; PROLOG-NEXT:    br i1 false, label %loop_exiting_bb2.3, label %exit1.loopexit
 ; PROLOG:       loop_exiting_bb2.3:
 ; PROLOG-NEXT:    br i1 false, label %loop_latch.3, label %exit3.loopexit
 ; PROLOG:       loop_latch.3:
-; PROLOG-NEXT:    %iv_next.3 = add i64 %iv_next.2, 1
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.4, label %loop_exiting_bb1.4
 ; PROLOG:       loop_exiting_bb1.4:
 ; PROLOG-NEXT:    br i1 false, label %loop_exiting_bb2.4, label %exit1.loopexit
 ; PROLOG:       loop_exiting_bb2.4:
 ; PROLOG-NEXT:    br i1 false, label %loop_latch.4, label %exit3.loopexit
 ; PROLOG:       loop_latch.4:
-; PROLOG-NEXT:    %iv_next.4 = add i64 %iv_next.3, 1
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.5, label %loop_exiting_bb1.5
 ; PROLOG:       loop_exiting_bb1.5:
 ; PROLOG-NEXT:    br i1 false, label %loop_exiting_bb2.5, label %exit1.loopexit
 ; PROLOG:       loop_exiting_bb2.5:
 ; PROLOG-NEXT:    br i1 false, label %loop_latch.5, label %exit3.loopexit
 ; PROLOG:       loop_latch.5:
-; PROLOG-NEXT:    %iv_next.5 = add i64 %iv_next.4, 1
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.6, label %loop_exiting_bb1.6
 ; PROLOG:       loop_exiting_bb1.6:
 ; PROLOG-NEXT:    br i1 false, label %loop_exiting_bb2.6, label %exit1.loopexit
 ; PROLOG:       loop_exiting_bb2.6:
 ; PROLOG-NEXT:    br i1 false, label %loop_latch.6, label %exit3.loopexit
 ; PROLOG:       loop_latch.6:
-; PROLOG-NEXT:    %iv_next.6 = add i64 %iv_next.5, 1
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.7, label %loop_exiting_bb1.7
 ; PROLOG:       loop_exiting_bb1.7:
 ; PROLOG-NEXT:    br i1 false, label %loop_exiting_bb2.7, label %exit1.loopexit
 ; PROLOG:       loop_exiting_bb2.7:
 ; PROLOG-NEXT:    br i1 false, label %loop_latch.7, label %exit3.loopexit
 ; PROLOG:       loop_latch.7:
-; PROLOG-NEXT:    %iv_next.7 = add i64 %iv_next.6, 1
+; PROLOG-NEXT:    %iv_next.7 = add i64 %iv, 8
 ; PROLOG-NEXT:    %cmp.7 = icmp ne i64 %iv_next.7, %trip
 ; PROLOG-NEXT:    br i1 %cmp.7, label %loop_header, label %exit2.loopexit.unr-lcssa
 ; PROLOG:       exit1.loopexit:
@@ -329,14 +306,13 @@ define void @test1(i64 %trip, i1 %cond) {
 ; PROLOG-BLOCK:       exit3:
 ; PROLOG-BLOCK-NEXT:    ret void
 ; PROLOG-BLOCK:       loop_latch:
-; PROLOG-BLOCK-NEXT:    %iv_next = add i64 %iv, 1
 ; PROLOG-BLOCK-NEXT:    br i1 %cond, label %loop_latch.1, label %loop_exiting_bb1.1
 ; PROLOG-BLOCK:       loop_exiting_bb1.1:
 ; PROLOG-BLOCK-NEXT:    br i1 false, label %loop_exiting_bb2.1, label %exit1.loopexit
 ; PROLOG-BLOCK:       loop_exiting_bb2.1:
 ; PROLOG-BLOCK-NEXT:    br i1 false, label %loop_latch.1, label %exit3.loopexit
 ; PROLOG-BLOCK:       loop_latch.1:
-; PROLOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv_next, 1
+; PROLOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv, 2
 ; PROLOG-BLOCK-NEXT:    %cmp.1 = icmp ne i64 %iv_next.1, %trip
 ; PROLOG-BLOCK-NEXT:    br i1 %cmp.1, label %loop_header, label %exit2.loopexit.unr-lcssa, !llvm.loop !0
 ; PROLOG-BLOCK:       exit1.loopexit:
@@ -407,7 +383,6 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; EPILOG-NEXT:    %3 = load i32, ptr %arrayidx, align 4
 ; EPILOG-NEXT:    %add = add nsw i32 %3, %sum.02
 ; EPILOG-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.1
 ; EPILOG:       for.exiting_block.1:
 ; EPILOG-NEXT:    %cmp.1 = icmp eq i64 %n, 42
@@ -416,8 +391,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; EPILOG-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; EPILOG-NEXT:    %4 = load i32, ptr %arrayidx.1, align 4
 ; EPILOG-NEXT:    %add.1 = add nsw i32 %4, %add
-; EPILOG-NEXT:    %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv.next, 1
-; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
+; EPILOG-NEXT:    %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv, 2
 ; EPILOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.2
 ; EPILOG:       for.exiting_block.2:
 ; EPILOG-NEXT:    %cmp.2 = icmp eq i64 %n, 42
@@ -426,8 +400,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; EPILOG-NEXT:    %arrayidx.2 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.1
 ; EPILOG-NEXT:    %5 = load i32, ptr %arrayidx.2, align 4
 ; EPILOG-NEXT:    %add.2 = add nsw i32 %5, %add.1
-; EPILOG-NEXT:    %indvars.iv.next.2 = add nuw nsw i64 %indvars.iv.next.1, 1
-; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
+; EPILOG-NEXT:    %indvars.iv.next.2 = add nuw nsw i64 %indvars.iv, 3
 ; EPILOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.3
 ; EPILOG:       for.exiting_block.3:
 ; EPILOG-NEXT:    %cmp.3 = icmp eq i64 %n, 42
@@ -436,8 +409,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; EPILOG-NEXT:    %arrayidx.3 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.2
 ; EPILOG-NEXT:    %6 = load i32, ptr %arrayidx.3, align 4
 ; EPILOG-NEXT:    %add.3 = add nsw i32 %6, %add.2
-; EPILOG-NEXT:    %indvars.iv.next.3 = add nuw nsw i64 %indvars.iv.next.2, 1
-; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
+; EPILOG-NEXT:    %indvars.iv.next.3 = add nuw nsw i64 %indvars.iv, 4
 ; EPILOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.4
 ; EPILOG:       for.exiting_block.4:
 ; EPILOG-NEXT:    %cmp.4 = icmp eq i64 %n, 42
@@ -446,8 +418,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; EPILOG-NEXT:    %arrayidx.4 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.3
 ; EPILOG-NEXT:    %7 = load i32, ptr %arrayidx.4, align 4
 ; EPILOG-NEXT:    %add.4 = add nsw i32 %7, %add.3
-; EPILOG-NEXT:    %indvars.iv.next.4 = add nuw nsw i64 %indvars.iv.next.3, 1
-; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
+; EPILOG-NEXT:    %indvars.iv.next.4 = add nuw nsw i64 %indvars.iv, 5
 ; EPILOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.5
 ; EPILOG:       for.exiting_block.5:
 ; EPILOG-NEXT:    %cmp.5 = icmp eq i64 %n, 42
@@ -456,8 +427,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; EPILOG-NEXT:    %arrayidx.5 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.4
 ; EPILOG-NEXT:    %8 = load i32, ptr %arrayidx.5, align 4
 ; EPILOG-NEXT:    %add.5 = add nsw i32 %8, %add.4
-; EPILOG-NEXT:    %indvars.iv.next.5 = add nuw nsw i64 %indvars.iv.next.4, 1
-; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
+; EPILOG-NEXT:    %indvars.iv.next.5 = add nuw nsw i64 %indvars.iv, 6
 ; EPILOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.6
 ; EPILOG:       for.exiting_block.6:
 ; EPILOG-NEXT:    %cmp.6 = icmp eq i64 %n, 42
@@ -466,8 +436,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; EPILOG-NEXT:    %arrayidx.6 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.5
 ; EPILOG-NEXT:    %9 = load i32, ptr %arrayidx.6, align 4
 ; EPILOG-NEXT:    %add.6 = add nsw i32 %9, %add.5
-; EPILOG-NEXT:    %indvars.iv.next.6 = add nuw nsw i64 %indvars.iv.next.5, 1
-; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
+; EPILOG-NEXT:    %indvars.iv.next.6 = add nuw nsw i64 %indvars.iv, 7
 ; EPILOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.7
 ; EPILOG:       for.exiting_block.7:
 ; EPILOG-NEXT:    %cmp.7 = icmp eq i64 %n, 42
@@ -476,8 +445,8 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; EPILOG-NEXT:    %arrayidx.7 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.6
 ; EPILOG-NEXT:    %10 = load i32, ptr %arrayidx.7, align 4
 ; EPILOG-NEXT:    %add.7 = add nsw i32 %10, %add.6
-; EPILOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv.next.6, 1
-; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; EPILOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv, 8
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter, 8
 ; EPILOG-NEXT:    %niter.ncmp.7 = icmp eq i64 %niter.next.7, %unroll_iter
 ; EPILOG-NEXT:    br i1 %niter.ncmp.7, label %for.end.unr-lcssa.loopexit, label %header
 ; EPILOG:       for.end.unr-lcssa.loopexit:
@@ -549,7 +518,6 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; EPILOG-BLOCK-NEXT:    %3 = load i32, ptr %arrayidx, align 4
 ; EPILOG-BLOCK-NEXT:    %add = add nsw i32 %3, %sum.02
 ; EPILOG-BLOCK-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-BLOCK-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-BLOCK-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.1
 ; EPILOG-BLOCK:       for.exiting_block.1:
 ; EPILOG-BLOCK-NEXT:    %cmp.1 = icmp eq i64 %n, 42
@@ -558,8 +526,8 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; EPILOG-BLOCK-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; EPILOG-BLOCK-NEXT:    %4 = load i32, ptr %arrayidx.1, align 4
 ; EPILOG-BLOCK-NEXT:    %add.1 = add nsw i32 %4, %add
-; EPILOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
-; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; EPILOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv, 2
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp eq i64 %niter.next.1, %unroll_iter
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1, label %for.end.unr-lcssa.loopexit, label %header, !llvm.loop !2
 ; EPILOG-BLOCK:       for.end.unr-lcssa.loopexit:
@@ -654,7 +622,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; PROLOG-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; PROLOG-NEXT:    %5 = load i32, ptr %arrayidx.1, align 4
 ; PROLOG-NEXT:    %add.1 = add nsw i32 %5, %add
-; PROLOG-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
+; PROLOG-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv, 2
 ; PROLOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.2
 ; PROLOG:       for.exiting_block.2:
 ; PROLOG-NEXT:    %cmp.2 = icmp eq i64 %n, 42
@@ -663,7 +631,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; PROLOG-NEXT:    %arrayidx.2 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.1
 ; PROLOG-NEXT:    %6 = load i32, ptr %arrayidx.2, align 4
 ; PROLOG-NEXT:    %add.2 = add nsw i32 %6, %add.1
-; PROLOG-NEXT:    %indvars.iv.next.2 = add i64 %indvars.iv.next.1, 1
+; PROLOG-NEXT:    %indvars.iv.next.2 = add i64 %indvars.iv, 3
 ; PROLOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.3
 ; PROLOG:       for.exiting_block.3:
 ; PROLOG-NEXT:    %cmp.3 = icmp eq i64 %n, 42
@@ -672,7 +640,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; PROLOG-NEXT:    %arrayidx.3 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.2
 ; PROLOG-NEXT:    %7 = load i32, ptr %arrayidx.3, align 4
 ; PROLOG-NEXT:    %add.3 = add nsw i32 %7, %add.2
-; PROLOG-NEXT:    %indvars.iv.next.3 = add i64 %indvars.iv.next.2, 1
+; PROLOG-NEXT:    %indvars.iv.next.3 = add i64 %indvars.iv, 4
 ; PROLOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.4
 ; PROLOG:       for.exiting_block.4:
 ; PROLOG-NEXT:    %cmp.4 = icmp eq i64 %n, 42
@@ -681,7 +649,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; PROLOG-NEXT:    %arrayidx.4 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.3
 ; PROLOG-NEXT:    %8 = load i32, ptr %arrayidx.4, align 4
 ; PROLOG-NEXT:    %add.4 = add nsw i32 %8, %add.3
-; PROLOG-NEXT:    %indvars.iv.next.4 = add i64 %indvars.iv.next.3, 1
+; PROLOG-NEXT:    %indvars.iv.next.4 = add i64 %indvars.iv, 5
 ; PROLOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.5
 ; PROLOG:       for.exiting_block.5:
 ; PROLOG-NEXT:    %cmp.5 = icmp eq i64 %n, 42
@@ -690,7 +658,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; PROLOG-NEXT:    %arrayidx.5 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.4
 ; PROLOG-NEXT:    %9 = load i32, ptr %arrayidx.5, align 4
 ; PROLOG-NEXT:    %add.5 = add nsw i32 %9, %add.4
-; PROLOG-NEXT:    %indvars.iv.next.5 = add i64 %indvars.iv.next.4, 1
+; PROLOG-NEXT:    %indvars.iv.next.5 = add i64 %indvars.iv, 6
 ; PROLOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.6
 ; PROLOG:       for.exiting_block.6:
 ; PROLOG-NEXT:    %cmp.6 = icmp eq i64 %n, 42
@@ -699,7 +667,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; PROLOG-NEXT:    %arrayidx.6 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.5
 ; PROLOG-NEXT:    %10 = load i32, ptr %arrayidx.6, align 4
 ; PROLOG-NEXT:    %add.6 = add nsw i32 %10, %add.5
-; PROLOG-NEXT:    %indvars.iv.next.6 = add i64 %indvars.iv.next.5, 1
+; PROLOG-NEXT:    %indvars.iv.next.6 = add i64 %indvars.iv, 7
 ; PROLOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.7
 ; PROLOG:       for.exiting_block.7:
 ; PROLOG-NEXT:    %cmp.7 = icmp eq i64 %n, 42
@@ -708,7 +676,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; PROLOG-NEXT:    %arrayidx.7 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.6
 ; PROLOG-NEXT:    %11 = load i32, ptr %arrayidx.7, align 4
 ; PROLOG-NEXT:    %add.7 = add nsw i32 %11, %add.6
-; PROLOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv.next.6, 1
+; PROLOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv, 8
 ; PROLOG-NEXT:    %exitcond.7 = icmp eq i64 %indvars.iv.next.7, %n
 ; PROLOG-NEXT:    br i1 %exitcond.7, label %for.end.unr-lcssa, label %header
 ; PROLOG:       for.end.unr-lcssa:
@@ -772,7 +740,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; PROLOG-BLOCK-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; PROLOG-BLOCK-NEXT:    %5 = load i32, ptr %arrayidx.1, align 4
 ; PROLOG-BLOCK-NEXT:    %add.1 = add nsw i32 %5, %add
-; PROLOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
+; PROLOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv, 2
 ; PROLOG-BLOCK-NEXT:    %exitcond.1 = icmp eq i64 %indvars.iv.next.1, %n
 ; PROLOG-BLOCK-NEXT:    br i1 %exitcond.1, label %for.end.unr-lcssa, label %header, !llvm.loop !2
 ; PROLOG-BLOCK:       for.end.unr-lcssa:
@@ -849,9 +817,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; EPILOG:       exit3:
 ; EPILOG-NEXT:    ret void
 ; EPILOG:       loop_latch:
-; EPILOG-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
 ; EPILOG-NEXT:    %sum.next = add i64 %sum, %add
-; EPILOG-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-NEXT:    br i1 undef, label %loop_latch.1, label %loop_exiting_bb1.1
 ; EPILOG:       loop_exiting_bb1.1:
 ; EPILOG-NEXT:    switch i64 %sum.next, label %loop_latch.1 [
@@ -859,9 +825,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; EPILOG-NEXT:    i64 42, label %exit3.loopexit
 ; EPILOG-NEXT:    ]
 ; EPILOG:       loop_latch.1:
-; EPILOG-NEXT:    %iv_next.1 = add nuw nsw i64 %iv_next, 1
 ; EPILOG-NEXT:    %sum.next.1 = add i64 %sum.next, %add
-; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
 ; EPILOG-NEXT:    br i1 undef, label %loop_latch.2, label %loop_exiting_bb1.2
 ; EPILOG:       loop_exiting_bb1.2:
 ; EPILOG-NEXT:    switch i64 %sum.next.1, label %loop_latch.2 [
@@ -869,9 +833,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; EPILOG-NEXT:    i64 42, label %exit3.loopexit
 ; EPILOG-NEXT:    ]
 ; EPILOG:       loop_latch.2:
-; EPILOG-NEXT:    %iv_next.2 = add nuw nsw i64 %iv_next.1, 1
 ; EPILOG-NEXT:    %sum.next.2 = add i64 %sum.next.1, %add
-; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
 ; EPILOG-NEXT:    br i1 undef, label %loop_latch.3, label %loop_exiting_bb1.3
 ; EPILOG:       loop_exiting_bb1.3:
 ; EPILOG-NEXT:    switch i64 %sum.next.2, label %loop_latch.3 [
@@ -879,9 +841,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; EPILOG-NEXT:    i64 42, label %exit3.loopexit
 ; EPILOG-NEXT:    ]
 ; EPILOG:       loop_latch.3:
-; EPILOG-NEXT:    %iv_next.3 = add nuw nsw i64 %iv_next.2, 1
 ; EPILOG-NEXT:    %sum.next.3 = add i64 %sum.next.2, %add
-; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
 ; EPILOG-NEXT:    br i1 undef, label %loop_latch.4, label %loop_exiting_bb1.4
 ; EPILOG:       loop_exiting_bb1.4:
 ; EPILOG-NEXT:    switch i64 %sum.next.3, label %loop_latch.4 [
@@ -889,9 +849,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; EPILOG-NEXT:    i64 42, label %exit3.loopexit
 ; EPILOG-NEXT:    ]
 ; EPILOG:       loop_latch.4:
-; EPILOG-NEXT:    %iv_next.4 = add nuw nsw i64 %iv_next.3, 1
 ; EPILOG-NEXT:    %sum.next.4 = add i64 %sum.next.3, %add
-; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
 ; EPILOG-NEXT:    br i1 undef, label %loop_latch.5, label %loop_exiting_bb1.5
 ; EPILOG:       loop_exiting_bb1.5:
 ; EPILOG-NEXT:    switch i64 %sum.next.4, label %loop_latch.5 [
@@ -899,9 +857,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; EPILOG-NEXT:    i64 42, label %exit3.loopexit
 ; EPILOG-NEXT:    ]
 ; EPILOG:       loop_latch.5:
-; EPILOG-NEXT:    %iv_next.5 = add nuw nsw i64 %iv_next.4, 1
 ; EPILOG-NEXT:    %sum.next.5 = add i64 %sum.next.4, %add
-; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
 ; EPILOG-NEXT:    br i1 undef, label %loop_latch.6, label %loop_exiting_bb1.6
 ; EPILOG:       loop_exiting_bb1.6:
 ; EPILOG-NEXT:    switch i64 %sum.next.5, label %loop_latch.6 [
@@ -909,9 +865,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; EPILOG-NEXT:    i64 42, label %exit3.loopexit
 ; EPILOG-NEXT:    ]
 ; EPILOG:       loop_latch.6:
-; EPILOG-NEXT:    %iv_next.6 = add nuw nsw i64 %iv_next.5, 1
 ; EPILOG-NEXT:    %sum.next.6 = add i64 %sum.next.5, %add
-; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
 ; EPILOG-NEXT:    br i1 undef, label %loop_latch.7, label %loop_exiting_bb1.7
 ; EPILOG:       loop_exiting_bb1.7:
 ; EPILOG-NEXT:    switch i64 %sum.next.6, label %loop_latch.7 [
@@ -919,9 +873,9 @@ define void @test3(i64 %trip, i64 %add) {
 ; EPILOG-NEXT:    i64 42, label %exit3.loopexit
 ; EPILOG-NEXT:    ]
 ; EPILOG:       loop_latch.7:
-; EPILOG-NEXT:    %iv_next.7 = add nuw nsw i64 %iv_next.6, 1
+; EPILOG-NEXT:    %iv_next.7 = add nuw nsw i64 %iv, 8
 ; EPILOG-NEXT:    %sum.next.7 = add i64 %sum.next.6, %add
-; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter, 8
 ; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.next.7, %unroll_iter
 ; EPILOG-NEXT:    br i1 %niter.ncmp.7, label %loop_header, label %exit2.loopexit.unr-lcssa.loopexit
 ; EPILOG:       exit1.loopexit:
@@ -988,9 +942,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; EPILOG-BLOCK:       exit3:
 ; EPILOG-BLOCK-NEXT:    ret void
 ; EPILOG-BLOCK:       loop_latch:
-; EPILOG-BLOCK-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
 ; EPILOG-BLOCK-NEXT:    %sum.next = add i64 %sum, %add
-; EPILOG-BLOCK-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-BLOCK-NEXT:    br i1 undef, label %loop_latch.1, label %loop_exiting_bb1.1
 ; EPILOG-BLOCK:       loop_exiting_bb1.1:
 ; EPILOG-BLOCK-NEXT:    switch i64 %sum.next, label %loop_latch.1 [
@@ -998,9 +950,9 @@ define void @test3(i64 %trip, i64 %add) {
 ; EPILOG-BLOCK-NEXT:    i64 42, label %exit3.loopexit
 ; EPILOG-BLOCK-NEXT:    ]
 ; EPILOG-BLOCK:       loop_latch.1:
-; EPILOG-BLOCK-NEXT:    %iv_next.1 = add nuw nsw i64 %iv_next, 1
+; EPILOG-BLOCK-NEXT:    %iv_next.1 = add nuw nsw i64 %iv, 2
 ; EPILOG-BLOCK-NEXT:    %sum.next.1 = add i64 %sum.next, %add
-; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.next.1, %unroll_iter
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1, label %loop_header, label %exit2.loopexit.unr-lcssa.loopexit, !llvm.loop !3
 ; EPILOG-BLOCK:       exit1.loopexit:
@@ -1081,7 +1033,6 @@ define void @test3(i64 %trip, i64 %add) {
 ; PROLOG:       exit3:
 ; PROLOG-NEXT:    ret void
 ; PROLOG:       loop_latch:
-; PROLOG-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
 ; PROLOG-NEXT:    %sum.next = add i64 %sum, %add
 ; PROLOG-NEXT:    br i1 undef, label %loop_latch.1, label %loop_exiting_bb1.1
 ; PROLOG:       loop_exiting_bb1.1:
@@ -1090,7 +1041,6 @@ define void @test3(i64 %trip, i64 %add) {
 ; PROLOG-NEXT:    i64 42, label %exit3.loopexit
 ; PROLOG-NEXT:    ]
 ; PROLOG:       loop_latch.1:
-; PROLOG-NEXT:    %iv_next.1 = add nuw nsw i64 %iv_next, 1
 ; PROLOG-NEXT:    %sum.next.1 = add i64 %sum.next, %add
 ; PROLOG-NEXT:    br i1 undef, label %loop_latch.2, label %loop_exiting_bb1.2
 ; PROLOG:       loop_exiting_bb1.2:
@@ -1099,7 +1049,6 @@ define void @test3(i64 %trip, i64 %add) {
 ; PROLOG-NEXT:    i64 42, label %exit3.loopexit
 ; PROLOG-NEXT:    ]
 ; PROLOG:       loop_latch.2:
-; PROLOG-NEXT:    %iv_next.2 = add nuw nsw i64 %iv_next.1, 1
 ; PROLOG-NEXT:    %sum.next.2 = add i64 %sum.next.1, %add
 ; PROLOG-NEXT:    br i1 undef, label %loop_latch.3, label %loop_exiting_bb1.3
 ; PROLOG:       loop_exiting_bb1.3:
@@ -1108,7 +1057,6 @@ define void @test3(i64 %trip, i64 %add) {
 ; PROLOG-NEXT:    i64 42, label %exit3.loopexit
 ; PROLOG-NEXT:    ]
 ; PROLOG:       loop_latch.3:
-; PROLOG-NEXT:    %iv_next.3 = add nuw nsw i64 %iv_next.2, 1
 ; PROLOG-NEXT:    %sum.next.3 = add i64 %sum.next.2, %add
 ; PROLOG-NEXT:    br i1 undef, label %loop_latch.4, label %loop_exiting_bb1.4
 ; PROLOG:       loop_exiting_bb1.4:
@@ -1117,7 +1065,6 @@ define void @test3(i64 %trip, i64 %add) {
 ; PROLOG-NEXT:    i64 42, label %exit3.loopexit
 ; PROLOG-NEXT:    ]
 ; PROLOG:       loop_latch.4:
-; PROLOG-NEXT:    %iv_next.4 = add nuw nsw i64 %iv_next.3, 1
 ; PROLOG-NEXT:    %sum.next.4 = add i64 %sum.next.3, %add
 ; PROLOG-NEXT:    br i1 undef, label %loop_latch.5, label %loop_exiting_bb1.5
 ; PROLOG:       loop_exiting_bb1.5:
@@ -1126,7 +1073,6 @@ define void @test3(i64 %trip, i64 %add) {
 ; PROLOG-NEXT:    i64 42, label %exit3.loopexit
 ; PROLOG-NEXT:    ]
 ; PROLOG:       loop_latch.5:
-; PROLOG-NEXT:    %iv_next.5 = add nuw nsw i64 %iv_next.4, 1
 ; PROLOG-NEXT:    %sum.next.5 = add i64 %sum.next.4, %add
 ; PROLOG-NEXT:    br i1 undef, label %loop_latch.6, label %loop_exiting_bb1.6
 ; PROLOG:       loop_exiting_bb1.6:
@@ -1135,7 +1081,6 @@ define void @test3(i64 %trip, i64 %add) {
 ; PROLOG-NEXT:    i64 42, label %exit3.loopexit
 ; PROLOG-NEXT:    ]
 ; PROLOG:       loop_latch.6:
-; PROLOG-NEXT:    %iv_next.6 = add nuw nsw i64 %iv_next.5, 1
 ; PROLOG-NEXT:    %sum.next.6 = add i64 %sum.next.5, %add
 ; PROLOG-NEXT:    br i1 undef, label %loop_latch.7, label %loop_exiting_bb1.7
 ; PROLOG:       loop_exiting_bb1.7:
@@ -1144,7 +1089,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; PROLOG-NEXT:    i64 42, label %exit3.loopexit
 ; PROLOG-NEXT:    ]
 ; PROLOG:       loop_latch.7:
-; PROLOG-NEXT:    %iv_next.7 = add nuw nsw i64 %iv_next.6, 1
+; PROLOG-NEXT:    %iv_next.7 = add nuw nsw i64 %iv, 8
 ; PROLOG-NEXT:    %sum.next.7 = add i64 %sum.next.6, %add
 ; PROLOG-NEXT:    %cmp.7 = icmp ne i64 %iv_next.7, %trip
 ; PROLOG-NEXT:    br i1 %cmp.7, label %loop_header, label %exit2.loopexit.unr-lcssa
@@ -1198,7 +1143,6 @@ define void @test3(i64 %trip, i64 %add) {
 ; PROLOG-BLOCK:       exit3:
 ; PROLOG-BLOCK-NEXT:    ret void
 ; PROLOG-BLOCK:       loop_latch:
-; PROLOG-BLOCK-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
 ; PROLOG-BLOCK-NEXT:    %sum.next = add i64 %sum, %add
 ; PROLOG-BLOCK-NEXT:    br i1 undef, label %loop_latch.1, label %loop_exiting_bb1.1
 ; PROLOG-BLOCK:       loop_exiting_bb1.1:
@@ -1207,7 +1151,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; PROLOG-BLOCK-NEXT:    i64 42, label %exit3.loopexit
 ; PROLOG-BLOCK-NEXT:    ]
 ; PROLOG-BLOCK:       loop_latch.1:
-; PROLOG-BLOCK-NEXT:    %iv_next.1 = add nuw nsw i64 %iv_next, 1
+; PROLOG-BLOCK-NEXT:    %iv_next.1 = add nuw nsw i64 %iv, 2
 ; PROLOG-BLOCK-NEXT:    %sum.next.1 = add i64 %sum.next, %add
 ; PROLOG-BLOCK-NEXT:    %cmp.1 = icmp ne i64 %iv_next.1, %trip
 ; PROLOG-BLOCK-NEXT:    br i1 %cmp.1, label %loop_header, label %exit2.loopexit.unr-lcssa, !llvm.loop !3
@@ -1277,7 +1221,6 @@ define i32 @hdr_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %3 = load i32, ptr %arrayidx, align 4
 ; EPILOG-NEXT:    %add = add nsw i32 %3, %sum.02
 ; EPILOG-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-NEXT:    br i1 %cond, label %latchExit.epilog-lcssa.loopexit, label %for.exiting_block.1
 ; EPILOG:       for.exiting_block.1:
 ; EPILOG-NEXT:    %cmp.1 = icmp eq i64 %n, 42
@@ -1286,8 +1229,7 @@ define i32 @hdr_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; EPILOG-NEXT:    %4 = load i32, ptr %arrayidx.1, align 4
 ; EPILOG-NEXT:    %add.1 = add nsw i32 %4, %add
-; EPILOG-NEXT:    %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv.next, 1
-; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
+; EPILOG-NEXT:    %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv, 2
 ; EPILOG-NEXT:    br i1 %cond, label %latchExit.epilog-lcssa.loopexit, label %for.exiting_block.2
 ; EPILOG:       for.exiting_block.2:
 ; EPILOG-NEXT:    %cmp.2 = icmp eq i64 %n, 42
@@ -1296,8 +1238,7 @@ define i32 @hdr_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.2 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.1
 ; EPILOG-NEXT:    %5 = load i32, ptr %arrayidx.2, align 4
 ; EPILOG-NEXT:    %add.2 = add nsw i32 %5, %add.1
-; EPILOG-NEXT:    %indvars.iv.next.2 = add nuw nsw i64 %indvars.iv.next.1, 1
-; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
+; EPILOG-NEXT:    %indvars.iv.next.2 = add nuw nsw i64 %indvars.iv, 3
 ; EPILOG-NEXT:    br i1 %cond, label %latchExit.epilog-lcssa.loopexit, label %for.exiting_block.3
 ; EPILOG:       for.exiting_block.3:
 ; EPILOG-NEXT:    %cmp.3 = icmp eq i64 %n, 42
@@ -1306,8 +1247,7 @@ define i32 @hdr_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.3 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.2
 ; EPILOG-NEXT:    %6 = load i32, ptr %arrayidx.3, align 4
 ; EPILOG-NEXT:    %add.3 = add nsw i32 %6, %add.2
-; EPILOG-NEXT:    %indvars.iv.next.3 = add nuw nsw i64 %indvars.iv.next.2, 1
-; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
+; EPILOG-NEXT:    %indvars.iv.next.3 = add nuw nsw i64 %indvars.iv, 4
 ; EPILOG-NEXT:    br i1 %cond, label %latchExit.epilog-lcssa.loopexit, label %for.exiting_block.4
 ; EPILOG:       for.exiting_block.4:
 ; EPILOG-NEXT:    %cmp.4 = icmp eq i64 %n, 42
@@ -1316,8 +1256,7 @@ define i32 @hdr_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.4 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.3
 ; EPILOG-NEXT:    %7 = load i32, ptr %arrayidx.4, align 4
 ; EPILOG-NEXT:    %add.4 = add nsw i32 %7, %add.3
