[runtime-unroll] Use incrementing IVs instead of decrementing ones

This is one of those wonderful "in theory X doesn't matter, but in practice is does" changes. In this particular case, we shift the IVs inserted by the runtime unroller to clamp iteration count of the loops* from decrementing to incrementing.

Why does this matter?  A couple of reasons:
* SCEV doesn't have a native subtract node.  Instead, all subtracts (A - B) are represented as A + -1 * B and drops any flags invalidated by such.  As a result, SCEV is slightly less good at reasoning about edge cases involving decrementing addrecs than incrementing ones.  (You can see this in the inferred flags in some of the test cases.)
* Other parts of the optimizer produce incrementing IVs, and they're common in idiomatic source language.  We do have support for reversing IVs, but in general if we produce one of each, the pair will persist surprisingly far through the optimizer before being coalesced.  (You can see this looking at nearby phis in the test cases.)

Note that if the hardware prefers decrementing (i.e. zero tested) loops, LSR should convert back immediately before codegen.

* Mostly irrelevant detail: The main loop of the prolog case is handled independently and will simple use the original IV with a changed start value.  We could in theory use this scheme for all iteration clamping, but that's a larger and more invasive change.

diff --git a/llvm/lib/Transforms/Utils/LoopUnrollRuntime.cpp b/llvm/lib/Transforms/Utils/LoopUnrollRuntime.cpp
index 22d6d5c948a00103c504149eea7771d29a1d41ae..0cde2bcfbca1ac34fd41b58d4103e0b365a5de78 100644 (file)
--- a/llvm/lib/Transforms/Utils/LoopUnrollRuntime.cpp
+++ b/llvm/lib/Transforms/Utils/LoopUnrollRuntime.cpp
@@ -353,20 +353,22 @@ CloneLoopBlocks(Loop *L, Value *NewIter, const bool UseEpilogRemainder,
     if (Latch == *BB) {
       // For the last block, create a loop back to cloned head.
       VMap.erase((*BB)->getTerminator());
+      // Use an incrementing IV.  Pre-incr/post-incr is backedge/trip count.
+      // Subtle: NewIter can be 0 if we wrapped when computing the trip count,
+      // thus we must compare the post-increment (wrapping) value.
       BasicBlock *FirstLoopBB = cast<BasicBlock>(VMap[Header]);
       BranchInst *LatchBR = cast<BranchInst>(NewBB->getTerminator());
       IRBuilder<> Builder(LatchBR);
       PHINode *NewIdx = PHINode::Create(NewIter->getType(), 2,
                                         suffix + ".iter",
                                         FirstLoopBB->getFirstNonPHI());
-      Value *IdxSub =
-        Builder.CreateSub(NewIdx, ConstantInt::get(NewIdx->getType(), 1),
-                          NewIdx->getName() + ".sub");
-      Value *IdxCmp =
-        Builder.CreateIsNotNull(IdxSub, NewIdx->getName() + ".cmp");
+      auto *Zero = ConstantInt::get(NewIdx->getType(), 0);
+      auto *One = ConstantInt::get(NewIdx->getType(), 1);
+      Value *IdxNext = Builder.CreateAdd(NewIdx, One, NewIdx->getName() + ".next");
+      Value *IdxCmp = Builder.CreateICmpNE(IdxNext, NewIter, NewIdx->getName() + ".cmp");
       Builder.CreateCondBr(IdxCmp, FirstLoopBB, InsertBot);
-      NewIdx->addIncoming(NewIter, InsertTop);
-      NewIdx->addIncoming(IdxSub, NewBB);
+      NewIdx->addIncoming(Zero, InsertTop);
+      NewIdx->addIncoming(IdxNext, NewBB);
       LatchBR->eraseFromParent();
     }
   }
@@ -920,23 +922,22 @@ bool llvm::UnrollRuntimeLoopRemainder(
                   PreserveLCSSA);
 
     // Update counter in loop for unrolling.
-    // I should be multiply of Count.
+    // Use an incrementing IV.  Pre-incr/post-incr is backedge/trip count.
+    // Subtle: TestVal can be 0 if we wrapped when computing the trip count,
+    // thus we must compare the post-increment (wrapping) value.
     IRBuilder<> B2(NewPreHeader->getTerminator());
     Value *TestVal = B2.CreateSub(TripCount, ModVal, "unroll_iter");
     BranchInst *LatchBR = cast<BranchInst>(Latch->getTerminator());
-    B2.SetInsertPoint(LatchBR);
     PHINode *NewIdx = PHINode::Create(TestVal->getType(), 2, "niter",
                                       Header->getFirstNonPHI());
-    Value *IdxSub =
-        B2.CreateSub(NewIdx, ConstantInt::get(NewIdx->getType(), 1),
-                     NewIdx->getName() + ".nsub");
-    Value *IdxCmp;
-    if (LatchBR->getSuccessor(0) == Header)
-      IdxCmp = B2.CreateIsNotNull(IdxSub, NewIdx->getName() + ".ncmp");
-    else
-      IdxCmp = B2.CreateIsNull(IdxSub, NewIdx->getName() + ".ncmp");
-    NewIdx->addIncoming(TestVal, NewPreHeader);
-    NewIdx->addIncoming(IdxSub, Latch);
+    B2.SetInsertPoint(LatchBR);
+    auto *Zero = ConstantInt::get(NewIdx->getType(), 0);
+    auto *One = ConstantInt::get(NewIdx->getType(), 1);
+    Value *IdxNext = B2.CreateAdd(NewIdx, One, NewIdx->getName() + ".next");
+    auto Pred = LatchBR->getSuccessor(0) == Header ? ICmpInst::ICMP_NE : ICmpInst::ICMP_EQ;
+    Value *IdxCmp = B2.CreateICmp(Pred, IdxNext, TestVal, NewIdx->getName() + ".ncmp");
+    NewIdx->addIncoming(Zero, NewPreHeader);
+    NewIdx->addIncoming(IdxNext, Latch);
     LatchBR->setCondition(IdxCmp);
   } else {
     // Connect the prolog code to the original loop and update the

diff --git a/llvm/test/DebugInfo/unrolled-loop-remainder.ll b/llvm/test/DebugInfo/unrolled-loop-remainder.ll
index ba4ce1f409f69ee638557974feb9f80d56ba2156..a1dbce56ad52fffc6e724e16b1b340f2807c133f 100644 (file)
--- a/llvm/test/DebugInfo/unrolled-loop-remainder.ll
+++ b/llvm/test/DebugInfo/unrolled-loop-remainder.ll
@@ -35,8 +35,7 @@ define i32 @func_c() local_unnamed_addr #0 !dbg !14 {
 ; CHECK-NEXT:    [[CONV_PROL:%.*]] = sext i32 [[TMP5]] to i64, !dbg [[DBG28]]
 ; CHECK-NEXT:    [[TMP6:%.*]] = inttoptr i64 [[CONV_PROL]] to i32*, !dbg [[DBG28]]
 ; CHECK-NEXT:    [[ADD_PROL:%.*]] = add nsw i32 [[DOTPR]], 2, !dbg [[DBG29:![0-9]+]]
-; CHECK-NEXT:    [[PROL_ITER_SUB:%.*]] = sub i32 [[XTRAITER]], 1, !dbg [[DBG24]]
-; CHECK-NEXT:    [[PROL_ITER_CMP:%.*]] = icmp ne i32 [[PROL_ITER_SUB]], 0, !dbg [[DBG24]]
+; CHECK-NEXT:    [[PROL_ITER_CMP:%.*]] = icmp ne i32 1, [[XTRAITER]], !dbg [[DBG24]]
 ; CHECK-NEXT:    br i1 [[PROL_ITER_CMP]], label [[FOR_BODY_PROL_1:%.*]], label [[FOR_BODY_PROL_LOOPEXIT_UNR_LCSSA:%.*]], !dbg [[DBG24]]
 ; CHECK:       for.body.prol.1:
 ; CHECK-NEXT:    [[ARRAYIDX_PROL_1:%.*]] = getelementptr inbounds i32, i32* [[TMP6]], i64 1, !dbg [[DBG28]]
@@ -44,8 +43,7 @@ define i32 @func_c() local_unnamed_addr #0 !dbg !14 {
 ; CHECK-NEXT:    [[CONV_PROL_1:%.*]] = sext i32 [[TMP7]] to i64, !dbg [[DBG28]]
 ; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[CONV_PROL_1]] to i32*, !dbg [[DBG28]]
 ; CHECK-NEXT:    [[ADD_PROL_1:%.*]] = add nsw i32 [[ADD_PROL]], 2, !dbg [[DBG29]]
-; CHECK-NEXT:    [[PROL_ITER_SUB_1:%.*]] = sub i32 [[PROL_ITER_SUB]], 1, !dbg [[DBG24]]
-; CHECK-NEXT:    [[PROL_ITER_CMP_1:%.*]] = icmp ne i32 [[PROL_ITER_SUB_1]], 0, !dbg [[DBG24]]
+; 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:
 ; CHECK-NEXT:    [[ARRAYIDX_PROL_2:%.*]] = getelementptr inbounds i32, i32* [[TMP8]], i64 1, !dbg [[DBG28]]

diff --git a/llvm/test/Transforms/LoopUnroll/AArch64/runtime-unroll-generic.ll b/llvm/test/Transforms/LoopUnroll/AArch64/runtime-unroll-generic.ll
index 5c8f9ca016795d6991bb1e1f5c5678779a2aa72d..4f0c07343bda81f3e3fa30da9643adbd3297af24 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/AArch64/runtime-unroll-generic.ll
+++ b/llvm/test/Transforms/LoopUnroll/AArch64/runtime-unroll-generic.ll
@@ -20,7 +20,7 @@ define void @runtime_unroll_generic(i32 %arg_0, i32* %arg_1, i16* %arg_2, i16* %
 ; CHECK-A55-NEXT:    [[UNROLL_ITER:%.*]] = and i32 [[ARG_0]], -4
 ; CHECK-A55-NEXT:    br label [[FOR_BODY6:%.*]]
 ; CHECK-A55:       for.body6:
-; CHECK-A55-NEXT:    [[NITER:%.*]] = phi i32 [ [[UNROLL_ITER]], [[FOR_BODY6_PREHEADER_NEW]] ], [ [[NITER_NSUB_3:%.*]], [[FOR_BODY6]] ]
+; CHECK-A55-NEXT:    [[NITER:%.*]] = phi i32 [ 0, [[FOR_BODY6_PREHEADER_NEW]] ], [ [[NITER_NEXT_3:%.*]], [[FOR_BODY6]] ]
 ; CHECK-A55-NEXT:    [[TMP2:%.*]] = load i16, i16* [[ARRAYIDX10]], align 2
 ; CHECK-A55-NEXT:    [[CONV:%.*]] = sext i16 [[TMP2]] to i32
 ; CHECK-A55-NEXT:    [[TMP3:%.*]] = load i16, i16* [[ARRAYIDX14]], align 2
@@ -50,8 +50,8 @@ define void @runtime_unroll_generic(i32 %arg_0, i32* %arg_1, i16* %arg_2, i16* %
 ; CHECK-A55-NEXT:    [[MUL16_3:%.*]] = mul nsw i32 [[CONV15_3]], [[CONV_3]]
 ; CHECK-A55-NEXT:    [[ADD21_3:%.*]] = add nsw i32 [[MUL16_3]], [[ADD21_2]]
 ; CHECK-A55-NEXT:    store i32 [[ADD21_3]], i32* [[ARRAYIDX20]], align 4
-; CHECK-A55-NEXT:    [[NITER_NSUB_3]] = add i32 [[NITER]], -4
-; CHECK-A55-NEXT:    [[NITER_NCMP_3_NOT:%.*]] = icmp eq i32 [[NITER_NSUB_3]], 0
+; CHECK-A55-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER]], 4
+; CHECK-A55-NEXT:    [[NITER_NCMP_3_NOT:%.*]] = icmp eq i32 [[NITER_NEXT_3]], [[UNROLL_ITER]]
 ; CHECK-A55-NEXT:    br i1 [[NITER_NCMP_3_NOT]], label [[FOR_END_LOOPEXIT_UNR_LCSSA]], label [[FOR_BODY6]], !llvm.loop [[LOOP0:![0-9]+]]
 ; CHECK-A55:       for.end.loopexit.unr-lcssa:
 ; CHECK-A55-NEXT:    [[LCMP_MOD_NOT:%.*]] = icmp eq i32 [[XTRAITER]], 0

diff --git a/llvm/test/Transforms/LoopUnroll/ARM/multi-blocks.ll b/llvm/test/Transforms/LoopUnroll/ARM/multi-blocks.ll
index 8c4257698ab7c2ba18ca2f9c2dd9dc4231213e61..9af0327f1a0469bfadf3ae4cd09760adb4ec1f08 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/ARM/multi-blocks.ll
+++ b/llvm/test/Transforms/LoopUnroll/ARM/multi-blocks.ll
@@ -42,8 +42,7 @@ define void @test_three_blocks(i32* nocapture %Output,
 ; CHECK:       for.inc.epil:
 ; CHECK-NEXT:    [[TEMP_1_EPIL:%.*]] = phi i32 [ [[ADD_EPIL]], [[IF_THEN_EPIL]] ], [ [[TEMP_09_UNR]], [[FOR_BODY_EPIL]] ]
 ; CHECK-NEXT:    [[INC_EPIL:%.*]] = add nuw i32 [[J_010_UNR]], 1
-; CHECK-NEXT:    [[EPIL_ITER_SUB:%.*]] = sub i32 [[XTRAITER]], 1
-; CHECK-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i32 [[EPIL_ITER_SUB]], 0
+; CHECK-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i32 1, [[XTRAITER]]
 ; CHECK-NEXT:    br i1 [[EPIL_ITER_CMP]], label [[FOR_BODY_EPIL_1:%.*]], label [[FOR_COND_CLEANUP_LOOPEXIT_EPILOG_LCSSA:%.*]]
 ; CHECK:       for.body.epil.1:
 ; CHECK-NEXT:    [[ARRAYIDX_EPIL_1:%.*]] = getelementptr inbounds i32, i32* [[CONDITION]], i32 [[INC_EPIL]]
@@ -58,8 +57,7 @@ define void @test_three_blocks(i32* nocapture %Output,
 ; 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:    [[EPIL_ITER_SUB_1:%.*]] = sub i32 [[EPIL_ITER_SUB]], 1
-; CHECK-NEXT:    [[EPIL_ITER_CMP_1:%.*]] = icmp ne i32 [[EPIL_ITER_SUB_1]], 0
+; 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:
 ; CHECK-NEXT:    [[ARRAYIDX_EPIL_2:%.*]] = getelementptr inbounds i32, i32* [[CONDITION]], i32 [[INC_EPIL_1]]
@@ -87,7 +85,7 @@ define void @test_three_blocks(i32* nocapture %Output,
 ; CHECK:       for.body:
 ; CHECK-NEXT:    [[J_010:%.*]] = phi i32 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[INC_3]], [[FOR_INC_3]] ]
 ; CHECK-NEXT:    [[TEMP_09:%.*]] = phi i32 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[TEMP_1_3]], [[FOR_INC_3]] ]
-; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ [[UNROLL_ITER]], [[FOR_BODY_PREHEADER_NEW]] ], [ [[NITER_NSUB_3:%.*]], [[FOR_INC_3]] ]
+; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[NITER_NEXT_3:%.*]], [[FOR_INC_3]] ]
 ; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[CONDITION]], i32 [[J_010]]
 ; CHECK-NEXT:    [[TMP8:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
 ; CHECK-NEXT:    [[TOBOOL:%.*]] = icmp eq i32 [[TMP8]], 0
@@ -100,7 +98,7 @@ define void @test_three_blocks(i32* nocapture %Output,
 ; 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_NSUB:%.*]] = sub i32 [[NITER]], 1
+; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i32 [[NITER]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, i32* [[CONDITION]], i32 [[INC]]
 ; CHECK-NEXT:    [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX_1]], align 4
 ; CHECK-NEXT:    [[TOBOOL_1:%.*]] = icmp eq i32 [[TMP10]], 0
@@ -113,7 +111,7 @@ define void @test_three_blocks(i32* nocapture %Output,
 ; 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_NSUB_1:%.*]] = sub i32 [[NITER_NSUB]], 1
+; CHECK-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i32 [[NITER_NEXT]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, i32* [[CONDITION]], i32 [[INC_1]]
 ; CHECK-NEXT:    [[TMP12:%.*]] = load i32, i32* [[ARRAYIDX_2]], align 4
 ; CHECK-NEXT:    [[TOBOOL_2:%.*]] = icmp eq i32 [[TMP12]], 0
@@ -126,7 +124,7 @@ define void @test_three_blocks(i32* nocapture %Output,
 ; 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_NSUB_2:%.*]] = sub i32 [[NITER_NSUB_1]], 1
+; CHECK-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i32 [[NITER_NEXT_1]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, i32* [[CONDITION]], i32 [[INC_2]]
 ; CHECK-NEXT:    [[TMP14:%.*]] = load i32, i32* [[ARRAYIDX_3]], align 4
 ; CHECK-NEXT:    [[TOBOOL_3:%.*]] = icmp eq i32 [[TMP14]], 0
@@ -139,8 +137,8 @@ define void @test_three_blocks(i32* nocapture %Output,
 ; 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_NSUB_3]] = sub i32 [[NITER_NSUB_2]], 1
-; CHECK-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i32 [[NITER_NSUB_3]], 0
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER_NEXT_2]], 1
+; 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]]
 ;
   i32* nocapture readonly %Condition,
@@ -414,8 +412,7 @@ define void @test_four_blocks(i32* nocapture %Output,
 ; CHECK:       for.inc.epil:
 ; CHECK-NEXT:    [[TEMP_1_EPIL:%.*]] = phi i32 [ [[ADD_EPIL]], [[IF_THEN_EPIL]] ], [ [[SUB13_EPIL]], [[IF_ELSE_EPIL]] ]
 ; CHECK-NEXT:    [[INC_EPIL:%.*]] = add nuw i32 [[J_027_UNR]], 1
-; CHECK-NEXT:    [[EPIL_ITER_SUB:%.*]] = sub i32 [[XTRAITER]], 1
-; CHECK-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i32 [[EPIL_ITER_SUB]], 0
+; CHECK-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i32 1, [[XTRAITER]]
 ; CHECK-NEXT:    br i1 [[EPIL_ITER_CMP]], label [[FOR_BODY_EPIL_1:%.*]], label [[FOR_COND_CLEANUP_LOOPEXIT_EPILOG_LCSSA:%.*]]
 ; CHECK:       for.body.epil.1:
 ; CHECK-NEXT:    [[ARRAYIDX_EPIL_1:%.*]] = getelementptr inbounds i32, i32* [[CONDITION]], i32 [[INC_EPIL]]
@@ -440,8 +437,7 @@ define void @test_four_blocks(i32* nocapture %Output,
 ; 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:    [[EPIL_ITER_SUB_1:%.*]] = sub i32 [[EPIL_ITER_SUB]], 1
-; CHECK-NEXT:    [[EPIL_ITER_CMP_1:%.*]] = icmp ne i32 [[EPIL_ITER_SUB_1]], 0
+; 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:
 ; CHECK-NEXT:    [[ARRAYIDX_EPIL_2:%.*]] = getelementptr inbounds i32, i32* [[CONDITION]], i32 [[INC_EPIL_1]]
@@ -480,7 +476,7 @@ define void @test_four_blocks(i32* nocapture %Output,
 ; CHECK-NEXT:    [[TMP12:%.*]] = phi i32 [ [[DOTPRE]], [[FOR_BODY_LR_PH_NEW]] ], [ [[TMP23]], [[FOR_INC_3]] ]
 ; CHECK-NEXT:    [[J_027:%.*]] = phi i32 [ 1, [[FOR_BODY_LR_PH_NEW]] ], [ [[INC_3]], [[FOR_INC_3]] ]
 ; CHECK-NEXT:    [[TEMP_026:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH_NEW]] ], [ [[TEMP_1_3]], [[FOR_INC_3]] ]
-; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ [[UNROLL_ITER]], [[FOR_BODY_LR_PH_NEW]] ], [ [[NITER_NSUB_3:%.*]], [[FOR_INC_3]] ]
+; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH_NEW]] ], [ [[NITER_NEXT_3:%.*]], [[FOR_INC_3]] ]
 ; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[CONDITION]], i32 [[J_027]]
 ; CHECK-NEXT:    [[TMP13:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
 ; CHECK-NEXT:    [[CMP1:%.*]] = icmp ugt i32 [[TMP13]], 65535
@@ -503,7 +499,7 @@ define void @test_four_blocks(i32* nocapture %Output,
 ; 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_NSUB:%.*]] = sub i32 [[NITER]], 1
+; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i32 [[NITER]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, i32* [[CONDITION]], i32 [[INC]]
 ; CHECK-NEXT:    [[TMP16:%.*]] = load i32, i32* [[ARRAYIDX_1]], align 4
 ; CHECK-NEXT:    [[CMP1_1:%.*]] = icmp ugt i32 [[TMP16]], 65535
@@ -526,7 +522,7 @@ define void @test_four_blocks(i32* nocapture %Output,
 ; 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_NSUB_1:%.*]] = sub i32 [[NITER_NSUB]], 1
+; CHECK-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i32 [[NITER_NEXT]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, i32* [[CONDITION]], i32 [[INC_1]]
 ; CHECK-NEXT:    [[TMP19:%.*]] = load i32, i32* [[ARRAYIDX_2]], align 4
 ; CHECK-NEXT:    [[CMP1_2:%.*]] = icmp ugt i32 [[TMP19]], 65535
@@ -549,7 +545,7 @@ define void @test_four_blocks(i32* nocapture %Output,
 ; 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_NSUB_2:%.*]] = sub i32 [[NITER_NSUB_1]], 1
+; CHECK-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i32 [[NITER_NEXT_1]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, i32* [[CONDITION]], i32 [[INC_2]]
 ; CHECK-NEXT:    [[TMP22:%.*]] = load i32, i32* [[ARRAYIDX_3]], align 4
 ; CHECK-NEXT:    [[CMP1_3:%.*]] = icmp ugt i32 [[TMP22]], 65535
@@ -572,8 +568,8 @@ define void @test_four_blocks(i32* nocapture %Output,
 ; 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_NSUB_3]] = sub i32 [[NITER_NSUB_2]], 1
-; CHECK-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i32 [[NITER_NSUB_3]], 0
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER_NEXT_2]], 1
+; 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]]
 ;
   i32* nocapture readonly %Condition,

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 8c91a521c5279692c5e46ff5024a0a5c1d959587..7c2da16b807aac4cf9988275064759971be5264f 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
@@ -31,7 +31,7 @@ define i8* @f(i8* returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK:       vector.body:
 ; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH_NEW]] ], [ [[INDEX_NEXT_1:%.*]], [[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[VEC_IND12:%.*]] = phi <16 x i32> [ <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>, [[VECTOR_PH_NEW]] ], [ [[VEC_IND_NEXT13_1:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT:    [[NITER:%.*]] = phi i64 [ [[UNROLL_ITER]], [[VECTOR_PH_NEW]] ], [ [[NITER_NSUB_1:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[NITER:%.*]] = phi i64 [ 0, [[VECTOR_PH_NEW]] ], [ [[NITER_NEXT_1:%.*]], [[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[TMP4:%.*]] = 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_IND12]]
 ; CHECK-NEXT:    [[TMP5:%.*]] = and <16 x i32> [[TMP4]], [[BROADCAST_SPLAT]]
 ; CHECK-NEXT:    [[TMP6:%.*]] = icmp eq <16 x i32> [[TMP5]], zeroinitializer
@@ -41,7 +41,7 @@ define i8* @f(i8* returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    store <16 x i8> [[TMP7]], <16 x i8>* [[TMP9]], 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_NSUB:%.*]] = sub i64 [[NITER]], 1
+; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i64 [[NITER]], 1
 ; CHECK-NEXT:    [[TMP10:%.*]] = 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:    [[TMP11:%.*]] = and <16 x i32> [[TMP10]], [[BROADCAST_SPLAT]]
 ; CHECK-NEXT:    [[TMP12:%.*]] = icmp eq <16 x i32> [[TMP11]], zeroinitializer
@@ -51,8 +51,8 @@ define i8* @f(i8* returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    store <16 x i8> [[TMP13]], <16 x i8>* [[TMP15]], 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_NSUB_1]] = sub i64 [[NITER_NSUB]], 1
-; CHECK-NEXT:    [[NITER_NCMP_1:%.*]] = icmp eq i64 [[NITER_NSUB_1]], 0
+; CHECK-NEXT:    [[NITER_NEXT_1]] = add i64 [[NITER_NEXT]], 1
+; 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:
 ; CHECK-NEXT:    [[INDEX_UNR_PH:%.*]] = phi i64 [ [[INDEX_NEXT_1]], [[VECTOR_BODY]] ]
@@ -89,7 +89,7 @@ define i8* @f(i8* returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    br label [[FOR_BODY_PROL:%.*]]
 ; CHECK:       for.body.prol:
 ; CHECK-NEXT:    [[INDVARS_IV_PROL:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_PROL:%.*]], [[FOR_BODY_PROL]] ], [ [[INDVARS_IV_PH]], [[FOR_BODY_PROL_PREHEADER]] ]
-; CHECK-NEXT:    [[PROL_ITER:%.*]] = phi i64 [ [[XTRAITER1]], [[FOR_BODY_PROL_PREHEADER]] ], [ [[PROL_ITER_SUB:%.*]], [[FOR_BODY_PROL]] ]
+; CHECK-NEXT:    [[PROL_ITER:%.*]] = phi i64 [ 0, [[FOR_BODY_PROL_PREHEADER]] ], [ [[PROL_ITER_NEXT:%.*]], [[FOR_BODY_PROL]] ]
 ; CHECK-NEXT:    [[TMP25:%.*]] = trunc i64 [[INDVARS_IV_PROL]] to i32
 ; CHECK-NEXT:    [[SHL_PROL:%.*]] = shl i32 1, [[TMP25]]
 ; CHECK-NEXT:    [[AND_PROL:%.*]] = and i32 [[SHL_PROL]], [[X]]
@@ -99,8 +99,8 @@ define i8* @f(i8* returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    store i8 [[CONV_PROL]], i8* [[ARRAYIDX_PROL]], align 1
 ; CHECK-NEXT:    [[INDVARS_IV_NEXT_PROL]] = add nuw nsw i64 [[INDVARS_IV_PROL]], 1
 ; CHECK-NEXT:    [[EXITCOND_PROL:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT_PROL]], [[WIDE_TRIP_COUNT]]
-; CHECK-NEXT:    [[PROL_ITER_SUB]] = sub i64 [[PROL_ITER]], 1
-; CHECK-NEXT:    [[PROL_ITER_CMP:%.*]] = icmp ne i64 [[PROL_ITER_SUB]], 0
+; CHECK-NEXT:    [[PROL_ITER_NEXT]] = add i64 [[PROL_ITER]], 1
+; CHECK-NEXT:    [[PROL_ITER_CMP:%.*]] = icmp ne i64 [[PROL_ITER_NEXT]], [[XTRAITER1]]
 ; CHECK-NEXT:    br i1 [[PROL_ITER_CMP]], label [[FOR_BODY_PROL]], label [[FOR_BODY_PROL_LOOPEXIT_UNR_LCSSA:%.*]], !llvm.loop [[LOOP0:![0-9]+]]
 ; CHECK:       for.body.prol.loopexit.unr-lcssa:
 ; CHECK-NEXT:    [[INDVARS_IV_UNR_PH:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_PROL]], [[FOR_BODY_PROL]] ]

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 20f6d91b0de361eb0e00bd5732015ff7473c63d8..dc5e12d09044bfbfdd29c3617cf092d9b71fb7a0 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/PowerPC/p8-unrolling-legalize-vectors.ll
+++ b/llvm/test/Transforms/LoopUnroll/PowerPC/p8-unrolling-legalize-vectors.ll
@@ -31,7 +31,7 @@ define i8* @f(i8* returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK:       vector.body:
 ; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH_NEW]] ], [ [[INDEX_NEXT_1:%.*]], [[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[VEC_IND12:%.*]] = phi <16 x i32> [ <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>, [[VECTOR_PH_NEW]] ], [ [[VEC_IND_NEXT13_1:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT:    [[NITER:%.*]] = phi i64 [ [[UNROLL_ITER]], [[VECTOR_PH_NEW]] ], [ [[NITER_NSUB_1:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[NITER:%.*]] = phi i64 [ 0, [[VECTOR_PH_NEW]] ], [ [[NITER_NEXT_1:%.*]], [[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[TMP4:%.*]] = 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_IND12]]
 ; CHECK-NEXT:    [[TMP5:%.*]] = and <16 x i32> [[TMP4]], [[BROADCAST_SPLAT]]
 ; CHECK-NEXT:    [[TMP6:%.*]] = icmp eq <16 x i32> [[TMP5]], zeroinitializer
@@ -41,7 +41,7 @@ define i8* @f(i8* returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    store <16 x i8> [[TMP7]], <16 x i8>* [[TMP9]], 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_NSUB:%.*]] = sub i64 [[NITER]], 1
+; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i64 [[NITER]], 1
 ; CHECK-NEXT:    [[TMP10:%.*]] = 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:    [[TMP11:%.*]] = and <16 x i32> [[TMP10]], [[BROADCAST_SPLAT]]
 ; CHECK-NEXT:    [[TMP12:%.*]] = icmp eq <16 x i32> [[TMP11]], zeroinitializer
@@ -51,8 +51,8 @@ define i8* @f(i8* returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    store <16 x i8> [[TMP13]], <16 x i8>* [[TMP15]], 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_NSUB_1]] = sub i64 [[NITER_NSUB]], 1
-; CHECK-NEXT:    [[NITER_NCMP_1:%.*]] = icmp eq i64 [[NITER_NSUB_1]], 0
+; CHECK-NEXT:    [[NITER_NEXT_1]] = add i64 [[NITER_NEXT]], 1
+; 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:
 ; CHECK-NEXT:    [[INDEX_UNR_PH:%.*]] = phi i64 [ [[INDEX_NEXT_1]], [[VECTOR_BODY]] ]
@@ -89,7 +89,7 @@ define i8* @f(i8* returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    br label [[FOR_BODY_PROL:%.*]]
 ; CHECK:       for.body.prol:
 ; CHECK-NEXT:    [[INDVARS_IV_PROL:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_PROL:%.*]], [[FOR_BODY_PROL]] ], [ [[INDVARS_IV_PH]], [[FOR_BODY_PROL_PREHEADER]] ]
-; CHECK-NEXT:    [[PROL_ITER:%.*]] = phi i64 [ [[XTRAITER1]], [[FOR_BODY_PROL_PREHEADER]] ], [ [[PROL_ITER_SUB:%.*]], [[FOR_BODY_PROL]] ]
+; CHECK-NEXT:    [[PROL_ITER:%.*]] = phi i64 [ 0, [[FOR_BODY_PROL_PREHEADER]] ], [ [[PROL_ITER_NEXT:%.*]], [[FOR_BODY_PROL]] ]
 ; CHECK-NEXT:    [[TMP25:%.*]] = trunc i64 [[INDVARS_IV_PROL]] to i32
 ; CHECK-NEXT:    [[SHL_PROL:%.*]] = shl i32 1, [[TMP25]]
 ; CHECK-NEXT:    [[AND_PROL:%.*]] = and i32 [[SHL_PROL]], [[X]]
@@ -99,8 +99,8 @@ define i8* @f(i8* returned %s, i32 zeroext %x, i32 signext %k) local_unnamed_add
 ; CHECK-NEXT:    store i8 [[CONV_PROL]], i8* [[ARRAYIDX_PROL]], align 1
 ; CHECK-NEXT:    [[INDVARS_IV_NEXT_PROL]] = add nuw nsw i64 [[INDVARS_IV_PROL]], 1
 ; CHECK-NEXT:    [[EXITCOND_PROL:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT_PROL]], [[WIDE_TRIP_COUNT]]
-; CHECK-NEXT:    [[PROL_ITER_SUB]] = sub i64 [[PROL_ITER]], 1
-; CHECK-NEXT:    [[PROL_ITER_CMP:%.*]] = icmp ne i64 [[PROL_ITER_SUB]], 0
+; CHECK-NEXT:    [[PROL_ITER_NEXT]] = add i64 [[PROL_ITER]], 1
+; CHECK-NEXT:    [[PROL_ITER_CMP:%.*]] = icmp ne i64 [[PROL_ITER_NEXT]], [[XTRAITER1]]
 ; CHECK-NEXT:    br i1 [[PROL_ITER_CMP]], label [[FOR_BODY_PROL]], label [[FOR_BODY_PROL_LOOPEXIT_UNR_LCSSA:%.*]], !llvm.loop [[LOOP0:![0-9]+]]
 ; CHECK:       for.body.prol.loopexit.unr-lcssa:
 ; CHECK-NEXT:    [[INDVARS_IV_UNR_PH:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_PROL]], [[FOR_BODY_PROL]] ]

diff --git a/llvm/test/Transforms/LoopUnroll/WebAssembly/basic-unrolling.ll b/llvm/test/Transforms/LoopUnroll/WebAssembly/basic-unrolling.ll
index e36f0693d01508f36341b62d39532d05c2532a2d..2ef342b80b92915101f18773b1181812ecb3d628 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/WebAssembly/basic-unrolling.ll
+++ b/llvm/test/Transforms/LoopUnroll/WebAssembly/basic-unrolling.ll
@@ -145,7 +145,7 @@ define hidden void @runtime(i32* nocapture %a, i32* nocapture readonly %b, i32*
 ; CHECK-NEXT:    ret void
 ; CHECK:       for.body:
 ; CHECK-NEXT:    [[I_09:%.*]] = phi i32 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[INC_1]], [[FOR_BODY]] ]
-; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ [[UNROLL_ITER]], [[FOR_BODY_PREHEADER_NEW]] ], [ [[NITER_NSUB_1:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[NITER_NEXT_1:%.*]], [[FOR_BODY]] ]
 ; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[B]], i32 [[I_09]]
 ; CHECK-NEXT:    [[I:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
 ; CHECK-NEXT:    [[ARRAYIDX1:%.*]] = getelementptr inbounds i32, i32* [[C]], i32 [[I_09]]
@@ -162,8 +162,8 @@ define hidden void @runtime(i32* nocapture %a, i32* nocapture readonly %b, i32*
 ; CHECK-NEXT:    [[ARRAYIDX2_1:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[INC]]
 ; CHECK-NEXT:    store i32 [[MUL_1]], i32* [[ARRAYIDX2_1]], align 4
 ; CHECK-NEXT:    [[INC_1]] = add nuw i32 [[I_09]], 2
-; CHECK-NEXT:    [[NITER_NSUB_1]] = add i32 [[NITER]], -2
-; CHECK-NEXT:    [[NITER_NCMP_1:%.*]] = icmp eq i32 [[NITER_NSUB_1]], 0
+; CHECK-NEXT:    [[NITER_NEXT_1]] = add i32 [[NITER]], 2
+; CHECK-NEXT:    [[NITER_NCMP_1:%.*]] = icmp eq i32 [[NITER_NEXT_1]], [[UNROLL_ITER]]
 ; CHECK-NEXT:    br i1 [[NITER_NCMP_1]], label [[FOR_COND_CLEANUP_LOOPEXIT_UNR_LCSSA]], label [[FOR_BODY]]
 ;
 entry:

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 b9e81c4adefa1d203d59ee3d3ea653526ccc7213..a2e3693aadd806a8190c18b9c9e36ab4703e1bf6 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
@@ -15,7 +15,7 @@ define i32 @test(i32* nocapture %a, i64 %n) {
 ; ENABLED:       header:
 ; ENABLED-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY_NEW]] ], [ [[INDVARS_IV_NEXT_7:%.*]], [[FOR_BODY_7:%.*]] ]
 ; ENABLED-NEXT:    [[SUM_02:%.*]] = phi i32 [ 0, [[ENTRY_NEW]] ], [ [[ADD_7:%.*]], [[FOR_BODY_7]] ]
-; ENABLED-NEXT:    [[NITER:%.*]] = phi i64 [ [[UNROLL_ITER]], [[ENTRY_NEW]] ], [ [[NITER_NSUB_7:%.*]], [[FOR_BODY_7]] ]
+; ENABLED-NEXT:    [[NITER:%.*]] = phi i64 [ 0, [[ENTRY_NEW]] ], [ [[NITER_NEXT_7:%.*]], [[FOR_BODY_7]] ]
 ; ENABLED-NEXT:    [[CMP:%.*]] = icmp eq i64 [[N]], 42
 ; ENABLED-NEXT:    br i1 [[CMP]], label [[FOR_EXIT2_LOOPEXIT:%.*]], label [[FOR_BODY:%.*]]
 ; ENABLED:       for.body:
@@ -23,7 +23,7 @@ define i32 @test(i32* nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[TMP2:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
 ; ENABLED-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], [[SUM_02]]
 ; ENABLED-NEXT:    [[INDVARS_IV_NEXT:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 1
-; ENABLED-NEXT:    [[NITER_NSUB:%.*]] = sub i64 [[NITER]], 1
+; ENABLED-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i64 [[NITER]], 1
 ; ENABLED-NEXT:    [[CMP_1:%.*]] = icmp eq i64 [[N]], 42
 ; ENABLED-NEXT:    br i1 [[CMP_1]], label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_1:%.*]]
 ; ENABLED:       for.body.1:
@@ -31,7 +31,7 @@ define i32 @test(i32* nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[TMP3:%.*]] = load i32, i32* [[ARRAYIDX_1]], align 4
 ; ENABLED-NEXT:    [[ADD_1:%.*]] = add nsw i32 [[TMP3]], [[ADD]]
 ; ENABLED-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT]], 1
-; ENABLED-NEXT:    [[NITER_NSUB_1:%.*]] = sub i64 [[NITER_NSUB]], 1
+; ENABLED-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i64 [[NITER_NEXT]], 1
 ; 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:
@@ -39,7 +39,7 @@ define i32 @test(i32* nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[TMP4:%.*]] = load i32, i32* [[ARRAYIDX_2]], align 4
 ; ENABLED-NEXT:    [[ADD_2:%.*]] = add nsw i32 [[TMP4]], [[ADD_1]]
 ; ENABLED-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_1]], 1
-; ENABLED-NEXT:    [[NITER_NSUB_2:%.*]] = sub i64 [[NITER_NSUB_1]], 1
+; ENABLED-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i64 [[NITER_NEXT_1]], 1
 ; ENABLED-NEXT:    [[CMP_3:%.*]] = icmp eq i64 [[N]], 42
 ; ENABLED-NEXT:    br i1 [[CMP_3]], label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_3:%.*]]
 ; ENABLED:       for.body.3:
@@ -47,7 +47,7 @@ define i32 @test(i32* nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[TMP5:%.*]] = load i32, i32* [[ARRAYIDX_3]], align 4
 ; ENABLED-NEXT:    [[ADD_3:%.*]] = add nsw i32 [[TMP5]], [[ADD_2]]
 ; ENABLED-NEXT:    [[INDVARS_IV_NEXT_3:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_2]], 1
-; ENABLED-NEXT:    [[NITER_NSUB_3:%.*]] = sub i64 [[NITER_NSUB_2]], 1
+; ENABLED-NEXT:    [[NITER_NEXT_3:%.*]] = add nuw nsw i64 [[NITER_NEXT_2]], 1
 ; 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:
@@ -55,7 +55,7 @@ define i32 @test(i32* nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[TMP6:%.*]] = load i32, i32* [[ARRAYIDX_4]], align 4
 ; ENABLED-NEXT:    [[ADD_4:%.*]] = add nsw i32 [[TMP6]], [[ADD_3]]
 ; ENABLED-NEXT:    [[INDVARS_IV_NEXT_4:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_3]], 1
-; ENABLED-NEXT:    [[NITER_NSUB_4:%.*]] = sub i64 [[NITER_NSUB_3]], 1
+; ENABLED-NEXT:    [[NITER_NEXT_4:%.*]] = add nuw nsw i64 [[NITER_NEXT_3]], 1
 ; ENABLED-NEXT:    [[CMP_5:%.*]] = icmp eq i64 [[N]], 42
 ; ENABLED-NEXT:    br i1 [[CMP_5]], label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_5:%.*]]
 ; ENABLED:       for.body.5:
@@ -63,7 +63,7 @@ define i32 @test(i32* nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[TMP7:%.*]] = load i32, i32* [[ARRAYIDX_5]], align 4
 ; ENABLED-NEXT:    [[ADD_5:%.*]] = add nsw i32 [[TMP7]], [[ADD_4]]
 ; ENABLED-NEXT:    [[INDVARS_IV_NEXT_5:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_4]], 1
-; ENABLED-NEXT:    [[NITER_NSUB_5:%.*]] = sub i64 [[NITER_NSUB_4]], 1
+; ENABLED-NEXT:    [[NITER_NEXT_5:%.*]] = add nuw nsw i64 [[NITER_NEXT_4]], 1
 ; 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:
@@ -71,7 +71,7 @@ define i32 @test(i32* nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[TMP8:%.*]] = load i32, i32* [[ARRAYIDX_6]], align 4
 ; ENABLED-NEXT:    [[ADD_6:%.*]] = add nsw i32 [[TMP8]], [[ADD_5]]
 ; ENABLED-NEXT:    [[INDVARS_IV_NEXT_6:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_5]], 1
-; ENABLED-NEXT:    [[NITER_NSUB_6:%.*]] = sub i64 [[NITER_NSUB_5]], 1
+; ENABLED-NEXT:    [[NITER_NEXT_6:%.*]] = add nuw nsw i64 [[NITER_NEXT_5]], 1
 ; ENABLED-NEXT:    [[CMP_7:%.*]] = icmp eq i64 [[N]], 42
 ; ENABLED-NEXT:    br i1 [[CMP_7]], label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_7]]
 ; ENABLED:       for.body.7:
@@ -79,8 +79,8 @@ define i32 @test(i32* nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[TMP9:%.*]] = load i32, i32* [[ARRAYIDX_7]], align 4
 ; ENABLED-NEXT:    [[ADD_7]] = add nsw i32 [[TMP9]], [[ADD_6]]
 ; ENABLED-NEXT:    [[INDVARS_IV_NEXT_7]] = add i64 [[INDVARS_IV_NEXT_6]], 1
-; ENABLED-NEXT:    [[NITER_NSUB_7]] = sub i64 [[NITER_NSUB_6]], 1
-; ENABLED-NEXT:    [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NSUB_7]], 0
+; ENABLED-NEXT:    [[NITER_NEXT_7]] = add i64 [[NITER_NEXT_6]], 1
+; 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:
 ; ENABLED-NEXT:    [[SUM_0_LCSSA_PH_PH:%.*]] = phi i32 [ [[ADD_7]], [[FOR_BODY_7]] ]
@@ -98,7 +98,7 @@ define i32 @test(i32* nocapture %a, i64 %n) {
 ; ENABLED:       header.epil:
 ; ENABLED-NEXT:    [[INDVARS_IV_EPIL:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_EPIL:%.*]], [[FOR_BODY_EPIL:%.*]] ], [ [[INDVARS_IV_UNR]], [[HEADER_EPIL_PREHEADER]] ]
 ; ENABLED-NEXT:    [[SUM_02_EPIL:%.*]] = phi i32 [ [[ADD_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[SUM_02_UNR]], [[HEADER_EPIL_PREHEADER]] ]
-; ENABLED-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[XTRAITER]], [[HEADER_EPIL_PREHEADER]] ], [ [[EPIL_ITER_SUB:%.*]], [[FOR_BODY_EPIL]] ]
+; ENABLED-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ 0, [[HEADER_EPIL_PREHEADER]] ], [ [[EPIL_ITER_NEXT:%.*]], [[FOR_BODY_EPIL]] ]
 ; ENABLED-NEXT:    [[CMP_EPIL:%.*]] = icmp eq i64 [[N]], 42
 ; ENABLED-NEXT:    br i1 [[CMP_EPIL]], label [[FOR_EXIT2_LOOPEXIT2:%.*]], label [[FOR_BODY_EPIL]]
 ; ENABLED:       for.body.epil:
@@ -107,8 +107,8 @@ define i32 @test(i32* nocapture %a, i64 %n) {
 ; ENABLED-NEXT:    [[ADD_EPIL]] = add nsw i32 [[TMP10]], [[SUM_02_EPIL]]
 ; ENABLED-NEXT:    [[INDVARS_IV_NEXT_EPIL]] = add i64 [[INDVARS_IV_EPIL]], 1
 ; ENABLED-NEXT:    [[EXITCOND_EPIL:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT_EPIL]], [[N]]
-; ENABLED-NEXT:    [[EPIL_ITER_SUB]] = sub i64 [[EPIL_ITER]], 1
-; ENABLED-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i64 [[EPIL_ITER_SUB]], 0
+; ENABLED-NEXT:    [[EPIL_ITER_NEXT]] = add i64 [[EPIL_ITER]], 1
+; ENABLED-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i64 [[EPIL_ITER_NEXT]], [[XTRAITER]]
 ; ENABLED-NEXT:    br i1 [[EPIL_ITER_CMP]], label [[HEADER_EPIL]], label [[FOR_END_EPILOG_LCSSA:%.*]], !llvm.loop [[LOOP0:![0-9]+]]
 ; ENABLED:       for.end.epilog-lcssa:
 ; ENABLED-NEXT:    [[SUM_0_LCSSA_PH1:%.*]] = phi i32 [ [[ADD_EPIL]], [[FOR_BODY_EPIL]] ]

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 472c550a2278ab77e145c7bb7a63d1054691f02d..f8dab0fa2c3cbd967f8f689e65cf2ae44a51ad81 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/runtime-loop-multiexit-dom-verify.ll
+++ b/llvm/test/Transforms/LoopUnroll/runtime-loop-multiexit-dom-verify.ll
@@ -258,7 +258,7 @@ define void @test4(i16 %c3) {
 ; CHECK-NEXT:    br label [[HEADER_PROL:%.*]]
 ; CHECK:       header.prol:
 ; CHECK-NEXT:    [[INDVARS_IV_PROL:%.*]] = phi i64 [ 0, [[HEADER_PROL_PREHEADER]] ], [ [[INDVARS_IV_NEXT_PROL:%.*]], [[LATCH_PROL:%.*]] ]
-; CHECK-NEXT:    [[PROL_ITER:%.*]] = phi i64 [ [[XTRAITER]], [[HEADER_PROL_PREHEADER]] ], [ [[PROL_ITER_SUB:%.*]], [[LATCH_PROL]] ]
+; CHECK-NEXT:    [[PROL_ITER:%.*]] = phi i64 [ 0, [[HEADER_PROL_PREHEADER]] ], [ [[PROL_ITER_NEXT:%.*]], [[LATCH_PROL]] ]
 ; CHECK-NEXT:    br label [[EXITING_PROL:%.*]]
 ; CHECK:       exiting.prol:
 ; CHECK-NEXT:    switch i16 [[C3:%.*]], label [[DEFAULT_LOOPEXIT_LOOPEXIT1:%.*]] [
@@ -268,8 +268,8 @@ define void @test4(i16 %c3) {
 ; CHECK:       latch.prol:
 ; CHECK-NEXT:    [[INDVARS_IV_NEXT_PROL]] = add nuw nsw i64 [[INDVARS_IV_PROL]], 1
 ; CHECK-NEXT:    [[C2_PROL:%.*]] = icmp ult i64 [[INDVARS_IV_NEXT_PROL]], [[C1]]
-; CHECK-NEXT:    [[PROL_ITER_SUB]] = sub i64 [[PROL_ITER]], 1
-; CHECK-NEXT:    [[PROL_ITER_CMP:%.*]] = icmp ne i64 [[PROL_ITER_SUB]], 0
+; CHECK-NEXT:    [[PROL_ITER_NEXT]] = add i64 [[PROL_ITER]], 1
+; CHECK-NEXT:    [[PROL_ITER_CMP:%.*]] = icmp ne i64 [[PROL_ITER_NEXT]], [[XTRAITER]]
 ; CHECK-NEXT:    br i1 [[PROL_ITER_CMP]], label [[HEADER_PROL]], label [[HEADER_PROL_LOOPEXIT_UNR_LCSSA:%.*]], !llvm.loop [[LOOP4:![0-9]+]]
 ; CHECK:       header.prol.loopexit.unr-lcssa:
 ; CHECK-NEXT:    [[INDVARS_IV_UNR_PH:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_PROL]], [[LATCH_PROL]] ]
@@ -377,7 +377,7 @@ define void @test5(i1 %c) {
 ; CHECK-NEXT:    br label [[OUTERH_PROL:%.*]]
 ; CHECK:       outerH.prol:
 ; CHECK-NEXT:    [[TMP4_PROL:%.*]] = phi i32 [ [[TMP6_PROL:%.*]], [[OUTERLATCH_PROL:%.*]] ], [ undef, [[OUTERH_PROL_PREHEADER]] ]
-; CHECK-NEXT:    [[PROL_ITER:%.*]] = phi i32 [ 0, [[OUTERH_PROL_PREHEADER]] ], [ [[PROL_ITER_SUB:%.*]], [[OUTERLATCH_PROL]] ]
+; CHECK-NEXT:    [[PROL_ITER:%.*]] = phi i32 [ 0, [[OUTERH_PROL_PREHEADER]] ], [ [[PROL_ITER_NEXT:%.*]], [[OUTERLATCH_PROL]] ]
 ; CHECK-NEXT:    br label [[INNERH_PROL:%.*]]
 ; CHECK:       innerH.prol:
 ; CHECK-NEXT:    br i1 [[C:%.*]], label [[INNEREXITING_PROL:%.*]], label [[OTHEREXITB_LOOPEXIT1:%.*]]
@@ -406,8 +406,8 @@ define void @test5(i1 %c) {
 ; CHECK:       outerLatch.prol:
 ; CHECK-NEXT:    [[TMP6_PROL]] = add i32 [[TMP4_PROL]], 1
 ; CHECK-NEXT:    [[TMP7_PROL:%.*]] = icmp sgt i32 [[TMP6_PROL]], 79
-; CHECK-NEXT:    [[PROL_ITER_SUB]] = sub i32 [[PROL_ITER]], 1
-; CHECK-NEXT:    [[PROL_ITER_CMP:%.*]] = icmp ne i32 [[PROL_ITER_SUB]], 0
+; CHECK-NEXT:    [[PROL_ITER_NEXT]] = add i32 [[PROL_ITER]], 1
+; CHECK-NEXT:    [[PROL_ITER_CMP:%.*]] = icmp ne i32 [[PROL_ITER_NEXT]], 0
 ; CHECK-NEXT:    br i1 [[PROL_ITER_CMP]], label [[OUTERH_PROL]], label [[OUTERH_PROL_LOOPEXIT_UNR_LCSSA:%.*]], !llvm.loop [[LOOP7:![0-9]+]]
 ; CHECK:       outerH.prol.loopexit.unr-lcssa:
 ; CHECK-NEXT:    [[TMP4_UNR_PH:%.*]] = phi i32 [ [[TMP6_PROL]], [[OUTERLATCH_PROL]] ]
@@ -620,13 +620,13 @@ define void @test6(i1 %c) {
 ; CHECK-NEXT:    br label [[HEADER_PROL:%.*]]
 ; CHECK:       header.prol:
 ; CHECK-NEXT:    [[INDVARS_IV_PROL:%.*]] = phi i64 [ undef, [[HEADER_PROL_PREHEADER]] ], [ [[INDVARS_IV_NEXT_PROL:%.*]], [[LATCH_PROL:%.*]] ]
-; CHECK-NEXT:    [[PROL_ITER:%.*]] = phi i64 [ [[XTRAITER]], [[HEADER_PROL_PREHEADER]] ], [ [[PROL_ITER_SUB:%.*]], [[LATCH_PROL]] ]
+; CHECK-NEXT:    [[PROL_ITER:%.*]] = phi i64 [ 0, [[HEADER_PROL_PREHEADER]] ], [ [[PROL_ITER_NEXT:%.*]], [[LATCH_PROL]] ]
 ; CHECK-NEXT:    br i1 [[C:%.*]], label [[LATCH_PROL]], label [[OTHEREXIT_LOOPEXIT1:%.*]]
 ; CHECK:       latch.prol:
 ; CHECK-NEXT:    [[INDVARS_IV_NEXT_PROL]] = add nsw i64 [[INDVARS_IV_PROL]], 2
 ; CHECK-NEXT:    [[TMP4:%.*]] = icmp slt i64 [[INDVARS_IV_NEXT_PROL]], 616
-; CHECK-NEXT:    [[PROL_ITER_SUB]] = sub i64 [[PROL_ITER]], 1
-; CHECK-NEXT:    [[PROL_ITER_CMP:%.*]] = icmp ne i64 [[PROL_ITER_SUB]], 0
+; CHECK-NEXT:    [[PROL_ITER_NEXT]] = add i64 [[PROL_ITER]], 1
+; CHECK-NEXT:    [[PROL_ITER_CMP:%.*]] = icmp ne i64 [[PROL_ITER_NEXT]], [[XTRAITER]]
 ; CHECK-NEXT:    br i1 [[PROL_ITER_CMP]], label [[HEADER_PROL]], label [[HEADER_PROL_LOOPEXIT_UNR_LCSSA:%.*]], !llvm.loop [[LOOP9:![0-9]+]]
 ; CHECK:       header.prol.loopexit.unr-lcssa:
 ; CHECK-NEXT:    [[INDVARS_IV_UNR_PH:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_PROL]], [[LATCH_PROL]] ]

