1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -loop-vectorize -force-vector-interleave=1 -force-vector-width=2 -S | FileCheck %s
3; RUN: opt < %s -loop-vectorize -force-vector-interleave=1 -force-vector-width=2 -instcombine -S | FileCheck %s --check-prefix=IND
4; RUN: opt < %s -loop-vectorize -force-vector-interleave=2 -force-vector-width=2 -instcombine -S | FileCheck %s --check-prefix=UNROLL
5; RUN: opt < %s -loop-vectorize -force-vector-interleave=2 -force-vector-width=2 -S | FileCheck %s --check-prefix=UNROLL-NO-IC
6; RUN: opt < %s -loop-vectorize -force-vector-interleave=2 -force-vector-width=4 -enable-interleaved-mem-accesses -instcombine -S | FileCheck %s --check-prefix=INTERLEAVE
7
8target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
9
10; Make sure that we can handle multiple integer induction variables.
11;
12define void @multi_int_induction(i32* %A, i32 %N) {
13; CHECK-LABEL: @multi_int_induction(
14; CHECK-NEXT:  for.body.lr.ph:
15; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[N:%.*]], -1
16; CHECK-NEXT:    [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
17; CHECK-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
18; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 2
19; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
20; CHECK:       vector.ph:
21; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 2
22; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]]
23; CHECK-NEXT:    [[CAST_VTC:%.*]] = trunc i64 [[N_VEC]] to i32
24; CHECK-NEXT:    [[IND_END:%.*]] = add i32 190, [[CAST_VTC]]
25; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
26; CHECK:       vector.body:
27; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
28; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 190, i32 191>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
29; CHECK-NEXT:    [[TMP3:%.*]] = add i64 [[INDEX]], 0
30; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP3]]
31; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[TMP4]], i32 0
32; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32* [[TMP5]] to <2 x i32>*
33; CHECK-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP6]], align 4
34; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
35; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
36; CHECK-NEXT:    [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
37; CHECK-NEXT:    br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
38; CHECK:       middle.block:
39; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
40; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
41; CHECK:       scalar.ph:
42; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_LR_PH:%.*]] ]
43; CHECK-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 190, [[FOR_BODY_LR_PH]] ]
44; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
45; CHECK:       for.body:
46; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
47; CHECK-NEXT:    [[COUNT_09:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ]
48; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV]]
49; CHECK-NEXT:    store i32 [[COUNT_09]], i32* [[ARRAYIDX2]], align 4
50; CHECK-NEXT:    [[INC]] = add nsw i32 [[COUNT_09]], 1
51; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1
52; CHECK-NEXT:    [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
53; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[LFTR_WIDEIV]], [[N]]
54; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP2:![0-9]+]]
55; CHECK:       for.end:
56; CHECK-NEXT:    ret void
57;
58; IND-LABEL: @multi_int_induction(
59; IND-NEXT:  for.body.lr.ph:
60; IND-NEXT:    [[TMP0:%.*]] = add i32 [[N:%.*]], -1
61; IND-NEXT:    [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
62; IND-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
63; IND-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp eq i32 [[TMP0]], 0
64; IND-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
65; IND:       vector.ph:
66; IND-NEXT:    [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934590
67; IND-NEXT:    [[CAST_VTC:%.*]] = trunc i64 [[N_VEC]] to i32
68; IND-NEXT:    [[IND_END:%.*]] = add i32 [[CAST_VTC]], 190
69; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
70; IND:       vector.body:
71; IND-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
72; IND-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 190, i32 191>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
73; IND-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[INDEX]]
74; IND-NEXT:    [[TMP4:%.*]] = bitcast i32* [[TMP3]] to <2 x i32>*
75; IND-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP4]], align 4
76; IND-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
77; IND-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
78; IND-NEXT:    [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
79; IND-NEXT:    br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
80; IND:       middle.block:
81; IND-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