-; EPILOG-NEXT:    %indvars.iv.next.4 = add nuw nsw i64 %indvars.iv.next.3, 1
-; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
+; EPILOG-NEXT:    %indvars.iv.next.4 = add nuw nsw i64 %indvars.iv, 5
 ; EPILOG-NEXT:    br i1 %cond, label %latchExit.epilog-lcssa.loopexit, label %for.exiting_block.5
 ; EPILOG:       for.exiting_block.5:
 ; EPILOG-NEXT:    %cmp.5 = icmp eq i64 %n, 42
@@ -1326,8 +1265,7 @@ define i32 @hdr_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.5 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.4
 ; EPILOG-NEXT:    %8 = load i32, ptr %arrayidx.5, align 4
 ; EPILOG-NEXT:    %add.5 = add nsw i32 %8, %add.4
-; EPILOG-NEXT:    %indvars.iv.next.5 = add nuw nsw i64 %indvars.iv.next.4, 1
-; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
+; EPILOG-NEXT:    %indvars.iv.next.5 = add nuw nsw i64 %indvars.iv, 6
 ; EPILOG-NEXT:    br i1 %cond, label %latchExit.epilog-lcssa.loopexit, label %for.exiting_block.6
 ; EPILOG:       for.exiting_block.6:
 ; EPILOG-NEXT:    %cmp.6 = icmp eq i64 %n, 42
@@ -1336,8 +1274,7 @@ define i32 @hdr_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.6 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.5
 ; EPILOG-NEXT:    %9 = load i32, ptr %arrayidx.6, align 4
 ; EPILOG-NEXT:    %add.6 = add nsw i32 %9, %add.5
-; EPILOG-NEXT:    %indvars.iv.next.6 = add nuw nsw i64 %indvars.iv.next.5, 1
-; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
+; EPILOG-NEXT:    %indvars.iv.next.6 = add nuw nsw i64 %indvars.iv, 7
 ; EPILOG-NEXT:    br i1 %cond, label %latchExit.epilog-lcssa.loopexit, label %for.exiting_block.7
 ; EPILOG:       for.exiting_block.7:
 ; EPILOG-NEXT:    %cmp.7 = icmp eq i64 %n, 42
@@ -1346,8 +1283,8 @@ define i32 @hdr_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.7 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.6
 ; EPILOG-NEXT:    %10 = load i32, ptr %arrayidx.7, align 4
 ; EPILOG-NEXT:    %add.7 = add nsw i32 %10, %add.6
-; EPILOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv.next.6, 1
-; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; EPILOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv, 8
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter, 8
 ; EPILOG-NEXT:    %niter.ncmp.7 = icmp eq i64 %niter.next.7, %unroll_iter
 ; EPILOG-NEXT:    br i1 %niter.ncmp.7, label %latchExit.unr-lcssa.loopexit, label %header
 ; EPILOG:       latchExit.unr-lcssa.loopexit:
@@ -1422,7 +1359,6 @@ define i32 @hdr_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    %3 = load i32, ptr %arrayidx, align 4
 ; EPILOG-BLOCK-NEXT:    %add = add nsw i32 %3, %sum.02
 ; EPILOG-BLOCK-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-BLOCK-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-BLOCK-NEXT:    br i1 %cond, label %latchExit.epilog-lcssa.loopexit, label %for.exiting_block.1
 ; EPILOG-BLOCK:       for.exiting_block.1:
 ; EPILOG-BLOCK-NEXT:    %cmp.1 = icmp eq i64 %n, 42
@@ -1431,8 +1367,8 @@ define i32 @hdr_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; EPILOG-BLOCK-NEXT:    %4 = load i32, ptr %arrayidx.1, align 4
 ; EPILOG-BLOCK-NEXT:    %add.1 = add nsw i32 %4, %add
-; EPILOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
-; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; EPILOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv, 2
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp eq i64 %niter.next.1, %unroll_iter
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1, label %latchExit.unr-lcssa.loopexit, label %header, !llvm.loop !4
 ; EPILOG-BLOCK:       latchExit.unr-lcssa.loopexit:
@@ -1531,7 +1467,7 @@ define i32 @hdr_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; PROLOG-NEXT:    %5 = load i32, ptr %arrayidx.1, align 4
 ; PROLOG-NEXT:    %add.1 = add nsw i32 %5, %add
-; PROLOG-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
+; PROLOG-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv, 2
 ; PROLOG-NEXT:    br i1 %cond, label %latchExit.unr-lcssa.loopexit, label %for.exiting_block.2
 ; PROLOG:       for.exiting_block.2:
 ; PROLOG-NEXT:    %cmp.2 = icmp eq i64 %n, 42
@@ -1540,7 +1476,7 @@ define i32 @hdr_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.2 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.1
 ; PROLOG-NEXT:    %6 = load i32, ptr %arrayidx.2, align 4
 ; PROLOG-NEXT:    %add.2 = add nsw i32 %6, %add.1
-; PROLOG-NEXT:    %indvars.iv.next.2 = add i64 %indvars.iv.next.1, 1
+; PROLOG-NEXT:    %indvars.iv.next.2 = add i64 %indvars.iv, 3
 ; PROLOG-NEXT:    br i1 %cond, label %latchExit.unr-lcssa.loopexit, label %for.exiting_block.3
 ; PROLOG:       for.exiting_block.3:
 ; PROLOG-NEXT:    %cmp.3 = icmp eq i64 %n, 42
@@ -1549,7 +1485,7 @@ define i32 @hdr_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.3 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.2
 ; PROLOG-NEXT:    %7 = load i32, ptr %arrayidx.3, align 4
 ; PROLOG-NEXT:    %add.3 = add nsw i32 %7, %add.2
-; PROLOG-NEXT:    %indvars.iv.next.3 = add i64 %indvars.iv.next.2, 1
+; PROLOG-NEXT:    %indvars.iv.next.3 = add i64 %indvars.iv, 4
 ; PROLOG-NEXT:    br i1 %cond, label %latchExit.unr-lcssa.loopexit, label %for.exiting_block.4
 ; PROLOG:       for.exiting_block.4:
 ; PROLOG-NEXT:    %cmp.4 = icmp eq i64 %n, 42
@@ -1558,7 +1494,7 @@ define i32 @hdr_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.4 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.3
 ; PROLOG-NEXT:    %8 = load i32, ptr %arrayidx.4, align 4
 ; PROLOG-NEXT:    %add.4 = add nsw i32 %8, %add.3
-; PROLOG-NEXT:    %indvars.iv.next.4 = add i64 %indvars.iv.next.3, 1
+; PROLOG-NEXT:    %indvars.iv.next.4 = add i64 %indvars.iv, 5
 ; PROLOG-NEXT:    br i1 %cond, label %latchExit.unr-lcssa.loopexit, label %for.exiting_block.5
 ; PROLOG:       for.exiting_block.5:
 ; PROLOG-NEXT:    %cmp.5 = icmp eq i64 %n, 42
@@ -1567,7 +1503,7 @@ define i32 @hdr_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.5 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.4
 ; PROLOG-NEXT:    %9 = load i32, ptr %arrayidx.5, align 4
 ; PROLOG-NEXT:    %add.5 = add nsw i32 %9, %add.4
-; PROLOG-NEXT:    %indvars.iv.next.5 = add i64 %indvars.iv.next.4, 1
+; PROLOG-NEXT:    %indvars.iv.next.5 = add i64 %indvars.iv, 6
 ; PROLOG-NEXT:    br i1 %cond, label %latchExit.unr-lcssa.loopexit, label %for.exiting_block.6
 ; PROLOG:       for.exiting_block.6:
 ; PROLOG-NEXT:    %cmp.6 = icmp eq i64 %n, 42
@@ -1576,7 +1512,7 @@ define i32 @hdr_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.6 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.5
 ; PROLOG-NEXT:    %10 = load i32, ptr %arrayidx.6, align 4
 ; PROLOG-NEXT:    %add.6 = add nsw i32 %10, %add.5
-; PROLOG-NEXT:    %indvars.iv.next.6 = add i64 %indvars.iv.next.5, 1
+; PROLOG-NEXT:    %indvars.iv.next.6 = add i64 %indvars.iv, 7
 ; PROLOG-NEXT:    br i1 %cond, label %latchExit.unr-lcssa.loopexit, label %for.exiting_block.7
 ; PROLOG:       for.exiting_block.7:
 ; PROLOG-NEXT:    %cmp.7 = icmp eq i64 %n, 42
@@ -1585,7 +1521,7 @@ define i32 @hdr_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.7 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.6
 ; PROLOG-NEXT:    %11 = load i32, ptr %arrayidx.7, align 4
 ; PROLOG-NEXT:    %add.7 = add nsw i32 %11, %add.6
-; PROLOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv.next.6, 1
+; PROLOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv, 8
 ; PROLOG-NEXT:    %exitcond.7 = icmp eq i64 %indvars.iv.next.7, %n
 ; PROLOG-NEXT:    br i1 %exitcond.7, label %latchExit.unr-lcssa.loopexit, label %header
 ; PROLOG:       latchExit.unr-lcssa.loopexit:
@@ -1652,7 +1588,7 @@ define i32 @hdr_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-BLOCK-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; PROLOG-BLOCK-NEXT:    %5 = load i32, ptr %arrayidx.1, align 4
 ; PROLOG-BLOCK-NEXT:    %add.1 = add nsw i32 %5, %add
-; PROLOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
+; PROLOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv, 2
 ; PROLOG-BLOCK-NEXT:    %exitcond.1 = icmp eq i64 %indvars.iv.next.1, %n
 ; PROLOG-BLOCK-NEXT:    br i1 %exitcond.1, label %latchExit.unr-lcssa.loopexit, label %header, !llvm.loop !4
 ; PROLOG-BLOCK:       latchExit.unr-lcssa.loopexit:
@@ -1725,7 +1661,6 @@ define i32 @otherblock_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %3 = load i32, ptr %arrayidx, align 4
 ; EPILOG-NEXT:    %add = add nsw i32 %3, %sum.02
 ; EPILOG-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.1
 ; EPILOG:       for.exiting_block.1:
 ; EPILOG-NEXT:    %cmp.1 = icmp eq i64 %n, 42
@@ -1734,8 +1669,7 @@ define i32 @otherblock_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; EPILOG-NEXT:    %4 = load i32, ptr %arrayidx.1, align 4
 ; EPILOG-NEXT:    %add.1 = add nsw i32 %4, %add
-; EPILOG-NEXT:    %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv.next, 1
-; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
+; EPILOG-NEXT:    %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv, 2
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.2
 ; EPILOG:       for.exiting_block.2:
 ; EPILOG-NEXT:    %cmp.2 = icmp eq i64 %n, 42
@@ -1744,8 +1678,7 @@ define i32 @otherblock_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.2 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.1
 ; EPILOG-NEXT:    %5 = load i32, ptr %arrayidx.2, align 4
 ; EPILOG-NEXT:    %add.2 = add nsw i32 %5, %add.1
-; EPILOG-NEXT:    %indvars.iv.next.2 = add nuw nsw i64 %indvars.iv.next.1, 1
-; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
+; EPILOG-NEXT:    %indvars.iv.next.2 = add nuw nsw i64 %indvars.iv, 3
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.3
 ; EPILOG:       for.exiting_block.3:
 ; EPILOG-NEXT:    %cmp.3 = icmp eq i64 %n, 42
@@ -1754,8 +1687,7 @@ define i32 @otherblock_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.3 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.2
 ; EPILOG-NEXT:    %6 = load i32, ptr %arrayidx.3, align 4
 ; EPILOG-NEXT:    %add.3 = add nsw i32 %6, %add.2
-; EPILOG-NEXT:    %indvars.iv.next.3 = add nuw nsw i64 %indvars.iv.next.2, 1
-; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
+; EPILOG-NEXT:    %indvars.iv.next.3 = add nuw nsw i64 %indvars.iv, 4
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.4
 ; EPILOG:       for.exiting_block.4:
 ; EPILOG-NEXT:    %cmp.4 = icmp eq i64 %n, 42
@@ -1764,8 +1696,7 @@ define i32 @otherblock_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.4 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.3
 ; EPILOG-NEXT:    %7 = load i32, ptr %arrayidx.4, align 4
 ; EPILOG-NEXT:    %add.4 = add nsw i32 %7, %add.3
-; EPILOG-NEXT:    %indvars.iv.next.4 = add nuw nsw i64 %indvars.iv.next.3, 1
-; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
+; EPILOG-NEXT:    %indvars.iv.next.4 = add nuw nsw i64 %indvars.iv, 5
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.5
 ; EPILOG:       for.exiting_block.5:
 ; EPILOG-NEXT:    %cmp.5 = icmp eq i64 %n, 42
@@ -1774,8 +1705,7 @@ define i32 @otherblock_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.5 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.4
 ; EPILOG-NEXT:    %8 = load i32, ptr %arrayidx.5, align 4
 ; EPILOG-NEXT:    %add.5 = add nsw i32 %8, %add.4
-; EPILOG-NEXT:    %indvars.iv.next.5 = add nuw nsw i64 %indvars.iv.next.4, 1
-; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
+; EPILOG-NEXT:    %indvars.iv.next.5 = add nuw nsw i64 %indvars.iv, 6
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.6
 ; EPILOG:       for.exiting_block.6:
 ; EPILOG-NEXT:    %cmp.6 = icmp eq i64 %n, 42
@@ -1784,8 +1714,7 @@ define i32 @otherblock_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.6 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.5
 ; EPILOG-NEXT:    %9 = load i32, ptr %arrayidx.6, align 4
 ; EPILOG-NEXT:    %add.6 = add nsw i32 %9, %add.5
-; EPILOG-NEXT:    %indvars.iv.next.6 = add nuw nsw i64 %indvars.iv.next.5, 1
-; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
+; EPILOG-NEXT:    %indvars.iv.next.6 = add nuw nsw i64 %indvars.iv, 7
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.7
 ; EPILOG:       for.exiting_block.7:
 ; EPILOG-NEXT:    %cmp.7 = icmp eq i64 %n, 42
@@ -1794,8 +1723,8 @@ define i32 @otherblock_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.7 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.6
 ; EPILOG-NEXT:    %10 = load i32, ptr %arrayidx.7, align 4
 ; EPILOG-NEXT:    %add.7 = add nsw i32 %10, %add.6
-; EPILOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv.next.6, 1
-; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; EPILOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv, 8
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter, 8
 ; EPILOG-NEXT:    %niter.ncmp.7 = icmp eq i64 %niter.next.7, %unroll_iter
 ; EPILOG-NEXT:    br i1 %niter.ncmp.7, label %latchExit.unr-lcssa.loopexit, label %header
 ; EPILOG:       latchExit.unr-lcssa.loopexit:
@@ -1870,7 +1799,6 @@ define i32 @otherblock_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    %3 = load i32, ptr %arrayidx, align 4
 ; EPILOG-BLOCK-NEXT:    %add = add nsw i32 %3, %sum.02
 ; EPILOG-BLOCK-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-BLOCK-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-BLOCK-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.1
 ; EPILOG-BLOCK:       for.exiting_block.1:
 ; EPILOG-BLOCK-NEXT:    %cmp.1 = icmp eq i64 %n, 42
@@ -1879,8 +1807,8 @@ define i32 @otherblock_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; EPILOG-BLOCK-NEXT:    %4 = load i32, ptr %arrayidx.1, align 4
 ; EPILOG-BLOCK-NEXT:    %add.1 = add nsw i32 %4, %add
-; EPILOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
-; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; EPILOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv, 2
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp eq i64 %niter.next.1, %unroll_iter
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1, label %latchExit.unr-lcssa.loopexit, label %header, !llvm.loop !5
 ; EPILOG-BLOCK:       latchExit.unr-lcssa.loopexit:
@@ -1979,7 +1907,7 @@ define i32 @otherblock_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; PROLOG-NEXT:    %5 = load i32, ptr %arrayidx.1, align 4
 ; PROLOG-NEXT:    %add.1 = add nsw i32 %5, %add
-; PROLOG-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
+; PROLOG-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv, 2
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.2
 ; PROLOG:       for.exiting_block.2:
 ; PROLOG-NEXT:    %cmp.2 = icmp eq i64 %n, 42
@@ -1988,7 +1916,7 @@ define i32 @otherblock_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.2 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.1
 ; PROLOG-NEXT:    %6 = load i32, ptr %arrayidx.2, align 4
 ; PROLOG-NEXT:    %add.2 = add nsw i32 %6, %add.1
-; PROLOG-NEXT:    %indvars.iv.next.2 = add i64 %indvars.iv.next.1, 1
+; PROLOG-NEXT:    %indvars.iv.next.2 = add i64 %indvars.iv, 3
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.3
 ; PROLOG:       for.exiting_block.3:
 ; PROLOG-NEXT:    %cmp.3 = icmp eq i64 %n, 42
@@ -1997,7 +1925,7 @@ define i32 @otherblock_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.3 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.2
 ; PROLOG-NEXT:    %7 = load i32, ptr %arrayidx.3, align 4
 ; PROLOG-NEXT:    %add.3 = add nsw i32 %7, %add.2
-; PROLOG-NEXT:    %indvars.iv.next.3 = add i64 %indvars.iv.next.2, 1
+; PROLOG-NEXT:    %indvars.iv.next.3 = add i64 %indvars.iv, 4
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.4
 ; PROLOG:       for.exiting_block.4:
 ; PROLOG-NEXT:    %cmp.4 = icmp eq i64 %n, 42
@@ -2006,7 +1934,7 @@ define i32 @otherblock_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.4 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.3
 ; PROLOG-NEXT:    %8 = load i32, ptr %arrayidx.4, align 4
 ; PROLOG-NEXT:    %add.4 = add nsw i32 %8, %add.3
-; PROLOG-NEXT:    %indvars.iv.next.4 = add i64 %indvars.iv.next.3, 1
+; PROLOG-NEXT:    %indvars.iv.next.4 = add i64 %indvars.iv, 5
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.5
 ; PROLOG:       for.exiting_block.5:
 ; PROLOG-NEXT:    %cmp.5 = icmp eq i64 %n, 42
@@ -2015,7 +1943,7 @@ define i32 @otherblock_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.5 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.4
 ; PROLOG-NEXT:    %9 = load i32, ptr %arrayidx.5, align 4
 ; PROLOG-NEXT:    %add.5 = add nsw i32 %9, %add.4
-; PROLOG-NEXT:    %indvars.iv.next.5 = add i64 %indvars.iv.next.4, 1
+; PROLOG-NEXT:    %indvars.iv.next.5 = add i64 %indvars.iv, 6
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.6
 ; PROLOG:       for.exiting_block.6:
 ; PROLOG-NEXT:    %cmp.6 = icmp eq i64 %n, 42
@@ -2024,7 +1952,7 @@ define i32 @otherblock_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.6 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.5
 ; PROLOG-NEXT:    %10 = load i32, ptr %arrayidx.6, align 4
 ; PROLOG-NEXT:    %add.6 = add nsw i32 %10, %add.5
-; PROLOG-NEXT:    %indvars.iv.next.6 = add i64 %indvars.iv.next.5, 1
+; PROLOG-NEXT:    %indvars.iv.next.6 = add i64 %indvars.iv, 7
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.7
 ; PROLOG:       for.exiting_block.7:
 ; PROLOG-NEXT:    %cmp.7 = icmp eq i64 %n, 42
@@ -2033,7 +1961,7 @@ define i32 @otherblock_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.7 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.6
 ; PROLOG-NEXT:    %11 = load i32, ptr %arrayidx.7, align 4
 ; PROLOG-NEXT:    %add.7 = add nsw i32 %11, %add.6
-; PROLOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv.next.6, 1
+; PROLOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv, 8
 ; PROLOG-NEXT:    %exitcond.7 = icmp eq i64 %indvars.iv.next.7, %n
 ; PROLOG-NEXT:    br i1 %exitcond.7, label %latchExit.unr-lcssa.loopexit, label %header
 ; PROLOG:       latchExit.unr-lcssa.loopexit:
@@ -2100,7 +2028,7 @@ define i32 @otherblock_latch_same_exit(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-BLOCK-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; PROLOG-BLOCK-NEXT:    %5 = load i32, ptr %arrayidx.1, align 4
 ; PROLOG-BLOCK-NEXT:    %add.1 = add nsw i32 %5, %add
-; PROLOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
+; PROLOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv, 2
 ; PROLOG-BLOCK-NEXT:    %exitcond.1 = icmp eq i64 %indvars.iv.next.1, %n
 ; PROLOG-BLOCK-NEXT:    br i1 %exitcond.1, label %latchExit.unr-lcssa.loopexit, label %header, !llvm.loop !5
 ; PROLOG-BLOCK:       latchExit.unr-lcssa.loopexit:
@@ -2174,7 +2102,6 @@ define i32 @otherblock_latch_same_exit2(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %3 = load i32, ptr %arrayidx, align 4
 ; EPILOG-NEXT:    %add = add nsw i32 %3, %sum.02
 ; EPILOG-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.1
 ; EPILOG:       for.exiting_block.1:
 ; EPILOG-NEXT:    %cmp.1 = icmp eq i64 %n, 42
@@ -2183,8 +2110,7 @@ define i32 @otherblock_latch_same_exit2(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; EPILOG-NEXT:    %4 = load i32, ptr %arrayidx.1, align 4
 ; EPILOG-NEXT:    %add.1 = add nsw i32 %4, %add
-; EPILOG-NEXT:    %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv.next, 1
-; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
+; EPILOG-NEXT:    %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv, 2
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.2
 ; EPILOG:       for.exiting_block.2:
 ; EPILOG-NEXT:    %cmp.2 = icmp eq i64 %n, 42
@@ -2193,8 +2119,7 @@ define i32 @otherblock_latch_same_exit2(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.2 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.1
 ; EPILOG-NEXT:    %5 = load i32, ptr %arrayidx.2, align 4
 ; EPILOG-NEXT:    %add.2 = add nsw i32 %5, %add.1
-; EPILOG-NEXT:    %indvars.iv.next.2 = add nuw nsw i64 %indvars.iv.next.1, 1
-; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
+; EPILOG-NEXT:    %indvars.iv.next.2 = add nuw nsw i64 %indvars.iv, 3
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.3
 ; EPILOG:       for.exiting_block.3:
 ; EPILOG-NEXT:    %cmp.3 = icmp eq i64 %n, 42
@@ -2203,8 +2128,7 @@ define i32 @otherblock_latch_same_exit2(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.3 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.2
 ; EPILOG-NEXT:    %6 = load i32, ptr %arrayidx.3, align 4
 ; EPILOG-NEXT:    %add.3 = add nsw i32 %6, %add.2
-; EPILOG-NEXT:    %indvars.iv.next.3 = add nuw nsw i64 %indvars.iv.next.2, 1
-; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
+; EPILOG-NEXT:    %indvars.iv.next.3 = add nuw nsw i64 %indvars.iv, 4
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.4
 ; EPILOG:       for.exiting_block.4:
 ; EPILOG-NEXT:    %cmp.4 = icmp eq i64 %n, 42
@@ -2213,8 +2137,7 @@ define i32 @otherblock_latch_same_exit2(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.4 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.3
 ; EPILOG-NEXT:    %7 = load i32, ptr %arrayidx.4, align 4
 ; EPILOG-NEXT:    %add.4 = add nsw i32 %7, %add.3
-; EPILOG-NEXT:    %indvars.iv.next.4 = add nuw nsw i64 %indvars.iv.next.3, 1
-; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
+; EPILOG-NEXT:    %indvars.iv.next.4 = add nuw nsw i64 %indvars.iv, 5
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.5
 ; EPILOG:       for.exiting_block.5:
 ; EPILOG-NEXT:    %cmp.5 = icmp eq i64 %n, 42
@@ -2223,8 +2146,7 @@ define i32 @otherblock_latch_same_exit2(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.5 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.4
 ; EPILOG-NEXT:    %8 = load i32, ptr %arrayidx.5, align 4
 ; EPILOG-NEXT:    %add.5 = add nsw i32 %8, %add.4
-; EPILOG-NEXT:    %indvars.iv.next.5 = add nuw nsw i64 %indvars.iv.next.4, 1
-; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
+; EPILOG-NEXT:    %indvars.iv.next.5 = add nuw nsw i64 %indvars.iv, 6
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.6
 ; EPILOG:       for.exiting_block.6:
 ; EPILOG-NEXT:    %cmp.6 = icmp eq i64 %n, 42
@@ -2233,8 +2155,7 @@ define i32 @otherblock_latch_same_exit2(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.6 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.5
 ; EPILOG-NEXT:    %9 = load i32, ptr %arrayidx.6, align 4
 ; EPILOG-NEXT:    %add.6 = add nsw i32 %9, %add.5
-; EPILOG-NEXT:    %indvars.iv.next.6 = add nuw nsw i64 %indvars.iv.next.5, 1
-; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
+; EPILOG-NEXT:    %indvars.iv.next.6 = add nuw nsw i64 %indvars.iv, 7
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.7
 ; EPILOG:       for.exiting_block.7:
 ; EPILOG-NEXT:    %cmp.7 = icmp eq i64 %n, 42
@@ -2243,8 +2164,8 @@ define i32 @otherblock_latch_same_exit2(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %arrayidx.7 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.6
 ; EPILOG-NEXT:    %10 = load i32, ptr %arrayidx.7, align 4
 ; EPILOG-NEXT:    %add.7 = add nsw i32 %10, %add.6
-; EPILOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv.next.6, 1
-; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; EPILOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv, 8
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter, 8
 ; EPILOG-NEXT:    %niter.ncmp.7 = icmp eq i64 %niter.next.7, %unroll_iter
 ; EPILOG-NEXT:    br i1 %niter.ncmp.7, label %latchExit.unr-lcssa.loopexit, label %header
 ; EPILOG:       latchExit.unr-lcssa.loopexit:
@@ -2319,7 +2240,6 @@ define i32 @otherblock_latch_same_exit2(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    %3 = load i32, ptr %arrayidx, align 4
 ; EPILOG-BLOCK-NEXT:    %add = add nsw i32 %3, %sum.02
 ; EPILOG-BLOCK-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-BLOCK-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-BLOCK-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.1
 ; EPILOG-BLOCK:       for.exiting_block.1:
 ; EPILOG-BLOCK-NEXT:    %cmp.1 = icmp eq i64 %n, 42
@@ -2328,8 +2248,8 @@ define i32 @otherblock_latch_same_exit2(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; EPILOG-BLOCK-NEXT:    %4 = load i32, ptr %arrayidx.1, align 4
 ; EPILOG-BLOCK-NEXT:    %add.1 = add nsw i32 %4, %add
-; EPILOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
-; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; EPILOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv, 2
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp eq i64 %niter.next.1, %unroll_iter
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1, label %latchExit.unr-lcssa.loopexit, label %header, !llvm.loop !6
 ; EPILOG-BLOCK:       latchExit.unr-lcssa.loopexit:
@@ -2428,7 +2348,7 @@ define i32 @otherblock_latch_same_exit2(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; PROLOG-NEXT:    %5 = load i32, ptr %arrayidx.1, align 4
 ; PROLOG-NEXT:    %add.1 = add nsw i32 %5, %add
-; PROLOG-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
+; PROLOG-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv, 2
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.2
 ; PROLOG:       for.exiting_block.2:
 ; PROLOG-NEXT:    %cmp.2 = icmp eq i64 %n, 42
@@ -2437,7 +2357,7 @@ define i32 @otherblock_latch_same_exit2(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.2 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.1
 ; PROLOG-NEXT:    %6 = load i32, ptr %arrayidx.2, align 4
 ; PROLOG-NEXT:    %add.2 = add nsw i32 %6, %add.1
-; PROLOG-NEXT:    %indvars.iv.next.2 = add i64 %indvars.iv.next.1, 1
+; PROLOG-NEXT:    %indvars.iv.next.2 = add i64 %indvars.iv, 3
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.3
 ; PROLOG:       for.exiting_block.3:
 ; PROLOG-NEXT:    %cmp.3 = icmp eq i64 %n, 42