diff --git a/llvm/test/Transforms/LoopUnroll/runtime-loop-multiple-exits.ll b/llvm/test/Transforms/LoopUnroll/runtime-loop-multiple-exits.ll
index 1b6dd4cf2838c7e4e451a9eb0f15b1a4abf8ee2a..bd8265ca9936cdb5230461e8d7701c4986cd4a92 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/runtime-loop-multiple-exits.ll
+++ b/llvm/test/Transforms/LoopUnroll/runtime-loop-multiple-exits.ll
@@ -20,7 +20,7 @@ define void @test1(i64 %trip, i1 %cond) {
 ; EPILOG-NEXT:    br label %loop_header
 ; EPILOG:       loop_header:
 ; EPILOG-NEXT:    %iv = phi i64 [ 0, %entry.new ], [ %iv_next.7, %loop_latch.7 ]
-; EPILOG-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.7, %loop_latch.7 ]
+; EPILOG-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.7, %loop_latch.7 ]
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch, label %loop_exiting_bb1
 ; EPILOG:       loop_exiting_bb1:
 ; EPILOG-NEXT:    br i1 false, label %loop_exiting_bb2, label %exit1.loopexit
@@ -34,7 +34,7 @@ define void @test1(i64 %trip, i1 %cond) {
 ; EPILOG-NEXT:    ret void
 ; EPILOG:       loop_latch:
 ; EPILOG-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
-; EPILOG-NEXT:    %niter.nsub = sub i64 %niter, 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
@@ -42,7 +42,7 @@ define void @test1(i64 %trip, i1 %cond) {
 ; 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.nsub.1 = sub i64 %niter.nsub, 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
@@ -50,7 +50,7 @@ define void @test1(i64 %trip, i1 %cond) {
 ; 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.nsub.2 = sub i64 %niter.nsub.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
@@ -58,7 +58,7 @@ define void @test1(i64 %trip, i1 %cond) {
 ; 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.nsub.3 = sub i64 %niter.nsub.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
@@ -66,7 +66,7 @@ define void @test1(i64 %trip, i1 %cond) {
 ; 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.nsub.4 = sub i64 %niter.nsub.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
@@ -74,7 +74,7 @@ define void @test1(i64 %trip, i1 %cond) {
 ; 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.nsub.5 = sub i64 %niter.nsub.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
@@ -82,7 +82,7 @@ define void @test1(i64 %trip, i1 %cond) {
 ; 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.nsub.6 = sub i64 %niter.nsub.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
@@ -90,8 +90,8 @@ define void @test1(i64 %trip, i1 %cond) {
 ; 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.nsub.7 = sub i64 %niter.nsub.6, 1
-; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.nsub.7, 0
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; 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:
 ; EPILOG-NEXT:    br label %exit1
@@ -110,7 +110,7 @@ define void @test1(i64 %trip, i1 %cond) {
 ; EPILOG-NEXT:    br label %loop_header.epil
 ; EPILOG:       loop_header.epil:
 ; EPILOG-NEXT:    %iv.epil = phi i64 [ %iv.unr, %loop_header.epil.preheader ], [ %iv_next.epil, %loop_latch.epil ]
-; EPILOG-NEXT:    %epil.iter = phi i64 [ %xtraiter, %loop_header.epil.preheader ], [ %epil.iter.sub, %loop_latch.epil ]
+; EPILOG-NEXT:    %epil.iter = phi i64 [ 0, %loop_header.epil.preheader ], [ %epil.iter.next, %loop_latch.epil ]
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.epil, label %loop_exiting_bb1.epil
 ; EPILOG:       loop_exiting_bb1.epil:
 ; EPILOG-NEXT:    br i1 false, label %loop_exiting_bb2.epil, label %exit1.loopexit1
@@ -119,8 +119,8 @@ define void @test1(i64 %trip, i1 %cond) {
 ; EPILOG:       loop_latch.epil:
 ; EPILOG-NEXT:    %iv_next.epil = add i64 %iv.epil, 1
 ; EPILOG-NEXT:    %cmp.epil = icmp ne i64 %iv_next.epil, %trip
-; EPILOG-NEXT:    %epil.iter.sub = sub i64 %epil.iter, 1
-; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.sub, 0
+; EPILOG-NEXT:    %epil.iter.next = add i64 %epil.iter, 1
+; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.next, %xtraiter
 ; EPILOG-NEXT:    br i1 %epil.iter.cmp, label %loop_header.epil, label %exit2.loopexit.epilog-lcssa, !llvm.loop !0
 ; EPILOG:       exit2.loopexit.epilog-lcssa:
 ; EPILOG-NEXT:    br label %exit2.loopexit
@@ -138,7 +138,7 @@ define void @test1(i64 %trip, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    br label %loop_header
 ; EPILOG-BLOCK:       loop_header:
 ; EPILOG-BLOCK-NEXT:    %iv = phi i64 [ 0, %entry.new ], [ %iv_next.1, %loop_latch.1 ]
-; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.1, %loop_latch.1 ]
+; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.1, %loop_latch.1 ]
 ; EPILOG-BLOCK-NEXT:    br i1 %cond, label %loop_latch, label %loop_exiting_bb1
 ; EPILOG-BLOCK:       loop_exiting_bb1:
 ; EPILOG-BLOCK-NEXT:    br i1 false, label %loop_exiting_bb2, label %exit1.loopexit
@@ -150,7 +150,7 @@ define void @test1(i64 %trip, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    ret void
 ; EPILOG-BLOCK:       loop_latch:
 ; EPILOG-BLOCK-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
-; EPILOG-BLOCK-NEXT:    %niter.nsub = sub i64 %niter, 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
@@ -158,8 +158,8 @@ define void @test1(i64 %trip, i1 %cond) {
 ; 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.nsub.1 = sub i64 %niter.nsub, 1
-; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.nsub.1, 0
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; 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:
 ; EPILOG-BLOCK-NEXT:    br label %exit1
@@ -193,7 +193,7 @@ define void @test1(i64 %trip, i1 %cond) {
 ; PROLOG-NEXT:    br label %loop_header.prol
 ; PROLOG:       loop_header.prol:
 ; PROLOG-NEXT:    %iv.prol = phi i64 [ 0, %loop_header.prol.preheader ], [ %iv_next.prol, %loop_latch.prol ]
-; PROLOG-NEXT:    %prol.iter = phi i64 [ %xtraiter, %loop_header.prol.preheader ], [ %prol.iter.sub, %loop_latch.prol ]
+; PROLOG-NEXT:    %prol.iter = phi i64 [ 0, %loop_header.prol.preheader ], [ %prol.iter.next, %loop_latch.prol ]
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.prol, label %loop_exiting_bb1.prol
 ; PROLOG:       loop_exiting_bb1.prol:
 ; PROLOG-NEXT:    br i1 false, label %loop_exiting_bb2.prol, label %exit1.loopexit1
@@ -202,8 +202,8 @@ define void @test1(i64 %trip, i1 %cond) {
 ; PROLOG:       loop_latch.prol:
 ; PROLOG-NEXT:    %iv_next.prol = add i64 %iv.prol, 1
 ; PROLOG-NEXT:    %cmp.prol = icmp ne i64 %iv_next.prol, %trip
-; PROLOG-NEXT:    %prol.iter.sub = sub i64 %prol.iter, 1
-; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.sub, 0
+; PROLOG-NEXT:    %prol.iter.next = add i64 %prol.iter, 1
+; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.next, %xtraiter
 ; PROLOG-NEXT:    br i1 %prol.iter.cmp, label %loop_header.prol, label %loop_header.prol.loopexit.unr-lcssa, !llvm.loop !0
 ; PROLOG:       loop_header.prol.loopexit.unr-lcssa:
 ; PROLOG-NEXT:    %iv.unr.ph = phi i64 [ %iv_next.prol, %loop_latch.prol ]
@@ -392,7 +392,7 @@ define i32 @test2(i32* nocapture %a, i64 %n) {
 ; EPILOG:       header:
 ; EPILOG-NEXT:    %indvars.iv = phi i64 [ 0, %entry.new ], [ %indvars.iv.next.7, %for.body.7 ]
 ; EPILOG-NEXT:    %sum.02 = phi i32 [ 0, %entry.new ], [ %add.7, %for.body.7 ]
-; EPILOG-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.7, %for.body.7 ]
+; EPILOG-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.7, %for.body.7 ]
 ; EPILOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block
 ; EPILOG:       for.exiting_block:
 ; EPILOG-NEXT:    %cmp = icmp eq i64 %n, 42
@@ -402,7 +402,7 @@ define i32 @test2(i32* nocapture %a, i64 %n) {
 ; EPILOG-NEXT:    %2 = load i32, i32* %arrayidx, align 4
 ; EPILOG-NEXT:    %add = add nsw i32 %2, %sum.02
 ; EPILOG-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-NEXT:    %niter.nsub = sub i64 %niter, 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
@@ -412,7 +412,7 @@ define i32 @test2(i32* nocapture %a, i64 %n) {
 ; EPILOG-NEXT:    %3 = load i32, i32* %arrayidx.1, align 4
 ; EPILOG-NEXT:    %add.1 = add nsw i32 %3, %add
 ; EPILOG-NEXT:    %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv.next, 1
-; EPILOG-NEXT:    %niter.nsub.1 = sub i64 %niter.nsub, 1
+; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
 ; 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
@@ -422,7 +422,7 @@ define i32 @test2(i32* nocapture %a, i64 %n) {
 ; EPILOG-NEXT:    %4 = load i32, i32* %arrayidx.2, align 4
 ; EPILOG-NEXT:    %add.2 = add nsw i32 %4, %add.1
 ; EPILOG-NEXT:    %indvars.iv.next.2 = add nuw nsw i64 %indvars.iv.next.1, 1
-; EPILOG-NEXT:    %niter.nsub.2 = sub i64 %niter.nsub.1, 1
+; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
 ; 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
@@ -432,7 +432,7 @@ define i32 @test2(i32* nocapture %a, i64 %n) {
 ; EPILOG-NEXT:    %5 = load i32, i32* %arrayidx.3, align 4
 ; EPILOG-NEXT:    %add.3 = add nsw i32 %5, %add.2
 ; EPILOG-NEXT:    %indvars.iv.next.3 = add nuw nsw i64 %indvars.iv.next.2, 1
-; EPILOG-NEXT:    %niter.nsub.3 = sub i64 %niter.nsub.2, 1
+; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
 ; 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
@@ -442,7 +442,7 @@ define i32 @test2(i32* nocapture %a, i64 %n) {
 ; EPILOG-NEXT:    %6 = load i32, i32* %arrayidx.4, align 4
 ; EPILOG-NEXT:    %add.4 = add nsw i32 %6, %add.3
 ; EPILOG-NEXT:    %indvars.iv.next.4 = add nuw nsw i64 %indvars.iv.next.3, 1
-; EPILOG-NEXT:    %niter.nsub.4 = sub i64 %niter.nsub.3, 1
+; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
 ; 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
@@ -452,7 +452,7 @@ define i32 @test2(i32* nocapture %a, i64 %n) {
 ; EPILOG-NEXT:    %7 = load i32, i32* %arrayidx.5, align 4
 ; EPILOG-NEXT:    %add.5 = add nsw i32 %7, %add.4
 ; EPILOG-NEXT:    %indvars.iv.next.5 = add nuw nsw i64 %indvars.iv.next.4, 1
-; EPILOG-NEXT:    %niter.nsub.5 = sub i64 %niter.nsub.4, 1
+; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
 ; 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
@@ -462,7 +462,7 @@ define i32 @test2(i32* nocapture %a, i64 %n) {
 ; EPILOG-NEXT:    %8 = load i32, i32* %arrayidx.6, align 4
 ; EPILOG-NEXT:    %add.6 = add nsw i32 %8, %add.5
 ; EPILOG-NEXT:    %indvars.iv.next.6 = add nuw nsw i64 %indvars.iv.next.5, 1
-; EPILOG-NEXT:    %niter.nsub.6 = sub i64 %niter.nsub.5, 1
+; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
 ; 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
@@ -472,8 +472,8 @@ define i32 @test2(i32* nocapture %a, i64 %n) {
 ; EPILOG-NEXT:    %9 = load i32, i32* %arrayidx.7, align 4
 ; EPILOG-NEXT:    %add.7 = add nsw i32 %9, %add.6
 ; EPILOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv.next.6, 1
-; EPILOG-NEXT:    %niter.nsub.7 = sub i64 %niter.nsub.6, 1
-; EPILOG-NEXT:    %niter.ncmp.7 = icmp eq i64 %niter.nsub.7, 0
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; 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:
 ; EPILOG-NEXT:    %sum.0.lcssa.ph.ph = phi i32 [ %add.7, %for.body.7 ]
@@ -491,7 +491,7 @@ define i32 @test2(i32* nocapture %a, i64 %n) {
 ; EPILOG:       header.epil:
 ; EPILOG-NEXT:    %indvars.iv.epil = phi i64 [ %indvars.iv.next.epil, %for.body.epil ], [ %indvars.iv.unr, %header.epil.preheader ]
 ; EPILOG-NEXT:    %sum.02.epil = phi i32 [ %add.epil, %for.body.epil ], [ %sum.02.unr, %header.epil.preheader ]
-; EPILOG-NEXT:    %epil.iter = phi i64 [ %xtraiter, %header.epil.preheader ], [ %epil.iter.sub, %for.body.epil ]
+; EPILOG-NEXT:    %epil.iter = phi i64 [ 0, %header.epil.preheader ], [ %epil.iter.next, %for.body.epil ]
 ; EPILOG-NEXT:    br i1 false, label %for.exit2.loopexit2, label %for.exiting_block.epil
 ; EPILOG:       for.exiting_block.epil:
 ; EPILOG-NEXT:    %cmp.epil = icmp eq i64 %n, 42
@@ -502,8 +502,8 @@ define i32 @test2(i32* nocapture %a, i64 %n) {
 ; EPILOG-NEXT:    %add.epil = add nsw i32 %10, %sum.02.epil
 ; EPILOG-NEXT:    %indvars.iv.next.epil = add i64 %indvars.iv.epil, 1
 ; EPILOG-NEXT:    %exitcond.epil = icmp eq i64 %indvars.iv.next.epil, %n
-; EPILOG-NEXT:    %epil.iter.sub = sub i64 %epil.iter, 1
-; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.sub, 0
+; EPILOG-NEXT:    %epil.iter.next = add i64 %epil.iter, 1
+; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.next, %xtraiter
 ; EPILOG-NEXT:    br i1 %epil.iter.cmp, label %header.epil, label %for.end.epilog-lcssa, !llvm.loop !2
 ; EPILOG:       for.end.epilog-lcssa:
 ; EPILOG-NEXT:    %sum.0.lcssa.ph1 = phi i32 [ %add.epil, %for.body.epil ]
@@ -533,7 +533,7 @@ define i32 @test2(i32* nocapture %a, i64 %n) {
 ; EPILOG-BLOCK:       header:
 ; EPILOG-BLOCK-NEXT:    %indvars.iv = phi i64 [ 0, %entry.new ], [ %indvars.iv.next.1, %for.body.1 ]
 ; EPILOG-BLOCK-NEXT:    %sum.02 = phi i32 [ 0, %entry.new ], [ %add.1, %for.body.1 ]
-; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.1, %for.body.1 ]
+; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.1, %for.body.1 ]
 ; EPILOG-BLOCK-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block
 ; EPILOG-BLOCK:       for.exiting_block:
 ; EPILOG-BLOCK-NEXT:    %cmp = icmp eq i64 %n, 42
@@ -543,7 +543,7 @@ define i32 @test2(i32* nocapture %a, i64 %n) {
 ; EPILOG-BLOCK-NEXT:    %2 = load i32, i32* %arrayidx, align 4
 ; EPILOG-BLOCK-NEXT:    %add = add nsw i32 %2, %sum.02
 ; EPILOG-BLOCK-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-BLOCK-NEXT:    %niter.nsub = sub i64 %niter, 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
@@ -553,8 +553,8 @@ define i32 @test2(i32* nocapture %a, i64 %n) {
 ; EPILOG-BLOCK-NEXT:    %3 = load i32, i32* %arrayidx.1, align 4
 ; EPILOG-BLOCK-NEXT:    %add.1 = add nsw i32 %3, %add
 ; EPILOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
-; EPILOG-BLOCK-NEXT:    %niter.nsub.1 = sub i64 %niter.nsub, 1
-; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp eq i64 %niter.nsub.1, 0
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; 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:
 ; EPILOG-BLOCK-NEXT:    %sum.0.lcssa.ph.ph = phi i32 [ %add.1, %for.body.1 ]
@@ -600,7 +600,7 @@ define i32 @test2(i32* nocapture %a, i64 %n) {
 ; PROLOG:       header.prol:
 ; PROLOG-NEXT:    %indvars.iv.prol = phi i64 [ %indvars.iv.next.prol, %for.body.prol ], [ 0, %header.prol.preheader ]
 ; PROLOG-NEXT:    %sum.02.prol = phi i32 [ %add.prol, %for.body.prol ], [ 0, %header.prol.preheader ]
-; PROLOG-NEXT:    %prol.iter = phi i64 [ %xtraiter, %header.prol.preheader ], [ %prol.iter.sub, %for.body.prol ]
+; PROLOG-NEXT:    %prol.iter = phi i64 [ 0, %header.prol.preheader ], [ %prol.iter.next, %for.body.prol ]
 ; PROLOG-NEXT:    br i1 false, label %for.exit2.loopexit1, label %for.exiting_block.prol
 ; PROLOG:       for.exiting_block.prol:
 ; PROLOG-NEXT:    %cmp.prol = icmp eq i64 %n, 42
@@ -611,8 +611,8 @@ define i32 @test2(i32* nocapture %a, i64 %n) {
 ; PROLOG-NEXT:    %add.prol = add nsw i32 %1, %sum.02.prol
 ; PROLOG-NEXT:    %indvars.iv.next.prol = add i64 %indvars.iv.prol, 1
 ; PROLOG-NEXT:    %exitcond.prol = icmp eq i64 %indvars.iv.next.prol, %n
-; PROLOG-NEXT:    %prol.iter.sub = sub i64 %prol.iter, 1
-; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.sub, 0
+; PROLOG-NEXT:    %prol.iter.next = add i64 %prol.iter, 1
+; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.next, %xtraiter
 ; PROLOG-NEXT:    br i1 %prol.iter.cmp, label %header.prol, label %header.prol.loopexit.unr-lcssa, !llvm.loop !2
 ; PROLOG:       header.prol.loopexit.unr-lcssa:
 ; PROLOG-NEXT:    %sum.0.lcssa.unr.ph = phi i32 [ %add.prol, %for.body.prol ]
@@ -826,7 +826,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; EPILOG:       loop_header:
 ; EPILOG-NEXT:    %iv = phi i64 [ 0, %entry.new ], [ %iv_next.7, %loop_latch.7 ]
 ; EPILOG-NEXT:    %sum = phi i64 [ 0, %entry.new ], [ %sum.next.7, %loop_latch.7 ]
-; EPILOG-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.7, %loop_latch.7 ]
+; EPILOG-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.7, %loop_latch.7 ]
 ; EPILOG-NEXT:    br i1 undef, label %loop_latch, label %loop_exiting_bb1
 ; EPILOG:       loop_exiting_bb1:
 ; EPILOG-NEXT:    switch i64 %sum, label %loop_latch [
@@ -842,7 +842,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; EPILOG:       loop_latch:
 ; EPILOG-NEXT:    %iv_next = add nuw nsw i64 %iv, 1
 ; EPILOG-NEXT:    %sum.next = add i64 %sum, %add
-; EPILOG-NEXT:    %niter.nsub = sub i64 %niter, 1
+; 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 [
@@ -852,7 +852,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; 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.nsub.1 = sub i64 %niter.nsub, 1
+; 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 [
@@ -862,7 +862,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; 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.nsub.2 = sub i64 %niter.nsub.1, 1
+; 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 [
@@ -872,7 +872,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; 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.nsub.3 = sub i64 %niter.nsub.2, 1
+; 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 [
@@ -882,7 +882,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; 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.nsub.4 = sub i64 %niter.nsub.3, 1
+; 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 [
@@ -892,7 +892,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; 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.nsub.5 = sub i64 %niter.nsub.4, 1
+; 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 [
@@ -902,7 +902,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; 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.nsub.6 = sub i64 %niter.nsub.5, 1
+; 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 [
@@ -912,8 +912,8 @@ define void @test3(i64 %trip, i64 %add) {
 ; EPILOG:       loop_latch.7:
 ; EPILOG-NEXT:    %iv_next.7 = add nuw nsw i64 %iv_next.6, 1
 ; EPILOG-NEXT:    %sum.next.7 = add i64 %sum.next.6, %add
-; EPILOG-NEXT:    %niter.nsub.7 = sub i64 %niter.nsub.6, 1
-; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.nsub.7, 0
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; 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:
 ; EPILOG-NEXT:    br label %exit1
@@ -935,7 +935,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; EPILOG:       loop_header.epil:
 ; EPILOG-NEXT:    %iv.epil = phi i64 [ %iv.unr, %loop_header.epil.preheader ], [ %iv_next.epil, %loop_latch.epil ]
 ; EPILOG-NEXT:    %sum.epil = phi i64 [ %sum.unr, %loop_header.epil.preheader ], [ %sum.next.epil, %loop_latch.epil ]
-; EPILOG-NEXT:    %epil.iter = phi i64 [ %xtraiter, %loop_header.epil.preheader ], [ %epil.iter.sub, %loop_latch.epil ]
+; EPILOG-NEXT:    %epil.iter = phi i64 [ 0, %loop_header.epil.preheader ], [ %epil.iter.next, %loop_latch.epil ]
 ; EPILOG-NEXT:    br i1 undef, label %loop_latch.epil, label %loop_exiting_bb1.epil
 ; EPILOG:       loop_exiting_bb1.epil:
 ; EPILOG-NEXT:    switch i64 %sum.epil, label %loop_latch.epil [
@@ -946,8 +946,8 @@ define void @test3(i64 %trip, i64 %add) {
 ; EPILOG-NEXT:    %iv_next.epil = add nuw nsw i64 %iv.epil, 1
 ; EPILOG-NEXT:    %sum.next.epil = add i64 %sum.epil, %add
 ; EPILOG-NEXT:    %cmp.epil = icmp ne i64 %iv_next.epil, %trip
-; EPILOG-NEXT:    %epil.iter.sub = sub i64 %epil.iter, 1
-; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.sub, 0
+; EPILOG-NEXT:    %epil.iter.next = add i64 %epil.iter, 1
+; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.next, %xtraiter
 ; EPILOG-NEXT:    br i1 %epil.iter.cmp, label %loop_header.epil, label %exit2.loopexit.epilog-lcssa, !llvm.loop !3
 ; EPILOG:       exit2.loopexit.epilog-lcssa:
 ; EPILOG-NEXT:    br label %exit2.loopexit
@@ -966,7 +966,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; EPILOG-BLOCK:       loop_header:
 ; EPILOG-BLOCK-NEXT:    %iv = phi i64 [ 0, %entry.new ], [ %iv_next.1, %loop_latch.1 ]
 ; EPILOG-BLOCK-NEXT:    %sum = phi i64 [ 0, %entry.new ], [ %sum.next.1, %loop_latch.1 ]
-; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.1, %loop_latch.1 ]
+; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.1, %loop_latch.1 ]
 ; EPILOG-BLOCK-NEXT:    br i1 undef, label %loop_latch, label %loop_exiting_bb1
 ; EPILOG-BLOCK:       loop_exiting_bb1:
 ; EPILOG-BLOCK-NEXT:    switch i64 %sum, label %loop_latch [
@@ -980,7 +980,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; 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.nsub = sub i64 %niter, 1
+; 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 [
@@ -990,8 +990,8 @@ define void @test3(i64 %trip, i64 %add) {
 ; EPILOG-BLOCK:       loop_latch.1:
 ; EPILOG-BLOCK-NEXT:    %iv_next.1 = add nuw nsw i64 %iv_next, 1
 ; EPILOG-BLOCK-NEXT:    %sum.next.1 = add i64 %sum.next, %add
-; EPILOG-BLOCK-NEXT:    %niter.nsub.1 = sub i64 %niter.nsub, 1
-; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.nsub.1, 0
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; 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:
 ; EPILOG-BLOCK-NEXT:    br label %exit1
@@ -1029,7 +1029,7 @@ define void @test3(i64 %trip, i64 %add) {
 ; PROLOG:       loop_header.prol:
 ; PROLOG-NEXT:    %iv.prol = phi i64 [ 0, %loop_header.prol.preheader ], [ %iv_next.prol, %loop_latch.prol ]
 ; PROLOG-NEXT:    %sum.prol = phi i64 [ 0, %loop_header.prol.preheader ], [ %sum.next.prol, %loop_latch.prol ]
-; PROLOG-NEXT:    %prol.iter = phi i64 [ %xtraiter, %loop_header.prol.preheader ], [ %prol.iter.sub, %loop_latch.prol ]
+; PROLOG-NEXT:    %prol.iter = phi i64 [ 0, %loop_header.prol.preheader ], [ %prol.iter.next, %loop_latch.prol ]
 ; PROLOG-NEXT:    br i1 undef, label %loop_latch.prol, label %loop_exiting_bb1.prol
 ; PROLOG:       loop_exiting_bb1.prol:
 ; PROLOG-NEXT:    switch i64 %sum.prol, label %loop_latch.prol [
@@ -1040,8 +1040,8 @@ define void @test3(i64 %trip, i64 %add) {
 ; PROLOG-NEXT:    %iv_next.prol = add nuw nsw i64 %iv.prol, 1
 ; PROLOG-NEXT:    %sum.next.prol = add i64 %sum.prol, %add
 ; PROLOG-NEXT:    %cmp.prol = icmp ne i64 %iv_next.prol, %trip
-; PROLOG-NEXT:    %prol.iter.sub = sub i64 %prol.iter, 1
-; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.sub, 0
+; PROLOG-NEXT:    %prol.iter.next = add i64 %prol.iter, 1
+; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.next, %xtraiter
 ; PROLOG-NEXT:    br i1 %prol.iter.cmp, label %loop_header.prol, label %loop_header.prol.loopexit.unr-lcssa, !llvm.loop !3
 ; PROLOG:       loop_header.prol.loopexit.unr-lcssa:
 ; PROLOG-NEXT:    %iv.unr.ph = phi i64 [ %iv_next.prol, %loop_latch.prol ]
@@ -1254,7 +1254,7 @@ define i32 @hdr_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG:       header:
 ; EPILOG-NEXT:    %indvars.iv = phi i64 [ 0, %entry.new ], [ %indvars.iv.next.7, %latch.7 ]
 ; EPILOG-NEXT:    %sum.02 = phi i32 [ 0, %entry.new ], [ %add.7, %latch.7 ]
-; EPILOG-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.7, %latch.7 ]
+; EPILOG-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.7, %latch.7 ]
 ; EPILOG-NEXT:    br i1 %cond, label %latchExit.epilog-lcssa.loopexit, label %for.exiting_block
 ; EPILOG:       for.exiting_block:
 ; EPILOG-NEXT:    %cmp = icmp eq i64 %n, 42
@@ -1264,7 +1264,7 @@ define i32 @hdr_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %2 = load i32, i32* %arrayidx, align 4
 ; EPILOG-NEXT:    %add = add nsw i32 %2, %sum.02
 ; EPILOG-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-NEXT:    %niter.nsub = sub i64 %niter, 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
@@ -1274,7 +1274,7 @@ define i32 @hdr_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %3 = load i32, i32* %arrayidx.1, align 4
 ; EPILOG-NEXT:    %add.1 = add nsw i32 %3, %add
 ; EPILOG-NEXT:    %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv.next, 1
-; EPILOG-NEXT:    %niter.nsub.1 = sub i64 %niter.nsub, 1
+; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
 ; 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
@@ -1284,7 +1284,7 @@ define i32 @hdr_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %4 = load i32, i32* %arrayidx.2, align 4
 ; EPILOG-NEXT:    %add.2 = add nsw i32 %4, %add.1
 ; EPILOG-NEXT:    %indvars.iv.next.2 = add nuw nsw i64 %indvars.iv.next.1, 1
-; EPILOG-NEXT:    %niter.nsub.2 = sub i64 %niter.nsub.1, 1
+; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
 ; 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
@@ -1294,7 +1294,7 @@ define i32 @hdr_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %5 = load i32, i32* %arrayidx.3, align 4
 ; EPILOG-NEXT:    %add.3 = add nsw i32 %5, %add.2
 ; EPILOG-NEXT:    %indvars.iv.next.3 = add nuw nsw i64 %indvars.iv.next.2, 1
-; EPILOG-NEXT:    %niter.nsub.3 = sub i64 %niter.nsub.2, 1
+; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
 ; 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
@@ -1304,7 +1304,7 @@ define i32 @hdr_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %6 = load i32, i32* %arrayidx.4, align 4
 ; EPILOG-NEXT:    %add.4 = add nsw i32 %6, %add.3
 ; EPILOG-NEXT:    %indvars.iv.next.4 = add nuw nsw i64 %indvars.iv.next.3, 1
-; EPILOG-NEXT:    %niter.nsub.4 = sub i64 %niter.nsub.3, 1
+; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
 ; 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
@@ -1314,7 +1314,7 @@ define i32 @hdr_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %7 = load i32, i32* %arrayidx.5, align 4
 ; EPILOG-NEXT:    %add.5 = add nsw i32 %7, %add.4
 ; EPILOG-NEXT:    %indvars.iv.next.5 = add nuw nsw i64 %indvars.iv.next.4, 1
-; EPILOG-NEXT:    %niter.nsub.5 = sub i64 %niter.nsub.4, 1
+; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
 ; 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
@@ -1324,7 +1324,7 @@ define i32 @hdr_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %8 = load i32, i32* %arrayidx.6, align 4
 ; EPILOG-NEXT:    %add.6 = add nsw i32 %8, %add.5
 ; EPILOG-NEXT:    %indvars.iv.next.6 = add nuw nsw i64 %indvars.iv.next.5, 1
-; EPILOG-NEXT:    %niter.nsub.6 = sub i64 %niter.nsub.5, 1
+; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
 ; 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
@@ -1334,8 +1334,8 @@ define i32 @hdr_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %9 = load i32, i32* %arrayidx.7, align 4
 ; EPILOG-NEXT:    %add.7 = add nsw i32 %9, %add.6
 ; EPILOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv.next.6, 1
-; EPILOG-NEXT:    %niter.nsub.7 = sub i64 %niter.nsub.6, 1
-; EPILOG-NEXT:    %niter.ncmp.7 = icmp eq i64 %niter.nsub.7, 0
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; 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:
 ; EPILOG-NEXT:    %result.ph.ph = phi i32 [ %add.7, %latch.7 ]
@@ -1353,7 +1353,7 @@ define i32 @hdr_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG:       header.epil:
 ; EPILOG-NEXT:    %indvars.iv.epil = phi i64 [ %indvars.iv.next.epil, %latch.epil ], [ %indvars.iv.unr, %header.epil.preheader ]
 ; EPILOG-NEXT:    %sum.02.epil = phi i32 [ %add.epil, %latch.epil ], [ %sum.02.unr, %header.epil.preheader ]
-; EPILOG-NEXT:    %epil.iter = phi i64 [ %xtraiter, %header.epil.preheader ], [ %epil.iter.sub, %latch.epil ]
+; EPILOG-NEXT:    %epil.iter = phi i64 [ 0, %header.epil.preheader ], [ %epil.iter.next, %latch.epil ]
 ; EPILOG-NEXT:    br i1 %cond, label %latchExit.epilog-lcssa.loopexit2, label %for.exiting_block.epil
 ; EPILOG:       for.exiting_block.epil:
 ; EPILOG-NEXT:    %cmp.epil = icmp eq i64 %n, 42
@@ -1364,8 +1364,8 @@ define i32 @hdr_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %add.epil = add nsw i32 %10, %sum.02.epil
 ; EPILOG-NEXT:    %indvars.iv.next.epil = add i64 %indvars.iv.epil, 1
 ; EPILOG-NEXT:    %exitcond.epil = icmp eq i64 %indvars.iv.next.epil, %n
-; EPILOG-NEXT:    %epil.iter.sub = sub i64 %epil.iter, 1
-; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.sub, 0
+; EPILOG-NEXT:    %epil.iter.next = add i64 %epil.iter, 1
+; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.next, %xtraiter
 ; EPILOG-NEXT:    br i1 %epil.iter.cmp, label %header.epil, label %latchExit.epilog-lcssa.loopexit2, !llvm.loop !4
 ; EPILOG:       latchExit.epilog-lcssa.loopexit:
 ; EPILOG-NEXT:    %result.ph1.ph = phi i32 [ 0, %header ], [ 0, %latch ], [ 0, %latch.1 ], [ 0, %latch.2 ], [ 0, %latch.3 ], [ 0, %latch.4 ], [ 0, %latch.5 ], [ 0, %latch.6 ]
@@ -1398,7 +1398,7 @@ define i32 @hdr_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-BLOCK:       header:
 ; EPILOG-BLOCK-NEXT:    %indvars.iv = phi i64 [ 0, %entry.new ], [ %indvars.iv.next.1, %latch.1 ]
 ; EPILOG-BLOCK-NEXT:    %sum.02 = phi i32 [ 0, %entry.new ], [ %add.1, %latch.1 ]
-; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.1, %latch.1 ]
+; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.1, %latch.1 ]
 ; EPILOG-BLOCK-NEXT:    br i1 %cond, label %latchExit.epilog-lcssa.loopexit, label %for.exiting_block
 ; EPILOG-BLOCK:       for.exiting_block:
 ; EPILOG-BLOCK-NEXT:    %cmp = icmp eq i64 %n, 42
@@ -1408,7 +1408,7 @@ define i32 @hdr_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    %2 = load i32, i32* %arrayidx, align 4
 ; EPILOG-BLOCK-NEXT:    %add = add nsw i32 %2, %sum.02
 ; EPILOG-BLOCK-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-BLOCK-NEXT:    %niter.nsub = sub i64 %niter, 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
@@ -1418,8 +1418,8 @@ define i32 @hdr_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    %3 = load i32, i32* %arrayidx.1, align 4
 ; EPILOG-BLOCK-NEXT:    %add.1 = add nsw i32 %3, %add
 ; EPILOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
-; EPILOG-BLOCK-NEXT:    %niter.nsub.1 = sub i64 %niter.nsub, 1
-; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp eq i64 %niter.nsub.1, 0
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; 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:
 ; EPILOG-BLOCK-NEXT:    %result.ph.ph = phi i32 [ %add.1, %latch.1 ]
@@ -1469,7 +1469,7 @@ define i32 @hdr_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG:       header.prol:
 ; PROLOG-NEXT:    %indvars.iv.prol = phi i64 [ %indvars.iv.next.prol, %latch.prol ], [ 0, %header.prol.preheader ]
 ; PROLOG-NEXT:    %sum.02.prol = phi i32 [ %add.prol, %latch.prol ], [ 0, %header.prol.preheader ]
-; PROLOG-NEXT:    %prol.iter = phi i64 [ %xtraiter, %header.prol.preheader ], [ %prol.iter.sub, %latch.prol ]
+; PROLOG-NEXT:    %prol.iter = phi i64 [ 0, %header.prol.preheader ], [ %prol.iter.next, %latch.prol ]
 ; PROLOG-NEXT:    br i1 %cond, label %latchExit.unr-lcssa.loopexit1, label %for.exiting_block.prol
 ; PROLOG:       for.exiting_block.prol:
 ; PROLOG-NEXT:    %cmp.prol = icmp eq i64 %n, 42
@@ -1480,8 +1480,8 @@ define i32 @hdr_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %add.prol = add nsw i32 %1, %sum.02.prol
 ; PROLOG-NEXT:    %indvars.iv.next.prol = add i64 %indvars.iv.prol, 1
 ; PROLOG-NEXT:    %exitcond.prol = icmp eq i64 %indvars.iv.next.prol, %n
-; PROLOG-NEXT:    %prol.iter.sub = sub i64 %prol.iter, 1
-; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.sub, 0
+; PROLOG-NEXT:    %prol.iter.next = add i64 %prol.iter, 1
+; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.next, %xtraiter
 ; PROLOG-NEXT:    br i1 %prol.iter.cmp, label %header.prol, label %header.prol.loopexit.unr-lcssa, !llvm.loop !4
 ; PROLOG:       header.prol.loopexit.unr-lcssa:
 ; PROLOG-NEXT:    %result.unr.ph = phi i32 [ %add.prol, %latch.prol ]
@@ -1698,7 +1698,7 @@ define i32 @otherblock_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG:       header:
 ; EPILOG-NEXT:    %indvars.iv = phi i64 [ 0, %entry.new ], [ %indvars.iv.next.7, %latch.7 ]
 ; EPILOG-NEXT:    %sum.02 = phi i32 [ 0, %entry.new ], [ %add.7, %latch.7 ]
-; EPILOG-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.7, %latch.7 ]
+; EPILOG-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.7, %latch.7 ]
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block
 ; EPILOG:       for.exiting_block:
 ; EPILOG-NEXT:    %cmp = icmp eq i64 %n, 42
@@ -1708,7 +1708,7 @@ define i32 @otherblock_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %2 = load i32, i32* %arrayidx, align 4
 ; EPILOG-NEXT:    %add = add nsw i32 %2, %sum.02
 ; EPILOG-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-NEXT:    %niter.nsub = sub i64 %niter, 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
@@ -1718,7 +1718,7 @@ define i32 @otherblock_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %3 = load i32, i32* %arrayidx.1, align 4
 ; EPILOG-NEXT:    %add.1 = add nsw i32 %3, %add
 ; EPILOG-NEXT:    %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv.next, 1
-; EPILOG-NEXT:    %niter.nsub.1 = sub i64 %niter.nsub, 1
+; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
 ; 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
@@ -1728,7 +1728,7 @@ define i32 @otherblock_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %4 = load i32, i32* %arrayidx.2, align 4
 ; EPILOG-NEXT:    %add.2 = add nsw i32 %4, %add.1
 ; EPILOG-NEXT:    %indvars.iv.next.2 = add nuw nsw i64 %indvars.iv.next.1, 1
-; EPILOG-NEXT:    %niter.nsub.2 = sub i64 %niter.nsub.1, 1
+; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
 ; 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
@@ -1738,7 +1738,7 @@ define i32 @otherblock_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %5 = load i32, i32* %arrayidx.3, align 4
 ; EPILOG-NEXT:    %add.3 = add nsw i32 %5, %add.2
 ; EPILOG-NEXT:    %indvars.iv.next.3 = add nuw nsw i64 %indvars.iv.next.2, 1
-; EPILOG-NEXT:    %niter.nsub.3 = sub i64 %niter.nsub.2, 1
+; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
 ; 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
@@ -1748,7 +1748,7 @@ define i32 @otherblock_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %6 = load i32, i32* %arrayidx.4, align 4
 ; EPILOG-NEXT:    %add.4 = add nsw i32 %6, %add.3
 ; EPILOG-NEXT:    %indvars.iv.next.4 = add nuw nsw i64 %indvars.iv.next.3, 1
-; EPILOG-NEXT:    %niter.nsub.4 = sub i64 %niter.nsub.3, 1
+; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
 ; 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
@@ -1758,7 +1758,7 @@ define i32 @otherblock_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %7 = load i32, i32* %arrayidx.5, align 4
 ; EPILOG-NEXT:    %add.5 = add nsw i32 %7, %add.4
 ; EPILOG-NEXT:    %indvars.iv.next.5 = add nuw nsw i64 %indvars.iv.next.4, 1
-; EPILOG-NEXT:    %niter.nsub.5 = sub i64 %niter.nsub.4, 1
+; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
 ; 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
@@ -1768,7 +1768,7 @@ define i32 @otherblock_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %8 = load i32, i32* %arrayidx.6, align 4
 ; EPILOG-NEXT:    %add.6 = add nsw i32 %8, %add.5
 ; EPILOG-NEXT:    %indvars.iv.next.6 = add nuw nsw i64 %indvars.iv.next.5, 1
-; EPILOG-NEXT:    %niter.nsub.6 = sub i64 %niter.nsub.5, 1
+; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
 ; 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
@@ -1778,8 +1778,8 @@ define i32 @otherblock_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %9 = load i32, i32* %arrayidx.7, align 4
 ; EPILOG-NEXT:    %add.7 = add nsw i32 %9, %add.6
 ; EPILOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv.next.6, 1
-; EPILOG-NEXT:    %niter.nsub.7 = sub i64 %niter.nsub.6, 1
-; EPILOG-NEXT:    %niter.ncmp.7 = icmp eq i64 %niter.nsub.7, 0
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; 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:
 ; EPILOG-NEXT:    %result.ph.ph = phi i32 [ %add.7, %latch.7 ]
@@ -1797,7 +1797,7 @@ define i32 @otherblock_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG:       header.epil:
 ; EPILOG-NEXT:    %indvars.iv.epil = phi i64 [ %indvars.iv.next.epil, %latch.epil ], [ %indvars.iv.unr, %header.epil.preheader ]
 ; EPILOG-NEXT:    %sum.02.epil = phi i32 [ %add.epil, %latch.epil ], [ %sum.02.unr, %header.epil.preheader ]
-; EPILOG-NEXT:    %epil.iter = phi i64 [ %xtraiter, %header.epil.preheader ], [ %epil.iter.sub, %latch.epil ]
+; EPILOG-NEXT:    %epil.iter = phi i64 [ 0, %header.epil.preheader ], [ %epil.iter.next, %latch.epil ]
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit2, label %for.exiting_block.epil
 ; EPILOG:       for.exiting_block.epil:
 ; EPILOG-NEXT:    %cmp.epil = icmp eq i64 %n, 42
@@ -1808,8 +1808,8 @@ define i32 @otherblock_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %add.epil = add nsw i32 %10, %sum.02.epil
 ; EPILOG-NEXT:    %indvars.iv.next.epil = add i64 %indvars.iv.epil, 1
 ; EPILOG-NEXT:    %exitcond.epil = icmp eq i64 %indvars.iv.next.epil, %n
-; EPILOG-NEXT:    %epil.iter.sub = sub i64 %epil.iter, 1
-; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.sub, 0
+; EPILOG-NEXT:    %epil.iter.next = add i64 %epil.iter, 1
+; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.next, %xtraiter
 ; EPILOG-NEXT:    br i1 %epil.iter.cmp, label %header.epil, label %latchExit.epilog-lcssa.loopexit3, !llvm.loop !5
 ; EPILOG:       latchExit.epilog-lcssa.loopexit:
 ; EPILOG-NEXT:    %result.ph1.ph = phi i32 [ 2, %for.exiting_block ], [ 2, %for.exiting_block.1 ], [ 2, %for.exiting_block.2 ], [ 2, %for.exiting_block.3 ], [ 2, %for.exiting_block.4 ], [ 2, %for.exiting_block.5 ], [ 2, %for.exiting_block.6 ], [ 2, %for.exiting_block.7 ]
@@ -1842,7 +1842,7 @@ define i32 @otherblock_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-BLOCK:       header:
 ; EPILOG-BLOCK-NEXT:    %indvars.iv = phi i64 [ 0, %entry.new ], [ %indvars.iv.next.1, %latch.1 ]
 ; EPILOG-BLOCK-NEXT:    %sum.02 = phi i32 [ 0, %entry.new ], [ %add.1, %latch.1 ]
-; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.1, %latch.1 ]
+; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.1, %latch.1 ]
 ; EPILOG-BLOCK-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block
 ; EPILOG-BLOCK:       for.exiting_block:
 ; EPILOG-BLOCK-NEXT:    %cmp = icmp eq i64 %n, 42
@@ -1852,7 +1852,7 @@ define i32 @otherblock_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    %2 = load i32, i32* %arrayidx, align 4
 ; EPILOG-BLOCK-NEXT:    %add = add nsw i32 %2, %sum.02
 ; EPILOG-BLOCK-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-BLOCK-NEXT:    %niter.nsub = sub i64 %niter, 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
@@ -1862,8 +1862,8 @@ define i32 @otherblock_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    %3 = load i32, i32* %arrayidx.1, align 4
 ; EPILOG-BLOCK-NEXT:    %add.1 = add nsw i32 %3, %add
 ; EPILOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
-; EPILOG-BLOCK-NEXT:    %niter.nsub.1 = sub i64 %niter.nsub, 1
-; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp eq i64 %niter.nsub.1, 0
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; 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:
 ; EPILOG-BLOCK-NEXT:    %result.ph.ph = phi i32 [ %add.1, %latch.1 ]
@@ -1913,7 +1913,7 @@ define i32 @otherblock_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG:       header.prol:
 ; PROLOG-NEXT:    %indvars.iv.prol = phi i64 [ %indvars.iv.next.prol, %latch.prol ], [ 0, %header.prol.preheader ]
 ; PROLOG-NEXT:    %sum.02.prol = phi i32 [ %add.prol, %latch.prol ], [ 0, %header.prol.preheader ]
-; PROLOG-NEXT:    %prol.iter = phi i64 [ %xtraiter, %header.prol.preheader ], [ %prol.iter.sub, %latch.prol ]
+; PROLOG-NEXT:    %prol.iter = phi i64 [ 0, %header.prol.preheader ], [ %prol.iter.next, %latch.prol ]
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit1, label %for.exiting_block.prol
 ; PROLOG:       for.exiting_block.prol:
 ; PROLOG-NEXT:    %cmp.prol = icmp eq i64 %n, 42
@@ -1924,8 +1924,8 @@ define i32 @otherblock_latch_same_exit(i32* nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %add.prol = add nsw i32 %1, %sum.02.prol
 ; PROLOG-NEXT:    %indvars.iv.next.prol = add i64 %indvars.iv.prol, 1
 ; PROLOG-NEXT:    %exitcond.prol = icmp eq i64 %indvars.iv.next.prol, %n
-; PROLOG-NEXT:    %prol.iter.sub = sub i64 %prol.iter, 1
-; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.sub, 0
+; PROLOG-NEXT:    %prol.iter.next = add i64 %prol.iter, 1
+; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.next, %xtraiter
 ; PROLOG-NEXT:    br i1 %prol.iter.cmp, label %header.prol, label %header.prol.loopexit.unr-lcssa, !llvm.loop !5
 ; PROLOG:       header.prol.loopexit.unr-lcssa:
 ; PROLOG-NEXT:    %result.unr.ph = phi i32 [ %add.prol, %latch.prol ]
@@ -2143,7 +2143,7 @@ define i32 @otherblock_latch_same_exit2(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG:       header:
 ; EPILOG-NEXT:    %indvars.iv = phi i64 [ 0, %entry.new ], [ %indvars.iv.next.7, %latch.7 ]
 ; EPILOG-NEXT:    %sum.02 = phi i32 [ 0, %entry.new ], [ %add.7, %latch.7 ]
-; EPILOG-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.7, %latch.7 ]
+; EPILOG-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.7, %latch.7 ]
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block
 ; EPILOG:       for.exiting_block:
 ; EPILOG-NEXT:    %cmp = icmp eq i64 %n, 42
@@ -2153,7 +2153,7 @@ define i32 @otherblock_latch_same_exit2(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %2 = load i32, i32* %arrayidx, align 4
 ; EPILOG-NEXT:    %add = add nsw i32 %2, %sum.02
 ; EPILOG-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-NEXT:    %niter.nsub = sub i64 %niter, 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
@@ -2163,7 +2163,7 @@ define i32 @otherblock_latch_same_exit2(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %3 = load i32, i32* %arrayidx.1, align 4
 ; EPILOG-NEXT:    %add.1 = add nsw i32 %3, %add
 ; EPILOG-NEXT:    %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv.next, 1
-; EPILOG-NEXT:    %niter.nsub.1 = sub i64 %niter.nsub, 1
+; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
 ; 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
@@ -2173,7 +2173,7 @@ define i32 @otherblock_latch_same_exit2(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %4 = load i32, i32* %arrayidx.2, align 4
 ; EPILOG-NEXT:    %add.2 = add nsw i32 %4, %add.1
 ; EPILOG-NEXT:    %indvars.iv.next.2 = add nuw nsw i64 %indvars.iv.next.1, 1
-; EPILOG-NEXT:    %niter.nsub.2 = sub i64 %niter.nsub.1, 1
+; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
 ; 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
@@ -2183,7 +2183,7 @@ define i32 @otherblock_latch_same_exit2(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %5 = load i32, i32* %arrayidx.3, align 4
 ; EPILOG-NEXT:    %add.3 = add nsw i32 %5, %add.2
 ; EPILOG-NEXT:    %indvars.iv.next.3 = add nuw nsw i64 %indvars.iv.next.2, 1
-; EPILOG-NEXT:    %niter.nsub.3 = sub i64 %niter.nsub.2, 1
+; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
 ; 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
@@ -2193,7 +2193,7 @@ define i32 @otherblock_latch_same_exit2(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %6 = load i32, i32* %arrayidx.4, align 4
 ; EPILOG-NEXT:    %add.4 = add nsw i32 %6, %add.3
 ; EPILOG-NEXT:    %indvars.iv.next.4 = add nuw nsw i64 %indvars.iv.next.3, 1
-; EPILOG-NEXT:    %niter.nsub.4 = sub i64 %niter.nsub.3, 1
+; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
 ; 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
@@ -2203,7 +2203,7 @@ define i32 @otherblock_latch_same_exit2(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %7 = load i32, i32* %arrayidx.5, align 4
 ; EPILOG-NEXT:    %add.5 = add nsw i32 %7, %add.4
 ; EPILOG-NEXT:    %indvars.iv.next.5 = add nuw nsw i64 %indvars.iv.next.4, 1
-; EPILOG-NEXT:    %niter.nsub.5 = sub i64 %niter.nsub.4, 1
+; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
 ; 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
@@ -2213,7 +2213,7 @@ define i32 @otherblock_latch_same_exit2(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %8 = load i32, i32* %arrayidx.6, align 4
 ; EPILOG-NEXT:    %add.6 = add nsw i32 %8, %add.5
 ; EPILOG-NEXT:    %indvars.iv.next.6 = add nuw nsw i64 %indvars.iv.next.5, 1
-; EPILOG-NEXT:    %niter.nsub.6 = sub i64 %niter.nsub.5, 1
+; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
 ; 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
@@ -2223,8 +2223,8 @@ define i32 @otherblock_latch_same_exit2(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %9 = load i32, i32* %arrayidx.7, align 4
 ; EPILOG-NEXT:    %add.7 = add nsw i32 %9, %add.6
 ; EPILOG-NEXT:    %indvars.iv.next.7 = add i64 %indvars.iv.next.6, 1
-; EPILOG-NEXT:    %niter.nsub.7 = sub i64 %niter.nsub.6, 1
-; EPILOG-NEXT:    %niter.ncmp.7 = icmp eq i64 %niter.nsub.7, 0
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; 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:
 ; EPILOG-NEXT:    %result.ph.ph = phi i32 [ %add.7, %latch.7 ]
@@ -2242,7 +2242,7 @@ define i32 @otherblock_latch_same_exit2(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG:       header.epil:
 ; EPILOG-NEXT:    %indvars.iv.epil = phi i64 [ %indvars.iv.next.epil, %latch.epil ], [ %indvars.iv.unr, %header.epil.preheader ]
 ; EPILOG-NEXT:    %sum.02.epil = phi i32 [ %add.epil, %latch.epil ], [ %sum.02.unr, %header.epil.preheader ]
-; EPILOG-NEXT:    %epil.iter = phi i64 [ %xtraiter, %header.epil.preheader ], [ %epil.iter.sub, %latch.epil ]
+; EPILOG-NEXT:    %epil.iter = phi i64 [ 0, %header.epil.preheader ], [ %epil.iter.next, %latch.epil ]
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit2, label %for.exiting_block.epil
 ; EPILOG:       for.exiting_block.epil:
 ; EPILOG-NEXT:    %cmp.epil = icmp eq i64 %n, 42
@@ -2253,8 +2253,8 @@ define i32 @otherblock_latch_same_exit2(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-NEXT:    %add.epil = add nsw i32 %10, %sum.02.epil
 ; EPILOG-NEXT:    %indvars.iv.next.epil = add i64 %indvars.iv.epil, 1
 ; EPILOG-NEXT:    %exitcond.epil = icmp eq i64 %indvars.iv.next.epil, %n
-; EPILOG-NEXT:    %epil.iter.sub = sub i64 %epil.iter, 1
-; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.sub, 0
+; EPILOG-NEXT:    %epil.iter.next = add i64 %epil.iter, 1
+; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.next, %xtraiter
 ; EPILOG-NEXT:    br i1 %epil.iter.cmp, label %header.epil, label %latchExit.epilog-lcssa.loopexit3, !llvm.loop !6
 ; EPILOG:       latchExit.epilog-lcssa.loopexit:
 ; EPILOG-NEXT:    %result.ph1.ph = phi i32 [ %sum.02, %for.exiting_block ], [ %add, %for.exiting_block.1 ], [ %add.1, %for.exiting_block.2 ], [ %add.2, %for.exiting_block.3 ], [ %add.3, %for.exiting_block.4 ], [ %add.4, %for.exiting_block.5 ], [ %add.5, %for.exiting_block.6 ], [ %add.6, %for.exiting_block.7 ]
@@ -2287,7 +2287,7 @@ define i32 @otherblock_latch_same_exit2(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-BLOCK:       header:
 ; EPILOG-BLOCK-NEXT:    %indvars.iv = phi i64 [ 0, %entry.new ], [ %indvars.iv.next.1, %latch.1 ]
 ; EPILOG-BLOCK-NEXT:    %sum.02 = phi i32 [ 0, %entry.new ], [ %add.1, %latch.1 ]
-; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.1, %latch.1 ]
+; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.1, %latch.1 ]
 ; EPILOG-BLOCK-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block
 ; EPILOG-BLOCK:       for.exiting_block:
 ; EPILOG-BLOCK-NEXT:    %cmp = icmp eq i64 %n, 42
@@ -2297,7 +2297,7 @@ define i32 @otherblock_latch_same_exit2(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    %2 = load i32, i32* %arrayidx, align 4
 ; EPILOG-BLOCK-NEXT:    %add = add nsw i32 %2, %sum.02
 ; EPILOG-BLOCK-NEXT:    %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-; EPILOG-BLOCK-NEXT:    %niter.nsub = sub i64 %niter, 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
@@ -2307,8 +2307,8 @@ define i32 @otherblock_latch_same_exit2(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    %3 = load i32, i32* %arrayidx.1, align 4
 ; EPILOG-BLOCK-NEXT:    %add.1 = add nsw i32 %3, %add
 ; EPILOG-BLOCK-NEXT:    %indvars.iv.next.1 = add i64 %indvars.iv.next, 1
-; EPILOG-BLOCK-NEXT:    %niter.nsub.1 = sub i64 %niter.nsub, 1
-; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp eq i64 %niter.nsub.1, 0
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; 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:
 ; EPILOG-BLOCK-NEXT:    %result.ph.ph = phi i32 [ %add.1, %latch.1 ]
@@ -2358,7 +2358,7 @@ define i32 @otherblock_latch_same_exit2(i32* nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG:       header.prol:
 ; PROLOG-NEXT:    %indvars.iv.prol = phi i64 [ %indvars.iv.next.prol, %latch.prol ], [ 0, %header.prol.preheader ]
 ; PROLOG-NEXT:    %sum.02.prol = phi i32 [ %add.prol, %latch.prol ], [ 0, %header.prol.preheader ]
-; PROLOG-NEXT:    %prol.iter = phi i64 [ %xtraiter, %header.prol.preheader ], [ %prol.iter.sub, %latch.prol ]
+; PROLOG-NEXT:    %prol.iter = phi i64 [ 0, %header.prol.preheader ], [ %prol.iter.next, %latch.prol ]
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit1, label %for.exiting_block.prol
 ; PROLOG:       for.exiting_block.prol:
 ; PROLOG-NEXT:    %cmp.prol = icmp eq i64 %n, 42
@@ -2369,8 +2369,8 @@ define i32 @otherblock_latch_same_exit2(i32* nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG-NEXT:    %add.prol = add nsw i32 %1, %sum.02.prol
 ; PROLOG-NEXT:    %indvars.iv.next.prol = add i64 %indvars.iv.prol, 1
 ; PROLOG-NEXT:    %exitcond.prol = icmp eq i64 %indvars.iv.next.prol, %n
-; PROLOG-NEXT:    %prol.iter.sub = sub i64 %prol.iter, 1
-; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.sub, 0
+; PROLOG-NEXT:    %prol.iter.next = add i64 %prol.iter, 1
+; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.next, %xtraiter
 ; PROLOG-NEXT:    br i1 %prol.iter.cmp, label %header.prol, label %header.prol.loopexit.unr-lcssa, !llvm.loop !6
 ; PROLOG:       header.prol.loopexit.unr-lcssa:
 ; PROLOG-NEXT:    %result.unr.ph = phi i32 [ %add.prol, %latch.prol ]
@@ -2589,7 +2589,7 @@ define i32 @otherblock_latch_same_exit3(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG:       header:
 ; EPILOG-NEXT:    %indvars.iv = phi i64 [ 0, %entry.new ], [ %indvars.iv.next.7, %latch.7 ]
 ; EPILOG-NEXT:    %sum.02 = phi i32 [ 0, %entry.new ], [ %add.7, %latch.7 ]
-; EPILOG-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.7, %latch.7 ]
+; EPILOG-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.7, %latch.7 ]
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block
 ; EPILOG:       for.exiting_block:
 ; EPILOG-NEXT:    %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv
@@ -2599,7 +2599,7 @@ define i32 @otherblock_latch_same_exit3(i32* 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.nsub = sub i64 %niter, 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, i32* %a, i64 %indvars.iv.next
@@ -2609,7 +2609,7 @@ define i32 @otherblock_latch_same_exit3(i32* nocapture %a, i64 %n, i1 %cond) {
 ; 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.nsub.1 = sub i64 %niter.nsub, 1
+; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
 ; 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, i32* %a, i64 %indvars.iv.next.1
@@ -2619,7 +2619,7 @@ define i32 @otherblock_latch_same_exit3(i32* nocapture %a, i64 %n, i1 %cond) {
 ; 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.nsub.2 = sub i64 %niter.nsub.1, 1
+; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
 ; 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, i32* %a, i64 %indvars.iv.next.2
@@ -2629,7 +2629,7 @@ define i32 @otherblock_latch_same_exit3(i32* nocapture %a, i64 %n, i1 %cond) {
 ; 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.nsub.3 = sub i64 %niter.nsub.2, 1
+; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
 ; 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, i32* %a, i64 %indvars.iv.next.3
@@ -2639,7 +2639,7 @@ define i32 @otherblock_latch_same_exit3(i32* nocapture %a, i64 %n, i1 %cond) {
 ; 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.nsub.4 = sub i64 %niter.nsub.3, 1
+; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
 ; 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, i32* %a, i64 %indvars.iv.next.4
@@ -2649,7 +2649,7 @@ define i32 @otherblock_latch_same_exit3(i32* nocapture %a, i64 %n, i1 %cond) {
 ; 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.nsub.5 = sub i64 %niter.nsub.4, 1
+; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
 ; 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, i32* %a, i64 %indvars.iv.next.5
@@ -2659,7 +2659,7 @@ define i32 @otherblock_latch_same_exit3(i32* nocapture %a, i64 %n, i1 %cond) {
 ; 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.nsub.6 = sub i64 %niter.nsub.5, 1
+; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
 ; 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, i32* %a, i64 %indvars.iv.next.6
@@ -2669,8 +2669,8 @@ define i32 @otherblock_latch_same_exit3(i32* nocapture %a, i64 %n, i1 %cond) {
 ; 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.nsub.7 = sub i64 %niter.nsub.6, 1
-; EPILOG-NEXT:    %niter.ncmp.7 = icmp eq i64 %niter.nsub.7, 0
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; 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:
 ; EPILOG-NEXT:    %result.ph.ph = phi i32 [ %add.7, %latch.7 ]
@@ -2688,7 +2688,7 @@ define i32 @otherblock_latch_same_exit3(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG:       header.epil:
 ; EPILOG-NEXT:    %indvars.iv.epil = phi i64 [ %indvars.iv.next.epil, %latch.epil ], [ %indvars.iv.unr, %header.epil.preheader ]
 ; EPILOG-NEXT:    %sum.02.epil = phi i32 [ %add.epil, %latch.epil ], [ %sum.02.unr, %header.epil.preheader ]
-; EPILOG-NEXT:    %epil.iter = phi i64 [ %xtraiter, %header.epil.preheader ], [ %epil.iter.sub, %latch.epil ]
+; EPILOG-NEXT:    %epil.iter = phi i64 [ 0, %header.epil.preheader ], [ %epil.iter.next, %latch.epil ]
 ; EPILOG-NEXT:    br i1 %cond, label %for.exit2.loopexit2, label %for.exiting_block.epil
 ; EPILOG:       for.exiting_block.epil:
 ; EPILOG-NEXT:    %arrayidx.epil = getelementptr inbounds i32, i32* %a, i64 %indvars.iv.epil
@@ -2699,8 +2699,8 @@ define i32 @otherblock_latch_same_exit3(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG:       latch.epil:
 ; EPILOG-NEXT:    %indvars.iv.next.epil = add i64 %indvars.iv.epil, 1
 ; EPILOG-NEXT:    %exitcond.epil = icmp eq i64 %indvars.iv.next.epil, %n
-; EPILOG-NEXT:    %epil.iter.sub = sub i64 %epil.iter, 1
-; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.sub, 0
+; EPILOG-NEXT:    %epil.iter.next = add i64 %epil.iter, 1
+; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.next, %xtraiter
 ; EPILOG-NEXT:    br i1 %epil.iter.cmp, label %header.epil, label %latchExit.epilog-lcssa.loopexit3, !llvm.loop !7
 ; EPILOG:       latchExit.epilog-lcssa.loopexit:
 ; EPILOG-NEXT:    %result.ph1.ph = phi i32 [ %sum.02, %for.exiting_block ], [ %add, %for.exiting_block.1 ], [ %add.1, %for.exiting_block.2 ], [ %add.2, %for.exiting_block.3 ], [ %add.3, %for.exiting_block.4 ], [ %add.4, %for.exiting_block.5 ], [ %add.5, %for.exiting_block.6 ], [ %add.6, %for.exiting_block.7 ]
@@ -2733,7 +2733,7 @@ define i32 @otherblock_latch_same_exit3(i32* nocapture %a, i64 %n, i1 %cond) {
 ; EPILOG-BLOCK:       header:
 ; EPILOG-BLOCK-NEXT:    %indvars.iv = phi i64 [ 0, %entry.new ], [ %indvars.iv.next.1, %latch.1 ]
 ; EPILOG-BLOCK-NEXT:    %sum.02 = phi i32 [ 0, %entry.new ], [ %add.1, %latch.1 ]
-; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.1, %latch.1 ]
+; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.1, %latch.1 ]
 ; EPILOG-BLOCK-NEXT:    br i1 %cond, label %for.exit2.loopexit, label %for.exiting_block
 ; EPILOG-BLOCK:       for.exiting_block:
 ; EPILOG-BLOCK-NEXT:    %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv
@@ -2743,7 +2743,7 @@ define i32 @otherblock_latch_same_exit3(i32* 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.nsub = sub i64 %niter, 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, i32* %a, i64 %indvars.iv.next
@@ -2753,8 +2753,8 @@ define i32 @otherblock_latch_same_exit3(i32* nocapture %a, i64 %n, i1 %cond) {
 ; 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.nsub.1 = sub i64 %niter.nsub, 1
-; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp eq i64 %niter.nsub.1, 0
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; 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:
 ; EPILOG-BLOCK-NEXT:    %result.ph.ph = phi i32 [ %add.1, %latch.1 ]
@@ -2804,7 +2804,7 @@ define i32 @otherblock_latch_same_exit3(i32* nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG:       header.prol:
 ; PROLOG-NEXT:    %indvars.iv.prol = phi i64 [ %indvars.iv.next.prol, %latch.prol ], [ 0, %header.prol.preheader ]
 ; PROLOG-NEXT:    %sum.02.prol = phi i32 [ %add.prol, %latch.prol ], [ 0, %header.prol.preheader ]
-; PROLOG-NEXT:    %prol.iter = phi i64 [ %xtraiter, %header.prol.preheader ], [ %prol.iter.sub, %latch.prol ]
+; PROLOG-NEXT:    %prol.iter = phi i64 [ 0, %header.prol.preheader ], [ %prol.iter.next, %latch.prol ]
 ; PROLOG-NEXT:    br i1 %cond, label %for.exit2.loopexit1, label %for.exiting_block.prol
 ; PROLOG:       for.exiting_block.prol:
 ; PROLOG-NEXT:    %arrayidx.prol = getelementptr inbounds i32, i32* %a, i64 %indvars.iv.prol
@@ -2815,8 +2815,8 @@ define i32 @otherblock_latch_same_exit3(i32* nocapture %a, i64 %n, i1 %cond) {
 ; PROLOG:       latch.prol:
 ; PROLOG-NEXT:    %indvars.iv.next.prol = add i64 %indvars.iv.prol, 1
 ; PROLOG-NEXT:    %exitcond.prol = icmp eq i64 %indvars.iv.next.prol, %n
-; PROLOG-NEXT:    %prol.iter.sub = sub i64 %prol.iter, 1
-; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.sub, 0
+; PROLOG-NEXT:    %prol.iter.next = add i64 %prol.iter, 1
+; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.next, %xtraiter
 ; PROLOG-NEXT:    br i1 %prol.iter.cmp, label %header.prol, label %header.prol.loopexit.unr-lcssa, !llvm.loop !7
 ; PROLOG:       header.prol.loopexit.unr-lcssa:
 ; PROLOG-NEXT:    %result.unr.ph = phi i32 [ %add.prol, %latch.prol ]
@@ -3078,7 +3078,7 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; EPILOG:       loop_header:
 ; EPILOG-NEXT:    %iv = phi i64 [ 0, %entry.new ], [ %iv_next.7, %loop_latch.7 ]
 ; EPILOG-NEXT:    %sum = phi i64 [ 0, %entry.new ], [ %sum.next.7, %loop_latch.7 ]
-; EPILOG-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.7, %loop_latch.7 ]
+; EPILOG-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.7, %loop_latch.7 ]
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch, label %loop_exiting
 ; EPILOG:       loop_exiting:
 ; EPILOG-NEXT:    %ivy = add i64 %iv, %add
@@ -3089,7 +3089,7 @@ 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.nsub = sub i64 %niter, 1
+; 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
@@ -3100,7 +3100,7 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; 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.nsub.1 = sub i64 %niter.nsub, 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.2
 ; EPILOG:       loop_exiting.2:
 ; EPILOG-NEXT:    %ivy.2 = add i64 %iv_next.1, %add
@@ -3111,7 +3111,7 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; 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.nsub.2 = sub i64 %niter.nsub.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.3
 ; EPILOG:       loop_exiting.3:
 ; EPILOG-NEXT:    %ivy.3 = add i64 %iv_next.2, %add
@@ -3122,7 +3122,7 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; 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.nsub.3 = sub i64 %niter.nsub.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.4
 ; EPILOG:       loop_exiting.4:
 ; EPILOG-NEXT:    %ivy.4 = add i64 %iv_next.3, %add
@@ -3133,7 +3133,7 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; 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.nsub.4 = sub i64 %niter.nsub.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.5
 ; EPILOG:       loop_exiting.5:
 ; EPILOG-NEXT:    %ivy.5 = add i64 %iv_next.4, %add
@@ -3144,7 +3144,7 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; 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.nsub.5 = sub i64 %niter.nsub.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.6
 ; EPILOG:       loop_exiting.6:
 ; EPILOG-NEXT:    %ivy.6 = add i64 %iv_next.5, %add
@@ -3155,7 +3155,7 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; 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.nsub.6 = sub i64 %niter.nsub.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.7
 ; EPILOG:       loop_exiting.7:
 ; EPILOG-NEXT:    %ivy.7 = add i64 %iv_next.6, %add
@@ -3166,8 +3166,8 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; EPILOG:       loop_latch.7:
 ; EPILOG-NEXT:    %iv_next.7 = add nuw nsw i64 %iv_next.6, 1
 ; EPILOG-NEXT:    %sum.next.7 = add i64 %sum.next.6, %add
-; EPILOG-NEXT:    %niter.nsub.7 = sub i64 %niter.nsub.6, 1
-; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.nsub.7, 0
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; 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:
 ; EPILOG-NEXT:    %result.ph = phi i64 [ %ivy, %loop_exiting ], [ %ivy, %loop_exiting ], [ %ivy.1, %loop_exiting.1 ], [ %ivy.1, %loop_exiting.1 ], [ %ivy.2, %loop_exiting.2 ], [ %ivy.2, %loop_exiting.2 ], [ %ivy.3, %loop_exiting.3 ], [ %ivy.3, %loop_exiting.3 ], [ %ivy.4, %loop_exiting.4 ], [ %ivy.4, %loop_exiting.4 ], [ %ivy.5, %loop_exiting.5 ], [ %ivy.5, %loop_exiting.5 ], [ %ivy.6, %loop_exiting.6 ], [ %ivy.6, %loop_exiting.6 ], [ %ivy.7, %loop_exiting.7 ], [ %ivy.7, %loop_exiting.7 ]
@@ -3194,7 +3194,7 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; EPILOG:       loop_header.epil:
 ; EPILOG-NEXT:    %iv.epil = phi i64 [ %iv.unr, %loop_header.epil.preheader ], [ %iv_next.epil, %loop_latch.epil ]
 ; EPILOG-NEXT:    %sum.epil = phi i64 [ %sum.unr, %loop_header.epil.preheader ], [ %sum.next.epil, %loop_latch.epil ]
-; EPILOG-NEXT:    %epil.iter = phi i64 [ %xtraiter, %loop_header.epil.preheader ], [ %epil.iter.sub, %loop_latch.epil ]
+; EPILOG-NEXT:    %epil.iter = phi i64 [ 0, %loop_header.epil.preheader ], [ %epil.iter.next, %loop_latch.epil ]
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.epil, label %loop_exiting.epil
 ; EPILOG:       loop_exiting.epil:
 ; EPILOG-NEXT:    %ivy.epil = add i64 %iv.epil, %add
@@ -3206,8 +3206,8 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; EPILOG-NEXT:    %iv_next.epil = add nuw nsw i64 %iv.epil, 1
 ; EPILOG-NEXT:    %sum.next.epil = add i64 %sum.epil, %add
 ; EPILOG-NEXT:    %cmp.epil = icmp ne i64 %iv_next.epil, %trip
-; EPILOG-NEXT:    %epil.iter.sub = sub i64 %epil.iter, 1
-; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.sub, 0
+; EPILOG-NEXT:    %epil.iter.next = add i64 %epil.iter, 1
+; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.next, %xtraiter
 ; EPILOG-NEXT:    br i1 %epil.iter.cmp, label %loop_header.epil, label %latchexit.epilog-lcssa, !llvm.loop !8
 ; EPILOG:       latchexit.epilog-lcssa:
 ; EPILOG-NEXT:    %sum.next.lcssa.ph1 = phi i64 [ %sum.next.epil, %loop_latch.epil ]
@@ -3228,7 +3228,7 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; EPILOG-BLOCK:       loop_header:
 ; EPILOG-BLOCK-NEXT:    %iv = phi i64 [ 0, %entry.new ], [ %iv_next.1, %loop_latch.1 ]
 ; EPILOG-BLOCK-NEXT:    %sum = phi i64 [ 0, %entry.new ], [ %sum.next.1, %loop_latch.1 ]
-; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.1, %loop_latch.1 ]
+; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.1, %loop_latch.1 ]
 ; EPILOG-BLOCK-NEXT:    br i1 %cond, label %loop_latch, label %loop_exiting
 ; EPILOG-BLOCK:       loop_exiting:
 ; EPILOG-BLOCK-NEXT:    %ivy = add i64 %iv, %add
@@ -3239,7 +3239,7 @@ 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.nsub = sub i64 %niter, 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.1
 ; EPILOG-BLOCK:       loop_exiting.1:
 ; EPILOG-BLOCK-NEXT:    %ivy.1 = add i64 %iv_next, %add
@@ -3250,8 +3250,8 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; EPILOG-BLOCK:       loop_latch.1:
 ; EPILOG-BLOCK-NEXT:    %iv_next.1 = add nuw nsw i64 %iv_next, 1
 ; EPILOG-BLOCK-NEXT:    %sum.next.1 = add i64 %sum.next, %add
-; EPILOG-BLOCK-NEXT:    %niter.nsub.1 = sub i64 %niter.nsub, 1
-; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.nsub.1, 0
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; 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 !8
 ; EPILOG-BLOCK:       exit1.loopexit:
 ; EPILOG-BLOCK-NEXT:    %result.ph = phi i64 [ %ivy, %loop_exiting ], [ %ivy, %loop_exiting ], [ %ivy.1, %loop_exiting.1 ], [ %ivy.1, %loop_exiting.1 ]
@@ -3298,7 +3298,7 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; PROLOG:       loop_header.prol:
 ; PROLOG-NEXT:    %iv.prol = phi i64 [ 0, %loop_header.prol.preheader ], [ %iv_next.prol, %loop_latch.prol ]
 ; PROLOG-NEXT:    %sum.prol = phi i64 [ 0, %loop_header.prol.preheader ], [ %sum.next.prol, %loop_latch.prol ]
-; PROLOG-NEXT:    %prol.iter = phi i64 [ %xtraiter, %loop_header.prol.preheader ], [ %prol.iter.sub, %loop_latch.prol ]
+; PROLOG-NEXT:    %prol.iter = phi i64 [ 0, %loop_header.prol.preheader ], [ %prol.iter.next, %loop_latch.prol ]
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.prol, label %loop_exiting.prol
 ; PROLOG:       loop_exiting.prol:
 ; PROLOG-NEXT:    %ivy.prol = add i64 %iv.prol, %add
@@ -3310,8 +3310,8 @@ define i64 @test5(i64 %trip, i64 %add, i1 %cond) {
 ; PROLOG-NEXT:    %iv_next.prol = add nuw nsw i64 %iv.prol, 1
 ; PROLOG-NEXT:    %sum.next.prol = add i64 %sum.prol, %add
 ; PROLOG-NEXT:    %cmp.prol = icmp ne i64 %iv_next.prol, %trip
-; PROLOG-NEXT:    %prol.iter.sub = sub i64 %prol.iter, 1
-; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.sub, 0
+; PROLOG-NEXT:    %prol.iter.next = add i64 %prol.iter, 1
+; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.next, %xtraiter
 ; PROLOG-NEXT:    br i1 %prol.iter.cmp, label %loop_header.prol, label %loop_header.prol.loopexit.unr-lcssa, !llvm.loop !8
 ; PROLOG:       loop_header.prol.loopexit.unr-lcssa:
 ; PROLOG-NEXT:    %iv.unr.ph = phi i64 [ %iv_next.prol, %loop_latch.prol ]
@@ -3534,7 +3534,7 @@ define i32 @test6(i32* nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG:       header:
 ; EPILOG-NEXT:    %indvars.iv = phi i64 [ 0, %entry.new ], [ %indvars.iv.next.7, %latch.7 ]
 ; EPILOG-NEXT:    %sum.02 = phi i32 [ 0, %entry.new ], [ %add.7, %latch.7 ]
-; EPILOG-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.7, %latch.7 ]
+; EPILOG-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.7, %latch.7 ]
 ; EPILOG-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block
 ; EPILOG:       for.exiting_block:
 ; EPILOG-NEXT:    %cmp = icmp eq i64 %n, 42
@@ -3544,7 +3544,7 @@ define i32 @test6(i32* nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-NEXT:    %load = load i32, i32* %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.nsub = sub i64 %niter, 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
@@ -3554,7 +3554,7 @@ define i32 @test6(i32* nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-NEXT:    %load.1 = load i32, i32* %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.nsub.1 = sub i64 %niter.nsub, 1
+; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
 ; 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
@@ -3564,7 +3564,7 @@ define i32 @test6(i32* nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-NEXT:    %load.2 = load i32, i32* %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.nsub.2 = sub i64 %niter.nsub.1, 1
+; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
 ; 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
@@ -3574,7 +3574,7 @@ define i32 @test6(i32* nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-NEXT:    %load.3 = load i32, i32* %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.nsub.3 = sub i64 %niter.nsub.2, 1
+; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
 ; 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
@@ -3584,7 +3584,7 @@ define i32 @test6(i32* nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-NEXT:    %load.4 = load i32, i32* %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.nsub.4 = sub i64 %niter.nsub.3, 1
+; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
 ; 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
@@ -3594,7 +3594,7 @@ define i32 @test6(i32* nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-NEXT:    %load.5 = load i32, i32* %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.nsub.5 = sub i64 %niter.nsub.4, 1
+; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
 ; 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
@@ -3604,7 +3604,7 @@ define i32 @test6(i32* nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-NEXT:    %load.6 = load i32, i32* %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.nsub.6 = sub i64 %niter.nsub.5, 1
+; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
 ; 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
@@ -3614,8 +3614,8 @@ define i32 @test6(i32* nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-NEXT:    %load.7 = load i32, i32* %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.nsub.7 = sub i64 %niter.nsub.6, 1
-; EPILOG-NEXT:    %niter.ncmp.7 = icmp eq i64 %niter.nsub.7, 0
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; 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:
 ; EPILOG-NEXT:    %sum.0.lcssa.ph.ph = phi i32 [ %add.7, %latch.7 ]
@@ -3633,7 +3633,7 @@ define i32 @test6(i32* nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG:       header.epil:
 ; EPILOG-NEXT:    %indvars.iv.epil = phi i64 [ %indvars.iv.next.epil, %latch.epil ], [ %indvars.iv.unr, %header.epil.preheader ]
 ; EPILOG-NEXT:    %sum.02.epil = phi i32 [ %add.epil, %latch.epil ], [ %sum.02.unr, %header.epil.preheader ]
-; EPILOG-NEXT:    %epil.iter = phi i64 [ %xtraiter, %header.epil.preheader ], [ %epil.iter.sub, %latch.epil ]
+; EPILOG-NEXT:    %epil.iter = phi i64 [ 0, %header.epil.preheader ], [ %epil.iter.next, %latch.epil ]
 ; EPILOG-NEXT:    br i1 false, label %for.exit2.loopexit2, label %for.exiting_block.epil
 ; EPILOG:       for.exiting_block.epil:
 ; EPILOG-NEXT:    %cmp.epil = icmp eq i64 %n, 42
@@ -3644,8 +3644,8 @@ define i32 @test6(i32* nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-NEXT:    %add.epil = add nsw i32 %load.epil, %sum.02.epil
 ; EPILOG-NEXT:    %indvars.iv.next.epil = add i64 %indvars.iv.epil, 1
 ; EPILOG-NEXT:    %exitcond.epil = icmp eq i64 %indvars.iv.next.epil, %n
-; EPILOG-NEXT:    %epil.iter.sub = sub i64 %epil.iter, 1
-; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.sub, 0
+; EPILOG-NEXT:    %epil.iter.next = add i64 %epil.iter, 1
+; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.next, %xtraiter
 ; EPILOG-NEXT:    br i1 %epil.iter.cmp, label %header.epil, label %latch_exit.epilog-lcssa, !llvm.loop !9
 ; EPILOG:       latch_exit.epilog-lcssa:
 ; EPILOG-NEXT:    %sum.0.lcssa.ph1 = phi i32 [ %add.epil, %latch.epil ]
@@ -3680,7 +3680,7 @@ define i32 @test6(i32* nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-BLOCK:       header:
 ; EPILOG-BLOCK-NEXT:    %indvars.iv = phi i64 [ 0, %entry.new ], [ %indvars.iv.next.1, %latch.1 ]
 ; EPILOG-BLOCK-NEXT:    %sum.02 = phi i32 [ 0, %entry.new ], [ %add.1, %latch.1 ]
-; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.1, %latch.1 ]
+; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.1, %latch.1 ]
 ; EPILOG-BLOCK-NEXT:    br i1 false, label %for.exit2.loopexit, label %for.exiting_block
 ; EPILOG-BLOCK:       for.exiting_block:
 ; EPILOG-BLOCK-NEXT:    %cmp = icmp eq i64 %n, 42
@@ -3690,7 +3690,7 @@ define i32 @test6(i32* nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-BLOCK-NEXT:    %load = load i32, i32* %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.nsub = sub i64 %niter, 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
@@ -3700,8 +3700,8 @@ define i32 @test6(i32* nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; EPILOG-BLOCK-NEXT:    %load.1 = load i32, i32* %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.nsub.1 = sub i64 %niter.nsub, 1
-; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp eq i64 %niter.nsub.1, 0
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; 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 !9
 ; EPILOG-BLOCK:       latch_exit.unr-lcssa.loopexit:
 ; EPILOG-BLOCK-NEXT:    %sum.0.lcssa.ph.ph = phi i32 [ %add.1, %latch.1 ]
@@ -3752,7 +3752,7 @@ define i32 @test6(i32* nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; PROLOG:       header.prol:
 ; PROLOG-NEXT:    %indvars.iv.prol = phi i64 [ %indvars.iv.next.prol, %latch.prol ], [ 0, %header.prol.preheader ]
 ; PROLOG-NEXT:    %sum.02.prol = phi i32 [ %add.prol, %latch.prol ], [ 0, %header.prol.preheader ]
-; PROLOG-NEXT:    %prol.iter = phi i64 [ %xtraiter, %header.prol.preheader ], [ %prol.iter.sub, %latch.prol ]
+; PROLOG-NEXT:    %prol.iter = phi i64 [ 0, %header.prol.preheader ], [ %prol.iter.next, %latch.prol ]
 ; PROLOG-NEXT:    br i1 false, label %for.exit2.loopexit1, label %for.exiting_block.prol
 ; PROLOG:       for.exiting_block.prol:
 ; PROLOG-NEXT:    %cmp.prol = icmp eq i64 %n, 42
@@ -3763,8 +3763,8 @@ define i32 @test6(i32* nocapture %a, i64 %n, i1 %cond, i32 %x) {
 ; PROLOG-NEXT:    %add.prol = add nsw i32 %load.prol, %sum.02.prol
 ; PROLOG-NEXT:    %indvars.iv.next.prol = add i64 %indvars.iv.prol, 1
 ; PROLOG-NEXT:    %exitcond.prol = icmp eq i64 %indvars.iv.next.prol, %n
-; PROLOG-NEXT:    %prol.iter.sub = sub i64 %prol.iter, 1
-; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.sub, 0
+; PROLOG-NEXT:    %prol.iter.next = add i64 %prol.iter, 1
+; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.next, %xtraiter
 ; PROLOG-NEXT:    br i1 %prol.iter.cmp, label %header.prol, label %header.prol.loopexit.unr-lcssa, !llvm.loop !9
 ; PROLOG:       header.prol.loopexit.unr-lcssa:
 ; PROLOG-NEXT:    %sum.0.lcssa.unr.ph = phi i32 [ %add.prol, %latch.prol ]
@@ -3998,40 +3998,40 @@ define i32 @test7(i32 %arg, i32 %arg1, i32 %arg2) {
 ; EPILOG-NEXT:    br label %header
 ; EPILOG:       header:
 ; EPILOG-NEXT:    %i6 = phi i64 [ 1, %preheader.new ], [ %add.7, %latch.7 ]
-; EPILOG-NEXT:    %niter = phi i64 [ %unroll_iter, %preheader.new ], [ %niter.nsub.7, %latch.7 ]
+; 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.nsub = sub i64 %niter, 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.nsub.1 = sub i64 %niter.nsub, 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.nsub.2 = sub i64 %niter.nsub.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.nsub.3 = sub i64 %niter.nsub.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.nsub.4 = sub i64 %niter.nsub.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.nsub.5 = sub i64 %niter.nsub.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.nsub.6 = sub i64 %niter.nsub.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.nsub.7 = sub i64 %niter.nsub.6, 1
-; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.nsub.7, 0
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; 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:
 ; EPILOG-NEXT:    %i6.unr.ph = phi i64 [ %add.7, %latch.7 ]
@@ -4044,13 +4044,13 @@ define i32 @test7(i32 %arg, i32 %arg1, i32 %arg2) {
 ; EPILOG-NEXT:    br label %header.epil
 ; EPILOG:       header.epil:
 ; EPILOG-NEXT:    %i6.epil = phi i64 [ %i6.unr, %header.epil.preheader ], [ %add.epil, %latch.epil ]
-; EPILOG-NEXT:    %epil.iter = phi i64 [ %xtraiter, %header.epil.preheader ], [ %epil.iter.sub, %latch.epil ]
+; EPILOG-NEXT:    %epil.iter = phi i64 [ 0, %header.epil.preheader ], [ %epil.iter.next, %latch.epil ]
 ; EPILOG-NEXT:    br i1 false, label %loopexit1.loopexit1, label %latch.epil
 ; EPILOG:       latch.epil:
 ; EPILOG-NEXT:    %add.epil = add nuw nsw i64 %i6.epil, 1
 ; EPILOG-NEXT:    %i9.epil = icmp slt i64 %add.epil, %sext
-; EPILOG-NEXT:    %epil.iter.sub = sub i64 %epil.iter, 1
-; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.sub, 0
+; EPILOG-NEXT:    %epil.iter.next = add i64 %epil.iter, 1
+; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.next, %xtraiter
 ; EPILOG-NEXT:    br i1 %epil.iter.cmp, label %header.epil, label %latchexit.epilog-lcssa, !llvm.loop !10
 ; EPILOG:       latchexit.epilog-lcssa:
 ; EPILOG-NEXT:    br label %latchexit
@@ -4085,16 +4085,16 @@ define i32 @test7(i32 %arg, i32 %arg1, i32 %arg2) {
 ; EPILOG-BLOCK-NEXT:    br label %header
 ; EPILOG-BLOCK:       header:
 ; EPILOG-BLOCK-NEXT:    %i6 = phi i64 [ 1, %preheader.new ], [ %add.1, %latch.1 ]
-; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ %unroll_iter, %preheader.new ], [ %niter.nsub.1, %latch.1 ]
+; 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.nsub = sub i64 %niter, 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.nsub.1 = sub i64 %niter.nsub, 1
-; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.nsub.1, 0
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; 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 !10
 ; EPILOG-BLOCK:       latchexit.unr-lcssa.loopexit:
 ; EPILOG-BLOCK-NEXT:    br label %latchexit.unr-lcssa
@@ -4134,13 +4134,13 @@ define i32 @test7(i32 %arg, i32 %arg1, i32 %arg2) {
 ; PROLOG-NEXT:    br label %header.prol
 ; PROLOG:       header.prol:
 ; PROLOG-NEXT:    %i6.prol = phi i64 [ 1, %header.prol.preheader ], [ %add.prol, %latch.prol ]
-; PROLOG-NEXT:    %prol.iter = phi i64 [ %xtraiter, %header.prol.preheader ], [ %prol.iter.sub, %latch.prol ]
+; PROLOG-NEXT:    %prol.iter = phi i64 [ 0, %header.prol.preheader ], [ %prol.iter.next, %latch.prol ]
 ; PROLOG-NEXT:    br i1 false, label %loopexit1.loopexit1, label %latch.prol
 ; PROLOG:       latch.prol:
 ; PROLOG-NEXT:    %add.prol = add nuw nsw i64 %i6.prol, 1
 ; PROLOG-NEXT:    %i9.prol = icmp slt i64 %add.prol, %sext
-; PROLOG-NEXT:    %prol.iter.sub = sub i64 %prol.iter, 1
-; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.sub, 0
+; PROLOG-NEXT:    %prol.iter.next = add i64 %prol.iter, 1
+; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.next, %xtraiter
 ; PROLOG-NEXT:    br i1 %prol.iter.cmp, label %header.prol, label %header.prol.loopexit.unr-lcssa, !llvm.loop !10
 ; PROLOG:       header.prol.loopexit.unr-lcssa:
 ; PROLOG-NEXT:    %i6.unr.ph = phi i64 [ %add.prol, %latch.prol ]
@@ -4296,40 +4296,40 @@ define void @test8() {
 ; EPILOG-NEXT:    br label %innerH
 ; EPILOG:       innerH:
 ; EPILOG-NEXT:    %i3 = phi i64 [ %i, %outerloop.new ], [ %i4.7, %latch.7 ]
-; EPILOG-NEXT:    %niter = phi i64 [ %unroll_iter, %outerloop.new ], [ %niter.nsub.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.nsub = sub i64 %niter, 1
+; 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.nsub.1 = sub i64 %niter.nsub, 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.nsub.2 = sub i64 %niter.nsub.1, 1
+; 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.nsub.3 = sub i64 %niter.nsub.2, 1
+; 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.nsub.4 = sub i64 %niter.nsub.3, 1
+; 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.nsub.5 = sub i64 %niter.nsub.4, 1
+; 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.nsub.6 = sub i64 %niter.nsub.5, 1
+; 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:    br i1 false, label %outerloop.loopexit.loopexit, label %latch.7
 ; EPILOG:       latch.7:
-; EPILOG-NEXT:    %niter.nsub.7 = sub i64 %niter.nsub.6, 1
-; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.nsub.7, 0
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; 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:
 ; EPILOG-NEXT:    %i3.unr.ph = phi i64 [ %i4.7, %latch.7 ]
@@ -4342,13 +4342,13 @@ define void @test8() {
 ; EPILOG-NEXT:    br label %innerH.epil
 ; EPILOG:       innerH.epil:
 ; EPILOG-NEXT:    %i3.epil = phi i64 [ %i4.epil, %latch.epil ], [ %i3.unr, %innerH.epil.preheader ]
-; EPILOG-NEXT:    %epil.iter = phi i64 [ %xtraiter, %innerH.epil.preheader ], [ %epil.iter.sub, %latch.epil ]
+; EPILOG-NEXT:    %epil.iter = phi i64 [ 0, %innerH.epil.preheader ], [ %epil.iter.next, %latch.epil ]
 ; EPILOG-NEXT:    %i4.epil = add nuw nsw i64 %i3.epil, 1
 ; EPILOG-NEXT:    br i1 false, label %outerloop.loopexit.loopexit1, label %latch.epil
 ; EPILOG:       latch.epil:
 ; EPILOG-NEXT:    %i6.epil = icmp ult i64 %i4.epil, 100
-; EPILOG-NEXT:    %epil.iter.sub = sub i64 %epil.iter, 1
-; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.sub, 0
