[SCEV] Add smin/umin support to getRangeRef

[reames]

Summary

We were failing to compute trip counts (both exact and maximum) for any loop which involved a comparison against either an umin or smin. It looks like this simply got missed when we added smin/umin to SCEV.

I'm more than a bit shocked that something this basis was missed for this long. Am I missing something?

[fhahn]

I've put up a patch for SCEVUmin D67177. I plan to land it tomorrow.

[reames]

Rebase over precommited tests.

[reames]

In D67514#1668282, @fhahn wrote:

I've put up a patch for SCEVUmin D67177. I plan to land it tomorrow.

Ok, could you review this one then? The umin portion looks identical, so assume I remove that after rebasing on your change.

[nikic]

LG. Note that umin/smin support is also missing from GetMinTrailingZerosImpl.

[nikic]

lib/Analysis/ScalarEvolution.cpp

5602

Just a side note: For operations that have an intrinsic sign preference, I think it would be better to pass on that specific signedness as the SignHint (here signed, below unsigned). Not sure if there's an assumption about a getRangeRef() chain always using the same signedness somewhere though.

[fhahn]

LGTM, thanks! I've just landed D67177, to avoid delays with this patch.

FYI my motivation for UMin support is improving SCEV results with information from loop guards: D67178.

lib/Analysis/ScalarEvolution.cpp

ScalarEvolution::getRangeRef(const SCEV *S,
   if (const SCEVUMaxExpr *UMax = dyn_cast<SCEVUMaxExpr>(S)) {
     ConstantRange X = getRangeRef(UMax->getOperand(0), SignHint);
     for (unsigned i = 1, e = UMax->getNumOperands(); i != e; ++i)
       X = X.umax(getRangeRef(UMax->getOperand(i), SignHint));
     return setRange(UMax, SignHint,
                     ConservativeResult.intersectWith(X, RangeType));
   }
 
+  if (const SCEVSMinExpr *SMin = dyn_cast<SCEVSMinExpr>(S)) {
+    ConstantRange X = getRangeRef(SMin->getOperand(0), SignHint);
+    for (unsigned i = 1, e = SMin->getNumOperands(); i != e; ++i)
+      X = X.smin(getRangeRef(SMin->getOperand(i), SignHint));
+    return setRange(SMin, SignHint,
+                    ConservativeResult.intersectWith(X, RangeType));
+  }
+
+  if (const SCEVUMinExpr *UMin = dyn_cast<SCEVUMinExpr>(S)) {
+    ConstantRange X = getRangeRef(UMin->getOperand(0), SignHint);
+    for (unsigned i = 1, e = UMin->getNumOperands(); i != e; ++i)
+      X = X.umin(getRangeRef(UMin->getOperand(i), SignHint));
+    return setRange(UMin, SignHint,
+                    ConservativeResult.intersectWith(X, RangeType));
+  }
+  
   if (const SCEVUDivExpr *UDiv = dyn_cast<SCEVUDivExpr>(S)) {
     ConstantRange X = getRangeRef(UDiv->getLHS(), SignHint);
     ConstantRange Y = getRangeRef(UDiv->getRHS(), SignHint);
     return setRange(UDiv, SignHint,
                     ConservativeResult.intersectWith(X.udiv(Y), RangeType));
   }
 
   if (const SCEVZeroExtendExpr *ZExt = dyn_cast<SCEVZeroExtendExpr>(S)) {

test/Analysis/ScalarEvolution/max-expr-cache.ll

bb4:
   %tmp40 = select i1 %tmp39, i32 %tmp38, i32 256
   %tmp41 = add nsw i32 %tmp40, 1
   %tmp42 = icmp sgt i32 %tmp8, %tmp41
   %tmp43 = select i1 %tmp42, i32 %tmp41, i32 %tmp8
   %tmp44 = add nsw i32 %tmp10, 7
   %tmp45 = icmp slt i32 %tmp43, 256
   %tmp46 = select i1 %tmp45, i32 %tmp43, i32 256
 ; CHECK:  %tmp46 = select i1 %tmp45, i32 %tmp43, i32 256
-; CHECK-NEXT:  -->  (256 smin (1 + (256 smin (1 + (256 smin (1 + (256 smin (1 + (256 smin (1 + (256 smin (1 + (256 smin (1 + (256 smin {%tmp3,+,-256}<%bb4>)) smin {%tmp3,+,-256}<%bb4>)) smin {%tmp3,+,-256}<%bb4>)) smin {%tmp3,+,-256}<%bb4>)) smin {%tmp3,+,-256}<%bb4>)) smin {%tmp3,+,-256}<%bb4>)) smin {%tmp3,+,-256}<%bb4>)) smin {%tmp3,+,-256}<%bb4>)
+; CHECK-NEXT:  -->  (256 smin (1 + (256 smin (1 + (256 smin (1 + (256 smin (1 + (256 smin (1 + (256 smin (1 + (256 smin (1 + (256 smin {%tmp3,+,-256}<%bb4>))<nsw> smin {%tmp3,+,-256}<%bb4>))<nsw> smin {%tmp3,+,-256}<%bb4>))<nsw> smin {%tmp3,+,-256}<%bb4>))<nsw> smin {%tmp3,+,-256}<%bb4>))<nsw> smin {%tmp3,+,-256}<%bb4>))<nsw> smin {%tmp3,+,-256}<%bb4>))<nsw> smin {%tmp3,+,-256}<%bb4>) U: [-2147483648,257) S: [-2147483648,257)
   %tmp47 = icmp sgt i32 %tmp44, %tmp46
   %tmp48 = select i1 %tmp47, i32 %tmp44, i32 %tmp46
   %tmp49 = ashr i32 %tmp48, 3
   %tmp50 = icmp sgt i32 %tmp49, 0
   %tmp51 = select i1 %tmp50, i32 %tmp49, i32 0
   %tmp52 = zext i32 %tmp51 to i64
   br label %bb53
 
bb4:
   %tmp40 = select i1 %tmp39, i32 %tmp38, i32 256
   %tmp41 = add nsw i32 %tmp40, 1
   %tmp42 = icmp ugt i32 %tmp8, %tmp41
   %tmp43 = select i1 %tmp42, i32 %tmp41, i32 %tmp8
   %tmp44 = add nsw i32 %tmp10, 7
   %tmp45 = icmp ult i32 %tmp43, 256
   %tmp46 = select i1 %tmp45, i32 %tmp43, i32 256
 ; CHECK:  %tmp46 = select i1 %tmp45, i32 %tmp43, i32 256
-; CHECK-NEXT:  --> (256 umin (1 + (256 umin (1 + (256 umin (1 + (256 umin (1 + (256 umin (1 + (256 umin (1 + (256 umin (1 + (256 umin {%tmp3,+,-256}<%bb4>)) umin {%tmp3,+,-256}<%bb4>)) umin {%tmp3,+,-256}<%bb4>)) umin {%tmp3,+,-256}<%bb4>)) umin {%tmp3,+,-256}<%bb4>)) umin {%tmp3,+,-256}<%bb4>)) umin {%tmp3,+,-256}<%bb4>)) umin {%tmp3,+,-256}<%bb4>) 
+; CHECK-NEXT:  --> (256 umin (1 + (256 umin (1 + (256 umin (1 + (256 umin (1 + (256 umin (1 + (256 umin (1 + (256 umin (1 + (256 umin {%tmp3,+,-256}<%bb4>))<nuw><nsw> umin {%tmp3,+,-256}<%bb4>))<nuw><nsw> umin {%tmp3,+,-256}<%bb4>))<nuw><nsw> umin {%tmp3,+,-256}<%bb4>))<nuw><nsw> umin {%tmp3,+,-256}<%bb4>))<nuw><nsw> umin {%tmp3,+,-256}<%bb4>))<nuw><nsw> umin {%tmp3,+,-256}<%bb4>))<nuw><nsw> umin {%tmp3,+,-256}<%bb4>) U: [0,257) S: [0,257)
   %tmp47 = icmp ugt i32 %tmp44, %tmp46
   %tmp48 = select i1 %tmp47, i32 %tmp44, i32 %tmp46
   %tmp49 = ashr i32 %tmp48, 3
   %tmp50 = icmp ugt i32 %tmp49, 0
   %tmp51 = select i1 %tmp50, i32 %tmp49, i32 0
   %tmp52 = zext i32 %tmp51 to i64
   br label %bb53
 