@@ -2446,7 +2366,7 @@ define i32 @otherblock_latch_same_exit2(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.3 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.2
 ; PROLOG-NEXT:    %7 = load i32, ptr %arrayidx.3, align 4
 ; PROLOG-NEXT:    %add.3 = add nsw i32 %7, %add.2
-; PROLOG-NEXT:    %indvars.iv.next.3 = add i64 %indvars.iv.next.2, 1
+; PROLOG-NEXT:    %indvars.iv.next.3 = add i64 %indvars.iv, 4
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.4
 ; PROLOG:       for.exiting_block.4:
 ; PROLOG-NEXT:    %cmp.4 = icmp eq i64 %n, 42
@@ -2455,7 +2375,7 @@ define i32 @otherblock_latch_same_exit2(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.4 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.3
 ; PROLOG-NEXT:    %8 = load i32, ptr %arrayidx.4, align 4
 ; PROLOG-NEXT:    %add.4 = add nsw i32 %8, %add.3
-; PROLOG-NEXT:    %indvars.iv.next.4 = add i64 %indvars.iv.next.3, 1
+; PROLOG-NEXT:    %indvars.iv.next.4 = add i64 %indvars.iv, 5
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.5
 ; PROLOG:       for.exiting_block.5:
 ; PROLOG-NEXT:    %cmp.5 = icmp eq i64 %n, 42
@@ -2464,7 +2384,7 @@ define i32 @otherblock_latch_same_exit2(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.5 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.4
 ; PROLOG-NEXT:    %9 = load i32, ptr %arrayidx.5, align 4
 ; PROLOG-NEXT:    %add.5 = add nsw i32 %9, %add.4
-; PROLOG-NEXT:    %indvars.iv.next.5 = add i64 %indvars.iv.next.4, 1
+; PROLOG-NEXT:    %indvars.iv.next.5 = add i64 %indvars.iv, 6
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.6
 ; PROLOG:       for.exiting_block.6:
 ; PROLOG-NEXT:    %cmp.6 = icmp eq i64 %n, 42
@@ -2473,7 +2393,7 @@ define i32 @otherblock_latch_same_exit2(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.6 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.5
 ; PROLOG-NEXT:    %10 = load i32, ptr %arrayidx.6, align 4
 ; PROLOG-NEXT:    %add.6 = add nsw i32 %10, %add.5
-; PROLOG-NEXT:    %indvars.iv.next.6 = add i64 %indvars.iv.next.5, 1
+; PROLOG-NEXT:    %indvars.iv.next.6 = add i64 %indvars.iv, 7
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.7
 ; PROLOG:       for.exiting_block.7:
 ; PROLOG-NEXT:    %cmp.7 = icmp eq i64 %n, 42
@@ -2482,7 +2402,7 @@ define i32 @otherblock_latch_same_exit2(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %arrayidx.7 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.6
 ; PROLOG-NEXT:    %11 = load i32, ptr %arrayidx.7, align 4
 ; PROLOG-NEXT:    %add.7 = add nsw i32 %11, %add.6
-; PROLOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv.next.6, 1
+; PROLOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv, 8
 ; PROLOG-NEXT:    %exitcond.7 = icmp eq i64 %indvars.iv.next.7, %n
 ; PROLOG-NEXT:    br i1 %exitcond.7, label %latchExit.unr-lcssa.loopexit, label %header
 ; PROLOG:       latchExit.unr-lcssa.loopexit:
@@ -2549,7 +2469,7 @@ define i32 @otherblock_latch_same_exit2(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-BLOCK-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; PROLOG-BLOCK-NEXT:    %5 = load i32, ptr %arrayidx.1, align 4
 ; PROLOG-BLOCK-NEXT:    %add.1 = add nsw i32 %5, %add
-; PROLOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
+; PROLOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv, 2
 ; PROLOG-BLOCK-NEXT:    %exitcond.1 = icmp eq i64 %indvars.iv.next.1, %n
 ; PROLOG-BLOCK-NEXT:    br i1 %exitcond.1, label %latchExit.unr-lcssa.loopexit, label %header, !llvm.loop !6
 ; PROLOG-BLOCK:       latchExit.unr-lcssa.loopexit:
@@ -2624,7 +2544,6 @@ define i32 @otherblock_latch_same_exit3(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    br i1 %cmp, label %latchExit.epilog-lcssa.loopexit, label %latch
 ; EPILOG:       latch:
 ; EPILOG-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.1
 ; EPILOG:       for.exiting_block.1:
 ; EPILOG-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
@@ -2633,8 +2552,7 @@ define i32 @otherblock_latch_same_exit3(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %cmp.1 = icmp eq i64 %n, 42
 ; EPILOG-NEXT:    br i1 %cmp.1, label %latchExit.epilog-lcssa.loopexit, label %latch.1
 ; EPILOG:       latch.1:
-; EPILOG-NEXT:    %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv.next, 1
-; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
+; EPILOG-NEXT:    %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv, 2
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.2
 ; EPILOG:       for.exiting_block.2:
 ; EPILOG-NEXT:    %arrayidx.2 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.1
@@ -2643,8 +2561,7 @@ define i32 @otherblock_latch_same_exit3(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %cmp.2 = icmp eq i64 %n, 42
 ; EPILOG-NEXT:    br i1 %cmp.2, label %latchExit.epilog-lcssa.loopexit, label %latch.2
 ; EPILOG:       latch.2:
-; EPILOG-NEXT:    %indvars.iv.next.2 = add nuw nsw i64 %indvars.iv.next.1, 1
-; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
+; EPILOG-NEXT:    %indvars.iv.next.2 = add nuw nsw i64 %indvars.iv, 3
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.3
 ; EPILOG:       for.exiting_block.3:
 ; EPILOG-NEXT:    %arrayidx.3 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.2
@@ -2653,8 +2570,7 @@ define i32 @otherblock_latch_same_exit3(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %cmp.3 = icmp eq i64 %n, 42
 ; EPILOG-NEXT:    br i1 %cmp.3, label %latchExit.epilog-lcssa.loopexit, label %latch.3
 ; EPILOG:       latch.3:
-; EPILOG-NEXT:    %indvars.iv.next.3 = add nuw nsw i64 %indvars.iv.next.2, 1
-; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
+; EPILOG-NEXT:    %indvars.iv.next.3 = add nuw nsw i64 %indvars.iv, 4
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.4
 ; EPILOG:       for.exiting_block.4:
 ; EPILOG-NEXT:    %arrayidx.4 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.3
@@ -2663,8 +2579,7 @@ define i32 @otherblock_latch_same_exit3(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %cmp.4 = icmp eq i64 %n, 42
 ; EPILOG-NEXT:    br i1 %cmp.4, label %latchExit.epilog-lcssa.loopexit, label %latch.4
 ; EPILOG:       latch.4:
-; EPILOG-NEXT:    %indvars.iv.next.4 = add nuw nsw i64 %indvars.iv.next.3, 1
-; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
+; EPILOG-NEXT:    %indvars.iv.next.4 = add nuw nsw i64 %indvars.iv, 5
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.5
 ; EPILOG:       for.exiting_block.5:
 ; EPILOG-NEXT:    %arrayidx.5 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.4
@@ -2673,8 +2588,7 @@ define i32 @otherblock_latch_same_exit3(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %cmp.5 = icmp eq i64 %n, 42
 ; EPILOG-NEXT:    br i1 %cmp.5, label %latchExit.epilog-lcssa.loopexit, label %latch.5
 ; EPILOG:       latch.5:
-; EPILOG-NEXT:    %indvars.iv.next.5 = add nuw nsw i64 %indvars.iv.next.4, 1
-; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
+; EPILOG-NEXT:    %indvars.iv.next.5 = add nuw nsw i64 %indvars.iv, 6
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.6
 ; EPILOG:       for.exiting_block.6:
 ; EPILOG-NEXT:    %arrayidx.6 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.5
@@ -2683,8 +2597,7 @@ define i32 @otherblock_latch_same_exit3(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %cmp.6 = icmp eq i64 %n, 42
 ; EPILOG-NEXT:    br i1 %cmp.6, label %latchExit.epilog-lcssa.loopexit, label %latch.6
 ; EPILOG:       latch.6:
-; EPILOG-NEXT:    %indvars.iv.next.6 = add nuw nsw i64 %indvars.iv.next.5, 1
-; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
+; EPILOG-NEXT:    %indvars.iv.next.6 = add nuw nsw i64 %indvars.iv, 7
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.7
 ; EPILOG:       for.exiting_block.7:
 ; EPILOG-NEXT:    %arrayidx.7 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.6
@@ -2693,8 +2606,8 @@ define i32 @otherblock_latch_same_exit3(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %cmp.7 = icmp eq i64 %n, 42
 ; EPILOG-NEXT:    br i1 %cmp.7, label %latchExit.epilog-lcssa.loopexit, label %latch.7
 ; EPILOG:       latch.7:
-; EPILOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv.next.6, 1
-; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; EPILOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv, 8
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter, 8
 ; EPILOG-NEXT:    %niter.ncmp.7 = icmp eq i64 %niter.next.7, %unroll_iter
 ; EPILOG-NEXT:    br i1 %niter.ncmp.7, label %latchExit.unr-lcssa.loopexit, label %header
 ; EPILOG:       latchExit.unr-lcssa.loopexit:
@@ -2769,7 +2682,6 @@ define i32 @otherblock_latch_same_exit3(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    br i1 %cmp, label %latchExit.epilog-lcssa.loopexit, label %latch
 ; EPILOG-BLOCK:       latch:
 ; EPILOG-BLOCK-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-BLOCK-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-BLOCK-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.1
 ; EPILOG-BLOCK:       for.exiting_block.1:
 ; EPILOG-BLOCK-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
@@ -2778,8 +2690,8 @@ define i32 @otherblock_latch_same_exit3(ptr nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    %cmp.1 = icmp eq i64 %n, 42
 ; EPILOG-BLOCK-NEXT:    br i1 %cmp.1, label %latchExit.epilog-lcssa.loopexit, label %latch.1
 ; EPILOG-BLOCK:       latch.1:
-; EPILOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
-; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; EPILOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv, 2
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp eq i64 %niter.next.1, %unroll_iter
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1, label %latchExit.unr-lcssa.loopexit, label %header, !llvm.loop !7
 ; EPILOG-BLOCK:       latchExit.unr-lcssa.loopexit:
@@ -2878,7 +2790,7 @@ define i32 @otherblock_latch_same_exit3(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %cmp.1 = icmp eq i64 %n, 42
 ; PROLOG-NEXT:    br i1 %cmp.1, label %latchExit.unr-lcssa.loopexit, label %latch.1
 ; PROLOG:       latch.1:
-; PROLOG-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
+; PROLOG-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv, 2
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.2
 ; PROLOG:       for.exiting_block.2:
 ; PROLOG-NEXT:    %arrayidx.2 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.1
@@ -2887,7 +2799,7 @@ define i32 @otherblock_latch_same_exit3(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %cmp.2 = icmp eq i64 %n, 42
 ; PROLOG-NEXT:    br i1 %cmp.2, label %latchExit.unr-lcssa.loopexit, label %latch.2
 ; PROLOG:       latch.2:
-; PROLOG-NEXT:    %indvars.iv.next.2 = add i64 %indvars.iv.next.1, 1
+; PROLOG-NEXT:    %indvars.iv.next.2 = add i64 %indvars.iv, 3
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.3
 ; PROLOG:       for.exiting_block.3:
 ; PROLOG-NEXT:    %arrayidx.3 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.2
@@ -2896,7 +2808,7 @@ define i32 @otherblock_latch_same_exit3(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %cmp.3 = icmp eq i64 %n, 42
 ; PROLOG-NEXT:    br i1 %cmp.3, label %latchExit.unr-lcssa.loopexit, label %latch.3
 ; PROLOG:       latch.3:
-; PROLOG-NEXT:    %indvars.iv.next.3 = add i64 %indvars.iv.next.2, 1
+; PROLOG-NEXT:    %indvars.iv.next.3 = add i64 %indvars.iv, 4
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.4
 ; PROLOG:       for.exiting_block.4:
 ; PROLOG-NEXT:    %arrayidx.4 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.3
@@ -2905,7 +2817,7 @@ define i32 @otherblock_latch_same_exit3(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %cmp.4 = icmp eq i64 %n, 42
 ; PROLOG-NEXT:    br i1 %cmp.4, label %latchExit.unr-lcssa.loopexit, label %latch.4
 ; PROLOG:       latch.4:
-; PROLOG-NEXT:    %indvars.iv.next.4 = add i64 %indvars.iv.next.3, 1
+; PROLOG-NEXT:    %indvars.iv.next.4 = add i64 %indvars.iv, 5
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.5
 ; PROLOG:       for.exiting_block.5:
 ; PROLOG-NEXT:    %arrayidx.5 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.4
@@ -2914,7 +2826,7 @@ define i32 @otherblock_latch_same_exit3(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %cmp.5 = icmp eq i64 %n, 42
 ; PROLOG-NEXT:    br i1 %cmp.5, label %latchExit.unr-lcssa.loopexit, label %latch.5
 ; PROLOG:       latch.5:
-; PROLOG-NEXT:    %indvars.iv.next.5 = add i64 %indvars.iv.next.4, 1
+; PROLOG-NEXT:    %indvars.iv.next.5 = add i64 %indvars.iv, 6
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.6
 ; PROLOG:       for.exiting_block.6:
 ; PROLOG-NEXT:    %arrayidx.6 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.5
@@ -2923,7 +2835,7 @@ define i32 @otherblock_latch_same_exit3(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %cmp.6 = icmp eq i64 %n, 42
 ; PROLOG-NEXT:    br i1 %cmp.6, label %latchExit.unr-lcssa.loopexit, label %latch.6
 ; PROLOG:       latch.6:
-; PROLOG-NEXT:    %indvars.iv.next.6 = add i64 %indvars.iv.next.5, 1
+; PROLOG-NEXT:    %indvars.iv.next.6 = add i64 %indvars.iv, 7
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block.7
 ; PROLOG:       for.exiting_block.7:
 ; PROLOG-NEXT:    %arrayidx.7 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.6
@@ -2932,7 +2844,7 @@ define i32 @otherblock_latch_same_exit3(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %cmp.7 = icmp eq i64 %n, 42
 ; PROLOG-NEXT:    br i1 %cmp.7, label %latchExit.unr-lcssa.loopexit, label %latch.7
 ; PROLOG:       latch.7:
-; PROLOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv.next.6, 1
+; PROLOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv, 8
 ; PROLOG-NEXT:    %exitcond.7 = icmp eq i64 %indvars.iv.next.7, %n
 ; PROLOG-NEXT:    br i1 %exitcond.7, label %latchExit.unr-lcssa.loopexit, label %header
 ; PROLOG:       latchExit.unr-lcssa.loopexit:
@@ -2999,7 +2911,7 @@ define i32 @otherblock_latch_same_exit3(ptr nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-BLOCK-NEXT:    %cmp.1 = icmp eq i64 %n, 42
 ; PROLOG-BLOCK-NEXT:    br i1 %cmp.1, label %latchExit.unr-lcssa.loopexit, label %latch.1
 ; PROLOG-BLOCK:       latch.1:
-; PROLOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
+; PROLOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv, 2
 ; PROLOG-BLOCK-NEXT:    %exitcond.1 = icmp eq i64 %indvars.iv.next.1, %n
 ; PROLOG-BLOCK-NEXT:    br i1 %exitcond.1, label %latchExit.unr-lcssa.loopexit, label %header, !llvm.loop !7
 ; PROLOG-BLOCK:       latchExit.unr-lcssa.loopexit:
@@ -3112,42 +3024,35 @@ define void @unique_exit(i32 %N, i32 %M) {
 ; EPILOG-NEXT:    %cmp1 = icmp ult i32 %inc, %N
 ; EPILOG-NEXT:    br i1 %cmp1, label %latch, label %latchExit.epilog-lcssa.loopexit
 ; EPILOG:       latch:
-; EPILOG-NEXT:    %niter.next = add nuw nsw i32 %niter, 1
-; EPILOG-NEXT:    %inc.1 = add nuw nsw i32 %inc, 1
+; EPILOG-NEXT:    %inc.1 = add nuw nsw i32 %i4, 2
 ; EPILOG-NEXT:    %cmp1.1 = icmp ult i32 %inc.1, %N
 ; EPILOG-NEXT:    br i1 %cmp1.1, label %latch.1, label %latchExit.epilog-lcssa.loopexit
 ; EPILOG:       latch.1:
-; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i32 %niter.next, 1
-; EPILOG-NEXT:    %inc.2 = add nuw nsw i32 %inc.1, 1
+; EPILOG-NEXT:    %inc.2 = add nuw nsw i32 %i4, 3
 ; EPILOG-NEXT:    %cmp1.2 = icmp ult i32 %inc.2, %N
 ; EPILOG-NEXT:    br i1 %cmp1.2, label %latch.2, label %latchExit.epilog-lcssa.loopexit
 ; EPILOG:       latch.2:
-; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i32 %niter.next.1, 1
-; EPILOG-NEXT:    %inc.3 = add nuw nsw i32 %inc.2, 1
+; EPILOG-NEXT:    %inc.3 = add nuw nsw i32 %i4, 4
 ; EPILOG-NEXT:    %cmp1.3 = icmp ult i32 %inc.3, %N
 ; EPILOG-NEXT:    br i1 %cmp1.3, label %latch.3, label %latchExit.epilog-lcssa.loopexit
 ; EPILOG:       latch.3:
-; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i32 %niter.next.2, 1
-; EPILOG-NEXT:    %inc.4 = add nuw nsw i32 %inc.3, 1
+; EPILOG-NEXT:    %inc.4 = add nuw nsw i32 %i4, 5
 ; EPILOG-NEXT:    %cmp1.4 = icmp ult i32 %inc.4, %N
 ; EPILOG-NEXT:    br i1 %cmp1.4, label %latch.4, label %latchExit.epilog-lcssa.loopexit
 ; EPILOG:       latch.4:
-; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i32 %niter.next.3, 1
-; EPILOG-NEXT:    %inc.5 = add nuw nsw i32 %inc.4, 1
+; EPILOG-NEXT:    %inc.5 = add nuw nsw i32 %i4, 6
 ; EPILOG-NEXT:    %cmp1.5 = icmp ult i32 %inc.5, %N
 ; EPILOG-NEXT:    br i1 %cmp1.5, label %latch.5, label %latchExit.epilog-lcssa.loopexit
 ; EPILOG:       latch.5:
-; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i32 %niter.next.4, 1
-; EPILOG-NEXT:    %inc.6 = add nuw nsw i32 %inc.5, 1
+; EPILOG-NEXT:    %inc.6 = add nuw nsw i32 %i4, 7
 ; EPILOG-NEXT:    %cmp1.6 = icmp ult i32 %inc.6, %N
 ; EPILOG-NEXT:    br i1 %cmp1.6, label %latch.6, label %latchExit.epilog-lcssa.loopexit
 ; EPILOG:       latch.6:
-; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i32 %niter.next.5, 1
-; EPILOG-NEXT:    %inc.7 = add nuw i32 %inc.6, 1
+; EPILOG-NEXT:    %inc.7 = add nuw i32 %i4, 8
 ; EPILOG-NEXT:    %cmp1.7 = icmp ult i32 %inc.7, %N
 ; EPILOG-NEXT:    br i1 %cmp1.7, label %latch.7, label %latchExit.epilog-lcssa.loopexit
 ; EPILOG:       latch.7:
-; EPILOG-NEXT:    %niter.next.7 = add nuw i32 %niter.next.6, 1
+; EPILOG-NEXT:    %niter.next.7 = add nuw i32 %niter, 8
 ; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i32 %niter.next.7, %unroll_iter
 ; EPILOG-NEXT:    br i1 %niter.ncmp.7, label %header, label %latchExit.unr-lcssa.loopexit
 ; EPILOG:       latchExit.unr-lcssa.loopexit:
@@ -3204,12 +3109,11 @@ define void @unique_exit(i32 %N, i32 %M) {
 ; EPILOG-BLOCK-NEXT:    %cmp1 = icmp ult i32 %inc, %N
 ; EPILOG-BLOCK-NEXT:    br i1 %cmp1, label %latch, label %latchExit.epilog-lcssa.loopexit
 ; EPILOG-BLOCK:       latch:
-; EPILOG-BLOCK-NEXT:    %niter.next = add nuw nsw i32 %niter, 1
-; EPILOG-BLOCK-NEXT:    %inc.1 = add nuw i32 %inc, 1
+; EPILOG-BLOCK-NEXT:    %inc.1 = add nuw i32 %i4, 2
 ; EPILOG-BLOCK-NEXT:    %cmp1.1 = icmp ult i32 %inc.1, %N
 ; EPILOG-BLOCK-NEXT:    br i1 %cmp1.1, label %latch.1, label %latchExit.epilog-lcssa.loopexit
 ; EPILOG-BLOCK:       latch.1:
-; EPILOG-BLOCK-NEXT:    %niter.next.1 = add nuw i32 %niter.next, 1
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add nuw i32 %niter, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i32 %niter.next.1, %unroll_iter
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1, label %header, label %latchExit.unr-lcssa.loopexit, !llvm.loop !8
 ; EPILOG-BLOCK:       latchExit.unr-lcssa.loopexit:
@@ -3278,31 +3182,31 @@ define void @unique_exit(i32 %N, i32 %M) {
 ; PROLOG-NEXT:    %cmp1 = icmp ult i32 %inc, %N
 ; PROLOG-NEXT:    br i1 %cmp1, label %latch, label %latchExit.unr-lcssa.loopexit
 ; PROLOG:       latch:
-; PROLOG-NEXT:    %inc.1 = add nuw i32 %inc, 1
+; PROLOG-NEXT:    %inc.1 = add nuw i32 %i4, 2
 ; PROLOG-NEXT:    %cmp1.1 = icmp ult i32 %inc.1, %N
 ; PROLOG-NEXT:    br i1 %cmp1.1, label %latch.1, label %latchExit.unr-lcssa.loopexit
 ; PROLOG:       latch.1:
-; PROLOG-NEXT:    %inc.2 = add nuw i32 %inc.1, 1
+; PROLOG-NEXT:    %inc.2 = add nuw i32 %i4, 3
 ; PROLOG-NEXT:    %cmp1.2 = icmp ult i32 %inc.2, %N
 ; PROLOG-NEXT:    br i1 %cmp1.2, label %latch.2, label %latchExit.unr-lcssa.loopexit
 ; PROLOG:       latch.2:
-; PROLOG-NEXT:    %inc.3 = add nuw i32 %inc.2, 1
+; PROLOG-NEXT:    %inc.3 = add nuw i32 %i4, 4
 ; PROLOG-NEXT:    %cmp1.3 = icmp ult i32 %inc.3, %N
 ; PROLOG-NEXT:    br i1 %cmp1.3, label %latch.3, label %latchExit.unr-lcssa.loopexit
 ; PROLOG:       latch.3:
-; PROLOG-NEXT:    %inc.4 = add nuw i32 %inc.3, 1
+; PROLOG-NEXT:    %inc.4 = add nuw i32 %i4, 5
 ; PROLOG-NEXT:    %cmp1.4 = icmp ult i32 %inc.4, %N
 ; PROLOG-NEXT:    br i1 %cmp1.4, label %latch.4, label %latchExit.unr-lcssa.loopexit
 ; PROLOG:       latch.4:
-; PROLOG-NEXT:    %inc.5 = add nuw i32 %inc.4, 1
+; PROLOG-NEXT:    %inc.5 = add nuw i32 %i4, 6
 ; PROLOG-NEXT:    %cmp1.5 = icmp ult i32 %inc.5, %N
 ; PROLOG-NEXT:    br i1 %cmp1.5, label %latch.5, label %latchExit.unr-lcssa.loopexit
 ; PROLOG:       latch.5:
-; PROLOG-NEXT:    %inc.6 = add nuw i32 %inc.5, 1
+; PROLOG-NEXT:    %inc.6 = add nuw i32 %i4, 7
 ; PROLOG-NEXT:    %cmp1.6 = icmp ult i32 %inc.6, %N
 ; PROLOG-NEXT:    br i1 %cmp1.6, label %latch.6, label %latchExit.unr-lcssa.loopexit
 ; PROLOG:       latch.6:
-; PROLOG-NEXT:    %inc.7 = add nuw i32 %inc.6, 1
+; PROLOG-NEXT:    %inc.7 = add nuw i32 %i4, 8
 ; PROLOG-NEXT:    %cmp1.7 = icmp ult i32 %inc.7, %N
 ; PROLOG-NEXT:    br i1 %cmp1.7, label %latch.7, label %latchExit.unr-lcssa.loopexit
 ; PROLOG:       latch.7:
@@ -3350,7 +3254,7 @@ define void @unique_exit(i32 %N, i32 %M) {
 ; PROLOG-BLOCK-NEXT:    %cmp1 = icmp ult i32 %inc, %N
 ; PROLOG-BLOCK-NEXT:    br i1 %cmp1, label %latch, label %latchExit.unr-lcssa.loopexit
 ; PROLOG-BLOCK:       latch:
-; PROLOG-BLOCK-NEXT:    %inc.1 = add nuw i32 %inc, 1
+; PROLOG-BLOCK-NEXT:    %inc.1 = add nuw i32 %i4, 2
 ; PROLOG-BLOCK-NEXT:    %cmp1.1 = icmp ult i32 %inc.1, %N
 ; PROLOG-BLOCK-NEXT:    br i1 %cmp1.1, label %latch.1, label %latchExit.unr-lcssa.loopexit
 ; PROLOG-BLOCK:       latch.1:
@@ -3414,7 +3318,6 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; EPILOG:       loop_latch:
 ; EPILOG-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
 ; EPILOG-NEXT:    %sum.next = add i64 %sum, %add
-; EPILOG-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.1, label %loop_exiting.1
 ; EPILOG:       loop_exiting.1:
 ; EPILOG-NEXT:    %ivy.1 = add i64 %iv_next, %add
@@ -3423,9 +3326,8 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; EPILOG-NEXT:    i64 42, label %exit1.loopexit
 ; EPILOG-NEXT:    ]
 ; EPILOG:       loop_latch.1:
-; EPILOG-NEXT:    %iv_next.1 = add nuw nsw i64 %iv_next, 1
+; EPILOG-NEXT:    %iv_next.1 = add nuw nsw i64 %iv, 2
 ; EPILOG-NEXT:    %sum.next.1 = add i64 %sum.next, %add
-; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.2, label %loop_exiting.2
 ; EPILOG:       loop_exiting.2:
 ; EPILOG-NEXT:    %ivy.2 = add i64 %iv_next.1, %add
@@ -3434,9 +3336,8 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; EPILOG-NEXT:    i64 42, label %exit1.loopexit
 ; EPILOG-NEXT:    ]
 ; EPILOG:       loop_latch.2:
-; EPILOG-NEXT:    %iv_next.2 = add nuw nsw i64 %iv_next.1, 1
+; EPILOG-NEXT:    %iv_next.2 = add nuw nsw i64 %iv, 3
 ; EPILOG-NEXT:    %sum.next.2 = add i64 %sum.next.1, %add
-; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.3, label %loop_exiting.3
 ; EPILOG:       loop_exiting.3:
 ; EPILOG-NEXT:    %ivy.3 = add i64 %iv_next.2, %add
@@ -3445,9 +3346,8 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; EPILOG-NEXT:    i64 42, label %exit1.loopexit
 ; EPILOG-NEXT:    ]
 ; EPILOG:       loop_latch.3:
-; EPILOG-NEXT:    %iv_next.3 = add nuw nsw i64 %iv_next.2, 1
+; EPILOG-NEXT:    %iv_next.3 = add nuw nsw i64 %iv, 4
 ; EPILOG-NEXT:    %sum.next.3 = add i64 %sum.next.2, %add
-; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.4, label %loop_exiting.4
 ; EPILOG:       loop_exiting.4:
 ; EPILOG-NEXT:    %ivy.4 = add i64 %iv_next.3, %add
@@ -3456,9 +3356,8 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; EPILOG-NEXT:    i64 42, label %exit1.loopexit
 ; EPILOG-NEXT:    ]
 ; EPILOG:       loop_latch.4:
-; EPILOG-NEXT:    %iv_next.4 = add nuw nsw i64 %iv_next.3, 1
+; EPILOG-NEXT:    %iv_next.4 = add nuw nsw i64 %iv, 5
 ; EPILOG-NEXT:    %sum.next.4 = add i64 %sum.next.3, %add
-; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.5, label %loop_exiting.5
 ; EPILOG:       loop_exiting.5:
 ; EPILOG-NEXT:    %ivy.5 = add i64 %iv_next.4, %add
@@ -3467,9 +3366,8 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; EPILOG-NEXT:    i64 42, label %exit1.loopexit
 ; EPILOG-NEXT:    ]
 ; EPILOG:       loop_latch.5:
-; EPILOG-NEXT:    %iv_next.5 = add nuw nsw i64 %iv_next.4, 1
+; EPILOG-NEXT:    %iv_next.5 = add nuw nsw i64 %iv, 6
 ; EPILOG-NEXT:    %sum.next.5 = add i64 %sum.next.4, %add
-; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.6, label %loop_exiting.6
 ; EPILOG:       loop_exiting.6:
 ; EPILOG-NEXT:    %ivy.6 = add i64 %iv_next.5, %add
@@ -3478,9 +3376,8 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; EPILOG-NEXT:    i64 42, label %exit1.loopexit
 ; EPILOG-NEXT:    ]
 ; EPILOG:       loop_latch.6:
-; EPILOG-NEXT:    %iv_next.6 = add nuw nsw i64 %iv_next.5, 1
+; EPILOG-NEXT:    %iv_next.6 = add nuw nsw i64 %iv, 7
 ; EPILOG-NEXT:    %sum.next.6 = add i64 %sum.next.5, %add
-; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.7, label %loop_exiting.7
 ; EPILOG:       loop_exiting.7:
 ; EPILOG-NEXT:    %ivy.7 = add i64 %iv_next.6, %add
@@ -3489,9 +3386,9 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; EPILOG-NEXT:    i64 42, label %exit1.loopexit
 ; EPILOG-NEXT:    ]
 ; EPILOG:       loop_latch.7:
-; EPILOG-NEXT:    %iv_next.7 = add nuw nsw i64 %iv_next.6, 1
+; EPILOG-NEXT:    %iv_next.7 = add nuw nsw i64 %iv, 8
 ; EPILOG-NEXT:    %sum.next.7 = add i64 %sum.next.6, %add
-; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter, 8
 ; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.next.7, %unroll_iter
 ; EPILOG-NEXT:    br i1 %niter.ncmp.7, label %loop_header, label %latchexit.unr-lcssa.loopexit
 ; EPILOG:       exit1.loopexit:
@@ -3565,7 +3462,6 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; EPILOG-BLOCK:       loop_latch:
 ; EPILOG-BLOCK-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
 ; EPILOG-BLOCK-NEXT:    %sum.next = add i64 %sum, %add
-; EPILOG-BLOCK-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-BLOCK-NEXT:    br i1 %cond, label %loop_latch.1, label %loop_exiting.1
 ; EPILOG-BLOCK:       loop_exiting.1:
 ; EPILOG-BLOCK-NEXT:    %ivy.1 = add i64 %iv_next, %add
@@ -3574,9 +3470,9 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    i64 42, label %exit1.loopexit
 ; EPILOG-BLOCK-NEXT:    ]