82; IND-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
83; IND:       scalar.ph:
84; IND-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_LR_PH:%.*]] ]
85; IND-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 190, [[FOR_BODY_LR_PH]] ]
86; IND-NEXT:    br label [[FOR_BODY:%.*]]
87; IND:       for.body:
88; IND-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
89; IND-NEXT:    [[COUNT_09:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ]
90; IND-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV]]
91; IND-NEXT:    store i32 [[COUNT_09]], i32* [[ARRAYIDX2]], align 4
92; IND-NEXT:    [[INC]] = add nsw i32 [[COUNT_09]], 1
93; IND-NEXT:    [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1
94; IND-NEXT:    [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
95; IND-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], [[N]]
96; IND-NEXT:    br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]]
97; IND:       for.end:
98; IND-NEXT:    ret void
99;
100; UNROLL-LABEL: @multi_int_induction(
101; UNROLL-NEXT:  for.body.lr.ph:
102; UNROLL-NEXT:    [[TMP0:%.*]] = add i32 [[N:%.*]], -1
103; UNROLL-NEXT:    [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
104; UNROLL-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
105; UNROLL-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 3
106; UNROLL-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
107; UNROLL:       vector.ph:
108; UNROLL-NEXT:    [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934588
109; UNROLL-NEXT:    [[CAST_VTC:%.*]] = trunc i64 [[N_VEC]] to i32
110; UNROLL-NEXT:    [[IND_END:%.*]] = add i32 [[CAST_VTC]], 190
111; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
112; UNROLL:       vector.body:
113; UNROLL-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
114; UNROLL-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 190, i32 191>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
115; UNROLL-NEXT:    [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
116; UNROLL-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[INDEX]]
117; UNROLL-NEXT:    [[TMP4:%.*]] = bitcast i32* [[TMP3]] to <2 x i32>*
118; UNROLL-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP4]], align 4
119; UNROLL-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[TMP3]], i64 2
120; UNROLL-NEXT:    [[TMP6:%.*]] = bitcast i32* [[TMP5]] to <2 x i32>*
121; UNROLL-NEXT:    store <2 x i32> [[STEP_ADD]], <2 x i32>* [[TMP6]], align 4
122; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
123; UNROLL-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 4, i32 4>
124; UNROLL-NEXT:    [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
125; UNROLL-NEXT:    br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
126; UNROLL:       middle.block:
127; UNROLL-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
128; UNROLL-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
129; UNROLL:       scalar.ph:
130; UNROLL-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_LR_PH:%.*]] ]
131; UNROLL-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 190, [[FOR_BODY_LR_PH]] ]
132; UNROLL-NEXT:    br label [[FOR_BODY:%.*]]
133; UNROLL:       for.body:
134; UNROLL-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
135; UNROLL-NEXT:    [[COUNT_09:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ]
136; UNROLL-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV]]
137; UNROLL-NEXT:    store i32 [[COUNT_09]], i32* [[ARRAYIDX2]], align 4
138; UNROLL-NEXT:    [[INC]] = add nsw i32 [[COUNT_09]], 1
139; UNROLL-NEXT:    [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1
140; UNROLL-NEXT:    [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
141; UNROLL-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], [[N]]
142; UNROLL-NEXT:    br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]]
143; UNROLL:       for.end:
144; UNROLL-NEXT:    ret void
145;
146; UNROLL-NO-IC-LABEL: @multi_int_induction(
147; UNROLL-NO-IC-NEXT:  for.body.lr.ph:
148; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = add i32 [[N:%.*]], -1
149; UNROLL-NO-IC-NEXT:    [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
150; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
151; UNROLL-NO-IC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 4
152; UNROLL-NO-IC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
153; UNROLL-NO-IC:       vector.ph:
154; UNROLL-NO-IC-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 4
155; UNROLL-NO-IC-NEXT:    [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]]
156; UNROLL-NO-IC-NEXT:    [[CAST_VTC:%.*]] = trunc i64 [[N_VEC]] to i32
157; UNROLL-NO-IC-NEXT:    [[IND_END:%.*]] = add i32 190, [[CAST_VTC]]
158; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
159; UNROLL-NO-IC:       vector.body:
160; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
161; UNROLL-NO-IC-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 190, i32 191>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