test/Analysis/ScalarEvolution/trip-count15.ll

 ; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
 ; RUN: opt -S -analyze -scalar-evolution < %s | FileCheck %s
 
 define void @umin_unsigned_check(i64 %n) {
 ; CHECK-LABEL: 'umin_unsigned_check'
 ; CHECK-NEXT:  Classifying expressions for: @umin_unsigned_check
 ; CHECK-NEXT:    %min.n = select i1 %min.cmp, i64 4096, i64 %n
-; CHECK-NEXT:    --> (4096 umin %n) U: full-set S: full-set
+; CHECK-NEXT:    --> (4096 umin %n) U: [0,4097) S: [0,4097)
 ; CHECK-NEXT:    %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
-; CHECK-NEXT:    --> {0,+,1}<%loop> U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
+; CHECK-NEXT:    --> {0,+,1}<%loop> U: [0,4098) S: [0,4098) Exits: (1 + (4096 umin %n))<nuw><nsw> LoopDispositions: { %loop: Computable }
 ; CHECK-NEXT:    %iv.next = add i64 %iv, 1
-; CHECK-NEXT:    --> {1,+,1}<%loop> U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
+; CHECK-NEXT:    --> {1,+,1}<%loop> U: [1,4099) S: [1,4099) Exits: (2 + (4096 umin %n)) LoopDispositions: { %loop: Computable }
 ; CHECK-NEXT:  Determining loop execution counts for: @umin_unsigned_check
-; CHECK-NEXT:  Loop %loop: Unpredictable backedge-taken count.
-; CHECK-NEXT:  Loop %loop: Unpredictable max backedge-taken count.
-; CHECK-NEXT:  Loop %loop: Unpredictable predicated backedge-taken count.
+; CHECK-NEXT:  Loop %loop: backedge-taken count is (1 + (4096 umin %n))<nuw><nsw>
+; CHECK-NEXT:  Loop %loop: max backedge-taken count is 4097
+; CHECK-NEXT:  Loop %loop: Predicated backedge-taken count is (1 + (4096 umin %n))<nuw><nsw>
+; CHECK-NEXT:   Predicates:
+; CHECK:       Loop %loop: Trip multiple is 1
 ;
 entry:
   %min.cmp = icmp ult i64 4096, %n
   %min.n = select i1 %min.cmp, i64 4096, i64 %n
   br label %loop
 loop:
   %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
   %iv.next = add i64 %iv, 1
   %exit = icmp ugt i64 %iv, %min.n
   br i1 %exit, label %loop_exit, label %loop
 
 loop_exit:
   ret void
 }
 
 define void @umin_signed_check(i64 %n) {
 ; CHECK-LABEL: 'umin_signed_check'
 ; CHECK-NEXT:  Classifying expressions for: @umin_signed_check
 ; CHECK-NEXT:    %min.n = select i1 %min.cmp, i64 4096, i64 %n
-; CHECK-NEXT:    --> (4096 umin %n) U: full-set S: full-set
+; CHECK-NEXT:    --> (4096 umin %n) U: [0,4097) S: [0,4097)
 ; CHECK-NEXT:    %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
-; CHECK-NEXT:    --> {0,+,1}<%loop> U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
+; CHECK-NEXT:    --> {0,+,1}<%loop> U: [0,4098) S: [0,4098) Exits: (1 + (4096 umin %n))<nuw><nsw> LoopDispositions: { %loop: Computable }
 ; CHECK-NEXT:    %iv.next = add i64 %iv, 1
-; CHECK-NEXT:    --> {1,+,1}<%loop> U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
+; CHECK-NEXT:    --> {1,+,1}<%loop> U: [1,4099) S: [1,4099) Exits: (2 + (4096 umin %n)) LoopDispositions: { %loop: Computable }
 ; CHECK-NEXT:  Determining loop execution counts for: @umin_signed_check
-; CHECK-NEXT:  Loop %loop: Unpredictable backedge-taken count.
-; CHECK-NEXT:  Loop %loop: Unpredictable max backedge-taken count.
-; CHECK-NEXT:  Loop %loop: Unpredictable predicated backedge-taken count.
+; CHECK-NEXT:  Loop %loop: backedge-taken count is (1 + (4096 umin %n))<nuw><nsw>
+; CHECK-NEXT:  Loop %loop: max backedge-taken count is 4097
+; CHECK-NEXT:  Loop %loop: Predicated backedge-taken count is (1 + (4096 umin %n))<nuw><nsw>
+; CHECK-NEXT:   Predicates:
+; CHECK:       Loop %loop: Trip multiple is 1
 ;
 entry:
   %min.cmp = icmp ult i64 4096, %n
   %min.n = select i1 %min.cmp, i64 4096, i64 %n
   br label %loop
 loop:
   %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
   %iv.next = add i64 %iv, 1
   %exit = icmp sgt i64 %iv, %min.n
   br i1 %exit, label %loop_exit, label %loop
 
 loop_exit:
   ret void
 }
 
 define void @smin_signed_check(i64 %n) {
 ; CHECK-LABEL: 'smin_signed_check'
 ; CHECK-NEXT:  Classifying expressions for: @smin_signed_check
 ; CHECK-NEXT:    %min.n = select i1 %min.cmp, i64 4096, i64 %n
-; CHECK-NEXT:    --> (4096 smin %n) U: full-set S: full-set
+; CHECK-NEXT:    --> (4096 smin %n) U: [-9223372036854775808,4097) S: [-9223372036854775808,4097)
 ; CHECK-NEXT:    %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
-; CHECK-NEXT:    --> {0,+,1}<%loop> U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
+; CHECK-NEXT:    --> {0,+,1}<%loop> U: [0,4098) S: [0,4098) Exits: (0 smax (1 + (4096 smin %n))<nsw>) LoopDispositions: { %loop: Computable }
 ; CHECK-NEXT:    %iv.next = add i64 %iv, 1
-; CHECK-NEXT:    --> {1,+,1}<%loop> U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
+; CHECK-NEXT:    --> {1,+,1}<%loop> U: [1,4099) S: [1,4099) Exits: (1 + (0 smax (1 + (4096 smin %n))<nsw>))<nuw><nsw> LoopDispositions: { %loop: Computable }
 ; CHECK-NEXT:  Determining loop execution counts for: @smin_signed_check
-; CHECK-NEXT:  Loop %loop: Unpredictable backedge-taken count.
-; CHECK-NEXT:  Loop %loop: Unpredictable max backedge-taken count.
-; CHECK-NEXT:  Loop %loop: Unpredictable predicated backedge-taken count.
+; CHECK-NEXT:  Loop %loop: backedge-taken count is (0 smax (1 + (4096 smin %n))<nsw>)
+; CHECK-NEXT:  Loop %loop: max backedge-taken count is 4097
+; CHECK-NEXT:  Loop %loop: Predicated backedge-taken count is (0 smax (1 + (4096 smin %n))<nsw>)
+; CHECK-NEXT:   Predicates:
+; CHECK:       Loop %loop: Trip multiple is 1
 ;
 entry:
   %min.cmp = icmp slt i64 4096, %n
   %min.n = select i1 %min.cmp, i64 4096, i64 %n
   br label %loop
 loop:
   %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
   %iv.next = add i64 %iv, 1
   %exit = icmp sgt i64 %iv, %min.n
   br i1 %exit, label %loop_exit, label %loop
 
 loop_exit:
   ret void
 }
 
 define void @smin_unsigned_check(i64 %n) {
 ; CHECK-LABEL: 'smin_unsigned_check'
 ; CHECK-NEXT:  Classifying expressions for: @smin_unsigned_check
 ; CHECK-NEXT:    %min.n = select i1 %min.cmp, i64 4096, i64 %n
-; CHECK-NEXT:    --> (4096 smin %n) U: full-set S: full-set
+; CHECK-NEXT:    --> (4096 smin %n) U: [-9223372036854775808,4097) S: [-9223372036854775808,4097)
 ; CHECK-NEXT:    %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
 ; CHECK-NEXT:    --> {0,+,1}<%loop> U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
 ; CHECK-NEXT:    %iv.next = add i64 %iv, 1
 ; CHECK-NEXT:    --> {1,+,1}<%loop> U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
 ; CHECK-NEXT:  Determining loop execution counts for: @smin_unsigned_check
 ; CHECK-NEXT:  Loop %loop: Unpredictable backedge-taken count.
 ; CHECK-NEXT:  Loop %loop: Unpredictable max backedge-taken count.
 ; CHECK-NEXT:  Loop %loop: Unpredictable predicated backedge-taken count.