 ; EPILOG-BLOCK:       loop_latch.1:
-; EPILOG-BLOCK-NEXT:    %iv_next.1 = add nuw nsw i64 %iv_next, 1
+; EPILOG-BLOCK-NEXT:    %iv_next.1 = add nuw nsw i64 %iv, 2
 ; EPILOG-BLOCK-NEXT:    %sum.next.1 = add i64 %sum.next, %add
-; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.next.1, %unroll_iter
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1, label %loop_header, label %latchexit.unr-lcssa.loopexit, !llvm.loop !9
 ; EPILOG-BLOCK:       exit1.loopexit:
@@ -3674,7 +3570,7 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; PROLOG-NEXT:    i64 42, label %exit1.loopexit
 ; PROLOG-NEXT:    ]
 ; PROLOG:       loop_latch.1:
-; PROLOG-NEXT:    %iv_next.1 = add nuw nsw i64 %iv_next, 1
+; PROLOG-NEXT:    %iv_next.1 = add nuw nsw i64 %iv, 2
 ; PROLOG-NEXT:    %sum.next.1 = add i64 %sum.next, %add
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.2, label %loop_exiting.2
 ; PROLOG:       loop_exiting.2:
@@ -3684,7 +3580,7 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; PROLOG-NEXT:    i64 42, label %exit1.loopexit
 ; PROLOG-NEXT:    ]
 ; PROLOG:       loop_latch.2:
-; PROLOG-NEXT:    %iv_next.2 = add nuw nsw i64 %iv_next.1, 1
+; PROLOG-NEXT:    %iv_next.2 = add nuw nsw i64 %iv, 3
 ; PROLOG-NEXT:    %sum.next.2 = add i64 %sum.next.1, %add
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.3, label %loop_exiting.3
 ; PROLOG:       loop_exiting.3:
@@ -3694,7 +3590,7 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; PROLOG-NEXT:    i64 42, label %exit1.loopexit
 ; PROLOG-NEXT:    ]
 ; PROLOG:       loop_latch.3:
-; PROLOG-NEXT:    %iv_next.3 = add nuw nsw i64 %iv_next.2, 1
+; PROLOG-NEXT:    %iv_next.3 = add nuw nsw i64 %iv, 4
 ; PROLOG-NEXT:    %sum.next.3 = add i64 %sum.next.2, %add
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.4, label %loop_exiting.4
 ; PROLOG:       loop_exiting.4:
@@ -3704,7 +3600,7 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; PROLOG-NEXT:    i64 42, label %exit1.loopexit
 ; PROLOG-NEXT:    ]
 ; PROLOG:       loop_latch.4:
-; PROLOG-NEXT:    %iv_next.4 = add nuw nsw i64 %iv_next.3, 1
+; PROLOG-NEXT:    %iv_next.4 = add nuw nsw i64 %iv, 5
 ; PROLOG-NEXT:    %sum.next.4 = add i64 %sum.next.3, %add
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.5, label %loop_exiting.5
 ; PROLOG:       loop_exiting.5:
@@ -3714,7 +3610,7 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; PROLOG-NEXT:    i64 42, label %exit1.loopexit
 ; PROLOG-NEXT:    ]
 ; PROLOG:       loop_latch.5:
-; PROLOG-NEXT:    %iv_next.5 = add nuw nsw i64 %iv_next.4, 1
+; PROLOG-NEXT:    %iv_next.5 = add nuw nsw i64 %iv, 6
 ; PROLOG-NEXT:    %sum.next.5 = add i64 %sum.next.4, %add
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.6, label %loop_exiting.6
 ; PROLOG:       loop_exiting.6:
@@ -3724,7 +3620,7 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; PROLOG-NEXT:    i64 42, label %exit1.loopexit
 ; PROLOG-NEXT:    ]
 ; PROLOG:       loop_latch.6:
-; PROLOG-NEXT:    %iv_next.6 = add nuw nsw i64 %iv_next.5, 1
+; PROLOG-NEXT:    %iv_next.6 = add nuw nsw i64 %iv, 7
 ; PROLOG-NEXT:    %sum.next.6 = add i64 %sum.next.5, %add
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.7, label %loop_exiting.7
 ; PROLOG:       loop_exiting.7:
@@ -3734,7 +3630,7 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; PROLOG-NEXT:    i64 42, label %exit1.loopexit
 ; PROLOG-NEXT:    ]
 ; PROLOG:       loop_latch.7:
-; PROLOG-NEXT:    %iv_next.7 = add nuw nsw i64 %iv_next.6, 1
+; PROLOG-NEXT:    %iv_next.7 = add nuw nsw i64 %iv, 8
 ; PROLOG-NEXT:    %sum.next.7 = add i64 %sum.next.6, %add
 ; PROLOG-NEXT:    %cmp.7 = icmp ne i64 %iv_next.7, %trip
 ; PROLOG-NEXT:    br i1 %cmp.7, label %loop_header, label %latchexit.unr-lcssa
@@ -3801,7 +3697,7 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; PROLOG-BLOCK-NEXT:    i64 42, label %exit1.loopexit
 ; PROLOG-BLOCK-NEXT:    ]
 ; PROLOG-BLOCK:       loop_latch.1:
-; PROLOG-BLOCK-NEXT:    %iv_next.1 = add nuw nsw i64 %iv_next, 1
+; PROLOG-BLOCK-NEXT:    %iv_next.1 = add nuw nsw i64 %iv, 2
 ; PROLOG-BLOCK-NEXT:    %sum.next.1 = add i64 %sum.next, %add
 ; PROLOG-BLOCK-NEXT:    %cmp.1 = icmp ne i64 %iv_next.1, %trip
 ; PROLOG-BLOCK-NEXT:    br i1 %cmp.1, label %loop_header, label %latchexit.unr-lcssa, !llvm.loop !9
@@ -3873,7 +3769,6 @@ define i32 @test6(ptr nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-NEXT:    %load = load i32, ptr %arrayidx, align 4
 ; EPILOG-NEXT:    %add = add nsw i32 %load, %sum.02
 ; EPILOG-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.1
 ; EPILOG:       for.exiting_block.1:
 ; EPILOG-NEXT:    %cmp.1 = icmp eq i64 %n, 42
@@ -3882,8 +3777,7 @@ define i32 @test6(ptr nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; EPILOG-NEXT:    %load.1 = load i32, ptr %arrayidx.1, align 4
 ; EPILOG-NEXT:    %add.1 = add nsw i32 %load.1, %add
-; EPILOG-NEXT:    %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv.next, 1
-; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
+; EPILOG-NEXT:    %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv, 2
 ; EPILOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.2
 ; EPILOG:       for.exiting_block.2:
 ; EPILOG-NEXT:    %cmp.2 = icmp eq i64 %n, 42
@@ -3892,8 +3786,7 @@ define i32 @test6(ptr nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-NEXT:    %arrayidx.2 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.1
 ; EPILOG-NEXT:    %load.2 = load i32, ptr %arrayidx.2, align 4
 ; EPILOG-NEXT:    %add.2 = add nsw i32 %load.2, %add.1
-; EPILOG-NEXT:    %indvars.iv.next.2 = add nuw nsw i64 %indvars.iv.next.1, 1
-; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
+; EPILOG-NEXT:    %indvars.iv.next.2 = add nuw nsw i64 %indvars.iv, 3
 ; EPILOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.3
 ; EPILOG:       for.exiting_block.3:
 ; EPILOG-NEXT:    %cmp.3 = icmp eq i64 %n, 42
@@ -3902,8 +3795,7 @@ define i32 @test6(ptr nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-NEXT:    %arrayidx.3 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.2
 ; EPILOG-NEXT:    %load.3 = load i32, ptr %arrayidx.3, align 4
 ; EPILOG-NEXT:    %add.3 = add nsw i32 %load.3, %add.2
-; EPILOG-NEXT:    %indvars.iv.next.3 = add nuw nsw i64 %indvars.iv.next.2, 1
-; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
+; EPILOG-NEXT:    %indvars.iv.next.3 = add nuw nsw i64 %indvars.iv, 4
 ; EPILOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.4
 ; EPILOG:       for.exiting_block.4:
 ; EPILOG-NEXT:    %cmp.4 = icmp eq i64 %n, 42
@@ -3912,8 +3804,7 @@ define i32 @test6(ptr nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-NEXT:    %arrayidx.4 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.3
 ; EPILOG-NEXT:    %load.4 = load i32, ptr %arrayidx.4, align 4
 ; EPILOG-NEXT:    %add.4 = add nsw i32 %load.4, %add.3
-; EPILOG-NEXT:    %indvars.iv.next.4 = add nuw nsw i64 %indvars.iv.next.3, 1
-; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
+; EPILOG-NEXT:    %indvars.iv.next.4 = add nuw nsw i64 %indvars.iv, 5
 ; EPILOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.5
 ; EPILOG:       for.exiting_block.5:
 ; EPILOG-NEXT:    %cmp.5 = icmp eq i64 %n, 42
@@ -3922,8 +3813,7 @@ define i32 @test6(ptr nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-NEXT:    %arrayidx.5 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.4
 ; EPILOG-NEXT:    %load.5 = load i32, ptr %arrayidx.5, align 4
 ; EPILOG-NEXT:    %add.5 = add nsw i32 %load.5, %add.4
-; EPILOG-NEXT:    %indvars.iv.next.5 = add nuw nsw i64 %indvars.iv.next.4, 1
-; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
+; EPILOG-NEXT:    %indvars.iv.next.5 = add nuw nsw i64 %indvars.iv, 6
 ; EPILOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.6
 ; EPILOG:       for.exiting_block.6:
 ; EPILOG-NEXT:    %cmp.6 = icmp eq i64 %n, 42
@@ -3932,8 +3822,7 @@ define i32 @test6(ptr nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-NEXT:    %arrayidx.6 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.5
 ; EPILOG-NEXT:    %load.6 = load i32, ptr %arrayidx.6, align 4
 ; EPILOG-NEXT:    %add.6 = add nsw i32 %load.6, %add.5
-; EPILOG-NEXT:    %indvars.iv.next.6 = add nuw nsw i64 %indvars.iv.next.5, 1
-; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
+; EPILOG-NEXT:    %indvars.iv.next.6 = add nuw nsw i64 %indvars.iv, 7
 ; EPILOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.7
 ; EPILOG:       for.exiting_block.7:
 ; EPILOG-NEXT:    %cmp.7 = icmp eq i64 %n, 42
@@ -3942,8 +3831,8 @@ define i32 @test6(ptr nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-NEXT:    %arrayidx.7 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.6
 ; EPILOG-NEXT:    %load.7 = load i32, ptr %arrayidx.7, align 4
 ; EPILOG-NEXT:    %add.7 = add nsw i32 %load.7, %add.6
-; EPILOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv.next.6, 1
-; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; EPILOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv, 8
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter, 8
 ; EPILOG-NEXT:    %niter.ncmp.7 = icmp eq i64 %niter.next.7, %unroll_iter
 ; EPILOG-NEXT:    br i1 %niter.ncmp.7, label %latch_exit.unr-lcssa.loopexit, label %header
 ; EPILOG:       latch_exit.unr-lcssa.loopexit:
@@ -4020,7 +3909,6 @@ define i32 @test6(ptr nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-BLOCK-NEXT:    %load = load i32, ptr %arrayidx, align 4
 ; EPILOG-BLOCK-NEXT:    %add = add nsw i32 %load, %sum.02
 ; EPILOG-BLOCK-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-BLOCK-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-BLOCK-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.1
 ; EPILOG-BLOCK:       for.exiting_block.1:
 ; EPILOG-BLOCK-NEXT:    %cmp.1 = icmp eq i64 %n, 42
@@ -4029,8 +3917,8 @@ define i32 @test6(ptr nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-BLOCK-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; EPILOG-BLOCK-NEXT:    %load.1 = load i32, ptr %arrayidx.1, align 4
 ; EPILOG-BLOCK-NEXT:    %add.1 = add nsw i32 %load.1, %add
-; EPILOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
-; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; EPILOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv, 2
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp eq i64 %niter.next.1, %unroll_iter
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1, label %latch_exit.unr-lcssa.loopexit, label %header, !llvm.loop !10
 ; EPILOG-BLOCK:       latch_exit.unr-lcssa.loopexit:
@@ -4130,7 +4018,7 @@ define i32 @test6(ptr nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; PROLOG-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; PROLOG-NEXT:    %load.1 = load i32, ptr %arrayidx.1, align 4
 ; PROLOG-NEXT:    %add.1 = add nsw i32 %load.1, %add
-; PROLOG-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
+; PROLOG-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv, 2
 ; PROLOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.2
 ; PROLOG:       for.exiting_block.2:
 ; PROLOG-NEXT:    %cmp.2 = icmp eq i64 %n, 42
@@ -4139,7 +4027,7 @@ define i32 @test6(ptr nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; PROLOG-NEXT:    %arrayidx.2 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.1
 ; PROLOG-NEXT:    %load.2 = load i32, ptr %arrayidx.2, align 4
 ; PROLOG-NEXT:    %add.2 = add nsw i32 %load.2, %add.1
-; PROLOG-NEXT:    %indvars.iv.next.2 = add i64 %indvars.iv.next.1, 1
+; PROLOG-NEXT:    %indvars.iv.next.2 = add i64 %indvars.iv, 3
 ; PROLOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.3
 ; PROLOG:       for.exiting_block.3:
 ; PROLOG-NEXT:    %cmp.3 = icmp eq i64 %n, 42
@@ -4148,7 +4036,7 @@ define i32 @test6(ptr nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; PROLOG-NEXT:    %arrayidx.3 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.2
 ; PROLOG-NEXT:    %load.3 = load i32, ptr %arrayidx.3, align 4
 ; PROLOG-NEXT:    %add.3 = add nsw i32 %load.3, %add.2
-; PROLOG-NEXT:    %indvars.iv.next.3 = add i64 %indvars.iv.next.2, 1
+; PROLOG-NEXT:    %indvars.iv.next.3 = add i64 %indvars.iv, 4
 ; PROLOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.4
 ; PROLOG:       for.exiting_block.4:
 ; PROLOG-NEXT:    %cmp.4 = icmp eq i64 %n, 42
@@ -4157,7 +4045,7 @@ define i32 @test6(ptr nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; PROLOG-NEXT:    %arrayidx.4 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.3
 ; PROLOG-NEXT:    %load.4 = load i32, ptr %arrayidx.4, align 4
 ; PROLOG-NEXT:    %add.4 = add nsw i32 %load.4, %add.3
-; PROLOG-NEXT:    %indvars.iv.next.4 = add i64 %indvars.iv.next.3, 1
+; PROLOG-NEXT:    %indvars.iv.next.4 = add i64 %indvars.iv, 5
 ; PROLOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.5
 ; PROLOG:       for.exiting_block.5:
 ; PROLOG-NEXT:    %cmp.5 = icmp eq i64 %n, 42
@@ -4166,7 +4054,7 @@ define i32 @test6(ptr nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; PROLOG-NEXT:    %arrayidx.5 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.4
 ; PROLOG-NEXT:    %load.5 = load i32, ptr %arrayidx.5, align 4
 ; PROLOG-NEXT:    %add.5 = add nsw i32 %load.5, %add.4
-; PROLOG-NEXT:    %indvars.iv.next.5 = add i64 %indvars.iv.next.4, 1
+; PROLOG-NEXT:    %indvars.iv.next.5 = add i64 %indvars.iv, 6
 ; PROLOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.6
 ; PROLOG:       for.exiting_block.6:
 ; PROLOG-NEXT:    %cmp.6 = icmp eq i64 %n, 42
@@ -4175,7 +4063,7 @@ define i32 @test6(ptr nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; PROLOG-NEXT:    %arrayidx.6 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.5
 ; PROLOG-NEXT:    %load.6 = load i32, ptr %arrayidx.6, align 4
 ; PROLOG-NEXT:    %add.6 = add nsw i32 %load.6, %add.5
-; PROLOG-NEXT:    %indvars.iv.next.6 = add i64 %indvars.iv.next.5, 1
+; PROLOG-NEXT:    %indvars.iv.next.6 = add i64 %indvars.iv, 7
 ; PROLOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block.7
 ; PROLOG:       for.exiting_block.7:
 ; PROLOG-NEXT:    %cmp.7 = icmp eq i64 %n, 42
@@ -4184,7 +4072,7 @@ define i32 @test6(ptr nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; PROLOG-NEXT:    %arrayidx.7 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next.6
 ; PROLOG-NEXT:    %load.7 = load i32, ptr %arrayidx.7, align 4
 ; PROLOG-NEXT:    %add.7 = add nsw i32 %load.7, %add.6
-; PROLOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv.next.6, 1
+; PROLOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv, 8
 ; PROLOG-NEXT:    %exitcond.7 = icmp eq i64 %indvars.iv.next.7, %n
 ; PROLOG-NEXT:    br i1 %exitcond.7, label %latch_exit.unr-lcssa, label %header
 ; PROLOG:       latch_exit.unr-lcssa:
@@ -4253,7 +4141,7 @@ define i32 @test6(ptr nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; PROLOG-BLOCK-NEXT:    %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next
 ; PROLOG-BLOCK-NEXT:    %load.1 = load i32, ptr %arrayidx.1, align 4
 ; PROLOG-BLOCK-NEXT:    %add.1 = add nsw i32 %load.1, %add
-; PROLOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
+; PROLOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv, 2
 ; PROLOG-BLOCK-NEXT:    %exitcond.1 = icmp eq i64 %indvars.iv.next.1, %n
 ; PROLOG-BLOCK-NEXT:    br i1 %exitcond.1, label %latch_exit.unr-lcssa, label %header, !llvm.loop !10
 ; PROLOG-BLOCK:       latch_exit.unr-lcssa:
@@ -4334,36 +4222,22 @@ define i32 @test7(i32 %arg, i32 %arg1, i32 %arg2) {
 ; EPILOG-NEXT:    %niter = phi i64 [ 0, %preheader.new ], [ %niter.next.7, %latch.7 ]
 ; EPILOG-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch
 ; EPILOG:       latch:
-; EPILOG-NEXT:    %add = add nuw nsw i64 %i6, 1
-; EPILOG-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch.1
 ; EPILOG:       latch.1:
-; EPILOG-NEXT:    %add.1 = add nuw nsw i64 %add, 1
-; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
 ; EPILOG-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch.2
 ; EPILOG:       latch.2:
-; EPILOG-NEXT:    %add.2 = add nuw nsw i64 %add.1, 1
-; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
 ; EPILOG-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch.3
 ; EPILOG:       latch.3:
-; EPILOG-NEXT:    %add.3 = add nuw nsw i64 %add.2, 1
-; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
 ; EPILOG-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch.4
 ; EPILOG:       latch.4:
-; EPILOG-NEXT:    %add.4 = add nuw nsw i64 %add.3, 1
-; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
 ; EPILOG-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch.5
 ; EPILOG:       latch.5:
-; EPILOG-NEXT:    %add.5 = add nuw nsw i64 %add.4, 1
-; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
 ; EPILOG-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch.6
 ; EPILOG:       latch.6:
-; EPILOG-NEXT:    %add.6 = add nuw nsw i64 %add.5, 1
-; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
 ; EPILOG-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch.7
 ; EPILOG:       latch.7:
-; EPILOG-NEXT:    %add.7 = add nuw nsw i64 %add.6, 1
-; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; EPILOG-NEXT:    %add.7 = add nuw nsw i64 %i6, 8
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter, 8
 ; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.next.7, %unroll_iter
 ; EPILOG-NEXT:    br i1 %niter.ncmp.7, label %header, label %latchexit.unr-lcssa.loopexit
 ; EPILOG:       latchexit.unr-lcssa.loopexit:
@@ -4422,12 +4296,10 @@ define i32 @test7(i32 %arg, i32 %arg1, i32 %arg2) {
 ; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ 0, %preheader.new ], [ %niter.next.1, %latch.1 ]
 ; EPILOG-BLOCK-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch
 ; EPILOG-BLOCK:       latch:
-; EPILOG-BLOCK-NEXT:    %add = add nuw nsw i64 %i6, 1
-; EPILOG-BLOCK-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-BLOCK-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch.1
 ; EPILOG-BLOCK:       latch.1:
-; EPILOG-BLOCK-NEXT:    %add.1 = add nuw nsw i64 %add, 1
-; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; EPILOG-BLOCK-NEXT:    %add.1 = add nuw nsw i64 %i6, 2
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.next.1, %unroll_iter
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1, label %header, label %latchexit.unr-lcssa.loopexit, !llvm.loop !11
 ; EPILOG-BLOCK:       latchexit.unr-lcssa.loopexit:
@@ -4490,28 +4362,21 @@ define i32 @test7(i32 %arg, i32 %arg1, i32 %arg2) {
 ; PROLOG-NEXT:    %i6 = phi i64 [ %i6.unr, %preheader.new ], [ %add.7, %latch.7 ]
 ; PROLOG-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch
 ; PROLOG:       latch:
-; PROLOG-NEXT:    %add = add nuw nsw i64 %i6, 1
 ; PROLOG-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch.1
 ; PROLOG:       latch.1:
-; PROLOG-NEXT:    %add.1 = add nuw nsw i64 %add, 1
 ; PROLOG-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch.2
 ; PROLOG:       latch.2:
-; PROLOG-NEXT:    %add.2 = add nuw nsw i64 %add.1, 1
 ; PROLOG-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch.3
 ; PROLOG:       latch.3:
-; PROLOG-NEXT:    %add.3 = add nuw nsw i64 %add.2, 1
 ; PROLOG-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch.4
 ; PROLOG:       latch.4:
-; PROLOG-NEXT:    %add.4 = add nuw nsw i64 %add.3, 1
 ; PROLOG-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch.5
 ; PROLOG:       latch.5:
-; PROLOG-NEXT:    %add.5 = add nuw nsw i64 %add.4, 1
 ; PROLOG-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch.6
 ; PROLOG:       latch.6:
-; PROLOG-NEXT:    %add.6 = add nuw nsw i64 %add.5, 1
 ; PROLOG-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch.7
 ; PROLOG:       latch.7:
-; PROLOG-NEXT:    %add.7 = add nuw nsw i64 %add.6, 1
+; PROLOG-NEXT:    %add.7 = add nuw nsw i64 %i6, 8
 ; PROLOG-NEXT:    %i9.7 = icmp slt i64 %add.7, %sext
 ; PROLOG-NEXT:    br i1 %i9.7, label %header, label %latchexit.unr-lcssa
 ; PROLOG:       latchexit.unr-lcssa:
@@ -4559,10 +4424,9 @@ define i32 @test7(i32 %arg, i32 %arg1, i32 %arg2) {
 ; PROLOG-BLOCK-NEXT:    %i6 = phi i64 [ %i6.unr, %preheader.new ], [ %add.1, %latch.1 ]
 ; PROLOG-BLOCK-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch
 ; PROLOG-BLOCK:       latch:
-; PROLOG-BLOCK-NEXT:    %add = add nuw nsw i64 %i6, 1
 ; PROLOG-BLOCK-NEXT:    br i1 false, label %loopexit1.loopexit, label %latch.1
 ; PROLOG-BLOCK:       latch.1:
-; PROLOG-BLOCK-NEXT:    %add.1 = add nuw nsw i64 %add, 1
+; PROLOG-BLOCK-NEXT:    %add.1 = add nuw nsw i64 %i6, 2
 ; PROLOG-BLOCK-NEXT:    %i9.1 = icmp slt i64 %add.1, %sext
 ; PROLOG-BLOCK-NEXT:    br i1 %i9.1, label %header, label %latchexit.unr-lcssa, !llvm.loop !11
 ; PROLOG-BLOCK:       latchexit.unr-lcssa:
@@ -4633,38 +4497,24 @@ define void @test8() {
 ; EPILOG:       innerH:
 ; EPILOG-NEXT:    %i3 = phi i64 [ %i, %outerloop.new ], [ %i4.7, %latch.7 ]
 ; EPILOG-NEXT:    %niter = phi i64 [ 0, %outerloop.new ], [ %niter.next.7, %latch.7 ]
-; EPILOG-NEXT:    %i4 = add nuw nsw i64 %i3, 1
 ; EPILOG-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch
 ; EPILOG:       latch:
-; EPILOG-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
-; EPILOG-NEXT:    %i4.1 = add nuw nsw i64 %i4, 1
 ; EPILOG-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch.1
 ; EPILOG:       latch.1:
-; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
-; EPILOG-NEXT:    %i4.2 = add nuw nsw i64 %i4.1, 1
 ; EPILOG-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch.2
 ; EPILOG:       latch.2:
-; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
-; EPILOG-NEXT:    %i4.3 = add nuw nsw i64 %i4.2, 1
 ; EPILOG-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch.3
 ; EPILOG:       latch.3:
-; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
-; EPILOG-NEXT:    %i4.4 = add nuw nsw i64 %i4.3, 1
 ; EPILOG-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch.4
 ; EPILOG:       latch.4:
-; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
-; EPILOG-NEXT:    %i4.5 = add nuw nsw i64 %i4.4, 1
 ; EPILOG-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch.5
 ; EPILOG:       latch.5:
-; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
-; EPILOG-NEXT:    %i4.6 = add nuw nsw i64 %i4.5, 1
 ; EPILOG-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch.6
 ; EPILOG:       latch.6:
-; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
-; EPILOG-NEXT:    %i4.7 = add nuw nsw i64 %i4.6, 1
+; EPILOG-NEXT:    %i4.7 = add nuw nsw i64 %i3, 8
 ; EPILOG-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch.7
 ; EPILOG:       latch.7:
-; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter, 8
 ; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.next.7, %unroll_iter
 ; EPILOG-NEXT:    br i1 %niter.ncmp.7, label %innerH, label %exit.unr-lcssa.loopexit
 ; EPILOG:       exit.unr-lcssa.loopexit:
@@ -4705,14 +4555,12 @@ define void @test8() {
 ; EPILOG-BLOCK:       innerH.1:
 ; EPILOG-BLOCK-NEXT:    %i3.1 = phi i64 [ 0, %outerloop.new.1 ], [ %i4.1.1, %latch.1.1 ]
 ; EPILOG-BLOCK-NEXT:    %niter.1 = phi i64 [ 0, %outerloop.new.1 ], [ %niter.next.1.1, %latch.1.1 ]
-; EPILOG-BLOCK-NEXT:    %i4.12 = add nuw nsw i64 %i3.1, 1
-; EPILOG-BLOCK-NEXT:    br i1 false, label %outerloop.loopexit.loopexit.1, label %latch.14
-; EPILOG-BLOCK:       latch.14:
-; EPILOG-BLOCK-NEXT:    %niter.next.13 = add nuw nsw i64 %niter.1, 1
-; EPILOG-BLOCK-NEXT:    %i4.1.1 = add nuw nsw i64 %i4.12, 1
+; EPILOG-BLOCK-NEXT:    br i1 false, label %outerloop.loopexit.loopexit.1, label %latch.12
+; EPILOG-BLOCK:       latch.12:
+; EPILOG-BLOCK-NEXT:    %i4.1.1 = add nuw nsw i64 %i3.1, 2
 ; EPILOG-BLOCK-NEXT:    br i1 false, label %outerloop.loopexit.loopexit.1, label %latch.1.1
 ; EPILOG-BLOCK:       latch.1.1:
-; EPILOG-BLOCK-NEXT:    %niter.next.1.1 = add i64 %niter.next.13, 1
+; EPILOG-BLOCK-NEXT:    %niter.next.1.1 = add i64 %niter.1, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1.1 = icmp ne i64 %niter.next.1.1, 100
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1.1, label %innerH.1, label %exit.unr-lcssa.loopexit.1, !llvm.loop !12
 ; EPILOG-BLOCK:       exit.unr-lcssa.loopexit.1:
@@ -4740,14 +4588,12 @@ define void @test8() {
 ; EPILOG-BLOCK:       innerH:
 ; EPILOG-BLOCK-NEXT:    %i3 = phi i64 [ %i, %outerloop.new ], [ %i4.1, %latch.1 ]
 ; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ 0, %outerloop.new ], [ %niter.next.1, %latch.1 ]
-; EPILOG-BLOCK-NEXT:    %i4 = add nuw nsw i64 %i3, 1
 ; EPILOG-BLOCK-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch
 ; EPILOG-BLOCK:       latch:
-; EPILOG-BLOCK-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
-; EPILOG-BLOCK-NEXT:    %i4.1 = add nuw nsw i64 %i4, 1
+; EPILOG-BLOCK-NEXT:    %i4.1 = add nuw nsw i64 %i3, 2
 ; EPILOG-BLOCK-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch.1
 ; EPILOG-BLOCK:       latch.1:
-; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.next.1, %unroll_iter
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1, label %innerH, label %exit.unr-lcssa.loopexit, !llvm.loop !12
 ; EPILOG-BLOCK:       exit.unr-lcssa.loopexit:
@@ -4805,28 +4651,21 @@ define void @test8() {
 ; PROLOG-NEXT:    br label %innerH
 ; PROLOG:       innerH:
 ; PROLOG-NEXT:    %i3 = phi i64 [ %i3.unr, %outerloop.new ], [ %i4.7, %latch.7 ]
-; PROLOG-NEXT:    %i4 = add nuw nsw i64 %i3, 1
 ; PROLOG-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch
 ; PROLOG:       latch:
-; PROLOG-NEXT:    %i4.1 = add nuw nsw i64 %i4, 1
 ; PROLOG-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch.1
 ; PROLOG:       latch.1:
-; PROLOG-NEXT:    %i4.2 = add nuw nsw i64 %i4.1, 1
 ; PROLOG-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch.2
 ; PROLOG:       latch.2:
-; PROLOG-NEXT:    %i4.3 = add nuw nsw i64 %i4.2, 1
 ; PROLOG-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch.3