+; EPILOG-NEXT:    %epil.iter.next = add i64 %epil.iter, 1
+; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.next, %xtraiter
 ; EPILOG-NEXT:    br i1 %epil.iter.cmp, label %innerH.epil, label %exit.epilog-lcssa, !llvm.loop !11
 ; EPILOG:       exit.epilog-lcssa:
 ; EPILOG-NEXT:    br label %exit
@@ -4368,16 +4368,16 @@ define void @test8() {
 ; EPILOG-BLOCK-NEXT:    br label %innerH.1
 ; 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 [ 100, %outerloop.new.1 ], [ %niter.nsub.1.1, %latch.1.1 ]
-; EPILOG-BLOCK-NEXT:    %i4.11 = add nuw nsw i64 %i3.1, 1
-; EPILOG-BLOCK-NEXT:    br i1 false, label %outerloop.loopexit.loopexit.1, label %latch.13
-; EPILOG-BLOCK:       latch.13:
-; EPILOG-BLOCK-NEXT:    %niter.nsub.12 = sub i64 %niter.1, 1
-; EPILOG-BLOCK-NEXT:    %i4.1.1 = add nuw nsw i64 %i4.11, 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.1.1
 ; EPILOG-BLOCK:       latch.1.1:
-; EPILOG-BLOCK-NEXT:    %niter.nsub.1.1 = sub i64 %niter.nsub.12, 1
-; EPILOG-BLOCK-NEXT:    %niter.ncmp.1.1 = icmp ne i64 %niter.nsub.1.1, 0
+; EPILOG-BLOCK-NEXT:    %niter.next.1.1 = add i64 %niter.next.13, 1
+; 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 !11
 ; EPILOG-BLOCK:       exit.unr-lcssa.loopexit.1:
 ; EPILOG-BLOCK-NEXT:    br label %exit.unr-lcssa.1
@@ -4403,16 +4403,16 @@ define void @test8() {
 ; EPILOG-BLOCK-NEXT:    br label %innerH
 ; EPILOG-BLOCK:       innerH:
 ; EPILOG-BLOCK-NEXT:    %i3 = phi i64 [ %i, %outerloop.new ], [ %i4.1, %latch.1 ]
-; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ %unroll_iter, %outerloop.new ], [ %niter.nsub.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.nsub = sub i64 %niter, 1
+; 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:    br i1 false, label %outerloop.loopexit.loopexit, label %latch.1
 ; EPILOG-BLOCK:       latch.1:
-; EPILOG-BLOCK-NEXT:    %niter.nsub.1 = sub i64 %niter.nsub, 1
-; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.nsub.1, 0
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; 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 !11
 ; EPILOG-BLOCK:       exit.unr-lcssa.loopexit:
 ; EPILOG-BLOCK-NEXT:    br label %exit.unr-lcssa
@@ -4450,13 +4450,13 @@ define void @test8() {
 ; PROLOG-NEXT:    br label %innerH.prol
 ; PROLOG:       innerH.prol:
 ; PROLOG-NEXT:    %i3.prol = phi i64 [ %i4.prol, %latch.prol ], [ %i, %innerH.prol.preheader ]
-; PROLOG-NEXT:    %prol.iter = phi i64 [ %xtraiter, %innerH.prol.preheader ], [ %prol.iter.sub, %latch.prol ]
+; PROLOG-NEXT:    %prol.iter = phi i64 [ 0, %innerH.prol.preheader ], [ %prol.iter.next, %latch.prol ]
 ; PROLOG-NEXT:    %i4.prol = add nuw nsw i64 %i3.prol, 1
 ; PROLOG-NEXT:    br i1 false, label %outerloop.loopexit.loopexit1, label %latch.prol
 ; PROLOG:       latch.prol:
 ; PROLOG-NEXT:    %i6.prol = icmp ult i64 %i4.prol, 100
-; PROLOG-NEXT:    %prol.iter.sub = sub i64 %prol.iter, 1
-; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.sub, 0
+; PROLOG-NEXT:    %prol.iter.next = add i64 %prol.iter, 1
+; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.next, %xtraiter
 ; PROLOG-NEXT:    br i1 %prol.iter.cmp, label %innerH.prol, label %innerH.prol.loopexit.unr-lcssa, !llvm.loop !11
 ; PROLOG:       innerH.prol.loopexit.unr-lcssa:
 ; PROLOG-NEXT:    %i3.unr.ph = phi i64 [ %i4.prol, %latch.prol ]
@@ -4620,7 +4620,7 @@ define i8 addrspace(1)* @test9(i8* nocapture readonly %arg, i32 %n) {
 ; EPILOG-NEXT:    br label %header
 ; EPILOG:       header:
 ; EPILOG-NEXT:    %phi = phi i64 [ %i4, %preheader.new ], [ %iv.next.7, %latch.7 ]
-; EPILOG-NEXT:    %niter = phi i32 [ %unroll_iter, %preheader.new ], [ %niter.nsub.7, %latch.7 ]
+; EPILOG-NEXT:    %niter = phi i32 [ 0, %preheader.new ], [ %niter.next.7, %latch.7 ]
 ; EPILOG-NEXT:    br i1 true, label %latch, label %innerexit.loopexit
 ; EPILOG:       innerexit.loopexit:
 ; EPILOG-NEXT:    %trip.lcssa.ph = phi i32 [ %trip, %header ], [ %trip, %latch ], [ %trip, %latch.1 ], [ %trip, %latch.2 ], [ %trip, %latch.3 ], [ %trip, %latch.4 ], [ %trip, %latch.5 ], [ %trip, %latch.6 ]
@@ -4634,36 +4634,36 @@ define i8 addrspace(1)* @test9(i8* nocapture readonly %arg, i32 %n) {
 ; EPILOG-NEXT:    ret i8 addrspace(1)* %i9
 ; EPILOG:       latch:
 ; EPILOG-NEXT:    %iv.next = add nuw nsw i64 %phi, 1
-; EPILOG-NEXT:    %niter.nsub = sub i32 %niter, 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.nsub.1 = sub i32 %niter.nsub, 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.nsub.2 = sub i32 %niter.nsub.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.nsub.3 = sub i32 %niter.nsub.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.nsub.4 = sub i32 %niter.nsub.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.nsub.5 = sub i32 %niter.nsub.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.nsub.6 = sub i32 %niter.nsub.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.nsub.7 = sub i32 %niter.nsub.6, 1
-; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i32 %niter.nsub.7, 0
+; EPILOG-NEXT:    %niter.next.7 = add i32 %niter.next.6, 1
+; 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:
 ; EPILOG-NEXT:    %phi.unr.ph = phi i64 [ %iv.next.7, %latch.7 ]
@@ -4676,15 +4676,15 @@ define i8 addrspace(1)* @test9(i8* nocapture readonly %arg, i32 %n) {
 ; EPILOG-NEXT:    br label %header.epil
 ; EPILOG:       header.epil:
 ; EPILOG-NEXT:    %phi.epil = phi i64 [ %phi.unr, %header.epil.preheader ], [ %iv.next.epil, %latch.epil ]
-; EPILOG-NEXT:    %epil.iter = phi i32 [ %xtraiter, %header.epil.preheader ], [ %epil.iter.sub, %latch.epil ]
+; EPILOG-NEXT:    %epil.iter = phi i32 [ 0, %header.epil.preheader ], [ %epil.iter.next, %latch.epil ]
 ; EPILOG-NEXT:    %i7.epil = trunc i64 %phi.epil to i32
 ; EPILOG-NEXT:    br i1 true, label %latch.epil, label %innerexit.loopexit1
 ; EPILOG:       latch.epil:
 ; EPILOG-NEXT:    %i11.epil = add nsw i32 %i7.epil, 1
 ; EPILOG-NEXT:    %innercnd.epil = icmp slt i32 %i11.epil, %trip
 ; EPILOG-NEXT:    %iv.next.epil = add nuw nsw i64 %phi.epil, 1
-; EPILOG-NEXT:    %epil.iter.sub = sub i32 %epil.iter, 1
-; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i32 %epil.iter.sub, 0
+; EPILOG-NEXT:    %epil.iter.next = add i32 %epil.iter, 1
+; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i32 %epil.iter.next, %xtraiter
 ; EPILOG-NEXT:    br i1 %epil.iter.cmp, label %header.epil, label %outerLatch.loopexit.epilog-lcssa, !llvm.loop !12
 ; EPILOG:       outerLatch.loopexit.epilog-lcssa:
 ; EPILOG-NEXT:    br label %outerLatch.loopexit
@@ -4711,7 +4711,7 @@ define i8 addrspace(1)* @test9(i8* nocapture readonly %arg, i32 %n) {
 ; EPILOG-BLOCK-NEXT:    br label %header
 ; EPILOG-BLOCK:       header:
 ; EPILOG-BLOCK-NEXT:    %phi = phi i64 [ 0, %preheader.new ], [ %iv.next.1, %latch.1 ]
-; EPILOG-BLOCK-NEXT:    %niter = phi i32 [ %unroll_iter, %preheader.new ], [ %niter.nsub.1, %latch.1 ]
+; EPILOG-BLOCK-NEXT:    %niter = phi i32 [ 0, %preheader.new ], [ %niter.next.1, %latch.1 ]
 ; EPILOG-BLOCK-NEXT:    br i1 true, label %latch, label %innerexit.loopexit.loopexit
 ; EPILOG-BLOCK:       innerexit.loopexit.loopexit:
 ; EPILOG-BLOCK-NEXT:    %trip.lcssa.ph.ph = phi i32 [ %trip, %latch ], [ %trip, %header ]
@@ -4731,12 +4731,12 @@ define i8 addrspace(1)* @test9(i8* nocapture readonly %arg, i32 %n) {
 ; EPILOG-BLOCK-NEXT:    ret i8 addrspace(1)* %i9
 ; EPILOG-BLOCK:       latch:
 ; EPILOG-BLOCK-NEXT:    %iv.next = add nuw nsw i64 %phi, 1
-; EPILOG-BLOCK-NEXT:    %niter.nsub = sub i32 %niter, 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.nsub.1 = sub i32 %niter.nsub, 1
-; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i32 %niter.nsub.1, 0
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i32 %niter.next, 1
+; 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 !13
 ; EPILOG-BLOCK:       outerLatch.loopexit.unr-lcssa.loopexit:
 ; EPILOG-BLOCK-NEXT:    br label %outerLatch.loopexit.unr-lcssa
@@ -4764,16 +4764,16 @@ define i8 addrspace(1)* @test9(i8* nocapture readonly %arg, i32 %n) {
 ; EPILOG-BLOCK-NEXT:    br label %header.1
 ; 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 [ %unroll_iter.1, %preheader.new.1 ], [ %niter.nsub.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.nsub.14 = sub i32 %niter.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:       latch.1.1:
 ; EPILOG-BLOCK-NEXT:    %iv.next.1.1 = add nuw nsw i64 %iv.next.13, 1
-; EPILOG-BLOCK-NEXT:    %niter.nsub.1.1 = sub i32 %niter.nsub.14, 1
-; EPILOG-BLOCK-NEXT:    %niter.ncmp.1.1 = icmp ne i32 %niter.nsub.1.1, 0
+; EPILOG-BLOCK-NEXT:    %niter.next.1.1 = add i32 %niter.next.14, 1
+; 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 !13
 ; EPILOG-BLOCK:       outerLatch.loopexit.unr-lcssa.loopexit.1:
 ; EPILOG-BLOCK-NEXT:    br label %outerLatch.loopexit.unr-lcssa.1
@@ -4809,15 +4809,15 @@ define i8 addrspace(1)* @test9(i8* nocapture readonly %arg, i32 %n) {
 ; PROLOG-NEXT:    br label %header.prol
 ; PROLOG:       header.prol:
 ; PROLOG-NEXT:    %phi.prol = phi i64 [ %i4, %header.prol.preheader ], [ %iv.next.prol, %latch.prol ]
-; PROLOG-NEXT:    %prol.iter = phi i32 [ %xtraiter, %header.prol.preheader ], [ %prol.iter.sub, %latch.prol ]
+; PROLOG-NEXT:    %prol.iter = phi i32 [ 0, %header.prol.preheader ], [ %prol.iter.next, %latch.prol ]
 ; PROLOG-NEXT:    %i7.prol = trunc i64 %phi.prol to i32
 ; PROLOG-NEXT:    br i1 true, label %latch.prol, label %innerexit.loopexit1
 ; PROLOG:       latch.prol:
 ; PROLOG-NEXT:    %i11.prol = add nsw i32 %i7.prol, 1
 ; PROLOG-NEXT:    %innercnd.prol = icmp slt i32 %i11.prol, %trip
 ; PROLOG-NEXT:    %iv.next.prol = add nuw nsw i64 %phi.prol, 1
-; PROLOG-NEXT:    %prol.iter.sub = sub i32 %prol.iter, 1
-; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i32 %prol.iter.sub, 0
+; PROLOG-NEXT:    %prol.iter.next = add i32 %prol.iter, 1
+; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i32 %prol.iter.next, %xtraiter
 ; PROLOG-NEXT:    br i1 %prol.iter.cmp, label %header.prol, label %header.prol.loopexit.unr-lcssa, !llvm.loop !12
 ; PROLOG:       header.prol.loopexit.unr-lcssa:
 ; PROLOG-NEXT:    %phi.unr.ph = phi i64 [ %iv.next.prol, %latch.prol ]
@@ -5019,56 +5019,56 @@ define void @test10(i64 %trip, i64 %trip2) {
 ; EPILOG-NEXT:    br label %loop_header
 ; EPILOG:       loop_header:
 ; EPILOG-NEXT:    %iv = phi i64 [ 0, %entry.new ], [ %iv_next.7, %loop_latch.7 ]
-; EPILOG-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.7, %loop_latch.7 ]
+; EPILOG-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.7, %loop_latch.7 ]
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early = icmp ne i64 %iv, %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.nsub = sub i64 %niter, 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.nsub.1 = sub i64 %niter.nsub, 1
+; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
 ; 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.nsub.2 = sub i64 %niter.nsub.1, 1
+; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
 ; 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.nsub.3 = sub i64 %niter.nsub.2, 1
+; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
 ; 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.nsub.4 = sub i64 %niter.nsub.3, 1
+; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
 ; 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.nsub.5 = sub i64 %niter.nsub.4, 1
+; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
 ; 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.nsub.6 = sub i64 %niter.nsub.5, 1
+; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
 ; 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.nsub.7 = sub i64 %niter.nsub.6, 1
-; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.nsub.7, 0
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; 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:
 ; EPILOG-NEXT:    br label %exit1
@@ -5087,15 +5087,15 @@ define void @test10(i64 %trip, i64 %trip2) {
 ; EPILOG-NEXT:    br label %loop_header.epil
 ; EPILOG:       loop_header.epil:
 ; EPILOG-NEXT:    %iv.epil = phi i64 [ %iv.unr, %loop_header.epil.preheader ], [ %iv_next.epil, %loop_latch.epil ]
-; EPILOG-NEXT:    %epil.iter = phi i64 [ %xtraiter, %loop_header.epil.preheader ], [ %epil.iter.sub, %loop_latch.epil ]
+; EPILOG-NEXT:    %epil.iter = phi i64 [ 0, %loop_header.epil.preheader ], [ %epil.iter.next, %loop_latch.epil ]
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.epil = icmp ne i64 %iv.epil, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.epil, label %loop_latch.epil, label %exit1.loopexit1
 ; EPILOG:       loop_latch.epil:
 ; EPILOG-NEXT:    %iv_next.epil = add i64 %iv.epil, 1
 ; EPILOG-NEXT:    %cmp.epil = icmp ne i64 %iv_next.epil, %trip
-; EPILOG-NEXT:    %epil.iter.sub = sub i64 %epil.iter, 1
-; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.sub, 0
+; EPILOG-NEXT:    %epil.iter.next = add i64 %epil.iter, 1
+; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.next, %xtraiter
 ; EPILOG-NEXT:    br i1 %epil.iter.cmp, label %loop_header.epil, label %exit2.epilog-lcssa, !llvm.loop !13
 ; EPILOG:       exit2.epilog-lcssa:
 ; EPILOG-NEXT:    br label %exit2
@@ -5113,20 +5113,20 @@ define void @test10(i64 %trip, i64 %trip2) {
 ; EPILOG-BLOCK-NEXT:    br label %loop_header
 ; EPILOG-BLOCK:       loop_header:
 ; EPILOG-BLOCK-NEXT:    %iv = phi i64 [ 0, %entry.new ], [ %iv_next.1, %loop_latch.1 ]
-; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.1, %loop_latch.1 ]
+; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.1, %loop_latch.1 ]
 ; EPILOG-BLOCK-NEXT:    call void @bar()
 ; EPILOG-BLOCK-NEXT:    %cmp_early = icmp ne i64 %iv, %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.nsub = sub i64 %niter, 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.nsub.1 = sub i64 %niter.nsub, 1
-; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.nsub.1, 0
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; 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 !15
 ; EPILOG-BLOCK:       exit1.loopexit:
 ; EPILOG-BLOCK-NEXT:    br label %exit1
@@ -5160,15 +5160,15 @@ define void @test10(i64 %trip, i64 %trip2) {
 ; PROLOG-NEXT:    br label %loop_header.prol
 ; PROLOG:       loop_header.prol:
 ; PROLOG-NEXT:    %iv.prol = phi i64 [ 0, %loop_header.prol.preheader ], [ %iv_next.prol, %loop_latch.prol ]
-; PROLOG-NEXT:    %prol.iter = phi i64 [ %xtraiter, %loop_header.prol.preheader ], [ %prol.iter.sub, %loop_latch.prol ]
+; PROLOG-NEXT:    %prol.iter = phi i64 [ 0, %loop_header.prol.preheader ], [ %prol.iter.next, %loop_latch.prol ]
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.prol = icmp ne i64 %iv.prol, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.prol, label %loop_latch.prol, label %exit1.loopexit1
 ; PROLOG:       loop_latch.prol:
 ; PROLOG-NEXT:    %iv_next.prol = add i64 %iv.prol, 1
 ; PROLOG-NEXT:    %cmp.prol = icmp ne i64 %iv_next.prol, %trip
-; PROLOG-NEXT:    %prol.iter.sub = sub i64 %prol.iter, 1
-; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.sub, 0
+; PROLOG-NEXT:    %prol.iter.next = add i64 %prol.iter, 1
+; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.next, %xtraiter
 ; PROLOG-NEXT:    br i1 %prol.iter.cmp, label %loop_header.prol, label %loop_header.prol.loopexit.unr-lcssa, !llvm.loop !13
 ; PROLOG:       loop_header.prol.loopexit.unr-lcssa:
 ; PROLOG-NEXT:    %iv.unr.ph = phi i64 [ %iv_next.prol, %loop_latch.prol ]
@@ -5312,48 +5312,48 @@ define void @test11(i64 %trip, i1 %cond) {
 ; EPILOG-NEXT:    br label %loop_header
 ; EPILOG:       loop_header:
 ; EPILOG-NEXT:    %iv = phi i64 [ 0, %entry.new ], [ %iv_next.7, %loop_latch.7 ]
-; EPILOG-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.7, %loop_latch.7 ]
+; EPILOG-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.7, %loop_latch.7 ]
 ; 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.nsub = sub i64 %niter, 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.nsub.1 = sub i64 %niter.nsub, 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.nsub.2 = sub i64 %niter.nsub.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.nsub.3 = sub i64 %niter.nsub.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.nsub.4 = sub i64 %niter.nsub.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.nsub.5 = sub i64 %niter.nsub.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.nsub.6 = sub i64 %niter.nsub.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.nsub.7 = sub i64 %niter.nsub.6, 1
-; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.nsub.7, 0
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; 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:
 ; EPILOG-NEXT:    br label %exit1
@@ -5372,14 +5372,14 @@ define void @test11(i64 %trip, i1 %cond) {
 ; EPILOG-NEXT:    br label %loop_header.epil
 ; EPILOG:       loop_header.epil:
 ; EPILOG-NEXT:    %iv.epil = phi i64 [ %iv.unr, %loop_header.epil.preheader ], [ %iv_next.epil, %loop_latch.epil ]
-; EPILOG-NEXT:    %epil.iter = phi i64 [ %xtraiter, %loop_header.epil.preheader ], [ %epil.iter.sub, %loop_latch.epil ]
+; EPILOG-NEXT:    %epil.iter = phi i64 [ 0, %loop_header.epil.preheader ], [ %epil.iter.next, %loop_latch.epil ]
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    br i1 %cond, label %loop_latch.epil, label %exit1.loopexit1
 ; EPILOG:       loop_latch.epil:
 ; EPILOG-NEXT:    %iv_next.epil = add i64 %iv.epil, 1
 ; EPILOG-NEXT:    %cmp.epil = icmp ne i64 %iv_next.epil, %trip
-; EPILOG-NEXT:    %epil.iter.sub = sub i64 %epil.iter, 1
-; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.sub, 0
+; EPILOG-NEXT:    %epil.iter.next = add i64 %epil.iter, 1
+; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.next, %xtraiter
 ; EPILOG-NEXT:    br i1 %epil.iter.cmp, label %loop_header.epil, label %exit2.epilog-lcssa, !llvm.loop !14
 ; EPILOG:       exit2.epilog-lcssa:
 ; EPILOG-NEXT:    br label %exit2
@@ -5397,18 +5397,18 @@ define void @test11(i64 %trip, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    br label %loop_header
 ; EPILOG-BLOCK:       loop_header:
 ; EPILOG-BLOCK-NEXT:    %iv = phi i64 [ 0, %entry.new ], [ %iv_next.1, %loop_latch.1 ]
-; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.1, %loop_latch.1 ]
+; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.1, %loop_latch.1 ]
 ; 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.nsub = sub i64 %niter, 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.nsub.1 = sub i64 %niter.nsub, 1
-; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.nsub.1, 0
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; 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:
 ; EPILOG-BLOCK-NEXT:    br label %exit1
@@ -5439,14 +5439,14 @@ define void @test11(i64 %trip, i1 %cond) {
 ; PROLOG-NEXT:    br label %loop_header.prol
 ; PROLOG:       loop_header.prol:
 ; PROLOG-NEXT:    %iv.prol = phi i64 [ 0, %loop_header.prol.preheader ], [ %iv_next.prol, %loop_latch.prol ]
-; PROLOG-NEXT:    %prol.iter = phi i64 [ %xtraiter, %loop_header.prol.preheader ], [ %prol.iter.sub, %loop_latch.prol ]
+; PROLOG-NEXT:    %prol.iter = phi i64 [ 0, %loop_header.prol.preheader ], [ %prol.iter.next, %loop_latch.prol ]
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    br i1 %cond, label %loop_latch.prol, label %exit1.loopexit1
 ; PROLOG:       loop_latch.prol:
 ; PROLOG-NEXT:    %iv_next.prol = add i64 %iv.prol, 1
 ; PROLOG-NEXT:    %cmp.prol = icmp ne i64 %iv_next.prol, %trip
-; PROLOG-NEXT:    %prol.iter.sub = sub i64 %prol.iter, 1
-; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.sub, 0
+; PROLOG-NEXT:    %prol.iter.next = add i64 %prol.iter, 1
+; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.next, %xtraiter
 ; PROLOG-NEXT:    br i1 %prol.iter.cmp, label %loop_header.prol, label %loop_header.prol.loopexit.unr-lcssa, !llvm.loop !14
 ; PROLOG:       loop_header.prol.loopexit.unr-lcssa:
 ; PROLOG-NEXT:    %iv.unr.ph = phi i64 [ %iv_next.prol, %loop_latch.prol ]
@@ -5578,7 +5578,7 @@ define void @test12(i64 %trip, i64 %trip2, i1 %cond) {
 ; EPILOG-NEXT:    br label %loop_header
 ; EPILOG:       loop_header:
 ; EPILOG-NEXT:    %iv = phi i64 [ 0, %entry.new ], [ %iv_next.7, %loop_latch.7 ]
-; EPILOG-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.7, %loop_latch.7 ]
+; EPILOG-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.7, %loop_latch.7 ]
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early = icmp ne i64 %iv, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early, label %loop_exiting_bb2, label %exit1.epilog-lcssa.loopexit
@@ -5586,7 +5586,7 @@ 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.nsub = sub i64 %niter, 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
@@ -5594,7 +5594,7 @@ define void @test12(i64 %trip, i64 %trip2, i1 %cond) {
 ; 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.nsub.1 = sub i64 %niter.nsub, 1
+; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
 ; 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
@@ -5602,7 +5602,7 @@ define void @test12(i64 %trip, i64 %trip2, i1 %cond) {
 ; 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.nsub.2 = sub i64 %niter.nsub.1, 1
+; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
 ; 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
@@ -5610,7 +5610,7 @@ define void @test12(i64 %trip, i64 %trip2, i1 %cond) {
 ; 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.nsub.3 = sub i64 %niter.nsub.2, 1
+; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
 ; 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
@@ -5618,7 +5618,7 @@ define void @test12(i64 %trip, i64 %trip2, i1 %cond) {
 ; 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.nsub.4 = sub i64 %niter.nsub.3, 1
+; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
 ; 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
@@ -5626,7 +5626,7 @@ define void @test12(i64 %trip, i64 %trip2, i1 %cond) {
 ; 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.nsub.5 = sub i64 %niter.nsub.4, 1
+; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
 ; 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
@@ -5634,7 +5634,7 @@ define void @test12(i64 %trip, i64 %trip2, i1 %cond) {
 ; 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.nsub.6 = sub i64 %niter.nsub.5, 1
+; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
 ; 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
@@ -5642,8 +5642,8 @@ define void @test12(i64 %trip, i64 %trip2, i1 %cond) {
 ; 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.nsub.7 = sub i64 %niter.nsub.6, 1
-; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.nsub.7, 0
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; 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:
 ; EPILOG-NEXT:    %iv.unr.ph = phi i64 [ %iv_next.7, %loop_latch.7 ]
@@ -5656,7 +5656,7 @@ define void @test12(i64 %trip, i64 %trip2, i1 %cond) {
 ; EPILOG-NEXT:    br label %loop_header.epil
 ; EPILOG:       loop_header.epil:
 ; EPILOG-NEXT:    %iv.epil = phi i64 [ %iv.unr, %loop_header.epil.preheader ], [ %iv_next.epil, %loop_latch.epil ]
-; EPILOG-NEXT:    %epil.iter = phi i64 [ %xtraiter, %loop_header.epil.preheader ], [ %epil.iter.sub, %loop_latch.epil ]
+; EPILOG-NEXT:    %epil.iter = phi i64 [ 0, %loop_header.epil.preheader ], [ %epil.iter.next, %loop_latch.epil ]
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.epil = icmp ne i64 %iv.epil, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.epil, label %loop_exiting_bb2.epil, label %exit1.epilog-lcssa.loopexit1
@@ -5665,8 +5665,8 @@ define void @test12(i64 %trip, i64 %trip2, i1 %cond) {
 ; EPILOG:       loop_latch.epil:
 ; EPILOG-NEXT:    %iv_next.epil = add i64 %iv.epil, 1
 ; EPILOG-NEXT:    %cmp.epil = icmp ne i64 %iv_next.epil, %trip
-; EPILOG-NEXT:    %epil.iter.sub = sub i64 %epil.iter, 1
-; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.sub, 0
+; EPILOG-NEXT:    %epil.iter.next = add i64 %epil.iter, 1
+; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.next, %xtraiter
 ; EPILOG-NEXT:    br i1 %epil.iter.cmp, label %loop_header.epil, label %exit1.epilog-lcssa.loopexit1, !llvm.loop !15
 ; EPILOG:       exit1.epilog-lcssa.loopexit:
 ; EPILOG-NEXT:    br label %exit1.epilog-lcssa
@@ -5688,7 +5688,7 @@ define void @test12(i64 %trip, i64 %trip2, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    br label %loop_header
 ; EPILOG-BLOCK:       loop_header:
 ; EPILOG-BLOCK-NEXT:    %iv = phi i64 [ 0, %entry.new ], [ %iv_next.1, %loop_latch.1 ]
-; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.1, %loop_latch.1 ]
+; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.1, %loop_latch.1 ]
 ; EPILOG-BLOCK-NEXT:    call void @bar()
 ; EPILOG-BLOCK-NEXT:    %cmp_early = icmp ne i64 %iv, %trip2
 ; EPILOG-BLOCK-NEXT:    br i1 %cmp_early, label %loop_exiting_bb2, label %exit1.epilog-lcssa.loopexit
@@ -5696,7 +5696,7 @@ 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.nsub = sub i64 %niter, 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
@@ -5704,8 +5704,8 @@ define void @test12(i64 %trip, i64 %trip2, i1 %cond) {
 ; 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.nsub.1 = sub i64 %niter.nsub, 1
-; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.nsub.1, 0
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; 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 !17
 ; EPILOG-BLOCK:       exit1.unr-lcssa.loopexit:
 ; EPILOG-BLOCK-NEXT:    %iv.unr.ph = phi i64 [ %iv_next.1, %loop_latch.1 ]
@@ -5741,7 +5741,7 @@ define void @test12(i64 %trip, i64 %trip2, i1 %cond) {
 ; PROLOG-NEXT:    br label %loop_header.prol
 ; PROLOG:       loop_header.prol:
 ; PROLOG-NEXT:    %iv.prol = phi i64 [ 0, %loop_header.prol.preheader ], [ %iv_next.prol, %loop_latch.prol ]
-; PROLOG-NEXT:    %prol.iter = phi i64 [ %xtraiter, %loop_header.prol.preheader ], [ %prol.iter.sub, %loop_latch.prol ]
+; PROLOG-NEXT:    %prol.iter = phi i64 [ 0, %loop_header.prol.preheader ], [ %prol.iter.next, %loop_latch.prol ]
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.prol = icmp ne i64 %iv.prol, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.prol, label %loop_exiting_bb2.prol, label %exit1.unr-lcssa.loopexit1
@@ -5750,8 +5750,8 @@ define void @test12(i64 %trip, i64 %trip2, i1 %cond) {
 ; PROLOG:       loop_latch.prol:
 ; PROLOG-NEXT:    %iv_next.prol = add i64 %iv.prol, 1
 ; PROLOG-NEXT:    %cmp.prol = icmp ne i64 %iv_next.prol, %trip
-; PROLOG-NEXT:    %prol.iter.sub = sub i64 %prol.iter, 1
-; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.sub, 0
+; PROLOG-NEXT:    %prol.iter.next = add i64 %prol.iter, 1
+; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.next, %xtraiter
 ; PROLOG-NEXT:    br i1 %prol.iter.cmp, label %loop_header.prol, label %loop_header.prol.loopexit.unr-lcssa, !llvm.loop !15
 ; PROLOG:       loop_header.prol.loopexit.unr-lcssa:
 ; PROLOG-NEXT:    %iv.unr.ph = phi i64 [ %iv_next.prol, %loop_latch.prol ]
@@ -5914,7 +5914,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; EPILOG-NEXT:    br label %loop_header
 ; EPILOG:       loop_header:
 ; EPILOG-NEXT:    %iv = phi i64 [ 0, %entry.new ], [ %iv_next.7, %loop_latch.7 ]
-; EPILOG-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.7, %loop_latch.7 ]
+; EPILOG-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.7, %loop_latch.7 ]
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early = icmp ne i64 %iv, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early, label %loop_exiting_bb2, label %exit1.epilog-lcssa.loopexit
@@ -5923,7 +5923,7 @@ 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.nsub = sub i64 %niter, 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
@@ -5932,7 +5932,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; 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.nsub.1 = sub i64 %niter.nsub, 1
+; EPILOG-NEXT:    %niter.next.1 = add nuw nsw i64 %niter.next, 1
 ; 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
@@ -5941,7 +5941,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; 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.nsub.2 = sub i64 %niter.nsub.1, 1
+; EPILOG-NEXT:    %niter.next.2 = add nuw nsw i64 %niter.next.1, 1
 ; 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
@@ -5950,7 +5950,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; 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.nsub.3 = sub i64 %niter.nsub.2, 1
+; EPILOG-NEXT:    %niter.next.3 = add nuw nsw i64 %niter.next.2, 1
 ; 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
@@ -5959,7 +5959,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; 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.nsub.4 = sub i64 %niter.nsub.3, 1
+; EPILOG-NEXT:    %niter.next.4 = add nuw nsw i64 %niter.next.3, 1
 ; 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
@@ -5968,7 +5968,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; 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.nsub.5 = sub i64 %niter.nsub.4, 1
+; EPILOG-NEXT:    %niter.next.5 = add nuw nsw i64 %niter.next.4, 1
 ; 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
@@ -5977,7 +5977,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; 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.nsub.6 = sub i64 %niter.nsub.5, 1
+; EPILOG-NEXT:    %niter.next.6 = add nuw nsw i64 %niter.next.5, 1
 ; 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
@@ -5986,8 +5986,8 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; 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.nsub.7 = sub i64 %niter.nsub.6, 1
-; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.nsub.7, 0
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; 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:
 ; EPILOG-NEXT:    %iv.unr.ph = phi i64 [ %iv_next.7, %loop_latch.7 ]
@@ -6000,7 +6000,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; EPILOG-NEXT:    br label %loop_header.epil
 ; EPILOG:       loop_header.epil:
 ; EPILOG-NEXT:    %iv.epil = phi i64 [ %iv.unr, %loop_header.epil.preheader ], [ %iv_next.epil, %loop_latch.epil ]
-; EPILOG-NEXT:    %epil.iter = phi i64 [ %xtraiter, %loop_header.epil.preheader ], [ %epil.iter.sub, %loop_latch.epil ]
+; EPILOG-NEXT:    %epil.iter = phi i64 [ 0, %loop_header.epil.preheader ], [ %epil.iter.next, %loop_latch.epil ]
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    %cmp_early.epil = icmp ne i64 %iv.epil, %trip2
 ; EPILOG-NEXT:    br i1 %cmp_early.epil, label %loop_exiting_bb2.epil, label %exit1.epilog-lcssa.loopexit1
@@ -6010,8 +6010,8 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; EPILOG:       loop_latch.epil:
 ; EPILOG-NEXT:    %iv_next.epil = add i64 %iv.epil, 1
 ; EPILOG-NEXT:    %cmp.epil = icmp ne i64 %iv_next.epil, %trip
-; EPILOG-NEXT:    %epil.iter.sub = sub i64 %epil.iter, 1
-; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.sub, 0
+; EPILOG-NEXT:    %epil.iter.next = add i64 %epil.iter, 1
+; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.next, %xtraiter
 ; EPILOG-NEXT:    br i1 %epil.iter.cmp, label %loop_header.epil, label %exit1.epilog-lcssa.loopexit1, !llvm.loop !16
 ; EPILOG:       exit1.epilog-lcssa.loopexit:
 ; EPILOG-NEXT:    br label %exit1.epilog-lcssa
@@ -6033,7 +6033,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; EPILOG-BLOCK-NEXT:    br label %loop_header
 ; EPILOG-BLOCK:       loop_header:
 ; EPILOG-BLOCK-NEXT:    %iv = phi i64 [ 0, %entry.new ], [ %iv_next.1, %loop_latch.1 ]
-; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.1, %loop_latch.1 ]
+; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.1, %loop_latch.1 ]
 ; EPILOG-BLOCK-NEXT:    call void @bar()
 ; EPILOG-BLOCK-NEXT:    %cmp_early = icmp ne i64 %iv, %trip2
 ; EPILOG-BLOCK-NEXT:    br i1 %cmp_early, label %loop_exiting_bb2, label %exit1.epilog-lcssa.loopexit
@@ -6042,7 +6042,7 @@ 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.nsub = sub i64 %niter, 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
@@ -6051,8 +6051,8 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; 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.nsub.1 = sub i64 %niter.nsub, 1
-; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.nsub.1, 0
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; 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:
 ; EPILOG-BLOCK-NEXT:    %iv.unr.ph = phi i64 [ %iv_next.1, %loop_latch.1 ]
@@ -6089,7 +6089,7 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; PROLOG-NEXT:    br label %loop_header.prol
 ; PROLOG:       loop_header.prol:
 ; PROLOG-NEXT:    %iv.prol = phi i64 [ 0, %loop_header.prol.preheader ], [ %iv_next.prol, %loop_latch.prol ]
-; PROLOG-NEXT:    %prol.iter = phi i64 [ %xtraiter, %loop_header.prol.preheader ], [ %prol.iter.sub, %loop_latch.prol ]
+; PROLOG-NEXT:    %prol.iter = phi i64 [ 0, %loop_header.prol.preheader ], [ %prol.iter.next, %loop_latch.prol ]
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    %cmp_early.prol = icmp ne i64 %iv.prol, %trip2
 ; PROLOG-NEXT:    br i1 %cmp_early.prol, label %loop_exiting_bb2.prol, label %exit1.unr-lcssa.loopexit1
@@ -6099,8 +6099,8 @@ define void @test13(i64 %trip, i64 %trip2) {
 ; PROLOG:       loop_latch.prol:
 ; PROLOG-NEXT:    %iv_next.prol = add i64 %iv.prol, 1
 ; PROLOG-NEXT:    %cmp.prol = icmp ne i64 %iv_next.prol, %trip
-; PROLOG-NEXT:    %prol.iter.sub = sub i64 %prol.iter, 1
-; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.sub, 0
+; PROLOG-NEXT:    %prol.iter.next = add i64 %prol.iter, 1
+; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.next, %xtraiter
 ; PROLOG-NEXT:    br i1 %prol.iter.cmp, label %loop_header.prol, label %loop_header.prol.loopexit.unr-lcssa, !llvm.loop !16
 ; PROLOG:       loop_header.prol.loopexit.unr-lcssa:
 ; PROLOG-NEXT:    %iv.unr.ph = phi i64 [ %iv_next.prol, %loop_latch.prol ]
@@ -6273,7 +6273,7 @@ define void @test14(i64 %trip, i1 %cond) {
 ; EPILOG-NEXT:    br label %loop_header
 ; EPILOG:       loop_header:
 ; EPILOG-NEXT:    %iv = phi i64 [ 0, %entry.new ], [ %iv_next.7, %loop_latch.7 ]
-; EPILOG-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.7, %loop_latch.7 ]
+; EPILOG-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.7, %loop_latch.7 ]
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    br i1 %cond, label %loop_exiting_bb2, label %exit1.epilog-lcssa.loopexit
 ; EPILOG:       loop_exiting_bb2:
@@ -6281,7 +6281,7 @@ define void @test14(i64 %trip, i1 %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.nsub = sub i64 %niter, 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:
@@ -6289,7 +6289,7 @@ define void @test14(i64 %trip, i1 %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.nsub.1 = sub i64 %niter.nsub, 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:
@@ -6297,7 +6297,7 @@ define void @test14(i64 %trip, i1 %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.nsub.2 = sub i64 %niter.nsub.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:
@@ -6305,7 +6305,7 @@ define void @test14(i64 %trip, i1 %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.nsub.3 = sub i64 %niter.nsub.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:
@@ -6313,7 +6313,7 @@ define void @test14(i64 %trip, i1 %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.nsub.4 = sub i64 %niter.nsub.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:
@@ -6321,7 +6321,7 @@ define void @test14(i64 %trip, i1 %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.nsub.5 = sub i64 %niter.nsub.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:
@@ -6329,7 +6329,7 @@ define void @test14(i64 %trip, i1 %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.nsub.6 = sub i64 %niter.nsub.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:
@@ -6337,8 +6337,8 @@ define void @test14(i64 %trip, i1 %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.nsub.7 = sub i64 %niter.nsub.6, 1
-; EPILOG-NEXT:    %niter.ncmp.7 = icmp ne i64 %niter.nsub.7, 0
+; EPILOG-NEXT:    %niter.next.7 = add i64 %niter.next.6, 1
+; 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:
 ; EPILOG-NEXT:    %iv.unr.ph = phi i64 [ %iv_next.7, %loop_latch.7 ]
@@ -6351,7 +6351,7 @@ define void @test14(i64 %trip, i1 %cond) {
 ; EPILOG-NEXT:    br label %loop_header.epil
 ; EPILOG:       loop_header.epil:
 ; EPILOG-NEXT:    %iv.epil = phi i64 [ %iv.unr, %loop_header.epil.preheader ], [ %iv_next.epil, %loop_latch.epil ]
-; EPILOG-NEXT:    %epil.iter = phi i64 [ %xtraiter, %loop_header.epil.preheader ], [ %epil.iter.sub, %loop_latch.epil ]
+; EPILOG-NEXT:    %epil.iter = phi i64 [ 0, %loop_header.epil.preheader ], [ %epil.iter.next, %loop_latch.epil ]
 ; EPILOG-NEXT:    call void @bar()
 ; EPILOG-NEXT:    br i1 %cond, label %loop_exiting_bb2.epil, label %exit1.epilog-lcssa.loopexit1
 ; EPILOG:       loop_exiting_bb2.epil:
@@ -6360,8 +6360,8 @@ define void @test14(i64 %trip, i1 %cond) {
 ; EPILOG:       loop_latch.epil:
 ; EPILOG-NEXT:    %iv_next.epil = add i64 %iv.epil, 1
 ; EPILOG-NEXT:    %cmp.epil = icmp ne i64 %iv_next.epil, %trip
-; EPILOG-NEXT:    %epil.iter.sub = sub i64 %epil.iter, 1
-; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.sub, 0
+; EPILOG-NEXT:    %epil.iter.next = add i64 %epil.iter, 1
+; EPILOG-NEXT:    %epil.iter.cmp = icmp ne i64 %epil.iter.next, %xtraiter
 ; EPILOG-NEXT:    br i1 %epil.iter.cmp, label %loop_header.epil, label %exit1.epilog-lcssa.loopexit1, !llvm.loop !17
 ; EPILOG:       exit1.epilog-lcssa.loopexit:
 ; EPILOG-NEXT:    br label %exit1.epilog-lcssa
@@ -6383,7 +6383,7 @@ define void @test14(i64 %trip, i1 %cond) {
 ; EPILOG-BLOCK-NEXT:    br label %loop_header
 ; EPILOG-BLOCK:       loop_header:
 ; EPILOG-BLOCK-NEXT:    %iv = phi i64 [ 0, %entry.new ], [ %iv_next.1, %loop_latch.1 ]
-; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ %unroll_iter, %entry.new ], [ %niter.nsub.1, %loop_latch.1 ]
+; EPILOG-BLOCK-NEXT:    %niter = phi i64 [ 0, %entry.new ], [ %niter.next.1, %loop_latch.1 ]
 ; EPILOG-BLOCK-NEXT:    call void @bar()
 ; EPILOG-BLOCK-NEXT:    br i1 %cond, label %loop_exiting_bb2, label %exit1.epilog-lcssa.loopexit
 ; EPILOG-BLOCK:       loop_exiting_bb2:
@@ -6391,7 +6391,7 @@ define void @test14(i64 %trip, i1 %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.nsub = sub i64 %niter, 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:
@@ -6399,8 +6399,8 @@ define void @test14(i64 %trip, i1 %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.nsub.1 = sub i64 %niter.nsub, 1
-; EPILOG-BLOCK-NEXT:    %niter.ncmp.1 = icmp ne i64 %niter.nsub.1, 0
+; EPILOG-BLOCK-NEXT:    %niter.next.1 = add i64 %niter.next, 1
+; 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:
 ; EPILOG-BLOCK-NEXT:    br label %exit1.unr-lcssa
@@ -6434,7 +6434,7 @@ define void @test14(i64 %trip, i1 %cond) {
 ; PROLOG-NEXT:    br label %loop_header.prol
 ; PROLOG:       loop_header.prol:
 ; PROLOG-NEXT:    %iv.prol = phi i64 [ 0, %loop_header.prol.preheader ], [ %iv_next.prol, %loop_latch.prol ]
-; PROLOG-NEXT:    %prol.iter = phi i64 [ %xtraiter, %loop_header.prol.preheader ], [ %prol.iter.sub, %loop_latch.prol ]
+; PROLOG-NEXT:    %prol.iter = phi i64 [ 0, %loop_header.prol.preheader ], [ %prol.iter.next, %loop_latch.prol ]
 ; PROLOG-NEXT:    call void @bar()
 ; PROLOG-NEXT:    br i1 %cond, label %loop_exiting_bb2.prol, label %exit1.unr-lcssa.loopexit1
 ; PROLOG:       loop_exiting_bb2.prol:
@@ -6443,8 +6443,8 @@ define void @test14(i64 %trip, i1 %cond) {
 ; PROLOG:       loop_latch.prol:
 ; PROLOG-NEXT:    %iv_next.prol = add i64 %iv.prol, 1
 ; PROLOG-NEXT:    %cmp.prol = icmp ne i64 %iv_next.prol, %trip
-; PROLOG-NEXT:    %prol.iter.sub = sub i64 %prol.iter, 1
-; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.sub, 0
+; PROLOG-NEXT:    %prol.iter.next = add i64 %prol.iter, 1
+; PROLOG-NEXT:    %prol.iter.cmp = icmp ne i64 %prol.iter.next, %xtraiter
 ; PROLOG-NEXT:    br i1 %prol.iter.cmp, label %loop_header.prol, label %loop_header.prol.loopexit.unr-lcssa, !llvm.loop !17
 ; PROLOG:       loop_header.prol.loopexit.unr-lcssa:
 ; PROLOG-NEXT:    %iv.unr.ph = phi i64 [ %iv_next.prol, %loop_latch.prol ]