test/Transforms/IRCE/rc-negative-bound.ll

 define void @test_03(i32 *%arr, i32 %n, i32 %bound) {
 ; CHECK-LABEL: @test_03(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[FIRST_ITR_CHECK:%.*]] = icmp sgt i32 [[N:%.*]], 0
 ; CHECK-NEXT:    br i1 [[FIRST_ITR_CHECK]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]]
 ; CHECK:       loop.preheader:
 ; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[BOUND:%.*]], -2147483647
 ; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt i32 [[TMP0]], 0
-; CHECK-NEXT:    [[SMIN:%.*]] = select i1 [[TMP1]], i32 [[TMP0]], i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = sub i32 [[BOUND]], [[SMIN]]
+; CHECK-NEXT:    [[SMAX:%.*]] = select i1 [[TMP1]], i32 [[TMP0]], i32 0
+; CHECK-NEXT:    [[TMP2:%.*]] = sub i32 [[BOUND]], [[SMAX]]
 ; CHECK-NEXT:    [[TMP3:%.*]] = icmp slt i32 [[BOUND]], 0
-; CHECK-NEXT:    [[SMAX:%.*]] = select i1 [[TMP3]], i32 [[BOUND]], i32 0
-; CHECK-NEXT:    [[TMP4:%.*]] = icmp sgt i32 [[SMAX]], -1
-; CHECK-NEXT:    [[SMIN1:%.*]] = select i1 [[TMP4]], i32 [[SMAX]], i32 -1
-; CHECK-NEXT:    [[TMP5:%.*]] = add i32 [[SMIN1]], 1
+; CHECK-NEXT:    [[SMIN:%.*]] = select i1 [[TMP3]], i32 [[BOUND]], i32 0
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp sgt i32 [[SMIN]], -1
+; CHECK-NEXT:    [[SMAX1:%.*]] = select i1 [[TMP4]], i32 [[SMIN]], i32 -1
+; CHECK-NEXT:    [[TMP5:%.*]] = add nsw i32 [[SMAX1]], 1
 ; CHECK-NEXT:    [[TMP6:%.*]] = mul i32 [[TMP2]], [[TMP5]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = icmp slt i32 [[N]], [[TMP6]]
-; CHECK-NEXT:    [[SMAX2:%.*]] = select i1 [[TMP7]], i32 [[N]], i32 [[TMP6]]
-; CHECK-NEXT:    [[TMP8:%.*]] = icmp sgt i32 [[SMAX2]], 0
-; CHECK-NEXT:    [[EXIT_MAINLOOP_AT:%.*]] = select i1 [[TMP8]], i32 [[SMAX2]], i32 0
+; CHECK-NEXT:    [[SMIN2:%.*]] = select i1 [[TMP7]], i32 [[N]], i32 [[TMP6]]
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp sgt i32 [[SMIN2]], 0
+; CHECK-NEXT:    [[EXIT_MAINLOOP_AT:%.*]] = select i1 [[TMP8]], i32 [[SMIN2]], i32 0
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp slt i32 0, [[EXIT_MAINLOOP_AT]]
 ; CHECK-NEXT:    br i1 [[TMP9]], label [[LOOP_PREHEADER4:%.*]], label [[MAIN_PSEUDO_EXIT:%.*]]
 ; CHECK:       loop.preheader4:
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
 ; CHECK-NEXT:    [[IDX:%.*]] = phi i32 [ [[IDX_NEXT:%.*]], [[IN_BOUNDS:%.*]] ], [ 0, [[LOOP_PREHEADER4]] ]
 ; CHECK-NEXT:    [[IDX_NEXT]] = add i32 [[IDX]], 1
 ; CHECK-NEXT:    [[ABC:%.*]] = icmp slt i32 [[IDX]], [[BOUND]]

 ; expand runtime checks against bound being positive or negative.
 define void @test_04(i32 *%arr, i32 %n, i32 %bound) {
 ; CHECK-LABEL: @test_04(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[FIRST_ITR_CHECK:%.*]] = icmp sgt i32 [[N:%.*]], 0
 ; CHECK-NEXT:    br i1 [[FIRST_ITR_CHECK]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]]
 ; CHECK:       loop.preheader:
 ; CHECK-NEXT:    [[TMP0:%.*]] = icmp slt i32 [[BOUND:%.*]], 0
-; CHECK-NEXT:    [[SMAX:%.*]] = select i1 [[TMP0]], i32 [[BOUND]], i32 0
-; CHECK-NEXT:    [[TMP1:%.*]] = sub i32 [[BOUND]], [[SMAX]]
-; CHECK-NEXT:    [[TMP2:%.*]] = icmp sgt i32 [[SMAX]], -1
-; CHECK-NEXT:    [[SMIN:%.*]] = select i1 [[TMP2]], i32 [[SMAX]], i32 -1
-; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[SMIN]], 1
+; CHECK-NEXT:    [[SMIN:%.*]] = select i1 [[TMP0]], i32 [[BOUND]], i32 0
+; CHECK-NEXT:    [[TMP1:%.*]] = sub i32 [[BOUND]], [[SMIN]]
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp sgt i32 [[SMIN]], -1
+; CHECK-NEXT:    [[SMAX:%.*]] = select i1 [[TMP2]], i32 [[SMIN]], i32 -1
+; CHECK-NEXT:    [[TMP3:%.*]] = add nsw i32 [[SMAX]], 1
 ; CHECK-NEXT:    [[TMP4:%.*]] = mul i32 [[TMP1]], [[TMP3]]
 ; CHECK-NEXT:    [[TMP5:%.*]] = icmp ult i32 [[N]], [[TMP4]]
 ; CHECK-NEXT:    [[EXIT_MAINLOOP_AT:%.*]] = select i1 [[TMP5]], i32 [[N]], i32 [[TMP4]]
 ; CHECK-NEXT:    [[TMP6:%.*]] = icmp ult i32 0, [[EXIT_MAINLOOP_AT]]
 ; CHECK-NEXT:    br i1 [[TMP6]], label [[LOOP_PREHEADER1:%.*]], label [[MAIN_PSEUDO_EXIT:%.*]]
 ; CHECK:       loop.preheader1:
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:

 define void @test_07(i32 *%arr, i32 %n, i32 %bound) {
 ; CHECK-LABEL: @test_07(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[FIRST_ITR_CHECK:%.*]] = icmp sgt i32 [[N:%.*]], 0
 ; CHECK-NEXT:    br i1 [[FIRST_ITR_CHECK]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]]
 ; CHECK:       loop.preheader:
 ; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[BOUND:%.*]], -2147483647
 ; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt i32 [[TMP0]], 0
-; CHECK-NEXT:    [[SMIN:%.*]] = select i1 [[TMP1]], i32 [[TMP0]], i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = sub i32 [[BOUND]], [[SMIN]]
+; CHECK-NEXT:    [[SMAX:%.*]] = select i1 [[TMP1]], i32 [[TMP0]], i32 0
+; CHECK-NEXT:    [[TMP2:%.*]] = sub i32 [[BOUND]], [[SMAX]]
 ; CHECK-NEXT:    [[TMP3:%.*]] = icmp slt i32 [[BOUND]], 0
-; CHECK-NEXT:    [[SMAX:%.*]] = select i1 [[TMP3]], i32 [[BOUND]], i32 0
-; CHECK-NEXT:    [[TMP4:%.*]] = icmp sgt i32 [[SMAX]], -1
-; CHECK-NEXT:    [[SMIN1:%.*]] = select i1 [[TMP4]], i32 [[SMAX]], i32 -1
-; CHECK-NEXT:    [[TMP5:%.*]] = add i32 [[SMIN1]], 1
+; CHECK-NEXT:    [[SMIN:%.*]] = select i1 [[TMP3]], i32 [[BOUND]], i32 0
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp sgt i32 [[SMIN]], -1
+; CHECK-NEXT:    [[SMAX1:%.*]] = select i1 [[TMP4]], i32 [[SMIN]], i32 -1
+; CHECK-NEXT:    [[TMP5:%.*]] = add nsw i32 [[SMAX1]], 1
 ; CHECK-NEXT:    [[TMP6:%.*]] = mul i32 [[TMP2]], [[TMP5]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = icmp slt i32 [[N]], [[TMP6]]
-; CHECK-NEXT:    [[SMAX2:%.*]] = select i1 [[TMP7]], i32 [[N]], i32 [[TMP6]]
-; CHECK-NEXT:    [[TMP8:%.*]] = icmp sgt i32 [[SMAX2]], 0
-; CHECK-NEXT:    [[EXIT_MAINLOOP_AT:%.*]] = select i1 [[TMP8]], i32 [[SMAX2]], i32 0
+; CHECK-NEXT:    [[SMIN2:%.*]] = select i1 [[TMP7]], i32 [[N]], i32 [[TMP6]]
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp sgt i32 [[SMIN2]], 0
+; CHECK-NEXT:    [[EXIT_MAINLOOP_AT:%.*]] = select i1 [[TMP8]], i32 [[SMIN2]], i32 0
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp slt i32 0, [[EXIT_MAINLOOP_AT]]
 ; CHECK-NEXT:    br i1 [[TMP9]], label [[LOOP_PREHEADER4:%.*]], label [[MAIN_PSEUDO_EXIT:%.*]]
 ; CHECK:       loop.preheader4:
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
 ; CHECK-NEXT:    [[IDX:%.*]] = phi i32 [ [[IDX_NEXT:%.*]], [[IN_BOUNDS:%.*]] ], [ 0, [[LOOP_PREHEADER4]] ]
 ; CHECK-NEXT:    [[IDX_NEXT]] = add i32 [[IDX]], 1
 ; CHECK-NEXT:    [[ABC:%.*]] = icmp ult i32 [[IDX]], [[BOUND]]