 ; PROLOG:       latch.3:
-; PROLOG-NEXT:    %i4.4 = add nuw nsw i64 %i4.3, 1
 ; PROLOG-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch.4
 ; PROLOG:       latch.4:
-; PROLOG-NEXT:    %i4.5 = add nuw nsw i64 %i4.4, 1
 ; PROLOG-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch.5
 ; PROLOG:       latch.5:
-; PROLOG-NEXT:    %i4.6 = add nuw nsw i64 %i4.5, 1
 ; PROLOG-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch.6
 ; PROLOG:       latch.6:
-; PROLOG-NEXT:    %i4.7 = add nuw nsw i64 %i4.6, 1
+; PROLOG-NEXT:    %i4.7 = add nuw nsw i64 %i3, 8
 ; PROLOG-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch.7
 ; PROLOG:       latch.7:
 ; PROLOG-NEXT:    %i6.7 = icmp ult i64 %i4.7, 100
@@ -4858,14 +4697,13 @@ define void @test8() {
 ; PROLOG-BLOCK-NEXT:    br label %innerH.1
 ; PROLOG-BLOCK:       innerH.1:
 ; PROLOG-BLOCK-NEXT:    %i3.1 = phi i64 [ %i3.unr.1, %outerloop.new.1 ], [ %i4.1.1, %latch.1.1 ]
-; PROLOG-BLOCK-NEXT:    %i4.11 = add nuw nsw i64 %i3.1, 1
-; PROLOG-BLOCK-NEXT:    br i1 false, label %outerloop.loopexit.loopexit.1, label %latch.12
-; PROLOG-BLOCK:       latch.12:
-; PROLOG-BLOCK-NEXT:    %i4.1.1 = add nuw nsw i64 %i4.11, 1
+; PROLOG-BLOCK-NEXT:    br i1 false, label %outerloop.loopexit.loopexit.1, label %latch.11
+; PROLOG-BLOCK:       latch.11:
+; PROLOG-BLOCK-NEXT:    %i4.1.1 = add nuw nsw i64 %i3.1, 2
 ; PROLOG-BLOCK-NEXT:    br i1 false, label %outerloop.loopexit.loopexit.1, label %latch.1.1
 ; PROLOG-BLOCK:       latch.1.1:
 ; PROLOG-BLOCK-NEXT:    %i6.1.1 = icmp ult i64 %i4.1.1, 100
-; PROLOG-BLOCK-NEXT:    br i1 %i6.1.1, label %innerH.1, label %exit.unr-lcssa.loopexit3, !llvm.loop !12
+; PROLOG-BLOCK-NEXT:    br i1 %i6.1.1, label %innerH.1, label %exit.unr-lcssa.loopexit2, !llvm.loop !12
 ; PROLOG-BLOCK:       outerloop.loopexit.loopexit.1:
 ; PROLOG-BLOCK-NEXT:    br label %outerloop.loopexit.1
 ; PROLOG-BLOCK:       outerloop.loopexit.1:
@@ -4892,17 +4730,16 @@ define void @test8() {
 ; PROLOG-BLOCK-NEXT:    br label %innerH
 ; PROLOG-BLOCK:       innerH:
 ; PROLOG-BLOCK-NEXT:    %i3 = phi i64 [ %i3.unr, %outerloop.new ], [ %i4.1, %latch.1 ]
-; PROLOG-BLOCK-NEXT:    %i4 = add nuw nsw i64 %i3, 1
 ; PROLOG-BLOCK-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch
 ; PROLOG-BLOCK:       latch:
-; PROLOG-BLOCK-NEXT:    %i4.1 = add nuw nsw i64 %i4, 1
+; PROLOG-BLOCK-NEXT:    %i4.1 = add nuw nsw i64 %i3, 2
 ; PROLOG-BLOCK-NEXT:    br i1 false, label %outerloop.loopexit.loopexit, label %latch.1
 ; PROLOG-BLOCK:       latch.1:
 ; PROLOG-BLOCK-NEXT:    %i6.1 = icmp ult i64 %i4.1, 100
 ; PROLOG-BLOCK-NEXT:    br i1 %i6.1, label %innerH, label %exit.unr-lcssa.loopexit, !llvm.loop !12
 ; PROLOG-BLOCK:       exit.unr-lcssa.loopexit:
 ; PROLOG-BLOCK-NEXT:    br label %exit.unr-lcssa
-; PROLOG-BLOCK:       exit.unr-lcssa.loopexit3:
+; PROLOG-BLOCK:       exit.unr-lcssa.loopexit2:
 ; PROLOG-BLOCK-NEXT:    br label %exit.unr-lcssa
 ; PROLOG-BLOCK:       exit.unr-lcssa:
 ; PROLOG-BLOCK-NEXT:    br label %exit
@@ -4970,36 +4807,22 @@ define ptr addrspace(1) @test9(ptr nocapture readonly %arg, i32 %n) {
 ; EPILOG-NEXT:    %i9 = call ptr addrspace(1) @foo(i32 %trip.lcssa)
 ; EPILOG-NEXT:    ret ptr addrspace(1) %i9
 ; EPILOG:       latch:
-; EPILOG-NEXT:    %iv.next = add nuw nsw i64 %phi, 1
-; EPILOG-NEXT:    %niter.next = add nuw nsw i32 %niter, 1
 ; EPILOG-NEXT:    br i1 true, label %latch.1, label %innerexit.loopexit
 ; EPILOG:       latch.1:
-; EPILOG-NEXT:    %iv.next.1 = add nuw nsw i64 %iv.next, 1
-; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i32 %niter.next, 1
 ; EPILOG-NEXT:    br i1 true, label %latch.2, label %innerexit.loopexit
 ; EPILOG:       latch.2:
-; EPILOG-NEXT:    %iv.next.2 = add nuw nsw i64 %iv.next.1, 1
-; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i32 %niter.next.1, 1
 ; EPILOG-NEXT:    br i1 true, label %latch.3, label %innerexit.loopexit
 ; EPILOG:       latch.3:
-; EPILOG-NEXT:    %iv.next.3 = add nuw nsw i64 %iv.next.2, 1
-; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i32 %niter.next.2, 1
 ; EPILOG-NEXT:    br i1 true, label %latch.4, label %innerexit.loopexit
 ; EPILOG:       latch.4:
-; EPILOG-NEXT:    %iv.next.4 = add nuw nsw i64 %iv.next.3, 1
-; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i32 %niter.next.3, 1
 ; EPILOG-NEXT:    br i1 true, label %latch.5, label %innerexit.loopexit
 ; EPILOG:       latch.5:
-; EPILOG-NEXT:    %iv.next.5 = add nuw nsw i64 %iv.next.4, 1
-; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i32 %niter.next.4, 1
 ; EPILOG-NEXT:    br i1 true, label %latch.6, label %innerexit.loopexit
 ; EPILOG:       latch.6:
-; EPILOG-NEXT:    %iv.next.6 = add nuw nsw i64 %iv.next.5, 1
-; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i32 %niter.next.5, 1
 ; EPILOG-NEXT:    br i1 true, label %latch.7, label %innerexit.loopexit
 ; EPILOG:       latch.7:
-; EPILOG-NEXT:    %iv.next.7 = add nuw nsw i64 %iv.next.6, 1
-; EPILOG-NEXT:    %niter.next.7 = add i32 %niter.next.6, 1
+; EPILOG-NEXT:    %iv.next.7 = add nuw nsw i64 %phi, 8
+; EPILOG-NEXT:    %niter.next.7 = add i32 %niter, 8
 ; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i32 %niter.next.7, %unroll_iter
 ; EPILOG-NEXT:    br i1 %niter.ncmp.7, label %header, label %outerLatch.loopexit.unr-lcssa.loopexit
 ; EPILOG:       outerLatch.loopexit.unr-lcssa.loopexit:
@@ -5054,11 +4877,11 @@ define ptr addrspace(1) @test9(ptr nocapture readonly %arg, i32 %n) {
 ; EPILOG-BLOCK:       innerexit.loopexit.loopexit:
 ; EPILOG-BLOCK-NEXT:    %trip.lcssa.ph.ph = phi i32 [ %trip, %latch ], [ %trip, %header ]
 ; EPILOG-BLOCK-NEXT:    br label %innerexit.loopexit
-; EPILOG-BLOCK:       innerexit.loopexit.loopexit6:
-; EPILOG-BLOCK-NEXT:    %trip.lcssa.ph.ph7 = phi i32 [ %trip.1, %latch.15 ], [ %trip.1, %header.1 ]
+; EPILOG-BLOCK:       innerexit.loopexit.loopexit4:
+; EPILOG-BLOCK-NEXT:    %trip.lcssa.ph.ph5 = phi i32 [ %trip.1, %latch.13 ], [ %trip.1, %header.1 ]
 ; EPILOG-BLOCK-NEXT:    br label %innerexit.loopexit
 ; EPILOG-BLOCK:       innerexit.loopexit:
-; EPILOG-BLOCK-NEXT:    %trip.lcssa.ph = phi i32 [ %trip.lcssa.ph.ph, %innerexit.loopexit.loopexit ], [ %trip.lcssa.ph.ph7, %innerexit.loopexit.loopexit6 ]
+; EPILOG-BLOCK-NEXT:    %trip.lcssa.ph = phi i32 [ %trip.lcssa.ph.ph, %innerexit.loopexit.loopexit ], [ %trip.lcssa.ph.ph5, %innerexit.loopexit.loopexit4 ]
 ; EPILOG-BLOCK-NEXT:    br label %innerexit
 ; EPILOG-BLOCK:       innerexit.loopexit1:
 ; EPILOG-BLOCK-NEXT:    %trip.lcssa.ph2 = phi i32 [ %trip, %header.epil ], [ %trip.1, %header.epil.1 ]
@@ -5068,12 +4891,10 @@ define ptr addrspace(1) @test9(ptr nocapture readonly %arg, i32 %n) {
 ; EPILOG-BLOCK-NEXT:    %i9 = call ptr addrspace(1) @foo(i32 %trip.lcssa)
 ; EPILOG-BLOCK-NEXT:    ret ptr addrspace(1) %i9
 ; EPILOG-BLOCK:       latch:
-; EPILOG-BLOCK-NEXT:    %iv.next = add nuw nsw i64 %phi, 1
-; EPILOG-BLOCK-NEXT:    %niter.next = add nuw nsw i32 %niter, 1
 ; EPILOG-BLOCK-NEXT:    br i1 true, label %latch.1, label %innerexit.loopexit.loopexit
 ; EPILOG-BLOCK:       latch.1:
-; EPILOG-BLOCK-NEXT:    %iv.next.1 = add nuw nsw i64 %iv.next, 1
-; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i32 %niter.next, 1
+; EPILOG-BLOCK-NEXT:    %iv.next.1 = add nuw nsw i64 %phi, 2
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i32 %niter, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i32 %niter.next.1, %unroll_iter
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1, label %header, label %outerLatch.loopexit.unr-lcssa.loopexit, !llvm.loop !14
 ; EPILOG-BLOCK:       outerLatch.loopexit.unr-lcssa.loopexit:
@@ -5105,14 +4926,12 @@ define ptr addrspace(1) @test9(ptr nocapture readonly %arg, i32 %n) {
 ; EPILOG-BLOCK:       header.1:
 ; EPILOG-BLOCK-NEXT:    %phi.1 = phi i64 [ 0, %preheader.new.1 ], [ %iv.next.1.1, %latch.1.1 ]
 ; EPILOG-BLOCK-NEXT:    %niter.1 = phi i32 [ 0, %preheader.new.1 ], [ %niter.next.1.1, %latch.1.1 ]
-; EPILOG-BLOCK-NEXT:    br i1 true, label %latch.15, label %innerexit.loopexit.loopexit6
-; EPILOG-BLOCK:       latch.15:
-; EPILOG-BLOCK-NEXT:    %iv.next.13 = add nuw nsw i64 %phi.1, 1
-; EPILOG-BLOCK-NEXT:    %niter.next.14 = add nuw nsw i32 %niter.1, 1
-; EPILOG-BLOCK-NEXT:    br i1 true, label %latch.1.1, label %innerexit.loopexit.loopexit6
+; EPILOG-BLOCK-NEXT:    br i1 true, label %latch.13, label %innerexit.loopexit.loopexit4
+; EPILOG-BLOCK:       latch.13:
+; EPILOG-BLOCK-NEXT:    br i1 true, label %latch.1.1, label %innerexit.loopexit.loopexit4
 ; EPILOG-BLOCK:       latch.1.1:
-; EPILOG-BLOCK-NEXT:    %iv.next.1.1 = add nuw nsw i64 %iv.next.13, 1
-; EPILOG-BLOCK-NEXT:    %niter.next.1.1 = add i32 %niter.next.14, 1
+; EPILOG-BLOCK-NEXT:    %iv.next.1.1 = add nuw nsw i64 %phi.1, 2
+; EPILOG-BLOCK-NEXT:    %niter.next.1.1 = add i32 %niter.1, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1.1 = icmp ne i32 %niter.next.1.1, %unroll_iter.1
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1.1, label %header.1, label %outerLatch.loopexit.unr-lcssa.loopexit.1, !llvm.loop !14
 ; EPILOG-BLOCK:       outerLatch.loopexit.unr-lcssa.loopexit.1:
@@ -5183,31 +5002,25 @@ define ptr addrspace(1) @test9(ptr nocapture readonly %arg, i32 %n) {
 ; PROLOG-NEXT:    %i9 = call ptr addrspace(1) @foo(i32 %trip.lcssa)
 ; PROLOG-NEXT:    ret ptr addrspace(1) %i9
 ; PROLOG:       latch:
-; PROLOG-NEXT:    %iv.next = add nuw nsw i64 %phi, 1
 ; PROLOG-NEXT:    br i1 true, label %latch.1, label %innerexit.loopexit
 ; PROLOG:       latch.1:
-; PROLOG-NEXT:    %iv.next.1 = add nuw nsw i64 %iv.next, 1
 ; PROLOG-NEXT:    br i1 true, label %latch.2, label %innerexit.loopexit
 ; PROLOG:       latch.2:
-; PROLOG-NEXT:    %iv.next.2 = add nuw nsw i64 %iv.next.1, 1
 ; PROLOG-NEXT:    br i1 true, label %latch.3, label %innerexit.loopexit
 ; PROLOG:       latch.3:
-; PROLOG-NEXT:    %iv.next.3 = add nuw nsw i64 %iv.next.2, 1
 ; PROLOG-NEXT:    br i1 true, label %latch.4, label %innerexit.loopexit
 ; PROLOG:       latch.4:
-; PROLOG-NEXT:    %iv.next.4 = add nuw nsw i64 %iv.next.3, 1
 ; PROLOG-NEXT:    br i1 true, label %latch.5, label %innerexit.loopexit
 ; PROLOG:       latch.5:
-; PROLOG-NEXT:    %iv.next.5 = add nuw nsw i64 %iv.next.4, 1
 ; PROLOG-NEXT:    br i1 true, label %latch.6, label %innerexit.loopexit
 ; PROLOG:       latch.6:
-; PROLOG-NEXT:    %iv.next.6 = add nuw nsw i64 %iv.next.5, 1
+; PROLOG-NEXT:    %iv.next.6 = add nuw nsw i64 %phi, 7
 ; PROLOG-NEXT:    %i7.7 = trunc i64 %iv.next.6 to i32
 ; PROLOG-NEXT:    br i1 true, label %latch.7, label %innerexit.loopexit
 ; PROLOG:       latch.7:
 ; PROLOG-NEXT:    %i11.7 = add nsw i32 %i7.7, 1
 ; PROLOG-NEXT:    %innercnd.7 = icmp slt i32 %i11.7, %trip
-; PROLOG-NEXT:    %iv.next.7 = add nuw nsw i64 %iv.next.6, 1
+; PROLOG-NEXT:    %iv.next.7 = add nuw nsw i64 %phi, 8
 ; PROLOG-NEXT:    br i1 %innercnd.7, label %header, label %outerLatch.loopexit.unr-lcssa
 ; PROLOG:       outerLatch.loopexit.unr-lcssa:
 ; PROLOG-NEXT:    br label %outerLatch.loopexit
@@ -5268,7 +5081,7 @@ define ptr addrspace(1) @test9(ptr nocapture readonly %arg, i32 %n) {
 ; PROLOG-BLOCK:       latch.1:
 ; PROLOG-BLOCK-NEXT:    %i11.1 = add nsw i32 %i7.1, 1
 ; PROLOG-BLOCK-NEXT:    %innercnd.1 = icmp slt i32 %i11.1, %trip
-; PROLOG-BLOCK-NEXT:    %iv.next.1 = add nuw nsw i64 %iv.next, 1
+; PROLOG-BLOCK-NEXT:    %iv.next.1 = add nuw nsw i64 %phi, 2
 ; PROLOG-BLOCK-NEXT:    br i1 %innercnd.1, label %header, label %outerLatch.loopexit.unr-lcssa, !llvm.loop !14
 ; PROLOG-BLOCK:       outerLatch.loopexit.unr-lcssa:
 ; PROLOG-BLOCK-NEXT:    br label %outerLatch.loopexit
@@ -5306,7 +5119,7 @@ define ptr addrspace(1) @test9(ptr nocapture readonly %arg, i32 %n) {
 ; PROLOG-BLOCK:       latch.1.1:
 ; PROLOG-BLOCK-NEXT:    %i11.1.1 = add nsw i32 %i7.1.1, 1
 ; PROLOG-BLOCK-NEXT:    %innercnd.1.1 = icmp slt i32 %i11.1.1, %trip.1
-; PROLOG-BLOCK-NEXT:    %iv.next.1.1 = add nuw nsw i64 %iv.next.13, 1
+; PROLOG-BLOCK-NEXT:    %iv.next.1.1 = add nuw nsw i64 %phi.1, 2
 ; PROLOG-BLOCK-NEXT:    br i1 %innercnd.1.1, label %header.1, label %outerLatch.loopexit.unr-lcssa.1, !llvm.loop !14
 ; PROLOG-BLOCK:       outerLatch.loopexit.unr-lcssa.1:
 ; PROLOG-BLOCK-NEXT:    br label %outerLatch.loopexit.1
@@ -5370,49 +5183,42 @@ define void @test10(i64 %trip, i64 %trip2) {
 ; EPILOG-NEXT:    br i1 %cmp_early, label %loop_latch, label %exit1.loopexit
 ; EPILOG:       loop_latch:
 ; EPILOG-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
-; EPILOG-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.1 = icmp ne i64 %iv_next, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.1, label %loop_latch.1, label %exit1.loopexit
 ; EPILOG:       loop_latch.1:
-; EPILOG-NEXT:    %iv_next.1 = add nuw nsw i64 %iv_next, 1
-; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
+; EPILOG-NEXT:    %iv_next.1 = add nuw nsw i64 %iv, 2
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.2 = icmp ne i64 %iv_next.1, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.2, label %loop_latch.2, label %exit1.loopexit
 ; EPILOG:       loop_latch.2:
-; EPILOG-NEXT:    %iv_next.2 = add nuw nsw i64 %iv_next.1, 1
-; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
+; EPILOG-NEXT:    %iv_next.2 = add nuw nsw i64 %iv, 3
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.3 = icmp ne i64 %iv_next.2, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.3, label %loop_latch.3, label %exit1.loopexit
 ; EPILOG:       loop_latch.3:
-; EPILOG-NEXT:    %iv_next.3 = add nuw nsw i64 %iv_next.2, 1
-; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
+; EPILOG-NEXT:    %iv_next.3 = add nuw nsw i64 %iv, 4
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.4 = icmp ne i64 %iv_next.3, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.4, label %loop_latch.4, label %exit1.loopexit
 ; EPILOG:       loop_latch.4:
-; EPILOG-NEXT:    %iv_next.4 = add nuw nsw i64 %iv_next.3, 1
-; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
+; EPILOG-NEXT:    %iv_next.4 = add nuw nsw i64 %iv, 5
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.5 = icmp ne i64 %iv_next.4, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.5, label %loop_latch.5, label %exit1.loopexit
 ; EPILOG:       loop_latch.5:
-; EPILOG-NEXT:    %iv_next.5 = add nuw nsw i64 %iv_next.4, 1
-; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
+; EPILOG-NEXT:    %iv_next.5 = add nuw nsw i64 %iv, 6
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.6 = icmp ne i64 %iv_next.5, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.6, label %loop_latch.6, label %exit1.loopexit
 ; EPILOG:       loop_latch.6:
-; EPILOG-NEXT:    %iv_next.6 = add nuw nsw i64 %iv_next.5, 1
-; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
+; EPILOG-NEXT:    %iv_next.6 = add nuw nsw i64 %iv, 7
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.7 = icmp ne i64 %iv_next.6, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.7, label %loop_latch.7, label %exit1.loopexit
 ; EPILOG:       loop_latch.7:
-; EPILOG-NEXT:    %iv_next.7 = add i64 %iv_next.6, 1
-; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; EPILOG-NEXT:    %iv_next.7 = add i64 %iv, 8
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter, 8
 ; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.next.7, %unroll_iter
 ; EPILOG-NEXT:    br i1 %niter.ncmp.7, label %loop_header, label %exit2.unr-lcssa.loopexit
 ; EPILOG:       exit1.loopexit:
@@ -5465,13 +5271,12 @@ define void @test10(i64 %trip, i64 %trip2) {
 ; EPILOG-BLOCK-NEXT:    br i1 %cmp_early, label %loop_latch, label %exit1.loopexit
 ; EPILOG-BLOCK:       loop_latch:
 ; EPILOG-BLOCK-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
-; EPILOG-BLOCK-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-BLOCK-NEXT:    call void @bar()
 ; EPILOG-BLOCK-NEXT:    %cmp_early.1 = icmp ne i64 %iv_next, %trip2
 ; EPILOG-BLOCK-NEXT:    br i1 %cmp_early.1, label %loop_latch.1, label %exit1.loopexit
 ; EPILOG-BLOCK:       loop_latch.1:
-; EPILOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv_next, 1
-; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; EPILOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv, 2
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.next.1, %unroll_iter
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1, label %loop_header, label %exit2.unr-lcssa.loopexit, !llvm.loop !16
 ; EPILOG-BLOCK:       exit1.loopexit:
@@ -5537,37 +5342,37 @@ define void @test10(i64 %trip, i64 %trip2) {
 ; PROLOG-NEXT:    %cmp_early.1 = icmp ne i64 %iv_next, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.1, label %loop_latch.1, label %exit1.loopexit
 ; PROLOG:       loop_latch.1:
-; PROLOG-NEXT:    %iv_next.1 = add i64 %iv_next, 1
+; PROLOG-NEXT:    %iv_next.1 = add i64 %iv, 2
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.2 = icmp ne i64 %iv_next.1, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.2, label %loop_latch.2, label %exit1.loopexit
 ; PROLOG:       loop_latch.2:
-; PROLOG-NEXT:    %iv_next.2 = add i64 %iv_next.1, 1
+; PROLOG-NEXT:    %iv_next.2 = add i64 %iv, 3
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.3 = icmp ne i64 %iv_next.2, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.3, label %loop_latch.3, label %exit1.loopexit
 ; PROLOG:       loop_latch.3:
-; PROLOG-NEXT:    %iv_next.3 = add i64 %iv_next.2, 1
+; PROLOG-NEXT:    %iv_next.3 = add i64 %iv, 4
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.4 = icmp ne i64 %iv_next.3, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.4, label %loop_latch.4, label %exit1.loopexit
 ; PROLOG:       loop_latch.4:
-; PROLOG-NEXT:    %iv_next.4 = add i64 %iv_next.3, 1
+; PROLOG-NEXT:    %iv_next.4 = add i64 %iv, 5
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.5 = icmp ne i64 %iv_next.4, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.5, label %loop_latch.5, label %exit1.loopexit
 ; PROLOG:       loop_latch.5:
-; PROLOG-NEXT:    %iv_next.5 = add i64 %iv_next.4, 1
+; PROLOG-NEXT:    %iv_next.5 = add i64 %iv, 6
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.6 = icmp ne i64 %iv_next.5, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.6, label %loop_latch.6, label %exit1.loopexit
 ; PROLOG:       loop_latch.6:
-; PROLOG-NEXT:    %iv_next.6 = add i64 %iv_next.5, 1
+; PROLOG-NEXT:    %iv_next.6 = add i64 %iv, 7
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.7 = icmp ne i64 %iv_next.6, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.7, label %loop_latch.7, label %exit1.loopexit
 ; PROLOG:       loop_latch.7:
-; PROLOG-NEXT:    %iv_next.7 = add i64 %iv_next.6, 1
+; PROLOG-NEXT:    %iv_next.7 = add i64 %iv, 8
 ; PROLOG-NEXT:    %cmp.7 = icmp ne i64 %iv_next.7, %trip
 ; PROLOG-NEXT:    br i1 %cmp.7, label %loop_header, label %exit2.unr-lcssa
 ; PROLOG:       exit1.loopexit:
@@ -5613,7 +5418,7 @@ define void @test10(i64 %trip, i64 %trip2) {
 ; PROLOG-BLOCK-NEXT:    %cmp_early.1 = icmp ne i64 %iv_next, %trip2
 ; PROLOG-BLOCK-NEXT:    br i1 %cmp_early.1, label %loop_latch.1, label %exit1.loopexit
 ; PROLOG-BLOCK:       loop_latch.1:
-; PROLOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv_next, 1
+; PROLOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv, 2
 ; PROLOG-BLOCK-NEXT:    %cmp.1 = icmp ne i64 %iv_next.1, %trip
 ; PROLOG-BLOCK-NEXT:    br i1 %cmp.1, label %loop_header, label %exit2.unr-lcssa, !llvm.loop !16
 ; PROLOG-BLOCK:       exit1.loopexit:
@@ -5665,43 +5470,29 @@ define void @test11(i64 %trip, i1 %cond) {
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch, label %exit1.loopexit
 ; EPILOG:       loop_latch:
-; EPILOG-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
-; EPILOG-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.1, label %exit1.loopexit
 ; EPILOG:       loop_latch.1:
-; EPILOG-NEXT:    %iv_next.1 = add nuw nsw i64 %iv_next, 1
-; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.2, label %exit1.loopexit
 ; EPILOG:       loop_latch.2:
-; EPILOG-NEXT:    %iv_next.2 = add nuw nsw i64 %iv_next.1, 1
-; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.3, label %exit1.loopexit
 ; EPILOG:       loop_latch.3:
-; EPILOG-NEXT:    %iv_next.3 = add nuw nsw i64 %iv_next.2, 1
-; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.4, label %exit1.loopexit
 ; EPILOG:       loop_latch.4:
-; EPILOG-NEXT:    %iv_next.4 = add nuw nsw i64 %iv_next.3, 1
-; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.5, label %exit1.loopexit
 ; EPILOG:       loop_latch.5:
-; EPILOG-NEXT:    %iv_next.5 = add nuw nsw i64 %iv_next.4, 1
-; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.6, label %exit1.loopexit
 ; EPILOG:       loop_latch.6:
-; EPILOG-NEXT:    %iv_next.6 = add nuw nsw i64 %iv_next.5, 1
-; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.7, label %exit1.loopexit
 ; EPILOG:       loop_latch.7:
-; EPILOG-NEXT:    %iv_next.7 = add i64 %iv_next.6, 1
-; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; EPILOG-NEXT:    %iv_next.7 = add i64 %iv, 8
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter, 8
 ; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.next.7, %unroll_iter
 ; EPILOG-NEXT:    br i1 %niter.ncmp.7, label %loop_header, label %exit2.unr-lcssa.loopexit
 ; EPILOG:       exit1.loopexit:
@@ -5751,13 +5542,11 @@ define void @test11(i64 %trip, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    call void @bar()
 ; EPILOG-BLOCK-NEXT:    br i1 %cond, label %loop_latch, label %exit1.loopexit
 ; EPILOG-BLOCK:       loop_latch:
-; EPILOG-BLOCK-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
-; EPILOG-BLOCK-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-BLOCK-NEXT:    call void @bar()
 ; EPILOG-BLOCK-NEXT:    br i1 %cond, label %loop_latch.1, label %exit1.loopexit
 ; EPILOG-BLOCK:       loop_latch.1:
-; EPILOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv_next, 1
-; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; EPILOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv, 2
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.next.1, %unroll_iter
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1, label %loop_header, label %exit2.unr-lcssa.loopexit, !llvm.loop !17
 ; EPILOG-BLOCK:       exit1.loopexit:
@@ -5813,35 +5602,28 @@ define void @test11(i64 %trip, i1 %cond) {
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch, label %exit1.loopexit
 ; PROLOG:       loop_latch:
-; PROLOG-NEXT:    %iv_next = add i64 %iv, 1
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.1, label %exit1.loopexit
 ; PROLOG:       loop_latch.1:
-; PROLOG-NEXT:    %iv_next.1 = add i64 %iv_next, 1
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.2, label %exit1.loopexit
 ; PROLOG:       loop_latch.2:
-; PROLOG-NEXT:    %iv_next.2 = add i64 %iv_next.1, 1
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.3, label %exit1.loopexit
 ; PROLOG:       loop_latch.3:
-; PROLOG-NEXT:    %iv_next.3 = add i64 %iv_next.2, 1
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.4, label %exit1.loopexit
 ; PROLOG:       loop_latch.4:
-; PROLOG-NEXT:    %iv_next.4 = add i64 %iv_next.3, 1
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.5, label %exit1.loopexit
 ; PROLOG:       loop_latch.5:
-; PROLOG-NEXT:    %iv_next.5 = add i64 %iv_next.4, 1
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.6, label %exit1.loopexit
 ; PROLOG:       loop_latch.6:
-; PROLOG-NEXT:    %iv_next.6 = add i64 %iv_next.5, 1
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.7, label %exit1.loopexit
 ; PROLOG:       loop_latch.7:
-; PROLOG-NEXT:    %iv_next.7 = add i64 %iv_next.6, 1
+; PROLOG-NEXT:    %iv_next.7 = add i64 %iv, 8
 ; PROLOG-NEXT:    %cmp.7 = icmp ne i64 %iv_next.7, %trip
 ; PROLOG-NEXT:    br i1 %cmp.7, label %loop_header, label %exit2.unr-lcssa
 ; PROLOG:       exit1.loopexit:
@@ -5880,11 +5662,10 @@ define void @test11(i64 %trip, i1 %cond) {
 ; PROLOG-BLOCK-NEXT:    call void @bar()
 ; PROLOG-BLOCK-NEXT:    br i1 %cond, label %loop_latch, label %exit1.loopexit
 ; PROLOG-BLOCK:       loop_latch:
-; PROLOG-BLOCK-NEXT:    %iv_next = add i64 %iv, 1
 ; PROLOG-BLOCK-NEXT:    call void @bar()
 ; PROLOG-BLOCK-NEXT:    br i1 %cond, label %loop_latch.1, label %exit1.loopexit
 ; PROLOG-BLOCK:       loop_latch.1:
-; PROLOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv_next, 1
+; PROLOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv, 2
 ; PROLOG-BLOCK-NEXT:    %cmp.1 = icmp ne i64 %iv_next.1, %trip
 ; PROLOG-BLOCK-NEXT:    br i1 %cmp.1, label %loop_header, label %exit2.unr-lcssa, !llvm.loop !17
 ; PROLOG-BLOCK:       exit1.loopexit:
@@ -5939,63 +5720,56 @@ define void @test12(i64 %trip, i64 %trip2, i1 %cond) {
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch:
 ; EPILOG-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
-; EPILOG-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.1 = icmp ne i64 %iv_next, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.1, label %loop_exiting_bb2.1, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_exiting_bb2.1:
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.1, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.1:
-; EPILOG-NEXT:    %iv_next.1 = add nuw nsw i64 %iv_next, 1
-; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
+; EPILOG-NEXT:    %iv_next.1 = add nuw nsw i64 %iv, 2
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.2 = icmp ne i64 %iv_next.1, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.2, label %loop_exiting_bb2.2, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_exiting_bb2.2:
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.2, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.2:
-; EPILOG-NEXT:    %iv_next.2 = add nuw nsw i64 %iv_next.1, 1
-; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