diff --git a/llvm/test/Transforms/LoopUnroll/runtime-loop.ll b/llvm/test/Transforms/LoopUnroll/runtime-loop.ll
index 98316905e1fefc3905e0b72de83e00c1b0bb07a0..3deda2f649860d513bb00a846cfdc2286457c538 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/runtime-loop.ll
+++ b/llvm/test/Transforms/LoopUnroll/runtime-loop.ll
@@ -32,8 +32,8 @@ target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f3
 
 ; EPILOG: for.body.epil:
 ; EPILOG: %indvars.iv.epil = phi i64 [ %indvars.iv.next.epil, %for.body.epil ],  [ %indvars.iv.unr, %for.body.epil.preheader ]
-; EPILOG:  %epil.iter.sub = sub i32 %epil.iter, 1
-; EPILOG:  %epil.iter.cmp = icmp ne i32 %epil.iter.sub, 0
+; EPILOG:  %epil.iter.next = add i32 %epil.iter, 1
+; EPILOG:  %epil.iter.cmp = icmp ne i32 %epil.iter.next, %xtraiter
 ; EPILOG:  br i1 %epil.iter.cmp, label %for.body.epil, label %for.end.loopexit.epilog-lcssa, !llvm.loop !0
 
 ; NOEPILOG: for.body:
@@ -41,8 +41,8 @@ target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f3
 
 ; PROLOG: for.body.prol:
 ; PROLOG: %indvars.iv.prol = phi i64 [ %indvars.iv.next.prol, %for.body.prol ], [ 0, %for.body.prol.preheader ]
-; PROLOG:  %prol.iter.sub = sub i32 %prol.iter, 1
-; PROLOG:  %prol.iter.cmp = icmp ne i32 %prol.iter.sub, 0
+; PROLOG:  %prol.iter.next = add i32 %prol.iter, 1
+; PROLOG:  %prol.iter.cmp = icmp ne i32 %prol.iter.next, %xtraiter
 ; PROLOG:  br i1 %prol.iter.cmp, label %for.body.prol, label %for.body.prol.loopexit.unr-lcssa, !llvm.loop !0
 
 ; NOPROLOG: for.body:

diff --git a/llvm/test/Transforms/LoopUnroll/runtime-loop5.ll b/llvm/test/Transforms/LoopUnroll/runtime-loop5.ll
index 28cd885dbc443519de8b9bae7d40deb60c1a719c..88ebdf93d69e08a783135814c501c31f8d8710ac 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/runtime-loop5.ll
+++ b/llvm/test/Transforms/LoopUnroll/runtime-loop5.ll
@@ -76,28 +76,28 @@ define i3 @test(i3* %a, i3 %n) {
 ; UNROLL-4:       for.body:
 ; UNROLL-4-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[INDVARS_IV_NEXT_3:%.*]], [[FOR_BODY]] ]
 ; UNROLL-4-NEXT:    [[SUM_02:%.*]] = phi i3 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[ADD_3:%.*]], [[FOR_BODY]] ]
-; UNROLL-4-NEXT:    [[NITER:%.*]] = phi i3 [ [[UNROLL_ITER]], [[FOR_BODY_PREHEADER_NEW]] ], [ [[NITER_NSUB_3:%.*]], [[FOR_BODY]] ]
+; UNROLL-4-NEXT:    [[NITER:%.*]] = phi i3 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[NITER_NEXT_3:%.*]], [[FOR_BODY]] ]
 ; UNROLL-4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i3, i3* [[A:%.*]], i64 [[INDVARS_IV]]
 ; UNROLL-4-NEXT:    [[TMP2:%.*]] = load i3, i3* [[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_NSUB:%.*]] = sub i3 [[NITER]], 1
+; UNROLL-4-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i3 [[NITER]], 1
 ; UNROLL-4-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i3, i3* [[A]], i64 [[INDVARS_IV_NEXT]]
 ; UNROLL-4-NEXT:    [[TMP3:%.*]] = load i3, i3* [[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_NSUB_1:%.*]] = sub i3 [[NITER_NSUB]], 1
+; UNROLL-4-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i3 [[NITER_NEXT]], 1
 ; UNROLL-4-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i3, i3* [[A]], i64 [[INDVARS_IV_NEXT_1]]
 ; UNROLL-4-NEXT:    [[TMP4:%.*]] = load i3, i3* [[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_NSUB_2:%.*]] = sub i3 [[NITER_NSUB_1]], 1
+; UNROLL-4-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i3 [[NITER_NEXT_1]], 1
 ; UNROLL-4-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i3, i3* [[A]], i64 [[INDVARS_IV_NEXT_2]]
 ; UNROLL-4-NEXT:    [[TMP5:%.*]] = load i3, i3* [[ARRAYIDX_3]], align 1
 ; UNROLL-4-NEXT:    [[ADD_3]] = add nsw i3 [[TMP5]], [[ADD_2]]
 ; UNROLL-4-NEXT:    [[INDVARS_IV_NEXT_3]] = add i64 [[INDVARS_IV_NEXT_2]], 1
-; UNROLL-4-NEXT:    [[NITER_NSUB_3]] = sub i3 [[NITER_NSUB_2]], 1
-; UNROLL-4-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i3 [[NITER_NSUB_3]], 0
+; UNROLL-4-NEXT:    [[NITER_NEXT_3]] = add i3 [[NITER_NEXT_2]], 1
+; 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:
 ; UNROLL-4-NEXT:    [[ADD_LCSSA_PH_PH:%.*]] = phi i3 [ [[ADD_3]], [[FOR_BODY]] ]
@@ -115,15 +115,15 @@ define i3 @test(i3* %a, i3 %n) {
 ; UNROLL-4:       for.body.epil:
 ; UNROLL-4-NEXT:    [[INDVARS_IV_EPIL:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[INDVARS_IV_UNR]], [[FOR_BODY_EPIL_PREHEADER]] ]
 ; UNROLL-4-NEXT:    [[SUM_02_EPIL:%.*]] = phi i3 [ [[ADD_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[SUM_02_UNR]], [[FOR_BODY_EPIL_PREHEADER]] ]
-; UNROLL-4-NEXT:    [[EPIL_ITER:%.*]] = phi i3 [ [[XTRAITER]], [[FOR_BODY_EPIL_PREHEADER]] ], [ [[EPIL_ITER_SUB:%.*]], [[FOR_BODY_EPIL]] ]
+; UNROLL-4-NEXT:    [[EPIL_ITER:%.*]] = phi i3 [ 0, [[FOR_BODY_EPIL_PREHEADER]] ], [ [[EPIL_ITER_NEXT:%.*]], [[FOR_BODY_EPIL]] ]
 ; UNROLL-4-NEXT:    [[ARRAYIDX_EPIL:%.*]] = getelementptr inbounds i3, i3* [[A]], i64 [[INDVARS_IV_EPIL]]
 ; UNROLL-4-NEXT:    [[TMP6:%.*]] = load i3, i3* [[ARRAYIDX_EPIL]], align 1
 ; UNROLL-4-NEXT:    [[ADD_EPIL]] = add nsw i3 [[TMP6]], [[SUM_02_EPIL]]
 ; UNROLL-4-NEXT:    [[INDVARS_IV_NEXT_EPIL]] = add i64 [[INDVARS_IV_EPIL]], 1
 ; UNROLL-4-NEXT:    [[LFTR_WIDEIV_EPIL:%.*]] = trunc i64 [[INDVARS_IV_NEXT_EPIL]] to i3
 ; UNROLL-4-NEXT:    [[EXITCOND_EPIL:%.*]] = icmp eq i3 [[LFTR_WIDEIV_EPIL]], [[N]]
-; UNROLL-4-NEXT:    [[EPIL_ITER_SUB]] = sub i3 [[EPIL_ITER]], 1
-; UNROLL-4-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i3 [[EPIL_ITER_SUB]], 0
+; UNROLL-4-NEXT:    [[EPIL_ITER_NEXT]] = add i3 [[EPIL_ITER]], 1
+; UNROLL-4-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i3 [[EPIL_ITER_NEXT]], [[XTRAITER]]
 ; UNROLL-4-NEXT:    br i1 [[EPIL_ITER_CMP]], label [[FOR_BODY_EPIL]], label [[FOR_END_LOOPEXIT_EPILOG_LCSSA:%.*]], !llvm.loop [[LOOP2:![0-9]+]]
 ; UNROLL-4:       for.end.loopexit.epilog-lcssa:
 ; UNROLL-4-NEXT:    [[ADD_LCSSA_PH1:%.*]] = phi i3 [ [[ADD_EPIL]], [[FOR_BODY_EPIL]] ]

diff --git a/llvm/test/Transforms/LoopUnroll/runtime-multiexit-heuristic.ll b/llvm/test/Transforms/LoopUnroll/runtime-multiexit-heuristic.ll
index 89aa49dd71a0c9d4ddd422ac4af38ef39efdc151..370e9a4c9e635827544964938a99a9fd1ddcea59 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/runtime-multiexit-heuristic.ll
+++ b/llvm/test/Transforms/LoopUnroll/runtime-multiexit-heuristic.ll
@@ -19,7 +19,7 @@ define i32 @test1(i32* nocapture %a, i64 %n) {
 ; CHECK:       header:
 ; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY_NEW]] ], [ [[INDVARS_IV_NEXT_7:%.*]], [[LATCH_7:%.*]] ]
 ; CHECK-NEXT:    [[SUM_02:%.*]] = phi i32 [ 0, [[ENTRY_NEW]] ], [ [[ADD_7:%.*]], [[LATCH_7]] ]
-; CHECK-NEXT:    [[NITER:%.*]] = phi i64 [ [[UNROLL_ITER]], [[ENTRY_NEW]] ], [ [[NITER_NSUB_7:%.*]], [[LATCH_7]] ]
+; CHECK-NEXT:    [[NITER:%.*]] = phi i64 [ 0, [[ENTRY_NEW]] ], [ [[NITER_NEXT_7:%.*]], [[LATCH_7]] ]
 ; CHECK-NEXT:    br label [[FOR_EXITING_BLOCK:%.*]]
 ; CHECK:       for.exiting_block:
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i64 [[N]], 42
@@ -92,8 +92,8 @@ define i32 @test1(i32* nocapture %a, i64 %n) {
 ; CHECK-NEXT:    [[TMP9:%.*]] = load i32, i32* [[ARRAYIDX_7]], align 4
 ; CHECK-NEXT:    [[ADD_7]] = add nsw i32 [[TMP9]], [[ADD_6]]
 ; CHECK-NEXT:    [[INDVARS_IV_NEXT_7]] = add i64 [[INDVARS_IV]], 8
-; CHECK-NEXT:    [[NITER_NSUB_7]] = add i64 [[NITER]], -8
-; CHECK-NEXT:    [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NSUB_7]], 0
+; 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 [[LATCHEXIT_UNR_LCSSA_LOOPEXIT:%.*]], label [[HEADER]]
 ; CHECK:       latchexit.unr-lcssa.loopexit:
 ; CHECK-NEXT:    br label [[LATCHEXIT_UNR_LCSSA]]
@@ -108,7 +108,7 @@ define i32 @test1(i32* nocapture %a, i64 %n) {
 ; CHECK:       header.epil:
 ; CHECK-NEXT:    [[INDVARS_IV_EPIL:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_EPIL:%.*]], [[LATCH_EPIL:%.*]] ], [ [[INDVARS_IV_UNR]], [[HEADER_EPIL_PREHEADER]] ]
 ; CHECK-NEXT:    [[SUM_02_EPIL:%.*]] = phi i32 [ [[ADD_EPIL:%.*]], [[LATCH_EPIL]] ], [ [[SUM_02_UNR]], [[HEADER_EPIL_PREHEADER]] ]
-; CHECK-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_SUB:%.*]], [[LATCH_EPIL]] ], [ [[XTRAITER]], [[HEADER_EPIL_PREHEADER]] ]
+; CHECK-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_NEXT:%.*]], [[LATCH_EPIL]] ], [ 0, [[HEADER_EPIL_PREHEADER]] ]
 ; CHECK-NEXT:    br label [[FOR_EXITING_BLOCK_EPIL:%.*]]
 ; CHECK:       for.exiting_block.epil:
 ; CHECK-NEXT:    [[CMP_EPIL:%.*]] = icmp eq i64 [[N]], 42
@@ -118,8 +118,8 @@ define i32 @test1(i32* nocapture %a, i64 %n) {
 ; CHECK-NEXT:    [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX_EPIL]], align 4
 ; CHECK-NEXT:    [[ADD_EPIL]] = add nsw i32 [[TMP10]], [[SUM_02_EPIL]]
 ; CHECK-NEXT:    [[INDVARS_IV_NEXT_EPIL]] = add i64 [[INDVARS_IV_EPIL]], 1
-; CHECK-NEXT:    [[EPIL_ITER_SUB]] = add i64 [[EPIL_ITER]], -1
-; CHECK-NEXT:    [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[EPIL_ITER_SUB]], 0
+; CHECK-NEXT:    [[EPIL_ITER_NEXT]] = add i64 [[EPIL_ITER]], 1
+; CHECK-NEXT:    [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[EPIL_ITER_NEXT]], [[XTRAITER]]
 ; CHECK-NEXT:    br i1 [[EPIL_ITER_CMP_NOT]], label [[LATCHEXIT_EPILOG_LCSSA:%.*]], label [[HEADER_EPIL]], !llvm.loop [[LOOP0:![0-9]+]]
 ; CHECK:       latchexit.epilog-lcssa:
 ; CHECK-NEXT:    br label [[LATCHEXIT]]