162; UNROLL-NO-IC-NEXT:    [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
163; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = add i64 [[INDEX]], 0
164; UNROLL-NO-IC-NEXT:    [[TMP4:%.*]] = add i64 [[INDEX]], 2
165; UNROLL-NO-IC-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP3]]
166; UNROLL-NO-IC-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP4]]
167; UNROLL-NO-IC-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i32, i32* [[TMP5]], i32 0
168; UNROLL-NO-IC-NEXT:    [[TMP8:%.*]] = bitcast i32* [[TMP7]] to <2 x i32>*
169; UNROLL-NO-IC-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP8]], align 4
170; UNROLL-NO-IC-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i32, i32* [[TMP5]], i32 2
171; UNROLL-NO-IC-NEXT:    [[TMP10:%.*]] = bitcast i32* [[TMP9]] to <2 x i32>*
172; UNROLL-NO-IC-NEXT:    store <2 x i32> [[STEP_ADD]], <2 x i32>* [[TMP10]], align 4
173; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
174; UNROLL-NO-IC-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[STEP_ADD]], <i32 2, i32 2>
175; UNROLL-NO-IC-NEXT:    [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
176; UNROLL-NO-IC-NEXT:    br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
177; UNROLL-NO-IC:       middle.block:
178; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
179; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
180; UNROLL-NO-IC:       scalar.ph:
181; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_LR_PH:%.*]] ]
182; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 190, [[FOR_BODY_LR_PH]] ]
183; UNROLL-NO-IC-NEXT:    br label [[FOR_BODY:%.*]]
184; UNROLL-NO-IC:       for.body:
185; UNROLL-NO-IC-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
186; UNROLL-NO-IC-NEXT:    [[COUNT_09:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ]
187; UNROLL-NO-IC-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV]]
188; UNROLL-NO-IC-NEXT:    store i32 [[COUNT_09]], i32* [[ARRAYIDX2]], align 4
189; UNROLL-NO-IC-NEXT:    [[INC]] = add nsw i32 [[COUNT_09]], 1
190; UNROLL-NO-IC-NEXT:    [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1
191; UNROLL-NO-IC-NEXT:    [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
192; UNROLL-NO-IC-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[LFTR_WIDEIV]], [[N]]
193; UNROLL-NO-IC-NEXT:    br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP2:![0-9]+]]
194; UNROLL-NO-IC:       for.end:
195; UNROLL-NO-IC-NEXT:    ret void
196;
197; INTERLEAVE-LABEL: @multi_int_induction(
198; INTERLEAVE-NEXT:  for.body.lr.ph:
199; INTERLEAVE-NEXT:    [[TMP0:%.*]] = add i32 [[N:%.*]], -1
200; INTERLEAVE-NEXT:    [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
201; INTERLEAVE-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
202; INTERLEAVE-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 7
203; INTERLEAVE-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
204; INTERLEAVE:       vector.ph:
205; INTERLEAVE-NEXT:    [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934584
206; INTERLEAVE-NEXT:    [[CAST_VTC:%.*]] = trunc i64 [[N_VEC]] to i32
207; INTERLEAVE-NEXT:    [[IND_END:%.*]] = add i32 [[CAST_VTC]], 190
208; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
209; INTERLEAVE:       vector.body:
210; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
211; INTERLEAVE-NEXT:    [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 190, i32 191, i32 192, i32 193>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
212; INTERLEAVE-NEXT:    [[STEP_ADD:%.*]] = add <4 x i32> [[VEC_IND]], <i32 4, i32 4, i32 4, i32 4>
213; INTERLEAVE-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[INDEX]]
214; INTERLEAVE-NEXT:    [[TMP4:%.*]] = bitcast i32* [[TMP3]] to <4 x i32>*
215; INTERLEAVE-NEXT:    store <4 x i32> [[VEC_IND]], <4 x i32>* [[TMP4]], align 4
216; INTERLEAVE-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[TMP3]], i64 4
217; INTERLEAVE-NEXT:    [[TMP6:%.*]] = bitcast i32* [[TMP5]] to <4 x i32>*
218; INTERLEAVE-NEXT:    store <4 x i32> [[STEP_ADD]], <4 x i32>* [[TMP6]], align 4
219; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
220; INTERLEAVE-NEXT:    [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 8, i32 8, i32 8, i32 8>
221; INTERLEAVE-NEXT:    [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
222; INTERLEAVE-NEXT:    br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
223; INTERLEAVE:       middle.block:
224; INTERLEAVE-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
225; INTERLEAVE-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
226; INTERLEAVE:       scalar.ph:
227; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_LR_PH:%.*]] ]
228; INTERLEAVE-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 190, [[FOR_BODY_LR_PH]] ]
229; INTERLEAVE-NEXT:    br label [[FOR_BODY:%.*]]
230; INTERLEAVE:       for.body:
231; INTERLEAVE-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
232; INTERLEAVE-NEXT:    [[COUNT_09:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ]
233; INTERLEAVE-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV]]