+; EPILOG-NEXT:    %iv_next.2 = add nuw nsw i64 %iv, 3
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.3 = icmp ne i64 %iv_next.2, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.3, label %loop_exiting_bb2.3, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_exiting_bb2.3:
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.3, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.3:
-; EPILOG-NEXT:    %iv_next.3 = add nuw nsw i64 %iv_next.2, 1
-; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
+; EPILOG-NEXT:    %iv_next.3 = add nuw nsw i64 %iv, 4
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.4 = icmp ne i64 %iv_next.3, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.4, label %loop_exiting_bb2.4, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_exiting_bb2.4:
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.4, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.4:
-; EPILOG-NEXT:    %iv_next.4 = add nuw nsw i64 %iv_next.3, 1
-; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
+; EPILOG-NEXT:    %iv_next.4 = add nuw nsw i64 %iv, 5
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.5 = icmp ne i64 %iv_next.4, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.5, label %loop_exiting_bb2.5, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_exiting_bb2.5:
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.5, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.5:
-; EPILOG-NEXT:    %iv_next.5 = add nuw nsw i64 %iv_next.4, 1
-; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
+; EPILOG-NEXT:    %iv_next.5 = add nuw nsw i64 %iv, 6
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.6 = icmp ne i64 %iv_next.5, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.6, label %loop_exiting_bb2.6, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_exiting_bb2.6:
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.6, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.6:
-; EPILOG-NEXT:    %iv_next.6 = add nuw nsw i64 %iv_next.5, 1
-; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
+; EPILOG-NEXT:    %iv_next.6 = add nuw nsw i64 %iv, 7
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.7 = icmp ne i64 %iv_next.6, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.7, label %loop_exiting_bb2.7, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_exiting_bb2.7:
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.7, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.7:
-; EPILOG-NEXT:    %iv_next.7 = add i64 %iv_next.6, 1
-; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; EPILOG-NEXT:    %iv_next.7 = add i64 %iv, 8
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter, 8
 ; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.next.7, %unroll_iter
 ; EPILOG-NEXT:    br i1 %niter.ncmp.7, label %loop_header, label %exit1.unr-lcssa.loopexit
 ; EPILOG:       exit1.unr-lcssa.loopexit:
@@ -6050,15 +5824,14 @@ define void @test12(i64 %trip, i64 %trip2, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    br i1 %cond, label %loop_latch, label %exit1.epilog-lcssa.loopexit
 ; EPILOG-BLOCK:       loop_latch:
 ; EPILOG-BLOCK-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
-; EPILOG-BLOCK-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-BLOCK-NEXT:    call void @bar()
 ; EPILOG-BLOCK-NEXT:    %cmp_early.1 = icmp ne i64 %iv_next, %trip2
 ; EPILOG-BLOCK-NEXT:    br i1 %cmp_early.1, label %loop_exiting_bb2.1, label %exit1.epilog-lcssa.loopexit
 ; EPILOG-BLOCK:       loop_exiting_bb2.1:
 ; EPILOG-BLOCK-NEXT:    br i1 %cond, label %loop_latch.1, label %exit1.epilog-lcssa.loopexit
 ; EPILOG-BLOCK:       loop_latch.1:
-; EPILOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv_next, 1
-; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; EPILOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv, 2
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.next.1, %unroll_iter
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1, label %loop_header, label %exit1.unr-lcssa.loopexit, !llvm.loop !18
 ; EPILOG-BLOCK:       exit1.unr-lcssa.loopexit:
@@ -6132,49 +5905,49 @@ define void @test12(i64 %trip, i64 %trip2, i1 %cond) {
 ; PROLOG:       loop_exiting_bb2.1:
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.1, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.1:
-; PROLOG-NEXT:    %iv_next.1 = add i64 %iv_next, 1
+; PROLOG-NEXT:    %iv_next.1 = add i64 %iv, 2
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.2 = icmp ne i64 %iv_next.1, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.2, label %loop_exiting_bb2.2, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_exiting_bb2.2:
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.2, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.2:
-; PROLOG-NEXT:    %iv_next.2 = add i64 %iv_next.1, 1
+; PROLOG-NEXT:    %iv_next.2 = add i64 %iv, 3
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.3 = icmp ne i64 %iv_next.2, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.3, label %loop_exiting_bb2.3, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_exiting_bb2.3:
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.3, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.3:
-; PROLOG-NEXT:    %iv_next.3 = add i64 %iv_next.2, 1
+; PROLOG-NEXT:    %iv_next.3 = add i64 %iv, 4
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.4 = icmp ne i64 %iv_next.3, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.4, label %loop_exiting_bb2.4, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_exiting_bb2.4:
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.4, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.4:
-; PROLOG-NEXT:    %iv_next.4 = add i64 %iv_next.3, 1
+; PROLOG-NEXT:    %iv_next.4 = add i64 %iv, 5
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.5 = icmp ne i64 %iv_next.4, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.5, label %loop_exiting_bb2.5, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_exiting_bb2.5:
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.5, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.5:
-; PROLOG-NEXT:    %iv_next.5 = add i64 %iv_next.4, 1
+; PROLOG-NEXT:    %iv_next.5 = add i64 %iv, 6
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.6 = icmp ne i64 %iv_next.5, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.6, label %loop_exiting_bb2.6, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_exiting_bb2.6:
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.6, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.6:
-; PROLOG-NEXT:    %iv_next.6 = add i64 %iv_next.5, 1
+; PROLOG-NEXT:    %iv_next.6 = add i64 %iv, 7
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.7 = icmp ne i64 %iv_next.6, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.7, label %loop_exiting_bb2.7, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_exiting_bb2.7:
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.7, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.7:
-; PROLOG-NEXT:    %iv_next.7 = add i64 %iv_next.6, 1
+; PROLOG-NEXT:    %iv_next.7 = add i64 %iv, 8
 ; PROLOG-NEXT:    %cmp.7 = icmp ne i64 %iv_next.7, %trip
 ; PROLOG-NEXT:    br i1 %cmp.7, label %loop_header, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       exit1.unr-lcssa.loopexit:
@@ -6224,7 +5997,7 @@ define void @test12(i64 %trip, i64 %trip2, i1 %cond) {
 ; PROLOG-BLOCK:       loop_exiting_bb2.1:
 ; PROLOG-BLOCK-NEXT:    br i1 %cond, label %loop_latch.1, label %exit1.unr-lcssa.loopexit
 ; PROLOG-BLOCK:       loop_latch.1:
-; PROLOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv_next, 1
+; PROLOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv, 2
 ; PROLOG-BLOCK-NEXT:    %cmp.1 = icmp ne i64 %iv_next.1, %trip
 ; PROLOG-BLOCK-NEXT:    br i1 %cmp.1, label %loop_header, label %exit1.unr-lcssa.loopexit, !llvm.loop !18
 ; PROLOG-BLOCK:       exit1.unr-lcssa.loopexit:
@@ -6280,7 +6053,6 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; EPILOG-NEXT:    br i1 %unknown, label %loop_latch, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch:
 ; EPILOG-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
-; EPILOG-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.1 = icmp ne i64 %iv_next, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.1, label %loop_exiting_bb2.1, label %exit1.epilog-lcssa.loopexit
@@ -6288,8 +6060,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; EPILOG-NEXT:    %unknown.1 = call i1 @unknown_cond()
 ; EPILOG-NEXT:    br i1 %unknown.1, label %loop_latch.1, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.1:
-; EPILOG-NEXT:    %iv_next.1 = add nuw nsw i64 %iv_next, 1
-; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
+; EPILOG-NEXT:    %iv_next.1 = add nuw nsw i64 %iv, 2
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.2 = icmp ne i64 %iv_next.1, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.2, label %loop_exiting_bb2.2, label %exit1.epilog-lcssa.loopexit
@@ -6297,8 +6068,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; EPILOG-NEXT:    %unknown.2 = call i1 @unknown_cond()
 ; EPILOG-NEXT:    br i1 %unknown.2, label %loop_latch.2, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.2:
-; EPILOG-NEXT:    %iv_next.2 = add nuw nsw i64 %iv_next.1, 1
-; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
+; EPILOG-NEXT:    %iv_next.2 = add nuw nsw i64 %iv, 3
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.3 = icmp ne i64 %iv_next.2, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.3, label %loop_exiting_bb2.3, label %exit1.epilog-lcssa.loopexit
@@ -6306,8 +6076,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; EPILOG-NEXT:    %unknown.3 = call i1 @unknown_cond()
 ; EPILOG-NEXT:    br i1 %unknown.3, label %loop_latch.3, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.3:
-; EPILOG-NEXT:    %iv_next.3 = add nuw nsw i64 %iv_next.2, 1
-; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
+; EPILOG-NEXT:    %iv_next.3 = add nuw nsw i64 %iv, 4
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.4 = icmp ne i64 %iv_next.3, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.4, label %loop_exiting_bb2.4, label %exit1.epilog-lcssa.loopexit
@@ -6315,8 +6084,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; EPILOG-NEXT:    %unknown.4 = call i1 @unknown_cond()
 ; EPILOG-NEXT:    br i1 %unknown.4, label %loop_latch.4, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.4:
-; EPILOG-NEXT:    %iv_next.4 = add nuw nsw i64 %iv_next.3, 1
-; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
+; EPILOG-NEXT:    %iv_next.4 = add nuw nsw i64 %iv, 5
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.5 = icmp ne i64 %iv_next.4, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.5, label %loop_exiting_bb2.5, label %exit1.epilog-lcssa.loopexit
@@ -6324,8 +6092,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; EPILOG-NEXT:    %unknown.5 = call i1 @unknown_cond()
 ; EPILOG-NEXT:    br i1 %unknown.5, label %loop_latch.5, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.5:
-; EPILOG-NEXT:    %iv_next.5 = add nuw nsw i64 %iv_next.4, 1
-; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
+; EPILOG-NEXT:    %iv_next.5 = add nuw nsw i64 %iv, 6
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.6 = icmp ne i64 %iv_next.5, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.6, label %loop_exiting_bb2.6, label %exit1.epilog-lcssa.loopexit
@@ -6333,8 +6100,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; EPILOG-NEXT:    %unknown.6 = call i1 @unknown_cond()
 ; EPILOG-NEXT:    br i1 %unknown.6, label %loop_latch.6, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.6:
-; EPILOG-NEXT:    %iv_next.6 = add nuw nsw i64 %iv_next.5, 1
-; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
+; EPILOG-NEXT:    %iv_next.6 = add nuw nsw i64 %iv, 7
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.7 = icmp ne i64 %iv_next.6, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.7, label %loop_exiting_bb2.7, label %exit1.epilog-lcssa.loopexit
@@ -6342,8 +6108,8 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; EPILOG-NEXT:    %unknown.7 = call i1 @unknown_cond()
 ; EPILOG-NEXT:    br i1 %unknown.7, label %loop_latch.7, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.7:
-; EPILOG-NEXT:    %iv_next.7 = add i64 %iv_next.6, 1
-; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; EPILOG-NEXT:    %iv_next.7 = add i64 %iv, 8
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter, 8
 ; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.next.7, %unroll_iter
 ; EPILOG-NEXT:    br i1 %niter.ncmp.7, label %loop_header, label %exit1.unr-lcssa.loopexit
 ; EPILOG:       exit1.unr-lcssa.loopexit:
@@ -6400,7 +6166,6 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; EPILOG-BLOCK-NEXT:    br i1 %unknown, label %loop_latch, label %exit1.epilog-lcssa.loopexit
 ; EPILOG-BLOCK:       loop_latch:
 ; EPILOG-BLOCK-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
-; EPILOG-BLOCK-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-BLOCK-NEXT:    call void @bar()
 ; EPILOG-BLOCK-NEXT:    %cmp_early.1 = icmp ne i64 %iv_next, %trip2
 ; EPILOG-BLOCK-NEXT:    br i1 %cmp_early.1, label %loop_exiting_bb2.1, label %exit1.epilog-lcssa.loopexit
@@ -6408,8 +6173,8 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; EPILOG-BLOCK-NEXT:    %unknown.1 = call i1 @unknown_cond()
 ; EPILOG-BLOCK-NEXT:    br i1 %unknown.1, label %loop_latch.1, label %exit1.epilog-lcssa.loopexit
 ; EPILOG-BLOCK:       loop_latch.1:
-; EPILOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv_next, 1
-; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; EPILOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv, 2
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.next.1, %unroll_iter
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1, label %loop_header, label %exit1.unr-lcssa.loopexit, !llvm.loop !19
 ; EPILOG-BLOCK:       exit1.unr-lcssa.loopexit:
@@ -6487,7 +6252,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; PROLOG-NEXT:    %unknown.1 = call i1 @unknown_cond()
 ; PROLOG-NEXT:    br i1 %unknown.1, label %loop_latch.1, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.1:
-; PROLOG-NEXT:    %iv_next.1 = add i64 %iv_next, 1
+; PROLOG-NEXT:    %iv_next.1 = add i64 %iv, 2
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.2 = icmp ne i64 %iv_next.1, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.2, label %loop_exiting_bb2.2, label %exit1.unr-lcssa.loopexit
@@ -6495,7 +6260,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; PROLOG-NEXT:    %unknown.2 = call i1 @unknown_cond()
 ; PROLOG-NEXT:    br i1 %unknown.2, label %loop_latch.2, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.2:
-; PROLOG-NEXT:    %iv_next.2 = add i64 %iv_next.1, 1
+; PROLOG-NEXT:    %iv_next.2 = add i64 %iv, 3
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.3 = icmp ne i64 %iv_next.2, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.3, label %loop_exiting_bb2.3, label %exit1.unr-lcssa.loopexit
@@ -6503,7 +6268,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; PROLOG-NEXT:    %unknown.3 = call i1 @unknown_cond()
 ; PROLOG-NEXT:    br i1 %unknown.3, label %loop_latch.3, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.3:
-; PROLOG-NEXT:    %iv_next.3 = add i64 %iv_next.2, 1
+; PROLOG-NEXT:    %iv_next.3 = add i64 %iv, 4
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.4 = icmp ne i64 %iv_next.3, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.4, label %loop_exiting_bb2.4, label %exit1.unr-lcssa.loopexit
@@ -6511,7 +6276,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; PROLOG-NEXT:    %unknown.4 = call i1 @unknown_cond()
 ; PROLOG-NEXT:    br i1 %unknown.4, label %loop_latch.4, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.4:
-; PROLOG-NEXT:    %iv_next.4 = add i64 %iv_next.3, 1
+; PROLOG-NEXT:    %iv_next.4 = add i64 %iv, 5
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.5 = icmp ne i64 %iv_next.4, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.5, label %loop_exiting_bb2.5, label %exit1.unr-lcssa.loopexit
@@ -6519,7 +6284,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; PROLOG-NEXT:    %unknown.5 = call i1 @unknown_cond()
 ; PROLOG-NEXT:    br i1 %unknown.5, label %loop_latch.5, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.5:
-; PROLOG-NEXT:    %iv_next.5 = add i64 %iv_next.4, 1
+; PROLOG-NEXT:    %iv_next.5 = add i64 %iv, 6
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.6 = icmp ne i64 %iv_next.5, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.6, label %loop_exiting_bb2.6, label %exit1.unr-lcssa.loopexit
@@ -6527,7 +6292,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; PROLOG-NEXT:    %unknown.6 = call i1 @unknown_cond()
 ; PROLOG-NEXT:    br i1 %unknown.6, label %loop_latch.6, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.6:
-; PROLOG-NEXT:    %iv_next.6 = add i64 %iv_next.5, 1
+; PROLOG-NEXT:    %iv_next.6 = add i64 %iv, 7
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.7 = icmp ne i64 %iv_next.6, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.7, label %loop_exiting_bb2.7, label %exit1.unr-lcssa.loopexit
@@ -6535,7 +6300,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; PROLOG-NEXT:    %unknown.7 = call i1 @unknown_cond()
 ; PROLOG-NEXT:    br i1 %unknown.7, label %loop_latch.7, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.7:
-; PROLOG-NEXT:    %iv_next.7 = add i64 %iv_next.6, 1
+; PROLOG-NEXT:    %iv_next.7 = add i64 %iv, 8
 ; PROLOG-NEXT:    %cmp.7 = icmp ne i64 %iv_next.7, %trip
 ; PROLOG-NEXT:    br i1 %cmp.7, label %loop_header, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       exit1.unr-lcssa.loopexit:
@@ -6588,7 +6353,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; PROLOG-BLOCK-NEXT:    %unknown.1 = call i1 @unknown_cond()
 ; PROLOG-BLOCK-NEXT:    br i1 %unknown.1, label %loop_latch.1, label %exit1.unr-lcssa.loopexit
 ; PROLOG-BLOCK:       loop_latch.1:
-; PROLOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv_next, 1
+; PROLOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv, 2
 ; PROLOG-BLOCK-NEXT:    %cmp.1 = icmp ne i64 %iv_next.1, %trip
 ; PROLOG-BLOCK-NEXT:    br i1 %cmp.1, label %loop_header, label %exit1.unr-lcssa.loopexit, !llvm.loop !19
 ; PROLOG-BLOCK:       exit1.unr-lcssa.loopexit:
@@ -6641,64 +6406,50 @@ define void @test14(i64 %trip, i1 %cond) {
 ; EPILOG-NEXT:    %unknown = call i1 @unknown_cond()
 ; EPILOG-NEXT:    br i1 %unknown, label %loop_latch, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch:
-; EPILOG-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
-; EPILOG-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    br i1 %cond, label %loop_exiting_bb2.1, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_exiting_bb2.1:
 ; EPILOG-NEXT:    %unknown.1 = call i1 @unknown_cond()
 ; EPILOG-NEXT:    br i1 %unknown.1, label %loop_latch.1, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.1:
-; EPILOG-NEXT:    %iv_next.1 = add nuw nsw i64 %iv_next, 1
-; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    br i1 %cond, label %loop_exiting_bb2.2, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_exiting_bb2.2:
 ; EPILOG-NEXT:    %unknown.2 = call i1 @unknown_cond()
 ; EPILOG-NEXT:    br i1 %unknown.2, label %loop_latch.2, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.2:
-; EPILOG-NEXT:    %iv_next.2 = add nuw nsw i64 %iv_next.1, 1
-; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    br i1 %cond, label %loop_exiting_bb2.3, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_exiting_bb2.3:
 ; EPILOG-NEXT:    %unknown.3 = call i1 @unknown_cond()
 ; EPILOG-NEXT:    br i1 %unknown.3, label %loop_latch.3, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.3:
-; EPILOG-NEXT:    %iv_next.3 = add nuw nsw i64 %iv_next.2, 1
-; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    br i1 %cond, label %loop_exiting_bb2.4, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_exiting_bb2.4:
 ; EPILOG-NEXT:    %unknown.4 = call i1 @unknown_cond()
 ; EPILOG-NEXT:    br i1 %unknown.4, label %loop_latch.4, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.4:
-; EPILOG-NEXT:    %iv_next.4 = add nuw nsw i64 %iv_next.3, 1
-; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    br i1 %cond, label %loop_exiting_bb2.5, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_exiting_bb2.5:
 ; EPILOG-NEXT:    %unknown.5 = call i1 @unknown_cond()
 ; EPILOG-NEXT:    br i1 %unknown.5, label %loop_latch.5, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.5:
-; EPILOG-NEXT:    %iv_next.5 = add nuw nsw i64 %iv_next.4, 1
-; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    br i1 %cond, label %loop_exiting_bb2.6, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_exiting_bb2.6:
 ; EPILOG-NEXT:    %unknown.6 = call i1 @unknown_cond()
 ; EPILOG-NEXT:    br i1 %unknown.6, label %loop_latch.6, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.6:
-; EPILOG-NEXT:    %iv_next.6 = add nuw nsw i64 %iv_next.5, 1
-; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    br i1 %cond, label %loop_exiting_bb2.7, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_exiting_bb2.7:
 ; EPILOG-NEXT:    %unknown.7 = call i1 @unknown_cond()
 ; EPILOG-NEXT:    br i1 %unknown.7, label %loop_latch.7, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_latch.7:
-; EPILOG-NEXT:    %iv_next.7 = add i64 %iv_next.6, 1
-; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; EPILOG-NEXT:    %iv_next.7 = add i64 %iv, 8
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter, 8
 ; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.next.7, %unroll_iter
 ; EPILOG-NEXT:    br i1 %niter.ncmp.7, label %loop_header, label %exit1.unr-lcssa.loopexit
 ; EPILOG:       exit1.unr-lcssa.loopexit:
@@ -6752,16 +6503,14 @@ define void @test14(i64 %trip, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    %unknown = call i1 @unknown_cond()
 ; EPILOG-BLOCK-NEXT:    br i1 %unknown, label %loop_latch, label %exit1.epilog-lcssa.loopexit
 ; EPILOG-BLOCK:       loop_latch:
-; EPILOG-BLOCK-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
-; EPILOG-BLOCK-NEXT:    %niter.next = add nuw nsw i64 %niter, 1
 ; EPILOG-BLOCK-NEXT:    call void @bar()
 ; EPILOG-BLOCK-NEXT:    br i1 %cond, label %loop_exiting_bb2.1, label %exit1.epilog-lcssa.loopexit
 ; EPILOG-BLOCK:       loop_exiting_bb2.1:
 ; EPILOG-BLOCK-NEXT:    %unknown.1 = call i1 @unknown_cond()
 ; EPILOG-BLOCK-NEXT:    br i1 %unknown.1, label %loop_latch.1, label %exit1.epilog-lcssa.loopexit
 ; EPILOG-BLOCK:       loop_latch.1:
-; EPILOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv_next, 1
-; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; EPILOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv, 2
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter, 2
 ; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.next.1, %unroll_iter
 ; EPILOG-BLOCK-NEXT:    br i1 %niter.ncmp.1, label %loop_header, label %exit1.unr-lcssa.loopexit, !llvm.loop !20
 ; EPILOG-BLOCK:       exit1.unr-lcssa.loopexit:
@@ -6826,56 +6575,49 @@ define void @test14(i64 %trip, i1 %cond) {
 ; PROLOG-NEXT:    %unknown = call i1 @unknown_cond()
 ; PROLOG-NEXT:    br i1 %unknown, label %loop_latch, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch:
-; PROLOG-NEXT:    %iv_next = add i64 %iv, 1
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    br i1 %cond, label %loop_exiting_bb2.1, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_exiting_bb2.1:
 ; PROLOG-NEXT:    %unknown.1 = call i1 @unknown_cond()
 ; PROLOG-NEXT:    br i1 %unknown.1, label %loop_latch.1, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.1:
-; PROLOG-NEXT:    %iv_next.1 = add i64 %iv_next, 1
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    br i1 %cond, label %loop_exiting_bb2.2, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_exiting_bb2.2:
 ; PROLOG-NEXT:    %unknown.2 = call i1 @unknown_cond()
 ; PROLOG-NEXT:    br i1 %unknown.2, label %loop_latch.2, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.2:
-; PROLOG-NEXT:    %iv_next.2 = add i64 %iv_next.1, 1
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    br i1 %cond, label %loop_exiting_bb2.3, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_exiting_bb2.3:
 ; PROLOG-NEXT:    %unknown.3 = call i1 @unknown_cond()
 ; PROLOG-NEXT:    br i1 %unknown.3, label %loop_latch.3, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.3:
-; PROLOG-NEXT:    %iv_next.3 = add i64 %iv_next.2, 1
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    br i1 %cond, label %loop_exiting_bb2.4, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_exiting_bb2.4:
 ; PROLOG-NEXT:    %unknown.4 = call i1 @unknown_cond()
 ; PROLOG-NEXT:    br i1 %unknown.4, label %loop_latch.4, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.4:
-; PROLOG-NEXT:    %iv_next.4 = add i64 %iv_next.3, 1
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    br i1 %cond, label %loop_exiting_bb2.5, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_exiting_bb2.5:
 ; PROLOG-NEXT:    %unknown.5 = call i1 @unknown_cond()
 ; PROLOG-NEXT:    br i1 %unknown.5, label %loop_latch.5, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.5:
-; PROLOG-NEXT:    %iv_next.5 = add i64 %iv_next.4, 1
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    br i1 %cond, label %loop_exiting_bb2.6, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_exiting_bb2.6:
 ; PROLOG-NEXT:    %unknown.6 = call i1 @unknown_cond()
 ; PROLOG-NEXT:    br i1 %unknown.6, label %loop_latch.6, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.6:
-; PROLOG-NEXT:    %iv_next.6 = add i64 %iv_next.5, 1
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    br i1 %cond, label %loop_exiting_bb2.7, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_exiting_bb2.7:
 ; PROLOG-NEXT:    %unknown.7 = call i1 @unknown_cond()
 ; PROLOG-NEXT:    br i1 %unknown.7, label %loop_latch.7, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       loop_latch.7:
-; PROLOG-NEXT:    %iv_next.7 = add i64 %iv_next.6, 1
+; PROLOG-NEXT:    %iv_next.7 = add i64 %iv, 8
 ; PROLOG-NEXT:    %cmp.7 = icmp ne i64 %iv_next.7, %trip
 ; PROLOG-NEXT:    br i1 %cmp.7, label %loop_header, label %exit1.unr-lcssa.loopexit
 ; PROLOG:       exit1.unr-lcssa.loopexit:
@@ -6918,14 +6660,13 @@ define void @test14(i64 %trip, i1 %cond) {
 ; PROLOG-BLOCK-NEXT:    %unknown = call i1 @unknown_cond()
 ; PROLOG-BLOCK-NEXT:    br i1 %unknown, label %loop_latch, label %exit1.unr-lcssa.loopexit
 ; PROLOG-BLOCK:       loop_latch:
-; PROLOG-BLOCK-NEXT:    %iv_next = add i64 %iv, 1
 ; PROLOG-BLOCK-NEXT:    call void @bar()
 ; PROLOG-BLOCK-NEXT:    br i1 %cond, label %loop_exiting_bb2.1, label %exit1.unr-lcssa.loopexit
 ; PROLOG-BLOCK:       loop_exiting_bb2.1:
 ; PROLOG-BLOCK-NEXT:    %unknown.1 = call i1 @unknown_cond()
 ; PROLOG-BLOCK-NEXT:    br i1 %unknown.1, label %loop_latch.1, label %exit1.unr-lcssa.loopexit
 ; PROLOG-BLOCK:       loop_latch.1:
-; PROLOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv_next, 1
+; PROLOG-BLOCK-NEXT:    %iv_next.1 = add i64 %iv, 2
 ; PROLOG-BLOCK-NEXT:    %cmp.1 = icmp ne i64 %iv_next.1, %trip
 ; PROLOG-BLOCK-NEXT:    br i1 %cmp.1, label %loop_header, label %exit1.unr-lcssa.loopexit, !llvm.loop !20
 ; PROLOG-BLOCK:       exit1.unr-lcssa.loopexit:

diff --git a/llvm/test/Transforms/LoopUnroll/runtime-loop5.ll b/llvm/test/Transforms/LoopUnroll/runtime-loop5.ll
index 2a66ccc..b44206a 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/runtime-loop5.ll
+++ b/llvm/test/Transforms/LoopUnroll/runtime-loop5.ll
@@ -81,22 +81,19 @@ define i3 @test(ptr %a, i3 %n) {
 ; UNROLL-4-NEXT:    [[TMP2:%.*]] = load i3, ptr [[ARRAYIDX]], align 1
 ; UNROLL-4-NEXT:    [[ADD:%.*]] = add nsw i3 [[TMP2]], [[SUM_02]]
 ; UNROLL-4-NEXT:    [[INDVARS_IV_NEXT:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 1
-; UNROLL-4-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i3 [[NITER]], 1
 ; UNROLL-4-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i3, ptr [[A]], i64 [[INDVARS_IV_NEXT]]
 ; UNROLL-4-NEXT:    [[TMP3:%.*]] = load i3, ptr [[ARRAYIDX_1]], align 1
 ; UNROLL-4-NEXT:    [[ADD_1:%.*]] = add nsw i3 [[TMP3]], [[ADD]]
-; UNROLL-4-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT]], 1
-; UNROLL-4-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i3 [[NITER_NEXT]], 1
+; UNROLL-4-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 2
 ; UNROLL-4-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i3, ptr [[A]], i64 [[INDVARS_IV_NEXT_1]]
 ; UNROLL-4-NEXT:    [[TMP4:%.*]] = load i3, ptr [[ARRAYIDX_2]], align 1
 ; UNROLL-4-NEXT:    [[ADD_2:%.*]] = add nsw i3 [[TMP4]], [[ADD_1]]
-; UNROLL-4-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_1]], 1
-; UNROLL-4-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i3 [[NITER_NEXT_1]], 1
+; UNROLL-4-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 3
 ; UNROLL-4-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i3, ptr [[A]], i64 [[INDVARS_IV_NEXT_2]]
 ; UNROLL-4-NEXT:    [[TMP5:%.*]] = load i3, ptr [[ARRAYIDX_3]], align 1
 ; UNROLL-4-NEXT:    [[ADD_3]] = add nsw i3 [[TMP5]], [[ADD_2]]
-; UNROLL-4-NEXT:    [[INDVARS_IV_NEXT_3]] = add nuw nsw i64 [[INDVARS_IV_NEXT_2]], 1
-; UNROLL-4-NEXT:    [[NITER_NEXT_3]] = add i3 [[NITER_NEXT_2]], 1
+; UNROLL-4-NEXT:    [[INDVARS_IV_NEXT_3]] = add nuw nsw i64 [[INDVARS_IV]], 4
+; UNROLL-4-NEXT:    [[NITER_NEXT_3]] = add i3 [[NITER]], -4
 ; UNROLL-4-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i3 [[NITER_NEXT_3]], [[UNROLL_ITER]]
 ; UNROLL-4-NEXT:    br i1 [[NITER_NCMP_3]], label [[FOR_END_LOOPEXIT_UNR_LCSSA_LOOPEXIT:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
 ; UNROLL-4:       for.end.loopexit.unr-lcssa.loopexit:

diff --git a/llvm/test/Transforms/LoopUnroll/runtime-multiexit-heuristic.ll b/llvm/test/Transforms/LoopUnroll/runtime-multiexit-heuristic.ll
index fbd4148..db2136b 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/runtime-multiexit-heuristic.ll
+++ b/llvm/test/Transforms/LoopUnroll/runtime-multiexit-heuristic.ll
@@ -191,7 +191,6 @@ define i32 @test1(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 ; ENABLED-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], [[SUM_02]]
 ; ENABLED-NEXT:    [[INDVARS_IV_NEXT:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 1
-; ENABLED-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i64 [[NITER]], 1
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_1:%.*]]
 ; ENABLED:       for.exiting_block.1:
 ; ENABLED-NEXT:    [[CMP_1:%.*]] = icmp eq i64 [[N]], 42
@@ -200,8 +199,7 @@ define i32 @test1(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT]]
 ; ENABLED-NEXT:    [[TMP4:%.*]] = load i32, ptr [[ARRAYIDX_1]], align 4
 ; ENABLED-NEXT:    [[ADD_1:%.*]] = add nsw i32 [[TMP4]], [[ADD]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i64 [[NITER_NEXT]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 2
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_2:%.*]]
 ; ENABLED:       for.exiting_block.2:
 ; ENABLED-NEXT:    [[CMP_2:%.*]] = icmp eq i64 [[N]], 42
@@ -210,8 +208,7 @@ define i32 @test1(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_1]]
 ; ENABLED-NEXT:    [[TMP5:%.*]] = load i32, ptr [[ARRAYIDX_2]], align 4
 ; ENABLED-NEXT:    [[ADD_2:%.*]] = add nsw i32 [[TMP5]], [[ADD_1]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_1]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i64 [[NITER_NEXT_1]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 3
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_3:%.*]]
 ; ENABLED:       for.exiting_block.3:
 ; ENABLED-NEXT:    [[CMP_3:%.*]] = icmp eq i64 [[N]], 42
@@ -220,8 +217,7 @@ define i32 @test1(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_2]]
 ; ENABLED-NEXT:    [[TMP6:%.*]] = load i32, ptr [[ARRAYIDX_3]], align 4
 ; ENABLED-NEXT:    [[ADD_3:%.*]] = add nsw i32 [[TMP6]], [[ADD_2]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_3:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_2]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_3:%.*]] = add nuw nsw i64 [[NITER_NEXT_2]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_3:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 4
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_4:%.*]]
 ; ENABLED:       for.exiting_block.4:
 ; ENABLED-NEXT:    [[CMP_4:%.*]] = icmp eq i64 [[N]], 42
@@ -230,8 +226,7 @@ define i32 @test1(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_4:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_3]]
 ; ENABLED-NEXT:    [[TMP7:%.*]] = load i32, ptr [[ARRAYIDX_4]], align 4
 ; ENABLED-NEXT:    [[ADD_4:%.*]] = add nsw i32 [[TMP7]], [[ADD_3]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_4:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_3]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_4:%.*]] = add nuw nsw i64 [[NITER_NEXT_3]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_4:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 5
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_5:%.*]]
 ; ENABLED:       for.exiting_block.5:
 ; ENABLED-NEXT:    [[CMP_5:%.*]] = icmp eq i64 [[N]], 42
@@ -240,8 +235,7 @@ define i32 @test1(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_5:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_4]]
 ; ENABLED-NEXT:    [[TMP8:%.*]] = load i32, ptr [[ARRAYIDX_5]], align 4
 ; ENABLED-NEXT:    [[ADD_5:%.*]] = add nsw i32 [[TMP8]], [[ADD_4]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_5:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_4]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_5:%.*]] = add nuw nsw i64 [[NITER_NEXT_4]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_5:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 6
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_6:%.*]]
 ; ENABLED:       for.exiting_block.6:
 ; ENABLED-NEXT:    [[CMP_6:%.*]] = icmp eq i64 [[N]], 42
@@ -250,8 +244,7 @@ define i32 @test1(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_6:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_5]]
 ; ENABLED-NEXT:    [[TMP9:%.*]] = load i32, ptr [[ARRAYIDX_6]], align 4
 ; ENABLED-NEXT:    [[ADD_6:%.*]] = add nsw i32 [[TMP9]], [[ADD_5]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_6:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_5]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_6:%.*]] = add nuw nsw i64 [[NITER_NEXT_5]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_6:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 7
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_7:%.*]]
 ; ENABLED:       for.exiting_block.7:
 ; ENABLED-NEXT:    [[CMP_7:%.*]] = icmp eq i64 [[N]], 42
@@ -260,8 +253,8 @@ define i32 @test1(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_7:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_6]]
 ; ENABLED-NEXT:    [[TMP10:%.*]] = load i32, ptr [[ARRAYIDX_7]], align 4
 ; ENABLED-NEXT:    [[ADD_7]] = add nsw i32 [[TMP10]], [[ADD_6]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_7]] = add i64 [[INDVARS_IV_NEXT_6]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_7]] = add i64 [[NITER_NEXT_6]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_7]] = add i64 [[INDVARS_IV]], 8
+; ENABLED-NEXT:    [[NITER_NEXT_7]] = add i64 [[NITER]], 8
 ; ENABLED-NEXT:    [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NEXT_7]], [[UNROLL_ITER]]
 ; ENABLED-NEXT:    br i1 [[NITER_NCMP_7]], label [[LATCHEXIT_UNR_LCSSA_LOOPEXIT:%.*]], label [[HEADER]]
 ; ENABLED:       latchexit.unr-lcssa.loopexit:
@@ -412,7 +405,6 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 ; ENABLED-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], [[SUM_02]]
 ; ENABLED-NEXT:    [[INDVARS_IV_NEXT:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 1
-; ENABLED-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i64 [[NITER]], 1
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_1:%.*]]
 ; ENABLED:       for.exiting_block.1:
 ; ENABLED-NEXT:    [[CMP_1:%.*]] = icmp eq i64 [[N]], 42
@@ -421,8 +413,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT]]
 ; ENABLED-NEXT:    [[TMP4:%.*]] = load i32, ptr [[ARRAYIDX_1]], align 4
 ; ENABLED-NEXT:    [[ADD_1:%.*]] = add nsw i32 [[TMP4]], [[ADD]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i64 [[NITER_NEXT]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 2
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_2:%.*]]
 ; ENABLED:       for.exiting_block.2:
 ; ENABLED-NEXT:    [[CMP_2:%.*]] = icmp eq i64 [[N]], 42
@@ -431,8 +422,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_1]]
 ; ENABLED-NEXT:    [[TMP5:%.*]] = load i32, ptr [[ARRAYIDX_2]], align 4
 ; ENABLED-NEXT:    [[ADD_2:%.*]] = add nsw i32 [[TMP5]], [[ADD_1]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_1]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i64 [[NITER_NEXT_1]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 3
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_3:%.*]]
 ; ENABLED:       for.exiting_block.3:
 ; ENABLED-NEXT:    [[CMP_3:%.*]] = icmp eq i64 [[N]], 42
@@ -441,8 +431,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_2]]
 ; ENABLED-NEXT:    [[TMP6:%.*]] = load i32, ptr [[ARRAYIDX_3]], align 4
 ; ENABLED-NEXT:    [[ADD_3:%.*]] = add nsw i32 [[TMP6]], [[ADD_2]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_3:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_2]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_3:%.*]] = add nuw nsw i64 [[NITER_NEXT_2]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_3:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 4
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_4:%.*]]
 ; ENABLED:       for.exiting_block.4:
 ; ENABLED-NEXT:    [[CMP_4:%.*]] = icmp eq i64 [[N]], 42
@@ -451,8 +440,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_4:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_3]]
 ; ENABLED-NEXT:    [[TMP7:%.*]] = load i32, ptr [[ARRAYIDX_4]], align 4
 ; ENABLED-NEXT:    [[ADD_4:%.*]] = add nsw i32 [[TMP7]], [[ADD_3]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_4:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_3]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_4:%.*]] = add nuw nsw i64 [[NITER_NEXT_3]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_4:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 5
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_5:%.*]]
 ; ENABLED:       for.exiting_block.5:
 ; ENABLED-NEXT:    [[CMP_5:%.*]] = icmp eq i64 [[N]], 42
@@ -461,8 +449,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_5:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_4]]
 ; ENABLED-NEXT:    [[TMP8:%.*]] = load i32, ptr [[ARRAYIDX_5]], align 4
 ; ENABLED-NEXT:    [[ADD_5:%.*]] = add nsw i32 [[TMP8]], [[ADD_4]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_5:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_4]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_5:%.*]] = add nuw nsw i64 [[NITER_NEXT_4]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_5:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 6
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_6:%.*]]
 ; ENABLED:       for.exiting_block.6:
 ; ENABLED-NEXT:    [[CMP_6:%.*]] = icmp eq i64 [[N]], 42
@@ -471,8 +458,7 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_6:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_5]]
 ; ENABLED-NEXT:    [[TMP9:%.*]] = load i32, ptr [[ARRAYIDX_6]], align 4
 ; ENABLED-NEXT:    [[ADD_6:%.*]] = add nsw i32 [[TMP9]], [[ADD_5]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_6:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_5]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_6:%.*]] = add nuw nsw i64 [[NITER_NEXT_5]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_6:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 7
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_7:%.*]]
 ; ENABLED:       for.exiting_block.7:
 ; ENABLED-NEXT:    [[CMP_7:%.*]] = icmp eq i64 [[N]], 42
@@ -481,8 +467,8 @@ define i32 @test2(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_7:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_6]]
 ; ENABLED-NEXT:    [[TMP10:%.*]] = load i32, ptr [[ARRAYIDX_7]], align 4
 ; ENABLED-NEXT:    [[ADD_7]] = add nsw i32 [[TMP10]], [[ADD_6]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_7]] = add i64 [[INDVARS_IV_NEXT_6]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_7]] = add i64 [[NITER_NEXT_6]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_7]] = add i64 [[INDVARS_IV]], 8
+; ENABLED-NEXT:    [[NITER_NEXT_7]] = add i64 [[NITER]], 8
 ; ENABLED-NEXT:    [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NEXT_7]], [[UNROLL_ITER]]
 ; ENABLED-NEXT:    br i1 [[NITER_NCMP_7]], label [[LATCHEXIT_UNR_LCSSA_LOOPEXIT:%.*]], label [[HEADER]]
 ; ENABLED:       latchexit.unr-lcssa.loopexit:
@@ -946,7 +932,6 @@ define i32 @test5(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 ; ENABLED-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], [[SUM_02]]
 ; ENABLED-NEXT:    [[INDVARS_IV_NEXT:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 1
-; ENABLED-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i64 [[NITER]], 1
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_1:%.*]]
 ; ENABLED:       for.exiting_block.1:
 ; ENABLED-NEXT:    [[CMP_1:%.*]] = icmp eq i64 [[N]], 42
@@ -955,8 +940,7 @@ define i32 @test5(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT]]
 ; ENABLED-NEXT:    [[TMP4:%.*]] = load i32, ptr [[ARRAYIDX_1]], align 4
 ; ENABLED-NEXT:    [[ADD_1:%.*]] = add nsw i32 [[TMP4]], [[ADD]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i64 [[NITER_NEXT]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 2
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_2:%.*]]
 ; ENABLED:       for.exiting_block.2:
 ; ENABLED-NEXT:    [[CMP_2:%.*]] = icmp eq i64 [[N]], 42
@@ -965,8 +949,7 @@ define i32 @test5(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_1]]
 ; ENABLED-NEXT:    [[TMP5:%.*]] = load i32, ptr [[ARRAYIDX_2]], align 4
 ; ENABLED-NEXT:    [[ADD_2:%.*]] = add nsw i32 [[TMP5]], [[ADD_1]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_1]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i64 [[NITER_NEXT_1]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 3
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_3:%.*]]
 ; ENABLED:       for.exiting_block.3:
 ; ENABLED-NEXT:    [[CMP_3:%.*]] = icmp eq i64 [[N]], 42
@@ -975,8 +958,7 @@ define i32 @test5(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_2]]
 ; ENABLED-NEXT:    [[TMP6:%.*]] = load i32, ptr [[ARRAYIDX_3]], align 4
 ; ENABLED-NEXT:    [[ADD_3:%.*]] = add nsw i32 [[TMP6]], [[ADD_2]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_3:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_2]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_3:%.*]] = add nuw nsw i64 [[NITER_NEXT_2]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_3:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 4
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_4:%.*]]
 ; ENABLED:       for.exiting_block.4:
 ; ENABLED-NEXT:    [[CMP_4:%.*]] = icmp eq i64 [[N]], 42
@@ -985,8 +967,7 @@ define i32 @test5(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_4:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_3]]
 ; ENABLED-NEXT:    [[TMP7:%.*]] = load i32, ptr [[ARRAYIDX_4]], align 4
 ; ENABLED-NEXT:    [[ADD_4:%.*]] = add nsw i32 [[TMP7]], [[ADD_3]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_4:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_3]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_4:%.*]] = add nuw nsw i64 [[NITER_NEXT_3]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_4:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 5
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_5:%.*]]
 ; ENABLED:       for.exiting_block.5:
 ; ENABLED-NEXT:    [[CMP_5:%.*]] = icmp eq i64 [[N]], 42
@@ -995,8 +976,7 @@ define i32 @test5(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_5:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_4]]
 ; ENABLED-NEXT:    [[TMP8:%.*]] = load i32, ptr [[ARRAYIDX_5]], align 4
 ; ENABLED-NEXT:    [[ADD_5:%.*]] = add nsw i32 [[TMP8]], [[ADD_4]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_5:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_4]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_5:%.*]] = add nuw nsw i64 [[NITER_NEXT_4]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_5:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 6
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_6:%.*]]
 ; ENABLED:       for.exiting_block.6:
 ; ENABLED-NEXT:    [[CMP_6:%.*]] = icmp eq i64 [[N]], 42
@@ -1005,8 +985,7 @@ define i32 @test5(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_6:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_5]]
 ; ENABLED-NEXT:    [[TMP9:%.*]] = load i32, ptr [[ARRAYIDX_6]], align 4
 ; ENABLED-NEXT:    [[ADD_6:%.*]] = add nsw i32 [[TMP9]], [[ADD_5]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_6:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_5]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_6:%.*]] = add nuw nsw i64 [[NITER_NEXT_5]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_6:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 7
 ; ENABLED-NEXT:    br label [[FOR_EXITING_BLOCK_7:%.*]]
 ; ENABLED:       for.exiting_block.7:
 ; ENABLED-NEXT:    [[CMP_7:%.*]] = icmp eq i64 [[N]], 42
@@ -1015,8 +994,8 @@ define i32 @test5(ptr nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ARRAYIDX_7:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT_6]]
 ; ENABLED-NEXT:    [[TMP10:%.*]] = load i32, ptr [[ARRAYIDX_7]], align 4
 ; ENABLED-NEXT:    [[ADD_7]] = add nsw i32 [[TMP10]], [[ADD_6]]
-; ENABLED-NEXT:    [[INDVARS_IV_NEXT_7]] = add i64 [[INDVARS_IV_NEXT_6]], 1
-; ENABLED-NEXT:    [[NITER_NEXT_7]] = add i64 [[NITER_NEXT_6]], 1
+; ENABLED-NEXT:    [[INDVARS_IV_NEXT_7]] = add i64 [[INDVARS_IV]], 8
+; ENABLED-NEXT:    [[NITER_NEXT_7]] = add i64 [[NITER]], 8
 ; ENABLED-NEXT:    [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NEXT_7]], [[UNROLL_ITER]]
 ; ENABLED-NEXT:    br i1 [[NITER_NCMP_7]], label [[LATCHEXIT_UNR_LCSSA_LOOPEXIT:%.*]], label [[HEADER]]
 ; ENABLED:       latchexit.unr-lcssa.loopexit:

diff --git a/llvm/test/Transforms/LoopUnroll/runtime-small-upperbound.ll b/llvm/test/Transforms/LoopUnroll/runtime-small-upperbound.ll
index 2d2a79c..247cdd6 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/runtime-small-upperbound.ll
+++ b/llvm/test/Transforms/LoopUnroll/runtime-small-upperbound.ll
@@ -108,29 +108,27 @@ define dso_local void @hoge_5(i8 %arg) {
 ; UPPER-NEXT:    [[TMP1:%.*]] = icmp ult i32 [[IV_NEXT]], 17
 ; UPPER-NEXT:    br i1 [[TMP1]], label [[LOOP_1:%.*]], label [[EXIT_LOOPEXIT:%.*]]
 ; UPPER:       loop.1:
-; UPPER-NEXT:    [[IV_NEXT_1:%.*]] = add nuw i32 [[IV_NEXT]], 4
+; UPPER-NEXT:    [[IV_NEXT_1:%.*]] = add nuw i32 [[X]], 8
 ; UPPER-NEXT:    [[PTR_NEXT_1:%.*]] = getelementptr inbounds i8, ptr [[PTR_NEXT]], i32 1
 ; UPPER-NEXT:    store i8 [[ARG]], ptr [[PTR_NEXT_1]], align 1
 ; UPPER-NEXT:    [[TMP2:%.*]] = icmp ult i32 [[IV_NEXT_1]], 17
 ; UPPER-NEXT:    br i1 [[TMP2]], label [[LOOP_2:%.*]], label [[EXIT_LOOPEXIT]]
 ; UPPER:       loop.2:
-; UPPER-NEXT:    [[IV_NEXT_2:%.*]] = add nuw i32 [[IV_NEXT_1]], 4
+; UPPER-NEXT:    [[IV_NEXT_2:%.*]] = add nuw i32 [[X]], 12
 ; UPPER-NEXT:    [[PTR_NEXT_2:%.*]] = getelementptr inbounds i8, ptr [[PTR_NEXT_1]], i32 1
 ; UPPER-NEXT:    store i8 [[ARG]], ptr [[PTR_NEXT_2]], align 1
 ; UPPER-NEXT:    [[TMP3:%.*]] = icmp ult i32 [[IV_NEXT_2]], 17
 ; UPPER-NEXT:    br i1 [[TMP3]], label [[LOOP_3:%.*]], label [[EXIT_LOOPEXIT]]
 ; UPPER:       loop.3:
-; UPPER-NEXT:    [[IV_NEXT_3:%.*]] = add nuw i32 [[IV_NEXT_2]], 4
+; UPPER-NEXT:    [[IV_NEXT_3:%.*]] = add nuw i32 [[X]], 16
 ; UPPER-NEXT:    [[PTR_NEXT_3:%.*]] = getelementptr inbounds i8, ptr [[PTR_NEXT_2]], i32 1
 ; UPPER-NEXT:    store i8 [[ARG]], ptr [[PTR_NEXT_3]], align 1
 ; UPPER-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[IV_NEXT_3]], 17
 ; UPPER-NEXT:    br i1 [[TMP4]], label [[LOOP_4:%.*]], label [[EXIT_LOOPEXIT]]
 ; UPPER:       loop.4:
-; UPPER-NEXT:    [[IV_NEXT_4:%.*]] = add nuw i32 [[IV_NEXT_3]], 4
 ; UPPER-NEXT:    [[PTR_NEXT_4:%.*]] = getelementptr inbounds i8, ptr [[PTR_NEXT_3]], i32 1
 ; UPPER-NEXT:    store i8 [[ARG]], ptr [[PTR_NEXT_4]], align 1
-; UPPER-NEXT:    [[TMP5:%.*]] = icmp ult i32 [[IV_NEXT_4]], 17
-; UPPER-NEXT:    br i1 [[TMP5]], label [[LOOP_5:%.*]], label [[EXIT_LOOPEXIT]]
+; UPPER-NEXT:    br i1 false, label [[LOOP_5:%.*]], label [[EXIT_LOOPEXIT]]
 ; UPPER:       loop.5:
 ; UPPER-NEXT:    [[PTR_NEXT_5:%.*]] = getelementptr inbounds i8, ptr [[PTR_NEXT_4]], i32 1
 ; UPPER-NEXT:    store i8 [[ARG]], ptr [[PTR_NEXT_5]], align 1

diff --git a/llvm/test/Transforms/LoopUnroll/runtime-unroll-assume-no-remainder.ll b/llvm/test/Transforms/LoopUnroll/runtime-unroll-assume-no-remainder.ll
index 6dc0d1c..81fceb6 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/runtime-unroll-assume-no-remainder.ll
+++ b/llvm/test/Transforms/LoopUnroll/runtime-unroll-assume-no-remainder.ll
@@ -33,7 +33,7 @@ define dso_local void @assumeDivisibleTC(ptr noalias nocapture %a, ptr noalias n
 ; CHECK-NEXT:    [[ADD_1:%.*]] = add i8 [[TMP1]], 3
 ; CHECK-NEXT:    [[ARRAYIDX4_1:%.*]] = getelementptr inbounds i8, ptr [[A]], i32 [[INC]]
 ; CHECK-NEXT:    store i8 [[ADD_1]], ptr [[ARRAYIDX4_1]], align 1
-; CHECK-NEXT:    [[INC_1]] = add nuw nsw i32 [[INC]], 1
+; CHECK-NEXT:    [[INC_1]] = add nuw nsw i32 [[I_011]], 2
 ; CHECK-NEXT:    [[CMP1_1:%.*]] = icmp slt i32 [[INC_1]], [[N]]
 ; CHECK-NEXT:    br i1 [[CMP1_1]], label [[FOR_BODY]], label [[EXIT_LOOPEXIT:%.*]], !llvm.loop [[LOOP0:![0-9]+]]
 ; CHECK:       exit.loopexit:
@@ -104,14 +104,13 @@ define dso_local void @cannotProveDivisibleTC(ptr noalias nocapture %a, ptr noal
 ; CHECK-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds i8, ptr [[A:%.*]], i32 [[I_011]]
 ; CHECK-NEXT:    store i8 [[ADD]], ptr [[ARRAYIDX4]], align 1
 ; CHECK-NEXT:    [[INC:%.*]] = add nuw nsw i32 [[I_011]], 1
-; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i32 [[NITER]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i8, ptr [[B]], i32 [[INC]]
 ; CHECK-NEXT:    [[TMP3:%.*]] = load i8, ptr [[ARRAYIDX_1]], align 1
 ; CHECK-NEXT:    [[ADD_1:%.*]] = add i8 [[TMP3]], 3
 ; CHECK-NEXT:    [[ARRAYIDX4_1:%.*]] = getelementptr inbounds i8, ptr [[A]], i32 [[INC]]
 ; CHECK-NEXT:    store i8 [[ADD_1]], ptr [[ARRAYIDX4_1]], align 1
-; CHECK-NEXT:    [[INC_1]] = add nuw nsw i32 [[INC]], 1
-; CHECK-NEXT:    [[NITER_NEXT_1]] = add i32 [[NITER_NEXT]], 1
+; CHECK-NEXT:    [[INC_1]] = add nuw nsw i32 [[I_011]], 2
+; CHECK-NEXT:    [[NITER_NEXT_1]] = add i32 [[NITER]], 2
 ; CHECK-NEXT:    [[NITER_NCMP_1:%.*]] = icmp ne i32 [[NITER_NEXT_1]], [[UNROLL_ITER]]
 ; CHECK-NEXT:    br i1 [[NITER_NCMP_1]], label [[FOR_BODY]], label [[EXIT_LOOPEXIT_UNR_LCSSA_LOOPEXIT:%.*]], !llvm.loop [[LOOP2:![0-9]+]]
 ; CHECK:       exit.loopexit.unr-lcssa.loopexit:

diff --git a/llvm/test/Transforms/LoopUnroll/scev-invalidation-lcssa.ll b/llvm/test/Transforms/LoopUnroll/scev-invalidation-lcssa.ll