 ; computeSafeIterationSpace).
 define void @test_08(i32 *%arr, i32 %n, i32 %bound) {
 ; CHECK-LABEL: @test_08(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[FIRST_ITR_CHECK:%.*]] = icmp sgt i32 [[N:%.*]], 0
 ; CHECK-NEXT:    br i1 [[FIRST_ITR_CHECK]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]]
 ; CHECK:       loop.preheader:
 ; CHECK-NEXT:    [[TMP0:%.*]] = icmp slt i32 [[BOUND:%.*]], 0
-; CHECK-NEXT:    [[SMAX:%.*]] = select i1 [[TMP0]], i32 [[BOUND]], i32 0
-; CHECK-NEXT:    [[TMP1:%.*]] = sub i32 [[BOUND]], [[SMAX]]
-; CHECK-NEXT:    [[TMP2:%.*]] = icmp sgt i32 [[SMAX]], -1
-; CHECK-NEXT:    [[SMIN:%.*]] = select i1 [[TMP2]], i32 [[SMAX]], i32 -1
-; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[SMIN]], 1
+; CHECK-NEXT:    [[SMIN:%.*]] = select i1 [[TMP0]], i32 [[BOUND]], i32 0
+; CHECK-NEXT:    [[TMP1:%.*]] = sub i32 [[BOUND]], [[SMIN]]
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp sgt i32 [[SMIN]], -1
+; CHECK-NEXT:    [[SMAX:%.*]] = select i1 [[TMP2]], i32 [[SMIN]], i32 -1
+; CHECK-NEXT:    [[TMP3:%.*]] = add nsw i32 [[SMAX]], 1
 ; CHECK-NEXT:    [[TMP4:%.*]] = mul i32 [[TMP1]], [[TMP3]]
 ; CHECK-NEXT:    [[TMP5:%.*]] = icmp ult i32 [[N]], [[TMP4]]
 ; CHECK-NEXT:    [[EXIT_MAINLOOP_AT:%.*]] = select i1 [[TMP5]], i32 [[N]], i32 [[TMP4]]
 ; CHECK-NEXT:    [[TMP6:%.*]] = icmp ult i32 0, [[EXIT_MAINLOOP_AT]]
 ; CHECK-NEXT:    br i1 [[TMP6]], label [[LOOP_PREHEADER1:%.*]], label [[MAIN_PSEUDO_EXIT:%.*]]
 ; CHECK:       loop.preheader1:
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:

test/Transforms/LoopVectorize/X86/load-deref-pred.ll

 define i32 @test_max_trip_count(i64 %len, i1* %test_base, i64 %n) {
 ; CHECK-LABEL: @test_max_trip_count(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[ALLOCA:%.*]] = alloca [4096 x i32]
 ; CHECK-NEXT:    [[BASE:%.*]] = bitcast [4096 x i32]* [[ALLOCA]] to i32*
 ; CHECK-NEXT:    call void @init(i32* [[BASE]])
 ; CHECK-NEXT:    [[MIN_CMP:%.*]] = icmp ult i64 4096, [[N:%.*]]
 ; CHECK-NEXT:    [[MIN_N:%.*]] = select i1 [[MIN_CMP]], i64 4096, i64 [[N]]
+; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[MIN_N]], 2
+; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 16
+; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK:       vector.ph:
+; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 16
+; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]
+; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
+; CHECK:       vector.body:
+; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP81:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI4:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP82:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI5:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP83:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI6:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP84:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> undef, i64 [[INDEX]], i32 0
+; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> undef, <4 x i32> zeroinitializer
+; CHECK-NEXT:    [[INDUCTION:%.*]] = add <4 x i64> [[BROADCAST_SPLAT]], <i64 0, i64 1, i64 2, i64 3>
+; CHECK-NEXT:    [[INDUCTION1:%.*]] = add <4 x i64> [[BROADCAST_SPLAT]], <i64 4, i64 5, i64 6, i64 7>
+; CHECK-NEXT:    [[INDUCTION2:%.*]] = add <4 x i64> [[BROADCAST_SPLAT]], <i64 8, i64 9, i64 10, i64 11>
+; CHECK-NEXT:    [[INDUCTION3:%.*]] = add <4 x i64> [[BROADCAST_SPLAT]], <i64 12, i64 13, i64 14, i64 15>
+; CHECK-NEXT:    [[TMP1:%.*]] = add i64 [[INDEX]], 0
+; CHECK-NEXT:    [[TMP2:%.*]] = add i64 [[INDEX]], 1
+; CHECK-NEXT:    [[TMP3:%.*]] = add i64 [[INDEX]], 2
+; CHECK-NEXT:    [[TMP4:%.*]] = add i64 [[INDEX]], 3
+; CHECK-NEXT:    [[TMP5:%.*]] = add i64 [[INDEX]], 4
+; CHECK-NEXT:    [[TMP6:%.*]] = add i64 [[INDEX]], 5
+; CHECK-NEXT:    [[TMP7:%.*]] = add i64 [[INDEX]], 6
+; CHECK-NEXT:    [[TMP8:%.*]] = add i64 [[INDEX]], 7
+; CHECK-NEXT:    [[TMP9:%.*]] = add i64 [[INDEX]], 8
+; CHECK-NEXT:    [[TMP10:%.*]] = add i64 [[INDEX]], 9
+; CHECK-NEXT:    [[TMP11:%.*]] = add i64 [[INDEX]], 10
+; CHECK-NEXT:    [[TMP12:%.*]] = add i64 [[INDEX]], 11
+; CHECK-NEXT:    [[TMP13:%.*]] = add i64 [[INDEX]], 12
+; CHECK-NEXT:    [[TMP14:%.*]] = add i64 [[INDEX]], 13
+; CHECK-NEXT:    [[TMP15:%.*]] = add i64 [[INDEX]], 14
+; CHECK-NEXT:    [[TMP16:%.*]] = add i64 [[INDEX]], 15
+; CHECK-NEXT:    [[TMP17:%.*]] = getelementptr inbounds i1, i1* [[TEST_BASE:%.*]], i64 [[TMP1]]
+; CHECK-NEXT:    [[TMP18:%.*]] = getelementptr inbounds i1, i1* [[TEST_BASE]], i64 [[TMP2]]
+; CHECK-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i1, i1* [[TEST_BASE]], i64 [[TMP3]]
+; CHECK-NEXT:    [[TMP20:%.*]] = getelementptr inbounds i1, i1* [[TEST_BASE]], i64 [[TMP4]]
+; CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds i1, i1* [[TEST_BASE]], i64 [[TMP5]]
+; CHECK-NEXT:    [[TMP22:%.*]] = getelementptr inbounds i1, i1* [[TEST_BASE]], i64 [[TMP6]]
+; CHECK-NEXT:    [[TMP23:%.*]] = getelementptr inbounds i1, i1* [[TEST_BASE]], i64 [[TMP7]]
+; CHECK-NEXT:    [[TMP24:%.*]] = getelementptr inbounds i1, i1* [[TEST_BASE]], i64 [[TMP8]]
+; CHECK-NEXT:    [[TMP25:%.*]] = getelementptr inbounds i1, i1* [[TEST_BASE]], i64 [[TMP9]]
+; CHECK-NEXT:    [[TMP26:%.*]] = getelementptr inbounds i1, i1* [[TEST_BASE]], i64 [[TMP10]]
+; CHECK-NEXT:    [[TMP27:%.*]] = getelementptr inbounds i1, i1* [[TEST_BASE]], i64 [[TMP11]]
+; CHECK-NEXT:    [[TMP28:%.*]] = getelementptr inbounds i1, i1* [[TEST_BASE]], i64 [[TMP12]]
+; CHECK-NEXT:    [[TMP29:%.*]] = getelementptr inbounds i1, i1* [[TEST_BASE]], i64 [[TMP13]]
+; CHECK-NEXT:    [[TMP30:%.*]] = getelementptr inbounds i1, i1* [[TEST_BASE]], i64 [[TMP14]]
+; CHECK-NEXT:    [[TMP31:%.*]] = getelementptr inbounds i1, i1* [[TEST_BASE]], i64 [[TMP15]]
+; CHECK-NEXT:    [[TMP32:%.*]] = getelementptr inbounds i1, i1* [[TEST_BASE]], i64 [[TMP16]]
+; CHECK-NEXT:    [[TMP33:%.*]] = load i1, i1* [[TMP17]]
+; CHECK-NEXT:    [[TMP34:%.*]] = load i1, i1* [[TMP18]]
+; CHECK-NEXT:    [[TMP35:%.*]] = load i1, i1* [[TMP19]]
+; CHECK-NEXT:    [[TMP36:%.*]] = load i1, i1* [[TMP20]]
+; CHECK-NEXT:    [[TMP37:%.*]] = insertelement <4 x i1> undef, i1 [[TMP33]], i32 0
+; CHECK-NEXT:    [[TMP38:%.*]] = insertelement <4 x i1> [[TMP37]], i1 [[TMP34]], i32 1
+; CHECK-NEXT:    [[TMP39:%.*]] = insertelement <4 x i1> [[TMP38]], i1 [[TMP35]], i32 2
+; CHECK-NEXT:    [[TMP40:%.*]] = insertelement <4 x i1> [[TMP39]], i1 [[TMP36]], i32 3
+; CHECK-NEXT:    [[TMP41:%.*]] = load i1, i1* [[TMP21]]
+; CHECK-NEXT:    [[TMP42:%.*]] = load i1, i1* [[TMP22]]
+; CHECK-NEXT:    [[TMP43:%.*]] = load i1, i1* [[TMP23]]
+; CHECK-NEXT:    [[TMP44:%.*]] = load i1, i1* [[TMP24]]
+; CHECK-NEXT:    [[TMP45:%.*]] = insertelement <4 x i1> undef, i1 [[TMP41]], i32 0
+; CHECK-NEXT:    [[TMP46:%.*]] = insertelement <4 x i1> [[TMP45]], i1 [[TMP42]], i32 1
+; CHECK-NEXT:    [[TMP47:%.*]] = insertelement <4 x i1> [[TMP46]], i1 [[TMP43]], i32 2
+; CHECK-NEXT:    [[TMP48:%.*]] = insertelement <4 x i1> [[TMP47]], i1 [[TMP44]], i32 3
+; CHECK-NEXT:    [[TMP49:%.*]] = load i1, i1* [[TMP25]]
+; CHECK-NEXT:    [[TMP50:%.*]] = load i1, i1* [[TMP26]]
+; CHECK-NEXT:    [[TMP51:%.*]] = load i1, i1* [[TMP27]]
+; CHECK-NEXT:    [[TMP52:%.*]] = load i1, i1* [[TMP28]]
+; CHECK-NEXT:    [[TMP53:%.*]] = insertelement <4 x i1> undef, i1 [[TMP49]], i32 0
+; CHECK-NEXT:    [[TMP54:%.*]] = insertelement <4 x i1> [[TMP53]], i1 [[TMP50]], i32 1
+; CHECK-NEXT:    [[TMP55:%.*]] = insertelement <4 x i1> [[TMP54]], i1 [[TMP51]], i32 2
+; CHECK-NEXT:    [[TMP56:%.*]] = insertelement <4 x i1> [[TMP55]], i1 [[TMP52]], i32 3
+; CHECK-NEXT:    [[TMP57:%.*]] = load i1, i1* [[TMP29]]
+; CHECK-NEXT:    [[TMP58:%.*]] = load i1, i1* [[TMP30]]
+; CHECK-NEXT:    [[TMP59:%.*]] = load i1, i1* [[TMP31]]
+; CHECK-NEXT:    [[TMP60:%.*]] = load i1, i1* [[TMP32]]
+; CHECK-NEXT:    [[TMP61:%.*]] = insertelement <4 x i1> undef, i1 [[TMP57]], i32 0
+; CHECK-NEXT:    [[TMP62:%.*]] = insertelement <4 x i1> [[TMP61]], i1 [[TMP58]], i32 1
+; CHECK-NEXT:    [[TMP63:%.*]] = insertelement <4 x i1> [[TMP62]], i1 [[TMP59]], i32 2
+; CHECK-NEXT:    [[TMP64:%.*]] = insertelement <4 x i1> [[TMP63]], i1 [[TMP60]], i32 3
+; CHECK-NEXT:    [[TMP65:%.*]] = getelementptr inbounds i32, i32* [[BASE]], i64 [[TMP1]]
+; CHECK-NEXT:    [[TMP66:%.*]] = getelementptr inbounds i32, i32* [[BASE]], i64 [[TMP5]]
+; CHECK-NEXT:    [[TMP67:%.*]] = getelementptr inbounds i32, i32* [[BASE]], i64 [[TMP9]]
+; CHECK-NEXT:    [[TMP68:%.*]] = getelementptr inbounds i32, i32* [[BASE]], i64 [[TMP13]]
+; CHECK-NEXT:    [[TMP69:%.*]] = getelementptr inbounds i32, i32* [[TMP65]], i32 0
+; CHECK-NEXT:    [[TMP70:%.*]] = bitcast i32* [[TMP69]] to <4 x i32>*
+; CHECK-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <4 x i32> @llvm.masked.load.v4i32.p0v4i32(<4 x i32>* [[TMP70]], i32 4, <4 x i1> [[TMP40]], <4 x i32> undef)
+; CHECK-NEXT:    [[TMP71:%.*]] = getelementptr inbounds i32, i32* [[TMP65]], i32 4
+; CHECK-NEXT:    [[TMP72:%.*]] = bitcast i32* [[TMP71]] to <4 x i32>*
+; CHECK-NEXT:    [[WIDE_MASKED_LOAD7:%.*]] = call <4 x i32> @llvm.masked.load.v4i32.p0v4i32(<4 x i32>* [[TMP72]], i32 4, <4 x i1> [[TMP48]], <4 x i32> undef)
+; CHECK-NEXT:    [[TMP73:%.*]] = getelementptr inbounds i32, i32* [[TMP65]], i32 8
+; CHECK-NEXT:    [[TMP74:%.*]] = bitcast i32* [[TMP73]] to <4 x i32>*
+; CHECK-NEXT:    [[WIDE_MASKED_LOAD8:%.*]] = call <4 x i32> @llvm.masked.load.v4i32.p0v4i32(<4 x i32>* [[TMP74]], i32 4, <4 x i1> [[TMP56]], <4 x i32> undef)
+; CHECK-NEXT:    [[TMP75:%.*]] = getelementptr inbounds i32, i32* [[TMP65]], i32 12
+; CHECK-NEXT:    [[TMP76:%.*]] = bitcast i32* [[TMP75]] to <4 x i32>*
+; CHECK-NEXT:    [[WIDE_MASKED_LOAD9:%.*]] = call <4 x i32> @llvm.masked.load.v4i32.p0v4i32(<4 x i32>* [[TMP76]], i32 4, <4 x i1> [[TMP64]], <4 x i32> undef)
+; CHECK-NEXT:    [[TMP77:%.*]] = xor <4 x i1> [[TMP40]], <i1 true, i1 true, i1 true, i1 true>
+; CHECK-NEXT:    [[TMP78:%.*]] = xor <4 x i1> [[TMP48]], <i1 true, i1 true, i1 true, i1 true>
+; CHECK-NEXT:    [[TMP79:%.*]] = xor <4 x i1> [[TMP56]], <i1 true, i1 true, i1 true, i1 true>
+; CHECK-NEXT:    [[TMP80:%.*]] = xor <4 x i1> [[TMP64]], <i1 true, i1 true, i1 true, i1 true>
+; CHECK-NEXT:    [[PREDPHI:%.*]] = select <4 x i1> [[TMP40]], <4 x i32> [[WIDE_MASKED_LOAD]], <4 x i32> zeroinitializer
+; CHECK-NEXT:    [[PREDPHI10:%.*]] = select <4 x i1> [[TMP48]], <4 x i32> [[WIDE_MASKED_LOAD7]], <4 x i32> zeroinitializer
+; CHECK-NEXT:    [[PREDPHI11:%.*]] = select <4 x i1> [[TMP56]], <4 x i32> [[WIDE_MASKED_LOAD8]], <4 x i32> zeroinitializer
+; CHECK-NEXT:    [[PREDPHI12:%.*]] = select <4 x i1> [[TMP64]], <4 x i32> [[WIDE_MASKED_LOAD9]], <4 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP81]] = add <4 x i32> [[VEC_PHI]], [[PREDPHI]]
+; CHECK-NEXT:    [[TMP82]] = add <4 x i32> [[VEC_PHI4]], [[PREDPHI10]]
+; CHECK-NEXT:    [[TMP83]] = add <4 x i32> [[VEC_PHI5]], [[PREDPHI11]]
+; CHECK-NEXT:    [[TMP84]] = add <4 x i32> [[VEC_PHI6]], [[PREDPHI12]]
+; CHECK-NEXT:    [[INDEX_NEXT]] = add i64 [[INDEX]], 16
+; CHECK-NEXT:    [[TMP85:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[TMP85]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !12
+; CHECK:       middle.block:
+; CHECK-NEXT:    [[BIN_RDX:%.*]] = add <4 x i32> [[TMP82]], [[TMP81]]
+; CHECK-NEXT:    [[BIN_RDX13:%.*]] = add <4 x i32> [[TMP83]], [[BIN_RDX]]
+; CHECK-NEXT:    [[BIN_RDX14:%.*]] = add <4 x i32> [[TMP84]], [[BIN_RDX13]]
+; CHECK-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[BIN_RDX14]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
+; CHECK-NEXT:    [[BIN_RDX15:%.*]] = add <4 x i32> [[BIN_RDX14]], [[RDX_SHUF]]
+; CHECK-NEXT:    [[RDX_SHUF16:%.*]] = shufflevector <4 x i32> [[BIN_RDX15]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[BIN_RDX17:%.*]] = add <4 x i32> [[BIN_RDX15]], [[RDX_SHUF16]]
+; CHECK-NEXT:    [[TMP86:%.*]] = extractelement <4 x i32> [[BIN_RDX17]], i32 0
+; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[CMP_N]], label [[LOOP_EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK:       scalar.ph:
+; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[TMP86]], [[MIDDLE_BLOCK]] ]
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LATCH:%.*]] ]
-; CHECK-NEXT:    [[ACCUM:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[ACCUM_NEXT:%.*]], [[LATCH]] ]
+; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LATCH:%.*]] ]
+; CHECK-NEXT:    [[ACCUM:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[ACCUM_NEXT:%.*]], [[LATCH]] ]
 ; CHECK-NEXT:    [[IV_NEXT]] = add i64 [[IV]], 1
-; CHECK-NEXT:    [[TEST_ADDR:%.*]] = getelementptr inbounds i1, i1* [[TEST_BASE:%.*]], i64 [[IV]]
+; CHECK-NEXT:    [[TEST_ADDR:%.*]] = getelementptr inbounds i1, i1* [[TEST_BASE]], i64 [[IV]]
 ; CHECK-NEXT:    [[EARLYCND:%.*]] = load i1, i1* [[TEST_ADDR]]
 ; CHECK-NEXT:    br i1 [[EARLYCND]], label [[PRED:%.*]], label [[LATCH]]
 ; CHECK:       pred:
 ; CHECK-NEXT:    [[ADDR:%.*]] = getelementptr inbounds i32, i32* [[BASE]], i64 [[IV]]
 ; CHECK-NEXT:    [[VAL:%.*]] = load i32, i32* [[ADDR]]
 ; CHECK-NEXT:    br label [[LATCH]]
 ; CHECK:       latch:
 ; CHECK-NEXT:    [[VAL_PHI:%.*]] = phi i32 [ 0, [[LOOP]] ], [ [[VAL]], [[PRED]] ]
 ; CHECK-NEXT:    [[ACCUM_NEXT]] = add i32 [[ACCUM]], [[VAL_PHI]]
 ; CHECK-NEXT:    [[EXIT:%.*]] = icmp ugt i64 [[IV]], [[MIN_N]]
-; CHECK-NEXT:    br i1 [[EXIT]], label [[LOOP_EXIT:%.*]], label [[LOOP]]
+; CHECK-NEXT:    br i1 [[EXIT]], label [[LOOP_EXIT]], label [[LOOP]], !llvm.loop !13
 ; CHECK:       loop_exit:
-; CHECK-NEXT:    [[ACCUM_NEXT_LCSSA:%.*]] = phi i32 [ [[ACCUM_NEXT]], [[LATCH]] ]
+; CHECK-NEXT:    [[ACCUM_NEXT_LCSSA:%.*]] = phi i32 [ [[ACCUM_NEXT]], [[LATCH]] ], [ [[TMP86]], [[MIDDLE_BLOCK]] ]
 ; CHECK-NEXT:    ret i32 [[ACCUM_NEXT_LCSSA]]
 ;
 entry:
   %alloca = alloca [4096 x i32]
   %base = bitcast [4096 x i32]* %alloca to i32*
   call void @init(i32* %base)
   %min.cmp = icmp ult i64 4096, %n
   %min.n = select i1 %min.cmp, i64 4096, i64 %n