234; INTERLEAVE-NEXT:    store i32 [[COUNT_09]], i32* [[ARRAYIDX2]], align 4
235; INTERLEAVE-NEXT:    [[INC]] = add nsw i32 [[COUNT_09]], 1
236; INTERLEAVE-NEXT:    [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1
237; INTERLEAVE-NEXT:    [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
238; INTERLEAVE-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], [[N]]
239; INTERLEAVE-NEXT:    br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]]
240; INTERLEAVE:       for.end:
241; INTERLEAVE-NEXT:    ret void
242;
243for.body.lr.ph:
244  br label %for.body
245
246for.body:
247  %indvars.iv = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next, %for.body ]
248  %count.09 = phi i32 [ 190, %for.body.lr.ph ], [ %inc, %for.body ]
249  %arrayidx2 = getelementptr inbounds i32, i32* %A, i64 %indvars.iv
250  store i32 %count.09, i32* %arrayidx2, align 4
251  %inc = add nsw i32 %count.09, 1
252  %indvars.iv.next = add i64 %indvars.iv, 1
253  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
254  %exitcond = icmp ne i32 %lftr.wideiv, %N
255  br i1 %exitcond, label %for.body, label %for.end
256
257for.end:
258  ret void
259}
260
261; Make sure we remove unneeded vectorization of induction variables.
262; In order for instcombine to cleanup the vectorized induction variables that we
263; create in the loop vectorizer we need to perform some form of redundancy
264; elimination to get rid of multiple uses.
265
266
267;   Vectorized induction variable.
268
269define void @scalar_use(float* %a, float %b, i64 %offset, i64 %offset2, i64 %n) {
270; CHECK-LABEL: @scalar_use(
271; CHECK-NEXT:  entry:
272; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 2
273; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
274; CHECK:       vector.memcheck:
275; CHECK-NEXT:    [[SCEVGEP:%.*]] = getelementptr float, float* [[A:%.*]], i64 [[OFFSET:%.*]]
276; CHECK-NEXT:    [[SCEVGEP1:%.*]] = bitcast float* [[SCEVGEP]] to i8*
277; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[N]], [[OFFSET]]
278; CHECK-NEXT:    [[SCEVGEP2:%.*]] = getelementptr float, float* [[A]], i64 [[TMP0]]
279; CHECK-NEXT:    [[SCEVGEP23:%.*]] = bitcast float* [[SCEVGEP2]] to i8*
280; CHECK-NEXT:    [[SCEVGEP4:%.*]] = getelementptr float, float* [[A]], i64 [[OFFSET2:%.*]]
281; CHECK-NEXT:    [[SCEVGEP45:%.*]] = bitcast float* [[SCEVGEP4]] to i8*
282; CHECK-NEXT:    [[TMP1:%.*]] = add i64 [[N]], [[OFFSET2]]
283; CHECK-NEXT:    [[SCEVGEP6:%.*]] = getelementptr float, float* [[A]], i64 [[TMP1]]
284; CHECK-NEXT:    [[SCEVGEP67:%.*]] = bitcast float* [[SCEVGEP6]] to i8*
285; CHECK-NEXT:    [[BOUND0:%.*]] = icmp ult i8* [[SCEVGEP1]], [[SCEVGEP67]]
286; CHECK-NEXT:    [[BOUND1:%.*]] = icmp ult i8* [[SCEVGEP45]], [[SCEVGEP23]]
287; CHECK-NEXT:    [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
288; CHECK-NEXT:    br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
289; CHECK:       vector.ph:
290; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], 2
291; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
292; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x float> poison, float [[B:%.*]], i32 0
293; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x float> [[BROADCAST_SPLATINSERT]], <2 x float> poison, <2 x i32> zeroinitializer
294; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
295; CHECK:       vector.body:
296; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
297; CHECK-NEXT:    [[TMP2:%.*]] = add i64 [[INDEX]], 0
298; CHECK-NEXT:    [[TMP3:%.*]] = add i64 [[TMP2]], [[OFFSET]]
299; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP3]]
300; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds float, float* [[TMP4]], i32 0
301; CHECK-NEXT:    [[TMP6:%.*]] = bitcast float* [[TMP5]] to <2 x float>*
302; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x float>, <2 x float>* [[TMP6]], align 4, !alias.scope !4, !noalias !7
303; CHECK-NEXT:    [[TMP7:%.*]] = add i64 [[TMP2]], [[OFFSET2]]
304; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP7]]
305; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds float, float* [[TMP8]], i32 0
306; CHECK-NEXT:    [[TMP10:%.*]] = bitcast float* [[TMP9]] to <2 x float>*
307; CHECK-NEXT:    [[WIDE_LOAD8:%.*]] = load <2 x float>, <2 x float>* [[TMP10]], align 4, !alias.scope !7
308; CHECK-NEXT:    [[TMP11:%.*]] = fmul fast <2 x float> [[BROADCAST_SPLAT]], [[WIDE_LOAD8]]
309; CHECK-NEXT:    [[TMP12:%.*]] = fadd fast <2 x float> [[WIDE_LOAD]], [[TMP11]]
310; CHECK-NEXT:    [[TMP13:%.*]] = bitcast float* [[TMP5]] to <2 x float>*
311; CHECK-NEXT:    store <2 x float> [[TMP12]], <2 x float>* [[TMP13]], align 4, !alias.scope !4, !noalias !7
312; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
313; CHECK-NEXT:    [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
314; CHECK-NEXT:    br i1 [[TMP14]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
315; CHECK:       middle.block:
316; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
317; CHECK-NEXT:    br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]]