index c8cb54a..ec71c67 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/scev-invalidation-lcssa.ll
+++ b/llvm/test/Transforms/LoopUnroll/scev-invalidation-lcssa.ll
@@ -37,22 +37,8 @@ define i32 @f(i1 %cond1) #0 !prof !0 {
 ; CHECK:       loop2:
 ; CHECK-NEXT:    [[PHI:%.*]] = phi i64 [ 0, [[ENTRY2_NEW]] ], [ [[INC_7:%.*]], [[LOOP2]] ]
 ; CHECK-NEXT:    [[NITER:%.*]] = phi i64 [ 0, [[ENTRY2_NEW]] ], [ [[NITER_NEXT_7:%.*]], [[LOOP2]] ]
-; CHECK-NEXT:    [[INC:%.*]] = add nuw nsw i64 [[PHI]], 1
-; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i64 [[NITER]], 1
-; CHECK-NEXT:    [[INC_1:%.*]] = add nuw nsw i64 [[INC]], 1
-; CHECK-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i64 [[NITER_NEXT]], 1
-; CHECK-NEXT:    [[INC_2:%.*]] = add nuw nsw i64 [[INC_1]], 1
-; CHECK-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i64 [[NITER_NEXT_1]], 1
-; CHECK-NEXT:    [[INC_3:%.*]] = add nuw nsw i64 [[INC_2]], 1
-; CHECK-NEXT:    [[NITER_NEXT_3:%.*]] = add nuw nsw i64 [[NITER_NEXT_2]], 1
-; CHECK-NEXT:    [[INC_4:%.*]] = add nuw nsw i64 [[INC_3]], 1
-; CHECK-NEXT:    [[NITER_NEXT_4:%.*]] = add nuw nsw i64 [[NITER_NEXT_3]], 1
-; CHECK-NEXT:    [[INC_5:%.*]] = add nuw nsw i64 [[INC_4]], 1
-; CHECK-NEXT:    [[NITER_NEXT_5:%.*]] = add nuw nsw i64 [[NITER_NEXT_4]], 1
-; CHECK-NEXT:    [[INC_6:%.*]] = add nuw nsw i64 [[INC_5]], 1
-; CHECK-NEXT:    [[NITER_NEXT_6:%.*]] = add nuw nsw i64 [[NITER_NEXT_5]], 1
-; CHECK-NEXT:    [[INC_7]] = add i64 [[INC_6]], 1
-; CHECK-NEXT:    [[NITER_NEXT_7]] = add i64 [[NITER_NEXT_6]], 1
+; CHECK-NEXT:    [[INC_7]] = add i64 [[PHI]], 8
+; CHECK-NEXT:    [[NITER_NEXT_7]] = add i64 [[NITER]], 8
 ; CHECK-NEXT:    [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NEXT_7]], [[UNROLL_ITER]]
 ; CHECK-NEXT:    br i1 [[NITER_NCMP_7]], label [[EXIT2_UNR_LCSSA_LOOPEXIT:%.*]], label [[LOOP2]]
 ; CHECK:       exit2.unr-lcssa.loopexit:

diff --git a/llvm/test/Transforms/LoopUnroll/shifted-tripcount.ll b/llvm/test/Transforms/LoopUnroll/shifted-tripcount.ll
index 62dc3d6..01ac4b7 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/shifted-tripcount.ll
+++ b/llvm/test/Transforms/LoopUnroll/shifted-tripcount.ll
@@ -17,7 +17,7 @@ define void @latch_exit(ptr nocapture %p, i64 %n) nounwind {
 ; CHECK-NEXT:    [[MUL9:%.*]] = fmul double [[TMP8]], [[TMP4]]
 ; CHECK-NEXT:    store double [[MUL9]], ptr [[ARRAYIDX7]], align 8
 ; CHECK-NEXT:    [[ARRAYIDX7_1:%.*]] = getelementptr double, ptr [[P]], i64 [[TMP16]]
-; CHECK-NEXT:    [[TMP16_1]] = add i64 [[TMP16]], 1
+; CHECK-NEXT:    [[TMP16_1]] = add i64 [[I_013]], 2
 ; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr double, ptr [[P]], i64 [[TMP16_1]]
 ; CHECK-NEXT:    [[TMP4_1:%.*]] = load double, ptr [[ARRAYIDX_1]], align 8
 ; CHECK-NEXT:    [[TMP8_1:%.*]] = load double, ptr [[ARRAYIDX7_1]], align 8
@@ -67,7 +67,7 @@ define void @non_latch_exit(ptr nocapture %p, i64 %n) nounwind {
 ; CHECK-NEXT:    br label [[LATCH:%.*]]
 ; CHECK:       latch:
 ; CHECK-NEXT:    [[ARRAYIDX7_1:%.*]] = getelementptr double, ptr [[P]], i64 [[TMP16]]
-; CHECK-NEXT:    [[TMP16_1]] = add i64 [[TMP16]], 1
+; CHECK-NEXT:    [[TMP16_1]] = add i64 [[I_013]], 2
 ; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr double, ptr [[P]], i64 [[TMP16_1]]
 ; CHECK-NEXT:    [[TMP4_1:%.*]] = load double, ptr [[ARRAYIDX_1]], align 8
 ; CHECK-NEXT:    [[TMP8_1:%.*]] = load double, ptr [[ARRAYIDX7_1]], align 8

diff --git a/llvm/test/Transforms/LoopUnroll/tripcount-overflow.ll b/llvm/test/Transforms/LoopUnroll/tripcount-overflow.ll
index 284fcb1..0e4c73e 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/tripcount-overflow.ll
+++ b/llvm/test/Transforms/LoopUnroll/tripcount-overflow.ll
@@ -25,9 +25,8 @@ define i32 @foo(i32 %N) {
 ; EPILOG-NEXT:    [[I:%.*]] = phi i32 [ 0, [[ENTRY_NEW]] ], [ [[INC_1:%.*]], [[WHILE_BODY]] ]
 ; EPILOG-NEXT:    [[NITER:%.*]] = phi i32 [ 0, [[ENTRY_NEW]] ], [ [[NITER_NEXT_1:%.*]], [[WHILE_BODY]] ]
 ; EPILOG-NEXT:    [[INC:%.*]] = add nuw nsw i32 [[I]], 1
-; EPILOG-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i32 [[NITER]], 1
-; EPILOG-NEXT:    [[INC_1]] = add i32 [[INC]], 1
-; EPILOG-NEXT:    [[NITER_NEXT_1]] = add i32 [[NITER_NEXT]], 1
+; EPILOG-NEXT:    [[INC_1]] = add i32 [[I]], 2
+; EPILOG-NEXT:    [[NITER_NEXT_1]] = add i32 [[NITER]], 2
 ; EPILOG-NEXT:    [[NITER_NCMP_1:%.*]] = icmp eq i32 [[NITER_NEXT_1]], [[UNROLL_ITER]]
 ; EPILOG-NEXT:    br i1 [[NITER_NCMP_1]], label [[WHILE_END_UNR_LCSSA_LOOPEXIT:%.*]], label [[WHILE_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
 ; EPILOG:       while.end.unr-lcssa.loopexit:
@@ -68,7 +67,7 @@ define i32 @foo(i32 %N) {
 ; PROLOG-NEXT:    [[I:%.*]] = phi i32 [ [[I_UNR]], [[ENTRY_NEW]] ], [ [[INC_1:%.*]], [[WHILE_BODY]] ]
 ; PROLOG-NEXT:    [[INC:%.*]] = add i32 [[I]], 1
 ; PROLOG-NEXT:    [[CMP_1:%.*]] = icmp eq i32 [[INC]], [[N]]
-; PROLOG-NEXT:    [[INC_1]] = add i32 [[INC]], 1
+; PROLOG-NEXT:    [[INC_1]] = add i32 [[I]], 2
 ; PROLOG-NEXT:    br i1 [[CMP_1]], label [[WHILE_END_UNR_LCSSA:%.*]], label [[WHILE_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
 ; PROLOG:       while.end.unr-lcssa:
 ; PROLOG-NEXT:    [[I_LCSSA_PH:%.*]] = phi i32 [ [[INC]], [[WHILE_BODY]] ]

diff --git a/llvm/test/Transforms/LoopUnroll/unroll-header-exiting-with-phis.ll b/llvm/test/Transforms/LoopUnroll/unroll-header-exiting-with-phis.ll
index b690f0d..ff8e6ef 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/unroll-header-exiting-with-phis.ll
+++ b/llvm/test/Transforms/LoopUnroll/unroll-header-exiting-with-phis.ll
@@ -65,11 +65,11 @@ define i16 @partial_unroll(ptr %A) {
 ; CHECK-NEXT:    [[TMP2_1]] = load i16, ptr [[PTR_1]], align 2
 ; CHECK-NEXT:    br label [[FOR_COND_CLEANUP3_1]]
 ; CHECK:       for.cond.cleanup3.1:
-; CHECK-NEXT:    [[INC9_1:%.*]] = add nuw nsw i64 [[INC9]], 1
+; CHECK-NEXT:    [[INC9_1:%.*]] = add nuw nsw i64 [[I_0]], 2
 ; CHECK-NEXT:    [[CMP_2:%.*]] = icmp ult i64 [[INC9_1]], 200
 ; CHECK-NEXT:    br i1 [[CMP_2]], label [[FOR_COND_CLEANUP3_2]], label [[FOR_COND_CLEANUP:%.*]]
 ; CHECK:       for.cond.cleanup3.2:
-; CHECK-NEXT:    [[INC9_2]] = add nuw nsw i64 [[INC9_1]], 1
+; CHECK-NEXT:    [[INC9_2]] = add nuw nsw i64 [[I_0]], 3
 ; CHECK-NEXT:    br label [[FOR_COND]]
 ;
 entry:

diff --git a/llvm/test/Transforms/LoopUnroll/unroll-opt-attribute.ll b/llvm/test/Transforms/LoopUnroll/unroll-opt-attribute.ll
index 6bac357..8221e6a 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/unroll-opt-attribute.ll
+++ b/llvm/test/Transforms/LoopUnroll/unroll-opt-attribute.ll
@@ -25,10 +25,7 @@ exit:
 
 ; CHECK_COUNT4-LABEL: @Test1
 ; CHECK_COUNT4:      phi
-; CHECK_COUNT4-NEXT: add
-; CHECK_COUNT4-NEXT: add
-; CHECK_COUNT4-NEXT: add
-; CHECK_COUNT4-NEXT: add
+; CHECK_COUNT4-NEXT: add{{.*}}, 4
 ; CHECK_COUNT4-NEXT: icmp
 
 

diff --git a/llvm/test/Transforms/LoopUnrollAndJam/dependencies_visit_order.ll b/llvm/test/Transforms/LoopUnrollAndJam/dependencies_visit_order.ll
index 9444d59..f1a5adf 100644 (file)
--- a/llvm/test/Transforms/LoopUnrollAndJam/dependencies_visit_order.ll
+++ b/llvm/test/Transforms/LoopUnrollAndJam/dependencies_visit_order.ll
@@ -15,25 +15,22 @@ define void @test1() {
 ; CHECK-NEXT:    [[I4017:%.*]] = zext i16 [[I10]] to i64
 ; CHECK-NEXT:    [[I13_I:%.*]] = add nuw nsw i64 [[I4017]], 1
 ; CHECK-NEXT:    [[I42:%.*]] = trunc i64 [[I13_I]] to i16
-; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i16 [[NITER]], 1
 ; CHECK-NEXT:    [[I4017_1:%.*]] = zext i16 [[I42]] to i64
 ; CHECK-NEXT:    [[I13_I_1:%.*]] = add nuw nsw i64 [[I4017_1]], 1
 ; CHECK-NEXT:    [[I42_1:%.*]] = trunc i64 [[I13_I_1]] to i16
-; CHECK-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i16 [[NITER_NEXT]], 1
 ; CHECK-NEXT:    [[I4017_2:%.*]] = zext i16 [[I42_1]] to i64
 ; CHECK-NEXT:    [[I13_I_2:%.*]] = add nuw nsw i64 [[I4017_2]], 1
 ; CHECK-NEXT:    [[I42_2:%.*]] = trunc i64 [[I13_I_2]] to i16
-; CHECK-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i16 [[NITER_NEXT_1]], 1
 ; CHECK-NEXT:    [[I4017_3:%.*]] = zext i16 [[I42_2]] to i64
 ; CHECK-NEXT:    [[I13_I_3:%.*]] = add nuw nsw i64 [[I4017_3]], 1
 ; CHECK-NEXT:    [[I42_3]] = trunc i64 [[I13_I_3]] to i16
-; CHECK-NEXT:    [[NITER_NEXT_3]] = add nuw i16 [[NITER_NEXT_2]], 1
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add nuw i16 [[NITER]], 4
 ; CHECK-NEXT:    br label [[BB10_PREHEADER:%.*]]
 ; CHECK:       bb10.preheader:
 ; CHECK-NEXT:    br i1 true, label [[BB38]], label [[BB10_PREHEADER]]
 ; CHECK:       bb38:
 ; CHECK-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i16 [[NITER_NEXT_3]], -28
-; CHECK-NEXT:    br i1 [[NITER_NCMP_3]], label [[BB1_BB43_CRIT_EDGE_UNR_LCSSA_LOOPEXIT:%.*]], label [[BB5_PREHEADER]], !llvm.loop [[LOOP1:![0-9]+]]
+; CHECK-NEXT:    br i1 [[NITER_NCMP_3]], label [[BB1_BB43_CRIT_EDGE_UNR_LCSSA_LOOPEXIT:%.*]], label [[BB5_PREHEADER]], !llvm.loop [[LOOP0:![0-9]+]]
 ; CHECK:       bb1.bb43_crit_edge.unr-lcssa.loopexit:
 ; CHECK-NEXT:    [[I10_UNR_PH:%.*]] = phi i16 [ [[I42_3]], [[BB38]] ]
 ; CHECK-NEXT:    br label [[BB1_BB43_CRIT_EDGE_UNR_LCSSA]]
@@ -56,7 +53,7 @@ define void @test1() {
 ; CHECK-NEXT:    [[I3_NOT_EPIL:%.*]] = icmp eq i16 [[I10_EPIL]], -27
 ; CHECK-NEXT:    [[EPIL_ITER_NEXT]] = add i16 [[EPIL_ITER]], 1
 ; CHECK-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i16 [[EPIL_ITER_NEXT]], 2
-; CHECK-NEXT:    br i1 [[EPIL_ITER_CMP]], label [[BB5_PREHEADER_EPIL]], label [[BB1_BB43_CRIT_EDGE_EPILOG_LCSSA:%.*]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK-NEXT:    br i1 [[EPIL_ITER_CMP]], label [[BB5_PREHEADER_EPIL]], label [[BB1_BB43_CRIT_EDGE_EPILOG_LCSSA:%.*]], !llvm.loop [[LOOP2:![0-9]+]]
 ; CHECK:       bb1.bb43_crit_edge.epilog-lcssa:
 ; CHECK-NEXT:    br label [[BB1_BB43_CRIT_EDGE]]
 ; CHECK:       bb1.bb43_crit_edge:

diff --git a/llvm/test/Transforms/LoopUnrollAndJam/unroll-and-jam.ll b/llvm/test/Transforms/LoopUnrollAndJam/unroll-and-jam.ll
index a2dd64e..f9f7da5 100644 (file)
--- a/llvm/test/Transforms/LoopUnrollAndJam/unroll-and-jam.ll
+++ b/llvm/test/Transforms/LoopUnrollAndJam/unroll-and-jam.ll
@@ -24,13 +24,10 @@ define void @test1(i32 %I, i32 %E, ptr noalias nocapture %A, ptr noalias nocaptu
 ; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[ADD8_3:%.*]], [[FOR_LATCH:%.*]] ], [ 0, [[FOR_OUTER_PREHEADER_NEW]] ]
 ; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ 0, [[FOR_OUTER_PREHEADER_NEW]] ], [ [[NITER_NEXT_3:%.*]], [[FOR_LATCH]] ]
 ; CHECK-NEXT:    [[ADD8:%.*]] = add nuw nsw i32 [[I]], 1
-; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i32 [[NITER]], 1
-; CHECK-NEXT:    [[ADD8_1:%.*]] = add nuw nsw i32 [[ADD8]], 1
-; CHECK-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i32 [[NITER_NEXT]], 1
-; CHECK-NEXT:    [[ADD8_2:%.*]] = add nuw nsw i32 [[ADD8_1]], 1
-; CHECK-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i32 [[NITER_NEXT_1]], 1
-; CHECK-NEXT:    [[ADD8_3]] = add nuw i32 [[ADD8_2]], 1
-; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER_NEXT_2]], 1
+; CHECK-NEXT:    [[ADD8_1:%.*]] = add nuw nsw i32 [[I]], 2
+; CHECK-NEXT:    [[ADD8_2:%.*]] = add nuw nsw i32 [[I]], 3
+; CHECK-NEXT:    [[ADD8_3]] = add nuw i32 [[I]], 4
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER]], 4
 ; CHECK-NEXT:    br label [[FOR_INNER:%.*]]
 ; CHECK:       for.inner:
 ; CHECK-NEXT:    [[J:%.*]] = phi i32 [ 0, [[FOR_OUTER]] ], [ [[INC:%.*]], [[FOR_INNER]] ]
@@ -116,7 +113,7 @@ define void @test1(i32 %I, i32 %E, ptr noalias nocapture %A, ptr noalias nocaptu
 ; CHECK-NEXT:    [[ADD_LCSSA_EPIL_1:%.*]] = phi i32 [ [[ADD_EPIL_1]], [[FOR_INNER_EPIL_1]] ]
 ; CHECK-NEXT:    [[ARRAYIDX6_EPIL_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD8_EPIL]]
 ; CHECK-NEXT:    store i32 [[ADD_LCSSA_EPIL_1]], ptr [[ARRAYIDX6_EPIL_1]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[ADD8_EPIL_1:%.*]] = add nuw i32 [[ADD8_EPIL]], 1
+; CHECK-NEXT:    [[ADD8_EPIL_1:%.*]] = add nuw i32 [[I_UNR]], 2
 ; CHECK-NEXT:    [[EPIL_ITER_CMP_1:%.*]] = icmp ne i32 2, [[XTRAITER]]
 ; CHECK-NEXT:    br i1 [[EPIL_ITER_CMP_1]], label [[FOR_OUTER_EPIL_2:%.*]], label [[FOR_END_LOOPEXIT_EPILOG_LCSSA]]
 ; CHECK:       for.outer.epil.2:
@@ -204,19 +201,16 @@ define void @test2(i32 %I, i32 %E, ptr noalias nocapture %A, ptr noalias nocaptu
 ; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i32 [[I]]
 ; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr [[ARRAYIDX]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ADD9:%.*]] = add nuw nsw i32 [[I]], 1
-; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i32 [[NITER]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD9]]
 ; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX_1]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[ADD9_1:%.*]] = add nuw nsw i32 [[ADD9]], 1
-; CHECK-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i32 [[NITER_NEXT]], 1
+; CHECK-NEXT:    [[ADD9_1:%.*]] = add nuw nsw i32 [[I]], 2
 ; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD9_1]]
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr [[ARRAYIDX_2]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[ADD9_2:%.*]] = add nuw nsw i32 [[ADD9_1]], 1
-; CHECK-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i32 [[NITER_NEXT_1]], 1
+; CHECK-NEXT:    [[ADD9_2:%.*]] = add nuw nsw i32 [[I]], 3
 ; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD9_2]]
 ; CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[ARRAYIDX_3]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[ADD9_3]] = add nuw i32 [[ADD9_2]], 1
-; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER_NEXT_2]], 1
+; CHECK-NEXT:    [[ADD9_3]] = add nuw i32 [[I]], 4
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER]], 4
 ; CHECK-NEXT:    br label [[FOR_INNER:%.*]]
 ; CHECK:       for.inner:
 ; CHECK-NEXT:    [[J:%.*]] = phi i32 [ 0, [[FOR_OUTER]] ], [ [[INC:%.*]], [[FOR_INNER]] ]
@@ -300,7 +294,7 @@ define void @test2(i32 %I, i32 %E, ptr noalias nocapture %A, ptr noalias nocaptu
 ; CHECK:       for.latch.epil.1:
 ; CHECK-NEXT:    [[ADD_LCSSA_EPIL_1:%.*]] = phi i32 [ [[ADD_EPIL_1]], [[FOR_INNER_EPIL_1]] ]
 ; CHECK-NEXT:    store i32 [[ADD_LCSSA_EPIL_1]], ptr [[ARRAYIDX_EPIL_1]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[ADD9_EPIL_1:%.*]] = add nuw i32 [[ADD9_EPIL]], 1
+; CHECK-NEXT:    [[ADD9_EPIL_1:%.*]] = add nuw i32 [[I_UNR]], 2
 ; CHECK-NEXT:    [[EPIL_ITER_CMP_1:%.*]] = icmp ne i32 2, [[XTRAITER]]
 ; CHECK-NEXT:    br i1 [[EPIL_ITER_CMP_1]], label [[FOR_OUTER_EPIL_2:%.*]], label [[FOR_END10_LOOPEXIT_EPILOG_LCSSA]]
 ; CHECK:       for.outer.epil.2:
@@ -623,14 +617,8 @@ define i32 @test6() #0 {
 ; CHECK:       for.outer:
 ; CHECK-NEXT:    [[INC5_SINK9:%.*]] = phi i32 [ 2, [[ENTRY_NEW]] ], [ [[INC5_3:%.*]], [[FOR_LATCH:%.*]] ]
 ; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ 0, [[ENTRY_NEW]] ], [ [[NITER_NEXT_3:%.*]], [[FOR_LATCH]] ]
-; CHECK-NEXT:    [[INC5:%.*]] = add nuw nsw i32 [[INC5_SINK9]], 1
-; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i32 [[NITER]], 1
-; CHECK-NEXT:    [[INC5_1:%.*]] = add nuw nsw i32 [[INC5]], 1
-; CHECK-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i32 [[NITER_NEXT]], 1
-; CHECK-NEXT:    [[INC5_2:%.*]] = add nuw nsw i32 [[INC5_1]], 1
-; CHECK-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i32 [[NITER_NEXT_1]], 1
-; CHECK-NEXT:    [[INC5_3]] = add nuw nsw i32 [[INC5_2]], 1
-; CHECK-NEXT:    [[NITER_NEXT_3]] = add nuw nsw i32 [[NITER_NEXT_2]], 1
+; CHECK-NEXT:    [[INC5_3]] = add nuw nsw i32 [[INC5_SINK9]], 4
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add nuw nsw i32 [[NITER]], 4
 ; CHECK-NEXT:    br label [[FOR_INNER:%.*]]
 ; CHECK:       for.inner:
 ; CHECK-NEXT:    [[INC_SINK8:%.*]] = phi i32 [ 0, [[FOR_OUTER]] ], [ [[INC:%.*]], [[FOR_INNER]] ]
@@ -727,25 +715,22 @@ define void @test7(i32 %I, i32 %E, ptr noalias nocapture %A, ptr noalias nocaptu
 ; CHECK-NEXT:    [[ADD:%.*]] = add nuw nsw i32 [[I]], 1
 ; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD]]
 ; CHECK-NEXT:    store i32 2, ptr [[ARRAYIDX2]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i32 [[NITER]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD]]
 ; CHECK-NEXT:    store i32 0, ptr [[ARRAYIDX_1]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[ADD_1:%.*]] = add nuw nsw i32 [[ADD]], 1
+; CHECK-NEXT:    [[ADD_1:%.*]] = add nuw nsw i32 [[I]], 2
 ; CHECK-NEXT:    [[ARRAYIDX2_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD_1]]
 ; CHECK-NEXT:    store i32 2, ptr [[ARRAYIDX2_1]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i32 [[NITER_NEXT]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD_1]]
 ; CHECK-NEXT:    store i32 0, ptr [[ARRAYIDX_2]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[ADD_2:%.*]] = add nuw nsw i32 [[ADD_1]], 1
+; CHECK-NEXT:    [[ADD_2:%.*]] = add nuw nsw i32 [[I]], 3
 ; CHECK-NEXT:    [[ARRAYIDX2_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD_2]]
 ; CHECK-NEXT:    store i32 2, ptr [[ARRAYIDX2_2]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i32 [[NITER_NEXT_1]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD_2]]
 ; CHECK-NEXT:    store i32 0, ptr [[ARRAYIDX_3]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[ADD_3]] = add nuw i32 [[ADD_2]], 1
+; CHECK-NEXT:    [[ADD_3]] = add nuw i32 [[I]], 4
 ; CHECK-NEXT:    [[ARRAYIDX2_3:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD_3]]
 ; CHECK-NEXT:    store i32 2, ptr [[ARRAYIDX2_3]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER_NEXT_2]], 1
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER]], 4
 ; CHECK-NEXT:    br label [[FOR_INNER:%.*]]
 ; CHECK:       for.latch:
 ; CHECK-NEXT:    [[ADD9_LCSSA:%.*]] = phi i32 [ [[ADD9:%.*]], [[FOR_INNER]] ]
@@ -818,7 +803,7 @@ define void @test7(i32 %I, i32 %E, ptr noalias nocapture %A, ptr noalias nocaptu
 ; CHECK:       for.outer.epil.1:
 ; CHECK-NEXT:    [[ARRAYIDX_EPIL_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD_EPIL]]
 ; CHECK-NEXT:    store i32 0, ptr [[ARRAYIDX_EPIL_1]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[ADD_EPIL_1:%.*]] = add nuw i32 [[ADD_EPIL]], 1
+; CHECK-NEXT:    [[ADD_EPIL_1:%.*]] = add nuw i32 [[I_UNR]], 2
 ; CHECK-NEXT:    [[ARRAYIDX2_EPIL_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD_EPIL_1]]
 ; CHECK-NEXT:    store i32 2, ptr [[ARRAYIDX2_EPIL_1]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    br label [[FOR_INNER_EPIL_1:%.*]]
@@ -839,7 +824,7 @@ define void @test7(i32 %I, i32 %E, ptr noalias nocapture %A, ptr noalias nocaptu
 ; CHECK:       for.outer.epil.2:
 ; CHECK-NEXT:    [[ARRAYIDX_EPIL_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD_EPIL_1]]
 ; CHECK-NEXT:    store i32 0, ptr [[ARRAYIDX_EPIL_2]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[ADD_EPIL_2:%.*]] = add nuw i32 [[ADD_EPIL_1]], 1
+; CHECK-NEXT:    [[ADD_EPIL_2:%.*]] = add nuw i32 [[I_UNR]], 3
 ; CHECK-NEXT:    [[ARRAYIDX2_EPIL_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD_EPIL_2]]
 ; CHECK-NEXT:    store i32 2, ptr [[ARRAYIDX2_EPIL_2]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    br label [[FOR_INNER_EPIL_2:%.*]]
@@ -928,25 +913,22 @@ define void @test8(i32 %I, i32 %E, ptr noalias nocapture %A, ptr noalias nocaptu
 ; CHECK-NEXT:    [[ADD:%.*]] = add nuw nsw i32 [[I]], 1
 ; CHECK-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD]]
 ; CHECK-NEXT:    store i32 2, ptr [[ARRAYIDX6]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i32 [[NITER]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD]]
 ; CHECK-NEXT:    store i32 0, ptr [[ARRAYIDX_1]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[ADD_1:%.*]] = add nuw nsw i32 [[ADD]], 1
+; CHECK-NEXT:    [[ADD_1:%.*]] = add nuw nsw i32 [[I]], 2
 ; CHECK-NEXT:    [[ARRAYIDX6_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD_1]]
 ; CHECK-NEXT:    store i32 2, ptr [[ARRAYIDX6_1]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i32 [[NITER_NEXT]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD_1]]
 ; CHECK-NEXT:    store i32 0, ptr [[ARRAYIDX_2]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[ADD_2:%.*]] = add nuw nsw i32 [[ADD_1]], 1
+; CHECK-NEXT:    [[ADD_2:%.*]] = add nuw nsw i32 [[I]], 3
 ; CHECK-NEXT:    [[ARRAYIDX6_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD_2]]
 ; CHECK-NEXT:    store i32 2, ptr [[ARRAYIDX6_2]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i32 [[NITER_NEXT_1]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD_2]]
 ; CHECK-NEXT:    store i32 0, ptr [[ARRAYIDX_3]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[ADD_3]] = add nuw i32 [[ADD_2]], 1
+; CHECK-NEXT:    [[ADD_3]] = add nuw i32 [[I]], 4
 ; CHECK-NEXT:    [[ARRAYIDX6_3:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD_3]]
 ; CHECK-NEXT:    store i32 2, ptr [[ARRAYIDX6_3]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER_NEXT_2]], 1
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER]], 4
 ; CHECK-NEXT:    br label [[FOR_INNER:%.*]]
 ; CHECK:       for.inner:
 ; CHECK-NEXT:    [[SUM:%.*]] = phi i32 [ 0, [[FOR_OUTER]] ], [ [[ADD9:%.*]], [[FOR_INNER]] ]
@@ -1019,7 +1001,7 @@ define void @test8(i32 %I, i32 %E, ptr noalias nocapture %A, ptr noalias nocaptu
 ; CHECK:       for.outer.epil.1:
 ; CHECK-NEXT:    [[ARRAYIDX_EPIL_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD_EPIL]]
 ; CHECK-NEXT:    store i32 0, ptr [[ARRAYIDX_EPIL_1]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[ADD_EPIL_1:%.*]] = add nuw i32 [[ADD_EPIL]], 1
+; CHECK-NEXT:    [[ADD_EPIL_1:%.*]] = add nuw i32 [[I_UNR]], 2
 ; CHECK-NEXT:    [[ARRAYIDX6_EPIL_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD_EPIL_1]]
 ; CHECK-NEXT:    store i32 2, ptr [[ARRAYIDX6_EPIL_1]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    br label [[FOR_INNER_EPIL_1:%.*]]
@@ -1040,7 +1022,7 @@ define void @test8(i32 %I, i32 %E, ptr noalias nocapture %A, ptr noalias nocaptu
 ; CHECK:       for.outer.epil.2:
 ; CHECK-NEXT:    [[ARRAYIDX_EPIL_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD_EPIL_1]]
 ; CHECK-NEXT:    store i32 0, ptr [[ARRAYIDX_EPIL_2]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[ADD_EPIL_2:%.*]] = add nuw i32 [[ADD_EPIL_1]], 1
+; CHECK-NEXT:    [[ADD_EPIL_2:%.*]] = add nuw i32 [[I_UNR]], 3
 ; CHECK-NEXT:    [[ARRAYIDX6_EPIL_2:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD_EPIL_2]]
 ; CHECK-NEXT:    store i32 2, ptr [[ARRAYIDX6_EPIL_2]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    br label [[FOR_INNER_EPIL_2:%.*]]
@@ -1135,13 +1117,10 @@ define void @test9(i32 %I, i32 %E, ptr nocapture %A, ptr nocapture readonly %B)
 ; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[ADD8_3:%.*]], [[FOR_LATCH:%.*]] ], [ 0, [[FOR_OUTER_PREHEADER_NEW]] ]
 ; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ 0, [[FOR_OUTER_PREHEADER_NEW]] ], [ [[NITER_NEXT_3:%.*]], [[FOR_LATCH]] ]
 ; CHECK-NEXT:    [[ADD8:%.*]] = add nuw nsw i32 [[I]], 1
-; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i32 [[NITER]], 1
-; CHECK-NEXT:    [[ADD8_1:%.*]] = add nuw nsw i32 [[ADD8]], 1
-; CHECK-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i32 [[NITER_NEXT]], 1
-; CHECK-NEXT:    [[ADD8_2:%.*]] = add nuw nsw i32 [[ADD8_1]], 1
-; CHECK-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i32 [[NITER_NEXT_1]], 1
-; CHECK-NEXT:    [[ADD8_3]] = add nuw i32 [[ADD8_2]], 1
-; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER_NEXT_2]], 1
+; CHECK-NEXT:    [[ADD8_1:%.*]] = add nuw nsw i32 [[I]], 2
+; CHECK-NEXT:    [[ADD8_2:%.*]] = add nuw nsw i32 [[I]], 3
+; CHECK-NEXT:    [[ADD8_3]] = add nuw i32 [[I]], 4
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER]], 4
 ; CHECK-NEXT:    br label [[FOR_INNER:%.*]]
 ; CHECK:       for.inner:
 ; CHECK-NEXT:    [[J:%.*]] = phi i32 [ 0, [[FOR_OUTER]] ], [ [[INC:%.*]], [[FOR_INNER]] ]
@@ -1233,7 +1212,7 @@ define void @test9(i32 %I, i32 %E, ptr nocapture %A, ptr nocapture readonly %B)
 ; CHECK-NEXT:    [[ADD_LCSSA_EPIL_1:%.*]] = phi i32 [ [[ADD_EPIL_1]], [[FOR_INNER_EPIL_1]] ]
 ; CHECK-NEXT:    [[ARRAYIDX6_EPIL_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[ADD8_EPIL]]
 ; CHECK-NEXT:    store i32 [[ADD_LCSSA_EPIL_1]], ptr [[ARRAYIDX6_EPIL_1]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[ADD8_EPIL_1:%.*]] = add nuw i32 [[ADD8_EPIL]], 1
+; CHECK-NEXT:    [[ADD8_EPIL_1:%.*]] = add nuw i32 [[I_UNR]], 2
 ; CHECK-NEXT:    [[EPIL_ITER_CMP_1:%.*]] = icmp ne i32 2, [[XTRAITER]]
 ; CHECK-NEXT:    br i1 [[EPIL_ITER_CMP_1]], label [[FOR_OUTER_EPIL_2:%.*]], label [[FOR_END_LOOPEXIT_EPILOG_LCSSA]]
 ; CHECK:       for.outer.epil.2:
@@ -1316,14 +1295,8 @@ define signext i16 @test10(i32 %k) #0 {
 ; CHECK:       for.body:
 ; CHECK-NEXT:    [[STOREMERGE82:%.*]] = phi i64 [ 0, [[ENTRY_NEW]] ], [ [[INC25_3:%.*]], [[FOR_INC24:%.*]] ]
 ; CHECK-NEXT:    [[NITER:%.*]] = phi i64 [ 0, [[ENTRY_NEW]] ], [ [[NITER_NEXT_3:%.*]], [[FOR_INC24]] ]
-; CHECK-NEXT:    [[INC25:%.*]] = add nuw nsw i64 [[STOREMERGE82]], 1
-; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i64 [[NITER]], 1
-; CHECK-NEXT:    [[INC25_1:%.*]] = add nuw nsw i64 [[INC25]], 1
-; CHECK-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i64 [[NITER_NEXT]], 1
-; CHECK-NEXT:    [[INC25_2:%.*]] = add nuw nsw i64 [[INC25_1]], 1
-; CHECK-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i64 [[NITER_NEXT_1]], 1
-; CHECK-NEXT:    [[INC25_3]] = add nuw nsw i64 [[INC25_2]], 1
-; CHECK-NEXT:    [[NITER_NEXT_3]] = add nuw nsw i64 [[NITER_NEXT_2]], 1
+; CHECK-NEXT:    [[INC25_3]] = add nuw nsw i64 [[STOREMERGE82]], 4
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add nuw nsw i64 [[NITER]], 4
 ; CHECK-NEXT:    br label [[FOR_BODY2:%.*]]
 ; CHECK:       for.body2:
 ; CHECK-NEXT:    [[STOREMERGE:%.*]] = phi i64 [ 4, [[FOR_BODY]] ], [ [[DEC:%.*]], [[FOR_INC21_3:%.*]] ]