 ; CHECK-NEXT:    [[PREDPHI11:%.*]] = select <4 x i1> [[TMP55]], <4 x i32> [[WIDE_MASKED_LOAD8]], <4 x i32> zeroinitializer
 ; CHECK-NEXT:    [[PREDPHI12:%.*]] = select <4 x i1> [[TMP63]], <4 x i32> [[WIDE_MASKED_LOAD9]], <4 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP80]] = add <4 x i32> [[VEC_PHI]], [[PREDPHI]]
 ; CHECK-NEXT:    [[TMP81]] = add <4 x i32> [[VEC_PHI4]], [[PREDPHI10]]
 ; CHECK-NEXT:    [[TMP82]] = add <4 x i32> [[VEC_PHI5]], [[PREDPHI11]]
 ; CHECK-NEXT:    [[TMP83]] = add <4 x i32> [[VEC_PHI6]], [[PREDPHI12]]
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add i64 [[INDEX]], 16
 ; CHECK-NEXT:    [[TMP84:%.*]] = icmp eq i64 [[INDEX_NEXT]], 3072
-; CHECK-NEXT:    br i1 [[TMP84]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !12
+; CHECK-NEXT:    br i1 [[TMP84]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !14
 ; CHECK:       middle.block:
 ; CHECK-NEXT:    [[BIN_RDX:%.*]] = add <4 x i32> [[TMP81]], [[TMP80]]
 ; CHECK-NEXT:    [[BIN_RDX13:%.*]] = add <4 x i32> [[TMP82]], [[BIN_RDX]]
 ; CHECK-NEXT:    [[BIN_RDX14:%.*]] = add <4 x i32> [[TMP83]], [[BIN_RDX13]]
 ; CHECK-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[BIN_RDX14]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[BIN_RDX15:%.*]] = add <4 x i32> [[BIN_RDX14]], [[RDX_SHUF]]
 ; CHECK-NEXT:    [[RDX_SHUF16:%.*]] = shufflevector <4 x i32> [[BIN_RDX15]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[BIN_RDX17:%.*]] = add <4 x i32> [[BIN_RDX15]], [[RDX_SHUF16]]