318; CHECK:       scalar.ph:
319; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_MEMCHECK]] ]
320; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
321; CHECK:       for.body:
322; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]
323; CHECK-NEXT:    [[IND_SUM:%.*]] = add i64 [[IV]], [[OFFSET]]
324; CHECK-NEXT:    [[ARR_IDX:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[IND_SUM]]
325; CHECK-NEXT:    [[L1:%.*]] = load float, float* [[ARR_IDX]], align 4
326; CHECK-NEXT:    [[IND_SUM2:%.*]] = add i64 [[IV]], [[OFFSET2]]
327; CHECK-NEXT:    [[ARR_IDX2:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[IND_SUM2]]
328; CHECK-NEXT:    [[L2:%.*]] = load float, float* [[ARR_IDX2]], align 4
329; CHECK-NEXT:    [[M:%.*]] = fmul fast float [[B]], [[L2]]
330; CHECK-NEXT:    [[AD:%.*]] = fadd fast float [[L1]], [[M]]
331; CHECK-NEXT:    store float [[AD]], float* [[ARR_IDX]], align 4
332; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
333; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
334; CHECK-NEXT:    br i1 [[EXITCOND]], label [[LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
335; CHECK:       loopexit:
336; CHECK-NEXT:    ret void
337;
338; IND-LABEL: @scalar_use(
339; IND-NEXT:  entry:
340; IND-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 2
341; IND-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
342; IND:       vector.memcheck:
343; IND-NEXT:    [[SCEVGEP:%.*]] = getelementptr float, float* [[A:%.*]], i64 [[OFFSET:%.*]]
344; IND-NEXT:    [[TMP0:%.*]] = add i64 [[N]], [[OFFSET]]
345; IND-NEXT:    [[SCEVGEP2:%.*]] = getelementptr float, float* [[A]], i64 [[TMP0]]
346; IND-NEXT:    [[SCEVGEP4:%.*]] = getelementptr float, float* [[A]], i64 [[OFFSET2:%.*]]
347; IND-NEXT:    [[TMP1:%.*]] = add i64 [[N]], [[OFFSET2]]
348; IND-NEXT:    [[SCEVGEP6:%.*]] = getelementptr float, float* [[A]], i64 [[TMP1]]
349; IND-NEXT:    [[BOUND0:%.*]] = icmp ult float* [[SCEVGEP]], [[SCEVGEP6]]
350; IND-NEXT:    [[BOUND1:%.*]] = icmp ult float* [[SCEVGEP4]], [[SCEVGEP2]]
351; IND-NEXT:    [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
352; IND-NEXT:    br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
353; IND:       vector.ph:
354; IND-NEXT:    [[N_VEC:%.*]] = and i64 [[N]], -2
355; IND-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x float> poison, float [[B:%.*]], i64 0
356; IND-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x float> [[BROADCAST_SPLATINSERT]], <2 x float> poison, <2 x i32> zeroinitializer
357; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
358; IND:       vector.body:
359; IND-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
360; IND-NEXT:    [[TMP2:%.*]] = add i64 [[INDEX]], [[OFFSET]]
361; IND-NEXT:    [[TMP3:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP2]]
362; IND-NEXT:    [[TMP4:%.*]] = bitcast float* [[TMP3]] to <2 x float>*
363; IND-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x float>, <2 x float>* [[TMP4]], align 4, !alias.scope !4, !noalias !7
364; IND-NEXT:    [[TMP5:%.*]] = add i64 [[INDEX]], [[OFFSET2]]
365; IND-NEXT:    [[TMP6:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP5]]
366; IND-NEXT:    [[TMP7:%.*]] = bitcast float* [[TMP6]] to <2 x float>*
367; IND-NEXT:    [[WIDE_LOAD8:%.*]] = load <2 x float>, <2 x float>* [[TMP7]], align 4, !alias.scope !7
368; IND-NEXT:    [[TMP8:%.*]] = fmul fast <2 x float> [[BROADCAST_SPLAT]], [[WIDE_LOAD8]]
369; IND-NEXT:    [[TMP9:%.*]] = fadd fast <2 x float> [[WIDE_LOAD]], [[TMP8]]
370; IND-NEXT:    [[TMP10:%.*]] = bitcast float* [[TMP3]] to <2 x float>*
371; IND-NEXT:    store <2 x float> [[TMP9]], <2 x float>* [[TMP10]], align 4, !alias.scope !4, !noalias !7
372; IND-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
373; IND-NEXT:    [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
374; IND-NEXT:    br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
375; IND:       middle.block:
376; IND-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[N]]
377; IND-NEXT:    br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]]
378; IND:       scalar.ph:
379; IND-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_MEMCHECK]] ]
380; IND-NEXT:    br label [[FOR_BODY:%.*]]
381; IND:       for.body:
382; IND-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]
383; IND-NEXT:    [[IND_SUM:%.*]] = add i64 [[IV]], [[OFFSET]]
384; IND-NEXT:    [[ARR_IDX:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[IND_SUM]]
385; IND-NEXT:    [[L1:%.*]] = load float, float* [[ARR_IDX]], align 4
386; IND-NEXT:    [[IND_SUM2:%.*]] = add i64 [[IV]], [[OFFSET2]]
387; IND-NEXT:    [[ARR_IDX2:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[IND_SUM2]]
388; IND-NEXT:    [[L2:%.*]] = load float, float* [[ARR_IDX2]], align 4
389; IND-NEXT:    [[M:%.*]] = fmul fast float [[L2]], [[B]]
390; IND-NEXT:    [[AD:%.*]] = fadd fast float [[L1]], [[M]]
391; IND-NEXT:    store float [[AD]], float* [[ARR_IDX]], align 4
392; IND-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