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 888656387dc3e37e432a70be966f0d926d9c7447..bcc3020ce6fd01b1386ba127f3659f5946de0ac2 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/runtime-unroll-assume-no-remainder.ll
+++ b/llvm/test/Transforms/LoopUnroll/runtime-unroll-assume-no-remainder.ll
@@ -97,22 +97,22 @@ define dso_local void @cannotProveDivisibleTC(i8* noalias nocapture %a, i8* noal
 ; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
 ; CHECK:       for.body:
 ; CHECK-NEXT:    [[I_011:%.*]] = phi i32 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[INC_1:%.*]], [[FOR_BODY]] ]
-; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ [[UNROLL_ITER]], [[FOR_BODY_PREHEADER_NEW]] ], [ [[NITER_NSUB_1:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[NITER_NEXT_1:%.*]], [[FOR_BODY]] ]
 ; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i8, i8* [[B:%.*]], i32 [[I_011]]
 ; CHECK-NEXT:    [[TMP2:%.*]] = load i8, i8* [[ARRAYIDX]], align 1
 ; CHECK-NEXT:    [[ADD:%.*]] = add i8 [[TMP2]], 3
 ; CHECK-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds i8, i8* [[A:%.*]], i32 [[I_011]]
 ; CHECK-NEXT:    store i8 [[ADD]], i8* [[ARRAYIDX4]], align 1
 ; CHECK-NEXT:    [[INC:%.*]] = add nuw nsw i32 [[I_011]], 1
-; CHECK-NEXT:    [[NITER_NSUB:%.*]] = sub i32 [[NITER]], 1
+; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i32 [[NITER]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i8, i8* [[B]], i32 [[INC]]
 ; CHECK-NEXT:    [[TMP3:%.*]] = load i8, i8* [[ARRAYIDX_1]], align 1
 ; CHECK-NEXT:    [[ADD_1:%.*]] = add i8 [[TMP3]], 3
 ; CHECK-NEXT:    [[ARRAYIDX4_1:%.*]] = getelementptr inbounds i8, i8* [[A]], i32 [[INC]]
 ; CHECK-NEXT:    store i8 [[ADD_1]], i8* [[ARRAYIDX4_1]], align 1
 ; CHECK-NEXT:    [[INC_1]] = add nuw nsw i32 [[INC]], 1
-; CHECK-NEXT:    [[NITER_NSUB_1]] = sub i32 [[NITER_NSUB]], 1
-; CHECK-NEXT:    [[NITER_NCMP_1:%.*]] = icmp ne i32 [[NITER_NSUB_1]], 0
+; CHECK-NEXT:    [[NITER_NEXT_1]] = add i32 [[NITER_NEXT]], 1
+; 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:
 ; CHECK-NEXT:    [[I_011_UNR_PH:%.*]] = phi i32 [ [[INC_1]], [[FOR_BODY]] ]

diff --git a/llvm/test/Transforms/LoopUnroll/runtime-unroll-remainder.ll b/llvm/test/Transforms/LoopUnroll/runtime-unroll-remainder.ll
index f0cbf13fba8068512a71c10e41fe8410b77adb8f..eb963303a36125060b9cb46a5e5212d3adc6b731 100644 (file)
--- a/llvm/test/Transforms/LoopUnroll/runtime-unroll-remainder.ll
+++ b/llvm/test/Transforms/LoopUnroll/runtime-unroll-remainder.ll
@@ -66,7 +66,7 @@ define i32 @unroll(i32* nocapture readonly %a, i32* nocapture readonly %b, i32 %
 ; CHECK:       for.body:
 ; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ 0, [[FOR_BODY_LR_PH_NEW]] ], [ [[INDVARS_IV_NEXT_3]], [[FOR_BODY]] ]
 ; CHECK-NEXT:    [[C_010:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH_NEW]] ], [ [[ADD_3]], [[FOR_BODY]] ]
-; CHECK-NEXT:    [[NITER:%.*]] = phi i64 [ [[UNROLL_ITER]], [[FOR_BODY_LR_PH_NEW]] ], [ [[NITER_NSUB_3:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[NITER:%.*]] = phi i64 [ 0, [[FOR_BODY_LR_PH_NEW]] ], [ [[NITER_NEXT_3:%.*]], [[FOR_BODY]] ]
 ; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV]]
 ; CHECK-NEXT:    [[TMP8:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
 ; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[INDVARS_IV]]
@@ -95,8 +95,8 @@ define i32 @unroll(i32* nocapture readonly %a, i32* nocapture readonly %b, i32 %
 ; CHECK-NEXT:    [[MUL_3:%.*]] = mul nsw i32 [[TMP15]], [[TMP14]]
 ; CHECK-NEXT:    [[ADD_3]] = add nsw i32 [[MUL_3]], [[ADD_2]]
 ; CHECK-NEXT:    [[INDVARS_IV_NEXT_3]] = add nuw nsw i64 [[INDVARS_IV]], 4
-; CHECK-NEXT:    [[NITER_NSUB_3]] = add i64 [[NITER]], -4
-; CHECK-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i64 [[NITER_NSUB_3]], 0
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add i64 [[NITER]], 4
+; CHECK-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i64 [[NITER_NEXT_3]], [[UNROLL_ITER]]
 ; CHECK-NEXT:    br i1 [[NITER_NCMP_3]], label [[FOR_COND_CLEANUP_LOOPEXIT_UNR_LCSSA_LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
 ;
 entry:

diff --git a/llvm/test/Transforms/LoopUnrollAndJam/unroll-and-jam.ll b/llvm/test/Transforms/LoopUnrollAndJam/unroll-and-jam.ll
index bdd1f4d4b921de8192552eef75a09235552a7cec..64e1774672b009d687bdf8bcb92cb89e96fb2e20 100644 (file)
--- a/llvm/test/Transforms/LoopUnrollAndJam/unroll-and-jam.ll
+++ b/llvm/test/Transforms/LoopUnrollAndJam/unroll-and-jam.ll
@@ -22,15 +22,15 @@ define void @test1(i32 %I, i32 %E, i32* noalias nocapture %A, i32* noalias nocap
 ; CHECK-NEXT:    br label [[FOR_OUTER:%.*]]
 ; CHECK:       for.outer:
 ; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[ADD8_3:%.*]], [[FOR_LATCH:%.*]] ], [ 0, [[FOR_OUTER_PREHEADER_NEW]] ]
-; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ [[UNROLL_ITER]], [[FOR_OUTER_PREHEADER_NEW]] ], [ [[NITER_NSUB_3:%.*]], [[FOR_LATCH]] ]
+; 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_NSUB:%.*]] = sub i32 [[NITER]], 1
+; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i32 [[NITER]], 1
 ; CHECK-NEXT:    [[ADD8_1:%.*]] = add nuw nsw i32 [[ADD8]], 1
-; CHECK-NEXT:    [[NITER_NSUB_1:%.*]] = sub i32 [[NITER_NSUB]], 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_NSUB_2:%.*]] = sub i32 [[NITER_NSUB_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_NSUB_3]] = sub i32 [[NITER_NSUB_2]], 1
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER_NEXT_2]], 1
 ; CHECK-NEXT:    br label [[FOR_INNER:%.*]]
 ; CHECK:       for.inner:
 ; CHECK-NEXT:    [[J:%.*]] = phi i32 [ 0, [[FOR_OUTER]] ], [ [[INC:%.*]], [[FOR_INNER]] ]
@@ -72,7 +72,7 @@ define void @test1(i32 %I, i32 %E, i32* noalias nocapture %A, i32* noalias nocap
 ; CHECK-NEXT:    store i32 [[ADD_LCSSA_2]], i32* [[ARRAYIDX6_2]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ARRAYIDX6_3:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD8_2]]
 ; CHECK-NEXT:    store i32 [[ADD_LCSSA_3]], i32* [[ARRAYIDX6_3]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i32 [[NITER_NSUB_3]], 0
+; CHECK-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i32 [[NITER_NEXT_3]], [[UNROLL_ITER]]
 ; CHECK-NEXT:    br i1 [[NITER_NCMP_3]], label [[FOR_END_LOOPEXIT_UNR_LCSSA_LOOPEXIT:%.*]], label [[FOR_OUTER]], !llvm.loop [[LOOP4:![0-9]+]]
 ; CHECK:       for.end.loopexit.unr-lcssa.loopexit:
 ; CHECK-NEXT:    [[I_UNR_PH:%.*]] = phi i32 [ [[ADD8_3]], [[FOR_LATCH]] ]
@@ -99,8 +99,7 @@ define void @test1(i32 %I, i32 %E, i32* noalias nocapture %A, i32* noalias nocap
 ; CHECK-NEXT:    [[ARRAYIDX6_EPIL:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[I_UNR]]
 ; CHECK-NEXT:    store i32 [[ADD_LCSSA_EPIL]], i32* [[ARRAYIDX6_EPIL]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ADD8_EPIL:%.*]] = add nuw i32 [[I_UNR]], 1
-; CHECK-NEXT:    [[EPIL_ITER_SUB:%.*]] = sub i32 [[XTRAITER]], 1
-; CHECK-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i32 [[EPIL_ITER_SUB]], 0
+; CHECK-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i32 1, [[XTRAITER]]
 ; CHECK-NEXT:    br i1 [[EPIL_ITER_CMP]], label [[FOR_OUTER_EPIL_1:%.*]], label [[FOR_END_LOOPEXIT_EPILOG_LCSSA:%.*]]
 ; CHECK:       for.outer.epil.1:
 ; CHECK-NEXT:    br label [[FOR_INNER_EPIL_1:%.*]]
@@ -118,8 +117,7 @@ define void @test1(i32 %I, i32 %E, i32* noalias nocapture %A, i32* noalias nocap
 ; CHECK-NEXT:    [[ARRAYIDX6_EPIL_1:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD8_EPIL]]
 ; CHECK-NEXT:    store i32 [[ADD_LCSSA_EPIL_1]], i32* [[ARRAYIDX6_EPIL_1]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ADD8_EPIL_1:%.*]] = add nuw i32 [[ADD8_EPIL]], 1
-; CHECK-NEXT:    [[EPIL_ITER_SUB_1:%.*]] = sub i32 [[EPIL_ITER_SUB]], 1
-; CHECK-NEXT:    [[EPIL_ITER_CMP_1:%.*]] = icmp ne i32 [[EPIL_ITER_SUB_1]], 0
+; 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:
 ; CHECK-NEXT:    br label [[FOR_INNER_EPIL_2:%.*]]
@@ -202,23 +200,23 @@ define void @test2(i32 %I, i32 %E, i32* noalias nocapture %A, i32* noalias nocap
 ; CHECK-NEXT:    br label [[FOR_OUTER:%.*]]
 ; CHECK:       for.outer:
 ; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[ADD9_3:%.*]], [[FOR_LATCH:%.*]] ], [ 0, [[FOR_OUTER_PREHEADER_NEW]] ]
-; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ [[UNROLL_ITER]], [[FOR_OUTER_PREHEADER_NEW]] ], [ [[NITER_NSUB_3:%.*]], [[FOR_LATCH]] ]
+; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ 0, [[FOR_OUTER_PREHEADER_NEW]] ], [ [[NITER_NEXT_3:%.*]], [[FOR_LATCH]] ]
 ; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 [[I]]
 ; CHECK-NEXT:    [[TMP2:%.*]] = load i32, i32* [[ARRAYIDX]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ADD9:%.*]] = add nuw nsw i32 [[I]], 1
-; CHECK-NEXT:    [[NITER_NSUB:%.*]] = sub i32 [[NITER]], 1
+; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i32 [[NITER]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD9]]
 ; CHECK-NEXT:    [[TMP3:%.*]] = load i32, i32* [[ARRAYIDX_1]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ADD9_1:%.*]] = add nuw nsw i32 [[ADD9]], 1
-; CHECK-NEXT:    [[NITER_NSUB_1:%.*]] = sub i32 [[NITER_NSUB]], 1
+; CHECK-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i32 [[NITER_NEXT]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD9_1]]
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i32, i32* [[ARRAYIDX_2]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ADD9_2:%.*]] = add nuw nsw i32 [[ADD9_1]], 1
-; CHECK-NEXT:    [[NITER_NSUB_2:%.*]] = sub i32 [[NITER_NSUB_1]], 1
+; CHECK-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i32 [[NITER_NEXT_1]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD9_2]]
 ; CHECK-NEXT:    [[TMP5:%.*]] = load i32, i32* [[ARRAYIDX_3]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ADD9_3]] = add nuw i32 [[ADD9_2]], 1
-; CHECK-NEXT:    [[NITER_NSUB_3]] = sub i32 [[NITER_NSUB_2]], 1
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER_NEXT_2]], 1
 ; CHECK-NEXT:    br label [[FOR_INNER:%.*]]
 ; CHECK:       for.inner:
 ; CHECK-NEXT:    [[J:%.*]] = phi i32 [ 0, [[FOR_OUTER]] ], [ [[INC:%.*]], [[FOR_INNER]] ]
@@ -256,7 +254,7 @@ define void @test2(i32 %I, i32 %E, i32* noalias nocapture %A, i32* noalias nocap
 ; CHECK-NEXT:    store i32 [[ADD_LCSSA_1]], i32* [[ARRAYIDX_1]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    store i32 [[ADD_LCSSA_2]], i32* [[ARRAYIDX_2]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    store i32 [[ADD_LCSSA_3]], i32* [[ARRAYIDX_3]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i32 [[NITER_NSUB_3]], 0
+; CHECK-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i32 [[NITER_NEXT_3]], [[UNROLL_ITER]]
 ; CHECK-NEXT:    br i1 [[NITER_NCMP_3]], label [[FOR_END10_LOOPEXIT_UNR_LCSSA_LOOPEXIT:%.*]], label [[FOR_OUTER]], !llvm.loop [[LOOP6:![0-9]+]]
 ; CHECK:       for.end10.loopexit.unr-lcssa.loopexit:
 ; CHECK-NEXT:    [[I_UNR_PH:%.*]] = phi i32 [ [[ADD9_3]], [[FOR_LATCH]] ]
@@ -284,8 +282,7 @@ define void @test2(i32 %I, i32 %E, i32* noalias nocapture %A, i32* noalias nocap
 ; CHECK-NEXT:    [[ADD_LCSSA_EPIL:%.*]] = phi i32 [ [[ADD_EPIL]], [[FOR_INNER_EPIL]] ]
 ; CHECK-NEXT:    store i32 [[ADD_LCSSA_EPIL]], i32* [[ARRAYIDX_EPIL]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ADD9_EPIL:%.*]] = add nuw i32 [[I_UNR]], 1
-; CHECK-NEXT:    [[EPIL_ITER_SUB:%.*]] = sub i32 [[XTRAITER]], 1
-; CHECK-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i32 [[EPIL_ITER_SUB]], 0
+; CHECK-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i32 1, [[XTRAITER]]
 ; CHECK-NEXT:    br i1 [[EPIL_ITER_CMP]], label [[FOR_OUTER_EPIL_1:%.*]], label [[FOR_END10_LOOPEXIT_EPILOG_LCSSA:%.*]]
 ; CHECK:       for.outer.epil.1:
 ; CHECK-NEXT:    [[ARRAYIDX_EPIL_1:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD9_EPIL]]
@@ -304,8 +301,7 @@ define void @test2(i32 %I, i32 %E, i32* noalias nocapture %A, i32* noalias nocap
 ; CHECK-NEXT:    [[ADD_LCSSA_EPIL_1:%.*]] = phi i32 [ [[ADD_EPIL_1]], [[FOR_INNER_EPIL_1]] ]
 ; CHECK-NEXT:    store i32 [[ADD_LCSSA_EPIL_1]], i32* [[ARRAYIDX_EPIL_1]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ADD9_EPIL_1:%.*]] = add nuw i32 [[ADD9_EPIL]], 1
-; CHECK-NEXT:    [[EPIL_ITER_SUB_1:%.*]] = sub i32 [[EPIL_ITER_SUB]], 1
-; CHECK-NEXT:    [[EPIL_ITER_CMP_1:%.*]] = icmp ne i32 [[EPIL_ITER_SUB_1]], 0
+; 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:
 ; CHECK-NEXT:    [[ARRAYIDX_EPIL_2:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD9_EPIL_1]]
@@ -626,15 +622,15 @@ define i32 @test6() #0 {
 ; CHECK-NEXT:    br label [[FOR_OUTER:%.*]]
 ; CHECK:       for.outer:
 ; CHECK-NEXT:    [[INC5_SINK9:%.*]] = phi i32 [ 2, [[ENTRY_NEW]] ], [ [[INC5_3:%.*]], [[FOR_LATCH:%.*]] ]
-; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ 4, [[ENTRY_NEW]] ], [ [[NITER_NSUB_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_NSUB:%.*]] = sub nuw nsw i32 [[NITER]], 1
+; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i32 [[NITER]], 1
 ; CHECK-NEXT:    [[INC5_1:%.*]] = add nuw nsw i32 [[INC5]], 1
-; CHECK-NEXT:    [[NITER_NSUB_1:%.*]] = sub nuw nsw i32 [[NITER_NSUB]], 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_NSUB_2:%.*]] = sub nuw nsw i32 [[NITER_NSUB_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_NSUB_3]] = sub nuw nsw i32 [[NITER_NSUB_2]], 1
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add nuw nsw i32 [[NITER_NEXT_2]], 1
 ; CHECK-NEXT:    br label [[FOR_INNER:%.*]]
 ; CHECK:       for.inner:
 ; CHECK-NEXT:    [[INC_SINK8:%.*]] = phi i32 [ 0, [[FOR_OUTER]] ], [ [[INC:%.*]], [[FOR_INNER]] ]
@@ -725,31 +721,31 @@ define void @test7(i32 %I, i32 %E, i32* noalias nocapture %A, i32* noalias nocap
 ; CHECK-NEXT:    br label [[FOR_OUTER:%.*]]
 ; CHECK:       for.outer:
 ; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[ADD_3:%.*]], [[FOR_LATCH:%.*]] ], [ 0, [[FOR_PREHEADER_NEW]] ]
-; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ [[UNROLL_ITER]], [[FOR_PREHEADER_NEW]] ], [ [[NITER_NSUB_3:%.*]], [[FOR_LATCH]] ]
+; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ 0, [[FOR_PREHEADER_NEW]] ], [ [[NITER_NEXT_3:%.*]], [[FOR_LATCH]] ]
 ; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 [[I]]
 ; CHECK-NEXT:    store i32 0, i32* [[ARRAYIDX]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ADD:%.*]] = add nuw nsw i32 [[I]], 1
 ; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD]]
 ; CHECK-NEXT:    store i32 2, i32* [[ARRAYIDX2]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NSUB:%.*]] = sub i32 [[NITER]], 1
+; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i32 [[NITER]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD]]
 ; CHECK-NEXT:    store i32 0, i32* [[ARRAYIDX_1]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ADD_1:%.*]] = add nuw nsw i32 [[ADD]], 1
 ; CHECK-NEXT:    [[ARRAYIDX2_1:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD_1]]
 ; CHECK-NEXT:    store i32 2, i32* [[ARRAYIDX2_1]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NSUB_1:%.*]] = sub i32 [[NITER_NSUB]], 1
+; CHECK-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i32 [[NITER_NEXT]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD_1]]
 ; CHECK-NEXT:    store i32 0, i32* [[ARRAYIDX_2]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ADD_2:%.*]] = add nuw nsw i32 [[ADD_1]], 1
 ; CHECK-NEXT:    [[ARRAYIDX2_2:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD_2]]
 ; CHECK-NEXT:    store i32 2, i32* [[ARRAYIDX2_2]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NSUB_2:%.*]] = sub i32 [[NITER_NSUB_1]], 1
+; CHECK-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i32 [[NITER_NEXT_1]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD_2]]
 ; CHECK-NEXT:    store i32 0, i32* [[ARRAYIDX_3]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ADD_3]] = add nuw i32 [[ADD_2]], 1
 ; CHECK-NEXT:    [[ARRAYIDX2_3:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD_3]]
 ; CHECK-NEXT:    store i32 2, i32* [[ARRAYIDX2_3]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NSUB_3]] = sub i32 [[NITER_NSUB_2]], 1
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER_NEXT_2]], 1
 ; CHECK-NEXT:    br label [[FOR_INNER:%.*]]
 ; CHECK:       for.latch:
 ; CHECK-NEXT:    [[ADD9_LCSSA:%.*]] = phi i32 [ [[ADD9:%.*]], [[FOR_INNER]] ]
@@ -760,7 +756,7 @@ define void @test7(i32 %I, i32 %E, i32* noalias nocapture %A, i32* noalias nocap
 ; CHECK-NEXT:    store i32 [[ADD9_LCSSA_1]], i32* [[ARRAYIDX_1]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    store i32 [[ADD9_LCSSA_2]], i32* [[ARRAYIDX_2]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    store i32 [[ADD9_LCSSA_3]], i32* [[ARRAYIDX_3]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i32 [[NITER_NSUB_3]], 0
+; CHECK-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i32 [[NITER_NEXT_3]], [[UNROLL_ITER]]
 ; CHECK-NEXT:    br i1 [[NITER_NCMP_3]], label [[FOR_END_LOOPEXIT_UNR_LCSSA_LOOPEXIT:%.*]], label [[FOR_OUTER]], !llvm.loop [[LOOP8:![0-9]+]]
 ; CHECK:       for.inner:
 ; CHECK-NEXT:    [[SUM:%.*]] = phi i32 [ 0, [[FOR_OUTER]] ], [ [[ADD9]], [[FOR_INNER]] ]
@@ -817,8 +813,7 @@ define void @test7(i32 %I, i32 %E, i32* noalias nocapture %A, i32* noalias nocap
 ; CHECK:       for.latch.epil:
 ; CHECK-NEXT:    [[ADD9_LCSSA_EPIL:%.*]] = phi i32 [ [[ADD9_EPIL]], [[FOR_INNER_EPIL]] ]
 ; CHECK-NEXT:    store i32 [[ADD9_LCSSA_EPIL]], i32* [[ARRAYIDX_EPIL]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[EPIL_ITER_SUB:%.*]] = sub i32 [[XTRAITER]], 1
-; CHECK-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i32 [[EPIL_ITER_SUB]], 0
+; CHECK-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i32 1, [[XTRAITER]]
 ; CHECK-NEXT:    br i1 [[EPIL_ITER_CMP]], label [[FOR_OUTER_EPIL_1:%.*]], label [[FOR_END_LOOPEXIT_EPILOG_LCSSA:%.*]]
 ; CHECK:       for.outer.epil.1:
 ; CHECK-NEXT:    [[ARRAYIDX_EPIL_1:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD_EPIL]]
@@ -839,8 +834,7 @@ define void @test7(i32 %I, i32 %E, i32* noalias nocapture %A, i32* noalias nocap
 ; CHECK:       for.latch.epil.1:
 ; CHECK-NEXT:    [[ADD9_LCSSA_EPIL_1:%.*]] = phi i32 [ [[ADD9_EPIL_1]], [[FOR_INNER_EPIL_1]] ]
 ; CHECK-NEXT:    store i32 [[ADD9_LCSSA_EPIL_1]], i32* [[ARRAYIDX_EPIL_1]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[EPIL_ITER_SUB_1:%.*]] = sub i32 [[EPIL_ITER_SUB]], 1
-; CHECK-NEXT:    [[EPIL_ITER_CMP_1:%.*]] = icmp ne i32 [[EPIL_ITER_SUB_1]], 0
+; 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:
 ; CHECK-NEXT:    [[ARRAYIDX_EPIL_2:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD_EPIL_1]]
@@ -928,31 +922,31 @@ define void @test8(i32 %I, i32 %E, i32* noalias nocapture %A, i32* noalias nocap
 ; CHECK-NEXT:    br label [[FOR_OUTER:%.*]]
 ; CHECK:       for.outer:
 ; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[ADD_3:%.*]], [[FOR_LATCH:%.*]] ], [ 0, [[FOR_OUTEST_NEW]] ]
-; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ [[UNROLL_ITER]], [[FOR_OUTEST_NEW]] ], [ [[NITER_NSUB_3:%.*]], [[FOR_LATCH]] ]
+; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ 0, [[FOR_OUTEST_NEW]] ], [ [[NITER_NEXT_3:%.*]], [[FOR_LATCH]] ]
 ; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 [[I]]
 ; CHECK-NEXT:    store i32 0, i32* [[ARRAYIDX]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ADD:%.*]] = add nuw nsw i32 [[I]], 1
 ; CHECK-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD]]
 ; CHECK-NEXT:    store i32 2, i32* [[ARRAYIDX6]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NSUB:%.*]] = sub i32 [[NITER]], 1
+; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i32 [[NITER]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD]]
 ; CHECK-NEXT:    store i32 0, i32* [[ARRAYIDX_1]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ADD_1:%.*]] = add nuw nsw i32 [[ADD]], 1
 ; CHECK-NEXT:    [[ARRAYIDX6_1:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD_1]]
 ; CHECK-NEXT:    store i32 2, i32* [[ARRAYIDX6_1]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NSUB_1:%.*]] = sub i32 [[NITER_NSUB]], 1
+; CHECK-NEXT:    [[NITER_NEXT_1:%.*]] = add nuw nsw i32 [[NITER_NEXT]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD_1]]
 ; CHECK-NEXT:    store i32 0, i32* [[ARRAYIDX_2]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ADD_2:%.*]] = add nuw nsw i32 [[ADD_1]], 1
 ; CHECK-NEXT:    [[ARRAYIDX6_2:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD_2]]
 ; CHECK-NEXT:    store i32 2, i32* [[ARRAYIDX6_2]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NSUB_2:%.*]] = sub i32 [[NITER_NSUB_1]], 1
+; CHECK-NEXT:    [[NITER_NEXT_2:%.*]] = add nuw nsw i32 [[NITER_NEXT_1]], 1
 ; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD_2]]
 ; CHECK-NEXT:    store i32 0, i32* [[ARRAYIDX_3]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ADD_3]] = add nuw i32 [[ADD_2]], 1
 ; CHECK-NEXT:    [[ARRAYIDX6_3:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD_3]]
 ; CHECK-NEXT:    store i32 2, i32* [[ARRAYIDX6_3]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NSUB_3]] = sub i32 [[NITER_NSUB_2]], 1
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER_NEXT_2]], 1
 ; CHECK-NEXT:    br label [[FOR_INNER:%.*]]
 ; CHECK:       for.inner:
 ; CHECK-NEXT:    [[SUM:%.*]] = phi i32 [ 0, [[FOR_OUTER]] ], [ [[ADD9:%.*]], [[FOR_INNER]] ]
@@ -990,7 +984,7 @@ define void @test8(i32 %I, i32 %E, i32* noalias nocapture %A, i32* noalias nocap
 ; CHECK-NEXT:    store i32 [[ADD9_LCSSA_1]], i32* [[ARRAYIDX_1]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    store i32 [[ADD9_LCSSA_2]], i32* [[ARRAYIDX_2]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    store i32 [[ADD9_LCSSA_3]], i32* [[ARRAYIDX_3]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i32 [[NITER_NSUB_3]], 0
+; CHECK-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i32 [[NITER_NEXT_3]], [[UNROLL_ITER]]
 ; CHECK-NEXT:    br i1 [[NITER_NCMP_3]], label [[FOR_CLEANUP_UNR_LCSSA_LOOPEXIT:%.*]], label [[FOR_OUTER]], !llvm.loop [[LOOP9:![0-9]+]]
 ; CHECK:       for.cleanup.unr-lcssa.loopexit:
 ; CHECK-NEXT:    [[I_UNR_PH:%.*]] = phi i32 [ [[ADD_3]], [[FOR_LATCH]] ]
@@ -1020,8 +1014,7 @@ define void @test8(i32 %I, i32 %E, i32* noalias nocapture %A, i32* noalias nocap
 ; CHECK:       for.latch.epil:
 ; CHECK-NEXT:    [[ADD9_LCSSA_EPIL:%.*]] = phi i32 [ [[ADD9_EPIL]], [[FOR_INNER_EPIL]] ]
 ; CHECK-NEXT:    store i32 [[ADD9_LCSSA_EPIL]], i32* [[ARRAYIDX_EPIL]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[EPIL_ITER_SUB:%.*]] = sub i32 [[XTRAITER]], 1
-; CHECK-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i32 [[EPIL_ITER_SUB]], 0
+; CHECK-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i32 1, [[XTRAITER]]
 ; CHECK-NEXT:    br i1 [[EPIL_ITER_CMP]], label [[FOR_OUTER_EPIL_1:%.*]], label [[FOR_CLEANUP_EPILOG_LCSSA:%.*]]
 ; CHECK:       for.outer.epil.1:
 ; CHECK-NEXT:    [[ARRAYIDX_EPIL_1:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD_EPIL]]
@@ -1042,8 +1035,7 @@ define void @test8(i32 %I, i32 %E, i32* noalias nocapture %A, i32* noalias nocap
 ; CHECK:       for.latch.epil.1:
 ; CHECK-NEXT:    [[ADD9_LCSSA_EPIL_1:%.*]] = phi i32 [ [[ADD9_EPIL_1]], [[FOR_INNER_EPIL_1]] ]
 ; CHECK-NEXT:    store i32 [[ADD9_LCSSA_EPIL_1]], i32* [[ARRAYIDX_EPIL_1]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[EPIL_ITER_SUB_1:%.*]] = sub i32 [[EPIL_ITER_SUB]], 1
-; CHECK-NEXT:    [[EPIL_ITER_CMP_1:%.*]] = icmp ne i32 [[EPIL_ITER_SUB_1]], 0
+; 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_CLEANUP_EPILOG_LCSSA]]
 ; CHECK:       for.outer.epil.2:
 ; CHECK-NEXT:    [[ARRAYIDX_EPIL_2:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD_EPIL_1]]
@@ -1141,15 +1133,15 @@ define void @test9(i32 %I, i32 %E, i32* nocapture %A, i16* nocapture readonly %B
 ; CHECK-NEXT:    br label [[FOR_OUTER:%.*]]
 ; CHECK:       for.outer:
 ; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[ADD8_3:%.*]], [[FOR_LATCH:%.*]] ], [ 0, [[FOR_OUTER_PREHEADER_NEW]] ]
-; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ [[UNROLL_ITER]], [[FOR_OUTER_PREHEADER_NEW]] ], [ [[NITER_NSUB_3:%.*]], [[FOR_LATCH]] ]
+; 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_NSUB:%.*]] = sub i32 [[NITER]], 1
+; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i32 [[NITER]], 1
 ; CHECK-NEXT:    [[ADD8_1:%.*]] = add nuw nsw i32 [[ADD8]], 1
-; CHECK-NEXT:    [[NITER_NSUB_1:%.*]] = sub i32 [[NITER_NSUB]], 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_NSUB_2:%.*]] = sub i32 [[NITER_NSUB_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_NSUB_3]] = sub i32 [[NITER_NSUB_2]], 1
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add i32 [[NITER_NEXT_2]], 1
 ; CHECK-NEXT:    br label [[FOR_INNER:%.*]]
 ; CHECK:       for.inner:
 ; CHECK-NEXT:    [[J:%.*]] = phi i32 [ 0, [[FOR_OUTER]] ], [ [[INC:%.*]], [[FOR_INNER]] ]
@@ -1195,7 +1187,7 @@ define void @test9(i32 %I, i32 %E, i32* nocapture %A, i16* nocapture readonly %B
 ; CHECK-NEXT:    store i32 [[ADD_LCSSA_2]], i32* [[ARRAYIDX6_2]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ARRAYIDX6_3:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD8_2]]
 ; CHECK-NEXT:    store i32 [[ADD_LCSSA_3]], i32* [[ARRAYIDX6_3]], align 4, !tbaa [[TBAA0]]
-; CHECK-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i32 [[NITER_NSUB_3]], 0
+; CHECK-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i32 [[NITER_NEXT_3]], [[UNROLL_ITER]]
 ; CHECK-NEXT:    br i1 [[NITER_NCMP_3]], label [[FOR_END_LOOPEXIT_UNR_LCSSA_LOOPEXIT:%.*]], label [[FOR_OUTER]], !llvm.loop [[LOOP12:![0-9]+]]
 ; CHECK:       for.end.loopexit.unr-lcssa.loopexit:
 ; CHECK-NEXT:    [[I_UNR_PH:%.*]] = phi i32 [ [[ADD8_3]], [[FOR_LATCH]] ]
@@ -1223,8 +1215,7 @@ define void @test9(i32 %I, i32 %E, i32* nocapture %A, i16* nocapture readonly %B
 ; CHECK-NEXT:    [[ARRAYIDX6_EPIL:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[I_UNR]]
 ; CHECK-NEXT:    store i32 [[ADD_LCSSA_EPIL]], i32* [[ARRAYIDX6_EPIL]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ADD8_EPIL:%.*]] = add nuw i32 [[I_UNR]], 1
-; CHECK-NEXT:    [[EPIL_ITER_SUB:%.*]] = sub i32 [[XTRAITER]], 1
-; CHECK-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i32 [[EPIL_ITER_SUB]], 0
+; CHECK-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i32 1, [[XTRAITER]]
 ; CHECK-NEXT:    br i1 [[EPIL_ITER_CMP]], label [[FOR_OUTER_EPIL_1:%.*]], label [[FOR_END_LOOPEXIT_EPILOG_LCSSA:%.*]]
 ; CHECK:       for.outer.epil.1:
 ; CHECK-NEXT:    br label [[FOR_INNER_EPIL_1:%.*]]
@@ -1243,8 +1234,7 @@ define void @test9(i32 %I, i32 %E, i32* nocapture %A, i16* nocapture readonly %B
 ; CHECK-NEXT:    [[ARRAYIDX6_EPIL_1:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD8_EPIL]]
 ; CHECK-NEXT:    store i32 [[ADD_LCSSA_EPIL_1]], i32* [[ARRAYIDX6_EPIL_1]], align 4, !tbaa [[TBAA0]]
 ; CHECK-NEXT:    [[ADD8_EPIL_1:%.*]] = add nuw i32 [[ADD8_EPIL]], 1
-; CHECK-NEXT:    [[EPIL_ITER_SUB_1:%.*]] = sub i32 [[EPIL_ITER_SUB]], 1
-; CHECK-NEXT:    [[EPIL_ITER_CMP_1:%.*]] = icmp ne i32 [[EPIL_ITER_SUB_1]], 0
+; 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:
 ; CHECK-NEXT:    br label [[FOR_INNER_EPIL_2:%.*]]
@@ -1325,15 +1315,15 @@ define signext i16 @test10(i32 %k) #0 {
 ; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
 ; CHECK:       for.body:
 ; CHECK-NEXT:    [[STOREMERGE82:%.*]] = phi i64 [ 0, [[ENTRY_NEW]] ], [ [[INC25_3:%.*]], [[FOR_INC24:%.*]] ]
-; CHECK-NEXT:    [[NITER:%.*]] = phi i64 [ 4, [[ENTRY_NEW]] ], [ [[NITER_NSUB_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_NSUB:%.*]] = sub nuw nsw i64 [[NITER]], 1
+; CHECK-NEXT:    [[NITER_NEXT:%.*]] = add nuw nsw i64 [[NITER]], 1
 ; CHECK-NEXT:    [[INC25_1:%.*]] = add nuw nsw i64 [[INC25]], 1
-; CHECK-NEXT:    [[NITER_NSUB_1:%.*]] = sub nuw nsw i64 [[NITER_NSUB]], 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_NSUB_2:%.*]] = sub nuw nsw i64 [[NITER_NSUB_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_NSUB_3]] = sub nuw nsw i64 [[NITER_NSUB_2]], 1
+; CHECK-NEXT:    [[NITER_NEXT_3]] = add nuw nsw i64 [[NITER_NEXT_2]], 1
 ; CHECK-NEXT:    br label [[FOR_BODY2:%.*]]
 ; CHECK:       for.body2:
 ; CHECK-NEXT:    [[STOREMERGE:%.*]] = phi i64 [ 4, [[FOR_BODY]] ], [ [[DEC:%.*]], [[FOR_INC21_3:%.*]] ]