 ; CHECK:       pred:
 ; CHECK-NEXT:    [[ADDR:%.*]] = getelementptr inbounds i32, i32* [[BASE]], i64 [[IV]]
 ; CHECK-NEXT:    [[VAL:%.*]] = load i32, i32* [[ADDR]]
 ; CHECK-NEXT:    br label [[LATCH]]
 ; CHECK:       latch:
 ; CHECK-NEXT:    [[VAL_PHI:%.*]] = phi i32 [ 0, [[LOOP]] ], [ [[VAL]], [[PRED]] ]
 ; CHECK-NEXT:    [[ACCUM_NEXT]] = add i32 [[ACCUM]], [[VAL_PHI]]
 ; CHECK-NEXT:    [[EXIT:%.*]] = icmp ugt i64 [[IV]], 4094
-; CHECK-NEXT:    br i1 [[EXIT]], label [[LOOP_EXIT]], label [[LOOP]], !llvm.loop !13
+; CHECK-NEXT:    br i1 [[EXIT]], label [[LOOP_EXIT]], label [[LOOP]], !llvm.loop !15
 ; CHECK:       loop_exit:
 ; CHECK-NEXT:    [[ACCUM_NEXT_LCSSA:%.*]] = phi i32 [ [[ACCUM_NEXT]], [[LATCH]] ], [ [[TMP85]], [[MIDDLE_BLOCK]] ]
 ; CHECK-NEXT:    ret i32 [[ACCUM_NEXT_LCSSA]]
 ;
 entry:
   %alloca = alloca [4096 x i32]
   %base = bitcast [4096 x i32]* %alloca to i32*
   call void @init(i32* %base)

 ; CHECK-NEXT:    [[PREDPHI38:%.*]] = select <4 x i1> [[TMP55]], <4 x i32> [[TMP123]], <4 x i32> zeroinitializer
 ; CHECK-NEXT:    [[PREDPHI39:%.*]] = select <4 x i1> [[TMP63]], <4 x i32> [[TMP143]], <4 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP148]] = add <4 x i32> [[VEC_PHI]], [[PREDPHI]]
 ; CHECK-NEXT:    [[TMP149]] = add <4 x i32> [[VEC_PHI4]], [[PREDPHI37]]
 ; CHECK-NEXT:    [[TMP150]] = add <4 x i32> [[VEC_PHI5]], [[PREDPHI38]]
 ; CHECK-NEXT:    [[TMP151]] = add <4 x i32> [[VEC_PHI6]], [[PREDPHI39]]
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add i64 [[INDEX]], 16
 ; CHECK-NEXT:    [[TMP152:%.*]] = icmp eq i64 [[INDEX_NEXT]], 2048
-; CHECK-NEXT:    br i1 [[TMP152]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !14
+; CHECK-NEXT:    br i1 [[TMP152]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !16
 ; CHECK:       middle.block:
 ; CHECK-NEXT:    [[BIN_RDX:%.*]] = add <4 x i32> [[TMP149]], [[TMP148]]
 ; CHECK-NEXT:    [[BIN_RDX40:%.*]] = add <4 x i32> [[TMP150]], [[BIN_RDX]]
 ; CHECK-NEXT:    [[BIN_RDX41:%.*]] = add <4 x i32> [[TMP151]], [[BIN_RDX40]]
 ; CHECK-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[BIN_RDX41]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[BIN_RDX42:%.*]] = add <4 x i32> [[BIN_RDX41]], [[RDX_SHUF]]
 ; CHECK-NEXT:    [[RDX_SHUF43:%.*]] = shufflevector <4 x i32> [[BIN_RDX42]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[BIN_RDX44:%.*]] = add <4 x i32> [[BIN_RDX42]], [[RDX_SHUF43]]

 ; CHECK:       pred:
 ; CHECK-NEXT:    [[ADDR:%.*]] = getelementptr inbounds i32, i32* [[BASE]], i64 [[IV]]
 ; CHECK-NEXT:    [[VAL:%.*]] = load i32, i32* [[ADDR]]
 ; CHECK-NEXT:    br label [[LATCH]]
 ; CHECK:       latch:
 ; CHECK-NEXT:    [[VAL_PHI:%.*]] = phi i32 [ 0, [[LOOP]] ], [ [[VAL]], [[PRED]] ]
 ; CHECK-NEXT:    [[ACCUM_NEXT]] = add i32 [[ACCUM]], [[VAL_PHI]]
 ; CHECK-NEXT:    [[EXIT:%.*]] = icmp ugt i64 [[IV]], 4093
-; CHECK-NEXT:    br i1 [[EXIT]], label [[LOOP_EXIT]], label [[LOOP]], !llvm.loop !15
+; CHECK-NEXT:    br i1 [[EXIT]], label [[LOOP_EXIT]], label [[LOOP]], !llvm.loop !17
 ; CHECK:       loop_exit:
 ; CHECK-NEXT:    [[ACCUM_NEXT_LCSSA:%.*]] = phi i32 [ [[ACCUM_NEXT]], [[LATCH]] ], [ [[TMP153]], [[MIDDLE_BLOCK]] ]
 ; CHECK-NEXT:    ret i32 [[ACCUM_NEXT_LCSSA]]
 ;
 entry:
   %alloca = alloca [4096 x i32]
   %base = bitcast [4096 x i32]* %alloca to i32*
   call void @init(i32* %base)

 ; CHECK-NEXT:    [[PREDPHI11:%.*]] = select <4 x i1> [[TMP55]], <4 x i32> [[WIDE_MASKED_LOAD8]], <4 x i32> zeroinitializer
 ; CHECK-NEXT:    [[PREDPHI12:%.*]] = select <4 x i1> [[TMP63]], <4 x i32> [[WIDE_MASKED_LOAD9]], <4 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP80]] = add <4 x i32> [[VEC_PHI]], [[PREDPHI]]
 ; CHECK-NEXT:    [[TMP81]] = add <4 x i32> [[VEC_PHI4]], [[PREDPHI10]]
 ; CHECK-NEXT:    [[TMP82]] = add <4 x i32> [[VEC_PHI5]], [[PREDPHI11]]
 ; CHECK-NEXT:    [[TMP83]] = add <4 x i32> [[VEC_PHI6]], [[PREDPHI12]]
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add i64 [[INDEX]], 16
 ; CHECK-NEXT:    [[TMP84:%.*]] = icmp eq i64 [[INDEX_NEXT]], 4096
-; CHECK-NEXT:    br i1 [[TMP84]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !16
+; CHECK-NEXT:    br i1 [[TMP84]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !18
 ; CHECK:       middle.block:
 ; CHECK-NEXT:    [[BIN_RDX:%.*]] = add <4 x i32> [[TMP81]], [[TMP80]]
 ; CHECK-NEXT:    [[BIN_RDX13:%.*]] = add <4 x i32> [[TMP82]], [[BIN_RDX]]
 ; CHECK-NEXT:    [[BIN_RDX14:%.*]] = add <4 x i32> [[TMP83]], [[BIN_RDX13]]
 ; CHECK-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[BIN_RDX14]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[BIN_RDX15:%.*]] = add <4 x i32> [[BIN_RDX14]], [[RDX_SHUF]]
 ; CHECK-NEXT:    [[RDX_SHUF16:%.*]] = shufflevector <4 x i32> [[BIN_RDX15]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[BIN_RDX17:%.*]] = add <4 x i32> [[BIN_RDX15]], [[RDX_SHUF16]]