393; IND-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
394; IND-NEXT:    br i1 [[EXITCOND]], label [[LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
395; IND:       loopexit:
396; IND-NEXT:    ret void
397;
398; UNROLL-LABEL: @scalar_use(
399; UNROLL-NEXT:  entry:
400; UNROLL-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4
401; UNROLL-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
402; UNROLL:       vector.memcheck:
403; UNROLL-NEXT:    [[SCEVGEP:%.*]] = getelementptr float, float* [[A:%.*]], i64 [[OFFSET:%.*]]
404; UNROLL-NEXT:    [[TMP0:%.*]] = add i64 [[N]], [[OFFSET]]
405; UNROLL-NEXT:    [[SCEVGEP2:%.*]] = getelementptr float, float* [[A]], i64 [[TMP0]]
406; UNROLL-NEXT:    [[SCEVGEP4:%.*]] = getelementptr float, float* [[A]], i64 [[OFFSET2:%.*]]
407; UNROLL-NEXT:    [[TMP1:%.*]] = add i64 [[N]], [[OFFSET2]]
408; UNROLL-NEXT:    [[SCEVGEP6:%.*]] = getelementptr float, float* [[A]], i64 [[TMP1]]
409; UNROLL-NEXT:    [[BOUND0:%.*]] = icmp ult float* [[SCEVGEP]], [[SCEVGEP6]]
410; UNROLL-NEXT:    [[BOUND1:%.*]] = icmp ult float* [[SCEVGEP4]], [[SCEVGEP2]]
411; UNROLL-NEXT:    [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
412; UNROLL-NEXT:    br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
413; UNROLL:       vector.ph:
414; UNROLL-NEXT:    [[N_VEC:%.*]] = and i64 [[N]], -4
415; UNROLL-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x float> poison, float [[B:%.*]], i64 0
416; UNROLL-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x float> [[BROADCAST_SPLATINSERT]], <2 x float> poison, <2 x i32> zeroinitializer
417; UNROLL-NEXT:    [[BROADCAST_SPLATINSERT11:%.*]] = insertelement <2 x float> poison, float [[B]], i64 0
418; UNROLL-NEXT:    [[BROADCAST_SPLAT12:%.*]] = shufflevector <2 x float> [[BROADCAST_SPLATINSERT11]], <2 x float> poison, <2 x i32> zeroinitializer
419; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
420; UNROLL:       vector.body:
421; UNROLL-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
422; UNROLL-NEXT:    [[TMP2:%.*]] = add i64 [[INDEX]], [[OFFSET]]
423; UNROLL-NEXT:    [[TMP3:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP2]]
424; UNROLL-NEXT:    [[TMP4:%.*]] = bitcast float* [[TMP3]] to <2 x float>*
425; UNROLL-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x float>, <2 x float>* [[TMP4]], align 4, !alias.scope !4, !noalias !7
426; UNROLL-NEXT:    [[TMP5:%.*]] = getelementptr inbounds float, float* [[TMP3]], i64 2
427; UNROLL-NEXT:    [[TMP6:%.*]] = bitcast float* [[TMP5]] to <2 x float>*
428; UNROLL-NEXT:    [[WIDE_LOAD8:%.*]] = load <2 x float>, <2 x float>* [[TMP6]], align 4, !alias.scope !4, !noalias !7
429; UNROLL-NEXT:    [[TMP7:%.*]] = add i64 [[INDEX]], [[OFFSET2]]
430; UNROLL-NEXT:    [[TMP8:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP7]]
431; UNROLL-NEXT:    [[TMP9:%.*]] = bitcast float* [[TMP8]] to <2 x float>*
432; UNROLL-NEXT:    [[WIDE_LOAD9:%.*]] = load <2 x float>, <2 x float>* [[TMP9]], align 4, !alias.scope !7
433; UNROLL-NEXT:    [[TMP10:%.*]] = getelementptr inbounds float, float* [[TMP8]], i64 2
434; UNROLL-NEXT:    [[TMP11:%.*]] = bitcast float* [[TMP10]] to <2 x float>*
435; UNROLL-NEXT:    [[WIDE_LOAD10:%.*]] = load <2 x float>, <2 x float>* [[TMP11]], align 4, !alias.scope !7
436; UNROLL-NEXT:    [[TMP12:%.*]] = fmul fast <2 x float> [[BROADCAST_SPLAT]], [[WIDE_LOAD9]]
437; UNROLL-NEXT:    [[TMP13:%.*]] = fmul fast <2 x float> [[BROADCAST_SPLAT12]], [[WIDE_LOAD10]]
438; UNROLL-NEXT:    [[TMP14:%.*]] = fadd fast <2 x float> [[WIDE_LOAD]], [[TMP12]]
439; UNROLL-NEXT:    [[TMP15:%.*]] = fadd fast <2 x float> [[WIDE_LOAD8]], [[TMP13]]
440; UNROLL-NEXT:    [[TMP16:%.*]] = bitcast float* [[TMP3]] to <2 x float>*
441; UNROLL-NEXT:    store <2 x float> [[TMP14]], <2 x float>* [[TMP16]], align 4, !alias.scope !4, !noalias !7
442; UNROLL-NEXT:    [[TMP17:%.*]] = bitcast float* [[TMP5]] to <2 x float>*
443; UNROLL-NEXT:    store <2 x float> [[TMP15]], <2 x float>* [[TMP17]], align 4, !alias.scope !4, !noalias !7
444; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
445; UNROLL-NEXT:    [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
446; UNROLL-NEXT:    br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
447; UNROLL:       middle.block:
448; UNROLL-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[N]]
449; UNROLL-NEXT:    br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]]
450; UNROLL:       scalar.ph:
451; UNROLL-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_MEMCHECK]] ]
452; UNROLL-NEXT:    br label [[FOR_BODY:%.*]]
453; UNROLL:       for.body:
454; UNROLL-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]
455; UNROLL-NEXT:    [[IND_SUM:%.*]] = add i64 [[IV]], [[OFFSET]]
456; UNROLL-NEXT:    [[ARR_IDX:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[IND_SUM]]
457; UNROLL-NEXT:    [[L1:%.*]] = load float, float* [[ARR_IDX]], align 4
458; UNROLL-NEXT:    [[IND_SUM2:%.*]] = add i64 [[IV]], [[OFFSET2]]
459; UNROLL-NEXT:    [[ARR_IDX2:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[IND_SUM2]]