diff --git a/llvm/test/Transforms/LoopVectorize/X86/float-induction-x86.ll b/llvm/test/Transforms/LoopVectorize/X86/float-induction-x86.ll
index 254a4c301483899a74f1d75b148a6bad149d3e05..f85a8c66ab9d05106ff36f1f7e56c0bc33397421 100644 (file)
--- a/llvm/test/Transforms/LoopVectorize/X86/float-induction-x86.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/float-induction-x86.ll
@@ -40,7 +40,7 @@ define void @fp_iv_loop1(float* noalias nocapture %A, i32 %N) #0 {
 ; AUTO_VEC:       vector.body:
 ; AUTO_VEC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH_NEW]] ], [ [[INDEX_NEXT_3:%.*]], [[VECTOR_BODY]] ]
 ; AUTO_VEC-NEXT:    [[VEC_IND:%.*]] = phi <8 x float> [ <float 1.000000e+00, float 1.500000e+00, float 2.000000e+00, float 2.500000e+00, float 3.000000e+00, float 3.500000e+00, float 4.000000e+00, float 4.500000e+00>, [[VECTOR_PH_NEW]] ], [ [[VEC_IND_NEXT_3:%.*]], [[VECTOR_BODY]] ]
-; AUTO_VEC-NEXT:    [[NITER:%.*]] = phi i64 [ [[UNROLL_ITER]], [[VECTOR_PH_NEW]] ], [ [[NITER_NSUB_3:%.*]], [[VECTOR_BODY]] ]
+; AUTO_VEC-NEXT:    [[NITER:%.*]] = phi i64 [ 0, [[VECTOR_PH_NEW]] ], [ [[NITER_NEXT_3:%.*]], [[VECTOR_BODY]] ]
 ; AUTO_VEC-NEXT:    [[STEP_ADD:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00>
 ; AUTO_VEC-NEXT:    [[STEP_ADD2:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00>
 ; AUTO_VEC-NEXT:    [[STEP_ADD3:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01>
@@ -109,8 +109,8 @@ define void @fp_iv_loop1(float* noalias nocapture %A, i32 %N) #0 {
 ; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD3_3]], <8 x float>* [[TMP36]], align 4
 ; AUTO_VEC-NEXT:    [[INDEX_NEXT_3]] = add nuw i64 [[INDEX]], 128
 ; AUTO_VEC-NEXT:    [[VEC_IND_NEXT_3]] = fadd fast <8 x float> [[VEC_IND]], <float 6.400000e+01, float 6.400000e+01, float 6.400000e+01, float 6.400000e+01, float 6.400000e+01, float 6.400000e+01, float 6.400000e+01, float 6.400000e+01>
-; AUTO_VEC-NEXT:    [[NITER_NSUB_3]] = add i64 [[NITER]], -4
-; AUTO_VEC-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i64 [[NITER_NSUB_3]], 0
+; AUTO_VEC-NEXT:    [[NITER_NEXT_3]] = add i64 [[NITER]], 4
+; AUTO_VEC-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i64 [[NITER_NEXT_3]], [[UNROLL_ITER]]
 ; AUTO_VEC-NEXT:    br i1 [[NITER_NCMP_3]], label [[MIDDLE_BLOCK_UNR_LCSSA]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
 ; AUTO_VEC:       middle.block.unr-lcssa:
 ; AUTO_VEC-NEXT:    [[INDEX_UNR:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT_3]], [[VECTOR_BODY]] ]
@@ -120,7 +120,7 @@ define void @fp_iv_loop1(float* noalias nocapture %A, i32 %N) #0 {
 ; AUTO_VEC:       vector.body.epil:
 ; AUTO_VEC-NEXT:    [[INDEX_EPIL:%.*]] = phi i64 [ [[INDEX_NEXT_EPIL:%.*]], [[VECTOR_BODY_EPIL]] ], [ [[INDEX_UNR]], [[MIDDLE_BLOCK_UNR_LCSSA]] ]
 ; AUTO_VEC-NEXT:    [[VEC_IND_EPIL:%.*]] = phi <8 x float> [ [[VEC_IND_NEXT_EPIL:%.*]], [[VECTOR_BODY_EPIL]] ], [ [[VEC_IND_UNR]], [[MIDDLE_BLOCK_UNR_LCSSA]] ]
-; AUTO_VEC-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_SUB:%.*]], [[VECTOR_BODY_EPIL]] ], [ [[XTRAITER]], [[MIDDLE_BLOCK_UNR_LCSSA]] ]
+; AUTO_VEC-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_NEXT:%.*]], [[VECTOR_BODY_EPIL]] ], [ 0, [[MIDDLE_BLOCK_UNR_LCSSA]] ]
 ; AUTO_VEC-NEXT:    [[STEP_ADD_EPIL:%.*]] = fadd fast <8 x float> [[VEC_IND_EPIL]], <float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00>
 ; AUTO_VEC-NEXT:    [[STEP_ADD2_EPIL:%.*]] = fadd fast <8 x float> [[VEC_IND_EPIL]], <float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00>
 ; AUTO_VEC-NEXT:    [[STEP_ADD3_EPIL:%.*]] = fadd fast <8 x float> [[VEC_IND_EPIL]], <float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01>
@@ -138,8 +138,8 @@ define void @fp_iv_loop1(float* noalias nocapture %A, i32 %N) #0 {
 ; AUTO_VEC-NEXT:    store <8 x float> [[STEP_ADD3_EPIL]], <8 x float>* [[TMP44]], align 4
 ; AUTO_VEC-NEXT:    [[INDEX_NEXT_EPIL]] = add nuw i64 [[INDEX_EPIL]], 32
 ; AUTO_VEC-NEXT:    [[VEC_IND_NEXT_EPIL]] = fadd fast <8 x float> [[VEC_IND_EPIL]], <float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01>
-; AUTO_VEC-NEXT:    [[EPIL_ITER_SUB]] = add i64 [[EPIL_ITER]], -1
-; AUTO_VEC-NEXT:    [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[EPIL_ITER_SUB]], 0
+; AUTO_VEC-NEXT:    [[EPIL_ITER_NEXT]] = add i64 [[EPIL_ITER]], 1
+; AUTO_VEC-NEXT:    [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[EPIL_ITER_NEXT]], [[XTRAITER]]
 ; AUTO_VEC-NEXT:    br i1 [[EPIL_ITER_CMP_NOT]], label [[MIDDLE_BLOCK]], label [[VECTOR_BODY_EPIL]], !llvm.loop [[LOOP2:![0-9]+]]
 ; AUTO_VEC:       middle.block:
 ; AUTO_VEC-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[ZEXT]]
@@ -208,7 +208,7 @@ define void @fp_iv_loop2(float* noalias nocapture %A, i32 %N) {
 ; AUTO_VEC:       for.body:
 ; AUTO_VEC-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[INDVARS_IV_NEXT_7:%.*]], [[FOR_BODY]] ]
 ; AUTO_VEC-NEXT:    [[X_06:%.*]] = phi float [ 1.000000e+00, [[FOR_BODY_PREHEADER_NEW]] ], [ [[CONV1_7:%.*]], [[FOR_BODY]] ]
-; AUTO_VEC-NEXT:    [[NITER:%.*]] = phi i64 [ [[UNROLL_ITER]], [[FOR_BODY_PREHEADER_NEW]] ], [ [[NITER_NSUB_7:%.*]], [[FOR_BODY]] ]
+; AUTO_VEC-NEXT:    [[NITER:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[NITER_NEXT_7:%.*]], [[FOR_BODY]] ]
 ; AUTO_VEC-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[INDVARS_IV]]
 ; AUTO_VEC-NEXT:    store float [[X_06]], float* [[ARRAYIDX]], align 4
 ; AUTO_VEC-NEXT:    [[CONV1:%.*]] = fadd float [[X_06]], 5.000000e-01
@@ -241,8 +241,8 @@ define void @fp_iv_loop2(float* noalias nocapture %A, i32 %N) {
 ; AUTO_VEC-NEXT:    store float [[CONV1_6]], float* [[ARRAYIDX_7]], align 4
 ; AUTO_VEC-NEXT:    [[CONV1_7]] = fadd float [[CONV1_6]], 5.000000e-01
 ; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT_7]] = add nuw nsw i64 [[INDVARS_IV]], 8
-; AUTO_VEC-NEXT:    [[NITER_NSUB_7]] = add i64 [[NITER]], -8
-; AUTO_VEC-NEXT:    [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NSUB_7]], 0
+; AUTO_VEC-NEXT:    [[NITER_NEXT_7]] = add i64 [[NITER]], 8
+; AUTO_VEC-NEXT:    [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NEXT_7]], [[UNROLL_ITER]]
 ; AUTO_VEC-NEXT:    br i1 [[NITER_NCMP_7]], label [[FOR_END_LOOPEXIT_UNR_LCSSA]], label [[FOR_BODY]]
 ; AUTO_VEC:       for.end.loopexit.unr-lcssa:
 ; AUTO_VEC-NEXT:    [[INDVARS_IV_UNR:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_NEXT_7]], [[FOR_BODY]] ]
@@ -252,13 +252,13 @@ define void @fp_iv_loop2(float* noalias nocapture %A, i32 %N) {
 ; AUTO_VEC:       for.body.epil:
 ; AUTO_VEC-NEXT:    [[INDVARS_IV_EPIL:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[INDVARS_IV_UNR]], [[FOR_END_LOOPEXIT_UNR_LCSSA]] ]
 ; AUTO_VEC-NEXT:    [[X_06_EPIL:%.*]] = phi float [ [[CONV1_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[X_06_UNR]], [[FOR_END_LOOPEXIT_UNR_LCSSA]] ]
-; AUTO_VEC-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_SUB:%.*]], [[FOR_BODY_EPIL]] ], [ [[XTRAITER]], [[FOR_END_LOOPEXIT_UNR_LCSSA]] ]
+; AUTO_VEC-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_NEXT:%.*]], [[FOR_BODY_EPIL]] ], [ 0, [[FOR_END_LOOPEXIT_UNR_LCSSA]] ]
 ; AUTO_VEC-NEXT:    [[ARRAYIDX_EPIL:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV_EPIL]]
 ; AUTO_VEC-NEXT:    store float [[X_06_EPIL]], float* [[ARRAYIDX_EPIL]], align 4
 ; AUTO_VEC-NEXT:    [[CONV1_EPIL]] = fadd float [[X_06_EPIL]], 5.000000e-01
 ; AUTO_VEC-NEXT:    [[INDVARS_IV_NEXT_EPIL]] = add nuw nsw i64 [[INDVARS_IV_EPIL]], 1
-; AUTO_VEC-NEXT:    [[EPIL_ITER_SUB]] = add i64 [[EPIL_ITER]], -1
-; AUTO_VEC-NEXT:    [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[EPIL_ITER_SUB]], 0
+; AUTO_VEC-NEXT:    [[EPIL_ITER_NEXT]] = add i64 [[EPIL_ITER]], 1
+; AUTO_VEC-NEXT:    [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[EPIL_ITER_NEXT]], [[XTRAITER]]
 ; AUTO_VEC-NEXT:    br i1 [[EPIL_ITER_CMP_NOT]], label [[FOR_END]], label [[FOR_BODY_EPIL]], !llvm.loop [[LOOP6:![0-9]+]]
 ; AUTO_VEC:       for.end:
 ; AUTO_VEC-NEXT:    ret void
@@ -310,7 +310,7 @@ define double @external_use_with_fast_math(double* %a, i64 %n) {
 ; AUTO_VEC:       vector.body:
 ; AUTO_VEC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH_NEW]] ], [ [[INDEX_NEXT_3:%.*]], [[VECTOR_BODY]] ]
 ; AUTO_VEC-NEXT:    [[VEC_IND:%.*]] = phi <4 x double> [ <double 0.000000e+00, double 3.000000e+00, double 6.000000e+00, double 9.000000e+00>, [[VECTOR_PH_NEW]] ], [ [[VEC_IND_NEXT_3:%.*]], [[VECTOR_BODY]] ]
-; AUTO_VEC-NEXT:    [[NITER:%.*]] = phi i64 [ [[UNROLL_ITER]], [[VECTOR_PH_NEW]] ], [ [[NITER_NSUB_3:%.*]], [[VECTOR_BODY]] ]
+; AUTO_VEC-NEXT:    [[NITER:%.*]] = phi i64 [ 0, [[VECTOR_PH_NEW]] ], [ [[NITER_NEXT_3:%.*]], [[VECTOR_BODY]] ]
 ; AUTO_VEC-NEXT:    [[STEP_ADD:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 1.200000e+01, double 1.200000e+01, double 1.200000e+01, double 1.200000e+01>
 ; AUTO_VEC-NEXT:    [[STEP_ADD2:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 2.400000e+01, double 2.400000e+01, double 2.400000e+01, double 2.400000e+01>
 ; AUTO_VEC-NEXT:    [[STEP_ADD3:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 3.600000e+01, double 3.600000e+01, double 3.600000e+01, double 3.600000e+01>
@@ -379,8 +379,8 @@ define double @external_use_with_fast_math(double* %a, i64 %n) {
 ; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD3_3]], <4 x double>* [[TMP36]], align 8
 ; AUTO_VEC-NEXT:    [[INDEX_NEXT_3]] = add nuw i64 [[INDEX]], 64
 ; AUTO_VEC-NEXT:    [[VEC_IND_NEXT_3]] = fadd fast <4 x double> [[VEC_IND]], <double 1.920000e+02, double 1.920000e+02, double 1.920000e+02, double 1.920000e+02>
-; AUTO_VEC-NEXT:    [[NITER_NSUB_3]] = add i64 [[NITER]], -4
-; AUTO_VEC-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i64 [[NITER_NSUB_3]], 0
+; AUTO_VEC-NEXT:    [[NITER_NEXT_3]] = add i64 [[NITER]], 4
+; AUTO_VEC-NEXT:    [[NITER_NCMP_3:%.*]] = icmp eq i64 [[NITER_NEXT_3]], [[UNROLL_ITER]]
 ; AUTO_VEC-NEXT:    br i1 [[NITER_NCMP_3]], label [[MIDDLE_BLOCK_UNR_LCSSA]], label [[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]]
 ; AUTO_VEC:       middle.block.unr-lcssa:
 ; AUTO_VEC-NEXT:    [[INDEX_UNR:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT_3]], [[VECTOR_BODY]] ]
@@ -390,7 +390,7 @@ define double @external_use_with_fast_math(double* %a, i64 %n) {
 ; AUTO_VEC:       vector.body.epil:
 ; AUTO_VEC-NEXT:    [[INDEX_EPIL:%.*]] = phi i64 [ [[INDEX_NEXT_EPIL:%.*]], [[VECTOR_BODY_EPIL]] ], [ [[INDEX_UNR]], [[MIDDLE_BLOCK_UNR_LCSSA]] ]
 ; AUTO_VEC-NEXT:    [[VEC_IND_EPIL:%.*]] = phi <4 x double> [ [[VEC_IND_NEXT_EPIL:%.*]], [[VECTOR_BODY_EPIL]] ], [ [[VEC_IND_UNR]], [[MIDDLE_BLOCK_UNR_LCSSA]] ]
-; AUTO_VEC-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_SUB:%.*]], [[VECTOR_BODY_EPIL]] ], [ [[XTRAITER]], [[MIDDLE_BLOCK_UNR_LCSSA]] ]
+; AUTO_VEC-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_NEXT:%.*]], [[VECTOR_BODY_EPIL]] ], [ 0, [[MIDDLE_BLOCK_UNR_LCSSA]] ]
 ; AUTO_VEC-NEXT:    [[STEP_ADD_EPIL:%.*]] = fadd fast <4 x double> [[VEC_IND_EPIL]], <double 1.200000e+01, double 1.200000e+01, double 1.200000e+01, double 1.200000e+01>
 ; AUTO_VEC-NEXT:    [[STEP_ADD2_EPIL:%.*]] = fadd fast <4 x double> [[VEC_IND_EPIL]], <double 2.400000e+01, double 2.400000e+01, double 2.400000e+01, double 2.400000e+01>
 ; AUTO_VEC-NEXT:    [[STEP_ADD3_EPIL:%.*]] = fadd fast <4 x double> [[VEC_IND_EPIL]], <double 3.600000e+01, double 3.600000e+01, double 3.600000e+01, double 3.600000e+01>
@@ -408,8 +408,8 @@ define double @external_use_with_fast_math(double* %a, i64 %n) {
 ; AUTO_VEC-NEXT:    store <4 x double> [[STEP_ADD3_EPIL]], <4 x double>* [[TMP44]], align 8
 ; AUTO_VEC-NEXT:    [[INDEX_NEXT_EPIL]] = add nuw i64 [[INDEX_EPIL]], 16
 ; AUTO_VEC-NEXT:    [[VEC_IND_NEXT_EPIL]] = fadd fast <4 x double> [[VEC_IND_EPIL]], <double 4.800000e+01, double 4.800000e+01, double 4.800000e+01, double 4.800000e+01>
-; AUTO_VEC-NEXT:    [[EPIL_ITER_SUB]] = add i64 [[EPIL_ITER]], -1
-; AUTO_VEC-NEXT:    [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[EPIL_ITER_SUB]], 0
+; AUTO_VEC-NEXT:    [[EPIL_ITER_NEXT]] = add i64 [[EPIL_ITER]], 1
+; AUTO_VEC-NEXT:    [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[EPIL_ITER_NEXT]], [[XTRAITER]]
 ; AUTO_VEC-NEXT:    br i1 [[EPIL_ITER_CMP_NOT]], label [[MIDDLE_BLOCK]], label [[VECTOR_BODY_EPIL]], !llvm.loop [[LOOP8:![0-9]+]]
 ; AUTO_VEC:       middle.block:
 ; AUTO_VEC-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
@@ -462,7 +462,7 @@ define double @external_use_without_fast_math(double* %a, i64 %n) {
 ; AUTO_VEC:       for.body:
 ; AUTO_VEC-NEXT:    [[I:%.*]] = phi i64 [ 0, [[ENTRY_NEW]] ], [ [[I_NEXT_7:%.*]], [[FOR_BODY]] ]
 ; AUTO_VEC-NEXT:    [[J:%.*]] = phi double [ 0.000000e+00, [[ENTRY_NEW]] ], [ [[J_NEXT_7:%.*]], [[FOR_BODY]] ]
-; AUTO_VEC-NEXT:    [[NITER:%.*]] = phi i64 [ [[UNROLL_ITER]], [[ENTRY_NEW]] ], [ [[NITER_NSUB_7:%.*]], [[FOR_BODY]] ]
+; AUTO_VEC-NEXT:    [[NITER:%.*]] = phi i64 [ 0, [[ENTRY_NEW]] ], [ [[NITER_NEXT_7:%.*]], [[FOR_BODY]] ]
 ; AUTO_VEC-NEXT:    [[T0:%.*]] = getelementptr double, double* [[A:%.*]], i64 [[I]]
 ; AUTO_VEC-NEXT:    store double [[J]], double* [[T0]], align 8
 ; AUTO_VEC-NEXT:    [[I_NEXT:%.*]] = or i64 [[I]], 1
@@ -495,8 +495,8 @@ define double @external_use_without_fast_math(double* %a, i64 %n) {
 ; AUTO_VEC-NEXT:    store double [[J_NEXT_6]], double* [[T0_7]], align 8
 ; AUTO_VEC-NEXT:    [[I_NEXT_7]] = add nuw nsw i64 [[I]], 8
 ; AUTO_VEC-NEXT:    [[J_NEXT_7]] = fadd double [[J_NEXT_6]], 3.000000e+00
-; AUTO_VEC-NEXT:    [[NITER_NSUB_7]] = add i64 [[NITER]], -8
-; AUTO_VEC-NEXT:    [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NSUB_7]], 0
+; AUTO_VEC-NEXT:    [[NITER_NEXT_7]] = add i64 [[NITER]], 8
+; AUTO_VEC-NEXT:    [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NEXT_7]], [[UNROLL_ITER]]
 ; AUTO_VEC-NEXT:    br i1 [[NITER_NCMP_7]], label [[FOR_END_UNR_LCSSA]], label [[FOR_BODY]]
 ; AUTO_VEC:       for.end.unr-lcssa:
 ; AUTO_VEC-NEXT:    [[J_LCSSA_PH:%.*]] = phi double [ undef, [[ENTRY:%.*]] ], [ [[J_NEXT_6]], [[FOR_BODY]] ]
@@ -507,13 +507,13 @@ define double @external_use_without_fast_math(double* %a, i64 %n) {
 ; AUTO_VEC:       for.body.epil:
 ; AUTO_VEC-NEXT:    [[I_EPIL:%.*]] = phi i64 [ [[I_NEXT_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[I_UNR]], [[FOR_END_UNR_LCSSA]] ]
 ; AUTO_VEC-NEXT:    [[J_EPIL:%.*]] = phi double [ [[J_NEXT_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[J_UNR]], [[FOR_END_UNR_LCSSA]] ]
-; AUTO_VEC-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_SUB:%.*]], [[FOR_BODY_EPIL]] ], [ [[XTRAITER]], [[FOR_END_UNR_LCSSA]] ]
+; AUTO_VEC-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_NEXT:%.*]], [[FOR_BODY_EPIL]] ], [ 0, [[FOR_END_UNR_LCSSA]] ]
 ; AUTO_VEC-NEXT:    [[T0_EPIL:%.*]] = getelementptr double, double* [[A]], i64 [[I_EPIL]]
 ; AUTO_VEC-NEXT:    store double [[J_EPIL]], double* [[T0_EPIL]], align 8
 ; AUTO_VEC-NEXT:    [[I_NEXT_EPIL]] = add nuw nsw i64 [[I_EPIL]], 1
 ; AUTO_VEC-NEXT:    [[J_NEXT_EPIL]] = fadd double [[J_EPIL]], 3.000000e+00
-; AUTO_VEC-NEXT:    [[EPIL_ITER_SUB]] = add i64 [[EPIL_ITER]], -1
-; AUTO_VEC-NEXT:    [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[EPIL_ITER_SUB]], 0
+; AUTO_VEC-NEXT:    [[EPIL_ITER_NEXT]] = add i64 [[EPIL_ITER]], 1
+; AUTO_VEC-NEXT:    [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[EPIL_ITER_NEXT]], [[XTRAITER]]
 ; AUTO_VEC-NEXT:    br i1 [[EPIL_ITER_CMP_NOT]], label [[FOR_END]], label [[FOR_BODY_EPIL]], !llvm.loop [[LOOP10:![0-9]+]]
 ; AUTO_VEC:       for.end:
 ; AUTO_VEC-NEXT:    [[J_LCSSA:%.*]] = phi double [ [[J_LCSSA_PH]], [[FOR_END_UNR_LCSSA]] ], [ [[J_EPIL]], [[FOR_BODY_EPIL]] ]
@@ -571,7 +571,7 @@ define void @fadd_reassoc_FMF(float* nocapture %p, i32 %N) {
 ; AUTO_VEC:       vector.body:
 ; AUTO_VEC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH_NEW]] ], [ [[INDEX_NEXT_1:%.*]], [[VECTOR_BODY]] ]
 ; AUTO_VEC-NEXT:    [[VEC_IND:%.*]] = phi <8 x float> [ <float 1.000000e+00, float 4.300000e+01, float 8.500000e+01, float 1.270000e+02, float 1.690000e+02, float 2.110000e+02, float 2.530000e+02, float 2.950000e+02>, [[VECTOR_PH_NEW]] ], [ [[VEC_IND_NEXT_1:%.*]], [[VECTOR_BODY]] ]
-; AUTO_VEC-NEXT:    [[NITER:%.*]] = phi i64 [ [[UNROLL_ITER]], [[VECTOR_PH_NEW]] ], [ [[NITER_NSUB_1:%.*]], [[VECTOR_BODY]] ]
+; AUTO_VEC-NEXT:    [[NITER:%.*]] = phi i64 [ 0, [[VECTOR_PH_NEW]] ], [ [[NITER_NEXT_1:%.*]], [[VECTOR_BODY]] ]
 ; AUTO_VEC-NEXT:    [[STEP_ADD:%.*]] = fadd reassoc <8 x float> [[VEC_IND]], <float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02>
 ; AUTO_VEC-NEXT:    [[STEP_ADD2:%.*]] = fadd reassoc <8 x float> [[STEP_ADD]], <float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02>
 ; AUTO_VEC-NEXT:    [[STEP_ADD3:%.*]] = fadd reassoc <8 x float> [[STEP_ADD2]], <float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02>
@@ -630,8 +630,8 @@ define void @fadd_reassoc_FMF(float* nocapture %p, i32 %N) {
 ; AUTO_VEC-NEXT:    store <8 x float> [[TMP33]], <8 x float>* [[TMP37]], align 4
 ; AUTO_VEC-NEXT:    [[INDEX_NEXT_1]] = add nuw i64 [[INDEX]], 64
 ; AUTO_VEC-NEXT:    [[VEC_IND_NEXT_1]] = fadd reassoc <8 x float> [[STEP_ADD3_1]], <float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02, float 3.360000e+02>
-; AUTO_VEC-NEXT:    [[NITER_NSUB_1]] = add i64 [[NITER]], -2
-; AUTO_VEC-NEXT:    [[NITER_NCMP_1:%.*]] = icmp eq i64 [[NITER_NSUB_1]], 0
+; AUTO_VEC-NEXT:    [[NITER_NEXT_1]] = add i64 [[NITER]], 2
+; AUTO_VEC-NEXT:    [[NITER_NCMP_1:%.*]] = icmp eq i64 [[NITER_NEXT_1]], [[UNROLL_ITER]]
 ; AUTO_VEC-NEXT:    br i1 [[NITER_NCMP_1]], label [[MIDDLE_BLOCK_UNR_LCSSA]], label [[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]]
 ; AUTO_VEC:       middle.block.unr-lcssa:
 ; AUTO_VEC-NEXT:    [[INDEX_UNR:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT_1]], [[VECTOR_BODY]] ]

diff --git a/llvm/test/Transforms/LoopVectorize/X86/gather_scatter.ll b/llvm/test/Transforms/LoopVectorize/X86/gather_scatter.ll
index 1ab06e8ba51b4882dd5269a567f84854b19375d4..77ab42885ff3b562ece4a146b963681ed56a7703 100644 (file)
--- a/llvm/test/Transforms/LoopVectorize/X86/gather_scatter.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/gather_scatter.ll
@@ -1450,7 +1450,7 @@ define void @test_gather_not_profitable_pr48429(i32 %d, float* readonly %ptr, fl
 ; AVX512:       vector.body:
 ; AVX512-NEXT:    [[POINTER_PHI:%.*]] = phi float* [ [[DEST]], [[VECTOR_PH_NEW]] ], [ [[PTR_IND_7:%.*]], [[VECTOR_BODY]] ]
 ; AVX512-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH_NEW]] ], [ [[INDEX_NEXT_7:%.*]], [[VECTOR_BODY]] ]
-; AVX512-NEXT:    [[NITER:%.*]] = phi i64 [ [[UNROLL_ITER]], [[VECTOR_PH_NEW]] ], [ [[NITER_NSUB_7:%.*]], [[VECTOR_BODY]] ]
+; AVX512-NEXT:    [[NITER:%.*]] = phi i64 [ 0, [[VECTOR_PH_NEW]] ], [ [[NITER_NEXT_7:%.*]], [[VECTOR_BODY]] ]
 ; AVX512-NEXT:    [[NEXT_GEP:%.*]] = getelementptr float, float* [[PTR]], i64 [[INDEX]]
 ; AVX512-NEXT:    [[TMP17:%.*]] = getelementptr float, float* [[POINTER_PHI]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
 ; AVX512-NEXT:    [[TMP18:%.*]] = getelementptr inbounds float, float* [[NEXT_GEP]], i64 [[IDXPROM]]
@@ -1547,8 +1547,8 @@ define void @test_gather_not_profitable_pr48429(i32 %d, float* readonly %ptr, fl
 ; AVX512-NEXT:    call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD15_7]], <16 x float*> [[TMP56]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
 ; AVX512-NEXT:    [[INDEX_NEXT_7]] = add nuw i64 [[INDEX]], 128
 ; AVX512-NEXT:    [[PTR_IND_7]] = getelementptr float, float* [[POINTER_PHI]], i64 2048
-; AVX512-NEXT:    [[NITER_NSUB_7]] = add i64 [[NITER]], -8
-; AVX512-NEXT:    [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NSUB_7]], 0
+; AVX512-NEXT:    [[NITER_NEXT_7]] = add i64 [[NITER]], 8
+; AVX512-NEXT:    [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NEXT_7]], [[UNROLL_ITER]]
 ; AVX512-NEXT:    br i1 [[NITER_NCMP_7]], label [[MIDDLE_BLOCK_UNR_LCSSA]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
 ; AVX512:       middle.block.unr-lcssa:
 ; AVX512-NEXT:    [[POINTER_PHI_UNR:%.*]] = phi float* [ [[DEST]], [[VECTOR_PH]] ], [ [[PTR_IND_7]], [[VECTOR_BODY]] ]
@@ -1558,7 +1558,7 @@ define void @test_gather_not_profitable_pr48429(i32 %d, float* readonly %ptr, fl
 ; AVX512:       vector.body.epil:
 ; AVX512-NEXT:    [[POINTER_PHI_EPIL:%.*]] = phi float* [ [[PTR_IND_EPIL:%.*]], [[VECTOR_BODY_EPIL]] ], [ [[POINTER_PHI_UNR]], [[MIDDLE_BLOCK_UNR_LCSSA]] ]
 ; AVX512-NEXT:    [[INDEX_EPIL:%.*]] = phi i64 [ [[INDEX_NEXT_EPIL:%.*]], [[VECTOR_BODY_EPIL]] ], [ [[INDEX_UNR]], [[MIDDLE_BLOCK_UNR_LCSSA]] ]
-; AVX512-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_SUB:%.*]], [[VECTOR_BODY_EPIL]] ], [ [[XTRAITER]], [[MIDDLE_BLOCK_UNR_LCSSA]] ]
+; AVX512-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_NEXT:%.*]], [[VECTOR_BODY_EPIL]] ], [ 0, [[MIDDLE_BLOCK_UNR_LCSSA]] ]
 ; AVX512-NEXT:    [[NEXT_GEP_EPIL:%.*]] = getelementptr float, float* [[PTR]], i64 [[INDEX_EPIL]]
 ; AVX512-NEXT:    [[TMP57:%.*]] = getelementptr float, float* [[POINTER_PHI_EPIL]], <16 x i64> <i64 0, i64 16, i64 32, i64 48, i64 64, i64 80, i64 96, i64 112, i64 128, i64 144, i64 160, i64 176, i64 192, i64 208, i64 224, i64 240>
 ; AVX512-NEXT:    [[TMP58:%.*]] = getelementptr inbounds float, float* [[NEXT_GEP_EPIL]], i64 [[IDXPROM]]
@@ -1571,8 +1571,8 @@ define void @test_gather_not_profitable_pr48429(i32 %d, float* readonly %ptr, fl
 ; AVX512-NEXT:    call void @llvm.masked.scatter.v16f32.v16p0f32(<16 x float> [[WIDE_LOAD15_EPIL]], <16 x float*> [[TMP61]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !5, !noalias !7
 ; AVX512-NEXT:    [[INDEX_NEXT_EPIL]] = add nuw i64 [[INDEX_EPIL]], 16
 ; AVX512-NEXT:    [[PTR_IND_EPIL]] = getelementptr float, float* [[POINTER_PHI_EPIL]], i64 256
-; AVX512-NEXT:    [[EPIL_ITER_SUB]] = add i64 [[EPIL_ITER]], -1
-; AVX512-NEXT:    [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[EPIL_ITER_SUB]], 0
+; AVX512-NEXT:    [[EPIL_ITER_NEXT]] = add i64 [[EPIL_ITER]], 1
+; AVX512-NEXT:    [[EPIL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[EPIL_ITER_NEXT]], [[XTRAITER]]
 ; AVX512-NEXT:    br i1 [[EPIL_ITER_CMP_NOT]], label [[MIDDLE_BLOCK]], label [[VECTOR_BODY_EPIL]], !llvm.loop [[LOOP11:![0-9]+]]
 ; AVX512:       middle.block:
 ; AVX512-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]

diff --git a/llvm/test/Transforms/PhaseOrdering/X86/vdiv.ll b/llvm/test/Transforms/PhaseOrdering/X86/vdiv.ll
index 70238d98b2cd7666bd444ffd63a6ff30697854a6..ba9d8c467c2e29aed01352dfe734d58945b1da16 100644 (file)
--- a/llvm/test/Transforms/PhaseOrdering/X86/vdiv.ll
+++ b/llvm/test/Transforms/PhaseOrdering/X86/vdiv.ll
@@ -90,15 +90,15 @@ define void @vdiv(double* %x, double* %y, double %a, i32 %N) #0 {
 ; CHECK-NEXT:    br label [[FOR_BODY_PROL:%.*]]
 ; CHECK:       for.body.prol:
 ; CHECK-NEXT:    [[INDVARS_IV_PROL:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_PROL:%.*]], [[FOR_BODY_PROL]] ], [ [[INDVARS_IV_PH]], [[FOR_BODY_PROL_PREHEADER]] ]
-; CHECK-NEXT:    [[PROL_ITER:%.*]] = phi i64 [ [[PROL_ITER_SUB:%.*]], [[FOR_BODY_PROL]] ], [ [[XTRAITER]], [[FOR_BODY_PROL_PREHEADER]] ]
+; CHECK-NEXT:    [[PROL_ITER:%.*]] = phi i64 [ [[PROL_ITER_NEXT:%.*]], [[FOR_BODY_PROL]] ], [ 0, [[FOR_BODY_PROL_PREHEADER]] ]
 ; CHECK-NEXT:    [[ARRAYIDX_PROL:%.*]] = getelementptr inbounds double, double* [[Y]], i64 [[INDVARS_IV_PROL]]
 ; CHECK-NEXT:    [[T0_PROL:%.*]] = load double, double* [[ARRAYIDX_PROL]], align 8, !tbaa [[TBAA3]]
 ; CHECK-NEXT:    [[TMP28:%.*]] = fmul fast double [[T0_PROL]], [[TMP27]]
 ; CHECK-NEXT:    [[ARRAYIDX2_PROL:%.*]] = getelementptr inbounds double, double* [[X]], i64 [[INDVARS_IV_PROL]]
 ; CHECK-NEXT:    store double [[TMP28]], double* [[ARRAYIDX2_PROL]], align 8, !tbaa [[TBAA3]]
 ; CHECK-NEXT:    [[INDVARS_IV_NEXT_PROL]] = add nuw nsw i64 [[INDVARS_IV_PROL]], 1
-; CHECK-NEXT:    [[PROL_ITER_SUB]] = add i64 [[PROL_ITER]], -1
-; CHECK-NEXT:    [[PROL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[PROL_ITER_SUB]], 0
+; CHECK-NEXT:    [[PROL_ITER_NEXT]] = add i64 [[PROL_ITER]], 1
+; CHECK-NEXT:    [[PROL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[PROL_ITER_NEXT]], [[XTRAITER]]
 ; CHECK-NEXT:    br i1 [[PROL_ITER_CMP_NOT]], label [[FOR_BODY_PROL_LOOPEXIT]], label [[FOR_BODY_PROL]], !llvm.loop [[LOOP14:![0-9]+]]
 ; CHECK:       for.body.prol.loopexit:
 ; CHECK-NEXT:    [[INDVARS_IV_UNR:%.*]] = phi i64 [ [[INDVARS_IV_PH]], [[FOR_BODY_PREHEADER18]] ], [ [[INDVARS_IV_NEXT_PROL]], [[FOR_BODY_PROL]] ]