 ; CHECK:       pred:
 ; CHECK-NEXT:    [[ADDR:%.*]] = getelementptr inbounds i32, i32* [[BASE]], i64 [[IV]]
 ; CHECK-NEXT:    [[VAL:%.*]] = load i32, i32* [[ADDR]]
 ; CHECK-NEXT:    br label [[LATCH]]
 ; CHECK:       latch:
 ; CHECK-NEXT:    [[VAL_PHI:%.*]] = phi i32 [ 0, [[LOOP]] ], [ [[VAL]], [[PRED]] ]
 ; CHECK-NEXT:    [[ACCUM_NEXT]] = add i32 [[ACCUM]], [[VAL_PHI]]
 ; CHECK-NEXT:    [[EXIT:%.*]] = icmp ugt i64 [[IV]], 4094
-; CHECK-NEXT:    br i1 [[EXIT]], label [[LOOP_EXIT]], label [[LOOP]], !llvm.loop !17
+; CHECK-NEXT:    br i1 [[EXIT]], label [[LOOP_EXIT]], label [[LOOP]], !llvm.loop !19
 ; CHECK:       loop_exit:
 ; CHECK-NEXT:    [[ACCUM_NEXT_LCSSA:%.*]] = phi i32 [ [[ACCUM_NEXT]], [[LATCH]] ], [ [[TMP85]], [[MIDDLE_BLOCK]] ]
 ; CHECK-NEXT:    ret i32 [[ACCUM_NEXT_LCSSA]]
 ;
 entry:
   %alloca = alloca [1024 x i32]
   %base = bitcast [1024 x i32]* %alloca to i32*
   call void @init(i32* %base)

 ; CHECK-NEXT:    [[PREDPHI11:%.*]] = select <4 x i1> [[TMP55]], <4 x i32> [[WIDE_MASKED_LOAD8]], <4 x i32> zeroinitializer
 ; CHECK-NEXT:    [[PREDPHI12:%.*]] = select <4 x i1> [[TMP63]], <4 x i32> [[WIDE_MASKED_LOAD9]], <4 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP80]] = add <4 x i32> [[VEC_PHI]], [[PREDPHI]]
 ; CHECK-NEXT:    [[TMP81]] = add <4 x i32> [[VEC_PHI4]], [[PREDPHI10]]
 ; CHECK-NEXT:    [[TMP82]] = add <4 x i32> [[VEC_PHI5]], [[PREDPHI11]]
 ; CHECK-NEXT:    [[TMP83]] = add <4 x i32> [[VEC_PHI6]], [[PREDPHI12]]
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add i64 [[INDEX]], 16
 ; CHECK-NEXT:    [[TMP84:%.*]] = icmp eq i64 [[INDEX_NEXT]], 4096
-; CHECK-NEXT:    br i1 [[TMP84]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !18
+; CHECK-NEXT:    br i1 [[TMP84]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !20
 ; CHECK:       middle.block:
 ; CHECK-NEXT:    [[BIN_RDX:%.*]] = add <4 x i32> [[TMP81]], [[TMP80]]
 ; CHECK-NEXT:    [[BIN_RDX13:%.*]] = add <4 x i32> [[TMP82]], [[BIN_RDX]]
 ; CHECK-NEXT:    [[BIN_RDX14:%.*]] = add <4 x i32> [[TMP83]], [[BIN_RDX13]]
 ; CHECK-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[BIN_RDX14]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[BIN_RDX15:%.*]] = add <4 x i32> [[BIN_RDX14]], [[RDX_SHUF]]
 ; CHECK-NEXT:    [[RDX_SHUF16:%.*]] = shufflevector <4 x i32> [[BIN_RDX15]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[BIN_RDX17:%.*]] = add <4 x i32> [[BIN_RDX15]], [[RDX_SHUF16]]

 ; CHECK:       pred:
 ; CHECK-NEXT:    [[ADDR:%.*]] = getelementptr inbounds i32, i32* [[BASE]], i64 [[IV]]
 ; CHECK-NEXT:    [[VAL:%.*]] = load i32, i32* [[ADDR]]
 ; CHECK-NEXT:    br label [[LATCH]]
 ; CHECK:       latch:
 ; CHECK-NEXT:    [[VAL_PHI:%.*]] = phi i32 [ 0, [[LOOP]] ], [ [[VAL]], [[PRED]] ]
 ; CHECK-NEXT:    [[ACCUM_NEXT]] = add i32 [[ACCUM]], [[VAL_PHI]]
 ; CHECK-NEXT:    [[EXIT:%.*]] = icmp ugt i64 [[IV]], 4094
-; CHECK-NEXT:    br i1 [[EXIT]], label [[LOOP_EXIT]], label [[LOOP]], !llvm.loop !19
+; CHECK-NEXT:    br i1 [[EXIT]], label [[LOOP_EXIT]], label [[LOOP]], !llvm.loop !21
 ; CHECK:       loop_exit:
 ; CHECK-NEXT:    [[ACCUM_NEXT_LCSSA:%.*]] = phi i32 [ [[ACCUM_NEXT]], [[LATCH]] ], [ [[TMP85]], [[MIDDLE_BLOCK]] ]
 ; CHECK-NEXT:    ret i32 [[ACCUM_NEXT_LCSSA]]
 ;
 entry:
   %alloca = alloca [4095 x i32]
   %base = bitcast [4095 x i32]* %alloca to i32*
   call void @init(i32* %base)

 ; CHECK-NEXT:    [[PREDPHI11:%.*]] = select <4 x i1> [[TMP55]], <4 x i32> [[WIDE_MASKED_LOAD8]], <4 x i32> zeroinitializer
 ; CHECK-NEXT:    [[PREDPHI12:%.*]] = select <4 x i1> [[TMP63]], <4 x i32> [[WIDE_MASKED_LOAD9]], <4 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP80]] = add <4 x i32> [[VEC_PHI]], [[PREDPHI]]
 ; CHECK-NEXT:    [[TMP81]] = add <4 x i32> [[VEC_PHI4]], [[PREDPHI10]]
 ; CHECK-NEXT:    [[TMP82]] = add <4 x i32> [[VEC_PHI5]], [[PREDPHI11]]
 ; CHECK-NEXT:    [[TMP83]] = add <4 x i32> [[VEC_PHI6]], [[PREDPHI12]]
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add i64 [[INDEX]], 16
 ; CHECK-NEXT:    [[TMP84:%.*]] = icmp eq i64 [[INDEX_NEXT]], 4096
-; CHECK-NEXT:    br i1 [[TMP84]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !20
+; CHECK-NEXT:    br i1 [[TMP84]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !22
 ; CHECK:       middle.block:
 ; CHECK-NEXT:    [[BIN_RDX:%.*]] = add <4 x i32> [[TMP81]], [[TMP80]]
 ; CHECK-NEXT:    [[BIN_RDX13:%.*]] = add <4 x i32> [[TMP82]], [[BIN_RDX]]
 ; CHECK-NEXT:    [[BIN_RDX14:%.*]] = add <4 x i32> [[TMP83]], [[BIN_RDX13]]
 ; CHECK-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[BIN_RDX14]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[BIN_RDX15:%.*]] = add <4 x i32> [[BIN_RDX14]], [[RDX_SHUF]]
 ; CHECK-NEXT:    [[RDX_SHUF16:%.*]] = shufflevector <4 x i32> [[BIN_RDX15]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[BIN_RDX17:%.*]] = add <4 x i32> [[BIN_RDX15]], [[RDX_SHUF16]]

 ; CHECK:       pred:
 ; CHECK-NEXT:    [[ADDR:%.*]] = getelementptr inbounds i32, i32* [[BASE]], i64 [[IV]]
 ; CHECK-NEXT:    [[VAL:%.*]] = load i32, i32* [[ADDR]]
 ; CHECK-NEXT:    br label [[LATCH]]
 ; CHECK:       latch:
 ; CHECK-NEXT:    [[VAL_PHI:%.*]] = phi i32 [ 0, [[LOOP]] ], [ [[VAL]], [[PRED]] ]
 ; CHECK-NEXT:    [[ACCUM_NEXT]] = add i32 [[ACCUM]], [[VAL_PHI]]
 ; CHECK-NEXT:    [[EXIT:%.*]] = icmp ugt i64 [[IV]], 4094
-; CHECK-NEXT:    br i1 [[EXIT]], label [[LOOP_EXIT]], label [[LOOP]], !llvm.loop !21
+; CHECK-NEXT:    br i1 [[EXIT]], label [[LOOP_EXIT]], label [[LOOP]], !llvm.loop !23
 ; CHECK:       loop_exit:
 ; CHECK-NEXT:    [[ACCUM_NEXT_LCSSA:%.*]] = phi i32 [ [[ACCUM_NEXT]], [[LATCH]] ], [ [[TMP85]], [[MIDDLE_BLOCK]] ]
 ; CHECK-NEXT:    ret i32 [[ACCUM_NEXT_LCSSA]]
 ;
 entry:
   %alloca = alloca [16383 x i8]
   %base = bitcast [16383 x i8]* %alloca to i32*
   call void @init(i32* %base)