460; UNROLL-NEXT:    [[L2:%.*]] = load float, float* [[ARR_IDX2]], align 4
461; UNROLL-NEXT:    [[M:%.*]] = fmul fast float [[L2]], [[B]]
462; UNROLL-NEXT:    [[AD:%.*]] = fadd fast float [[L1]], [[M]]
463; UNROLL-NEXT:    store float [[AD]], float* [[ARR_IDX]], align 4
464; UNROLL-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
465; UNROLL-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
466; UNROLL-NEXT:    br i1 [[EXITCOND]], label [[LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
467; UNROLL:       loopexit:
468; UNROLL-NEXT:    ret void
469;
470; UNROLL-NO-IC-LABEL: @scalar_use(
471; UNROLL-NO-IC-NEXT:  entry:
472; UNROLL-NO-IC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4
473; UNROLL-NO-IC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
474; UNROLL-NO-IC:       vector.memcheck:
475; UNROLL-NO-IC-NEXT:    [[SCEVGEP:%.*]] = getelementptr float, float* [[A:%.*]], i64 [[OFFSET:%.*]]
476; UNROLL-NO-IC-NEXT:    [[SCEVGEP1:%.*]] = bitcast float* [[SCEVGEP]] to i8*
477; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = add i64 [[N]], [[OFFSET]]
478; UNROLL-NO-IC-NEXT:    [[SCEVGEP2:%.*]] = getelementptr float, float* [[A]], i64 [[TMP0]]
479; UNROLL-NO-IC-NEXT:    [[SCEVGEP23:%.*]] = bitcast float* [[SCEVGEP2]] to i8*
480; UNROLL-NO-IC-NEXT:    [[SCEVGEP4:%.*]] = getelementptr float, float* [[A]], i64 [[OFFSET2:%.*]]
481; UNROLL-NO-IC-NEXT:    [[SCEVGEP45:%.*]] = bitcast float* [[SCEVGEP4]] to i8*
482; UNROLL-NO-IC-NEXT:    [[TMP1:%.*]] = add i64 [[N]], [[OFFSET2]]
483; UNROLL-NO-IC-NEXT:    [[SCEVGEP6:%.*]] = getelementptr float, float* [[A]], i64 [[TMP1]]
484; UNROLL-NO-IC-NEXT:    [[SCEVGEP67:%.*]] = bitcast float* [[SCEVGEP6]] to i8*
485; UNROLL-NO-IC-NEXT:    [[BOUND0:%.*]] = icmp ult i8* [[SCEVGEP1]], [[SCEVGEP67]]
486; UNROLL-NO-IC-NEXT:    [[BOUND1:%.*]] = icmp ult i8* [[SCEVGEP45]], [[SCEVGEP23]]
487; UNROLL-NO-IC-NEXT:    [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
488; UNROLL-NO-IC-NEXT:    br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
489; UNROLL-NO-IC:       vector.ph:
490; UNROLL-NO-IC-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], 4
491; UNROLL-NO-IC-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
492; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x float> poison, float [[B:%.*]], i32 0
493; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x float> [[BROADCAST_SPLATINSERT]], <2 x float> poison, <2 x i32> zeroinitializer
494; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLATINSERT11:%.*]] = insertelement <2 x float> poison, float [[B]], i32 0
495; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLAT12:%.*]] = shufflevector <2 x float> [[BROADCAST_SPLATINSERT11]], <2 x float> poison, <2 x i32> zeroinitializer
496; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
497; UNROLL-NO-IC:       vector.body:
498; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
499; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = add i64 [[INDEX]], 0
500; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = add i64 [[INDEX]], 2
501; UNROLL-NO-IC-NEXT:    [[TMP4:%.*]] = add i64 [[TMP2]], [[OFFSET]]
502; UNROLL-NO-IC-NEXT:    [[TMP5:%.*]] = add i64 [[TMP3]], [[OFFSET]]
503; UNROLL-NO-IC-NEXT:    [[TMP6:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP4]]
504; UNROLL-NO-IC-NEXT:    [[TMP7:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP5]]
505; UNROLL-NO-IC-NEXT:    [[TMP8:%.*]] = getelementptr inbounds float, float* [[TMP6]], i32 0
506; UNROLL-NO-IC-NEXT:    [[TMP9:%.*]] = bitcast float* [[TMP8]] to <2 x float>*
507; UNROLL-NO-IC-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x float>, <2 x float>* [[TMP9]], align 4, !alias.scope !4, !noalias !7
508; UNROLL-NO-IC-NEXT:    [[TMP10:%.*]] = getelementptr inbounds float, float* [[TMP6]], i32 2
509; UNROLL-NO-IC-NEXT:    [[TMP11:%.*]] = bitcast float* [[TMP10]] to <2 x float>*
510; UNROLL-NO-IC-NEXT:    [[WIDE_LOAD8:%.*]] = load <2 x float>, <2 x float>* [[TMP11]], align 4, !alias.scope !4, !noalias !7
511; UNROLL-NO-IC-NEXT:    [[TMP12:%.*]] = add i64 [[TMP2]], [[OFFSET2]]
512; UNROLL-NO-IC-NEXT:    [[TMP13:%.*]] = add i64 [[TMP3]], [[OFFSET2]]
513; UNROLL-NO-IC-NEXT:    [[TMP14:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP12]]
514; UNROLL-NO-IC-NEXT:    [[TMP15:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP13]]
515; UNROLL-NO-IC-NEXT:    [[TMP16:%.*]] = getelementptr inbounds float, float* [[TMP14]], i32 0
516; UNROLL-NO-IC-NEXT:    [[TMP17:%.*]] = bitcast float* [[TMP16]] to <2 x float>*
517; UNROLL-NO-IC-NEXT:    [[WIDE_LOAD9:%.*]] = load <2 x float>, <2 x float>* [[TMP17]], align 4, !alias.scope !7
518; UNROLL-NO-IC-NEXT:    [[TMP18:%.*]] = getelementptr inbounds float, float* [[TMP14]], i32 2
519; UNROLL-NO-IC-NEXT:    [[TMP19:%.*]] = bitcast float* [[TMP18]] to <2 x float>*
520; UNROLL-NO-IC-NEXT:    [[WIDE_LOAD10:%.*]] = load <2 x float>, <2 x float>* [[TMP19]], align 4, !alias.scope !7
521; UNROLL-NO-IC-NEXT:    [[TMP20:%.*]] = fmul fast <2 x float> [[BROADCAST_SPLAT]], [[WIDE_LOAD9]]
522; UNROLL-NO-IC-NEXT:    [[TMP21:%.*]] = fmul fast <2 x float> [[BROADCAST_SPLAT12]], [[WIDE_LOAD10]]
523; UNROLL-NO-IC-NEXT:    [[TMP22:%.*]] = fadd fast <2 x float> [[WIDE_LOAD]], [[TMP20]]
524; UNROLL-NO-IC-NEXT:    [[TMP23:%.*]] = fadd fast <2 x float> [[WIDE_LOAD8]], [[TMP21]]
525; UNROLL-NO-IC-NEXT:    [[TMP24:%.*]] = bitcast float* [[TMP8]] to <2 x float>*
526; UNROLL-NO-IC-NEXT:    store <2 x float> [[TMP22]], <2 x float>* [[TMP24]], align 4, !alias.scope !4, !noalias !7
527; UNROLL-NO-IC-NEXT:    [[TMP25:%.*]] = bitcast float* [[TMP10]] to <2 x float>*
528; UNROLL-NO-IC-NEXT:    store <2 x float> [[TMP23]], <2 x float>* [[TMP25]], align 4, !alias.scope !4, !noalias !7
529; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
530; UNROLL-NO-IC-NEXT:    [[TMP26:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
531; UNROLL-NO-IC-NEXT:    br i1 [[TMP26]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
532; UNROLL-NO-IC:       middle.block:
533; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
534; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]]
535; UNROLL-NO-IC:       scalar.ph:
536; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_MEMCHECK]] ]
537; UNROLL-NO-IC-NEXT:    br label [[FOR_BODY:%.*]]
538; UNROLL-NO-IC:       for.body:
539; UNROLL-NO-IC-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]
540; UNROLL-NO-IC-NEXT:    [[IND_SUM:%.*]] = add i64 [[IV]], [[OFFSET]]
541; UNROLL-NO-IC-NEXT:    [[ARR_IDX:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[IND_SUM]]
542; UNROLL-NO-IC-NEXT:    [[L1:%.*]] = load float, float* [[ARR_IDX]], align 4
543; UNROLL-NO-IC-NEXT:    [[IND_SUM2:%.*]] = add i64 [[IV]], [[OFFSET2]]
544; UNROLL-NO-IC-NEXT:    [[ARR_IDX2:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[IND_SUM2]]
545; UNROLL-NO-IC-NEXT:    [[L2:%.*]] = load float, float* [[ARR_IDX2]], align 4
546; UNROLL-NO-IC-NEXT:    [[M:%.*]] = fmul fast float [[B]], [[L2]]
547; UNROLL-NO-IC-NEXT:    [[AD:%.*]] = fadd fast float [[L1]], [[M]]
548; UNROLL-NO-IC-NEXT:    store float [[AD]], float* [[ARR_IDX]], align 4
549; UNROLL-NO-IC-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
550; UNROLL-NO-IC-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
551; UNROLL-NO-IC-NEXT:    br i1 [[EXITCOND]], label [[LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
552; UNROLL-NO-IC:       loopexit:
553; UNROLL-NO-IC-NEXT:    ret void
554;
555; INTERLEAVE-LABEL: @scalar_use(
556; INTERLEAVE-NEXT:  entry:
557; INTERLEAVE-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 8
558; INTERLEAVE-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
559; INTERLEAVE:       vector.memcheck:
560; INTERLEAVE-NEXT:    [[SCEVGEP:%.*]] = getelementptr float, float* [[A:%.*]], i64 [[OFFSET:%.*]]
561; INTERLEAVE-NEXT:    [[TMP0:%.*]] = add i64 [[N]], [[OFFSET]]
562; INTERLEAVE-NEXT:    [[SCEVGEP2:%.*]] = getelementptr float, float* [[A]], i64 [[TMP0]]
563; INTERLEAVE-NEXT:    [[SCEVGEP4:%.*]] = getelementptr float, float* [[A]], i64 [[OFFSET2:%.*]]
564; INTERLEAVE-NEXT:    [[TMP1:%.*]] = add i64 [[N]], [[OFFSET2]]
565; INTERLEAVE-NEXT:    [[SCEVGEP6:%.*]] = getelementptr float, float* [[A]], i64 [[TMP1]]
566; INTERLEAVE-NEXT:    [[BOUND0:%.*]] = icmp ult float* [[SCEVGEP]], [[SCEVGEP6]]
567; INTERLEAVE-NEXT:    [[BOUND1:%.*]] = icmp ult float* [[SCEVGEP4]], [[SCEVGEP2]]
568; INTERLEAVE-NEXT:    [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
569; INTERLEAVE-NEXT:    br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
570; INTERLEAVE:       vector.ph:
571; INTERLEAVE-NEXT:    [[N_VEC:%.*]] = and i64 [[N]], -8
572; INTERLEAVE-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x float> poison, float [[B:%.*]], i64 0
573; INTERLEAVE-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT]], <4 x float> poison, <4 x i32> zeroinitializer
574; INTERLEAVE-NEXT:    [[BROADCAST_SPLATINSERT11:%.*]] = insertelement <4 x float> poison, float [[B]], i64 0
575; INTERLEAVE-NEXT:    [[BROADCAST_SPLAT12:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT11]], <4 x float> poison, <4 x i32> zeroinitializer
576; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
577; INTERLEAVE:       vector.body:
578; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
579; INTERLEAVE-NEXT:    [[TMP2:%.*]] = add i64 [[INDEX]], [[OFFSET]]
580; INTERLEAVE-NEXT:    [[TMP3:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP2]]
581; INTERLEAVE-NEXT:    [[TMP4:%.*]] = bitcast float* [[TMP3]] to <4 x float>*
582; INTERLEAVE-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x float>, <4 x float>* [[TMP4]], align 4, !alias.scope !4, !noalias !7
583; INTERLEAVE-NEXT:    [[TMP5:%.*]] = getelementptr inbounds float, float* [[TMP3]], i64 4
584; INTERLEAVE-NEXT:    [[TMP6:%.*]] = bitcast float* [[TMP5]] to <4 x float>*
585; INTERLEAVE-NEXT:    [[WIDE_LOAD8:%.*]] = load <4 x float>, <4 x float>* [[TMP6]], align 4, !alias.scope !4, !noalias !7
586; INTERLEAVE-NEXT:    [[TMP7:%.*]] = add i64 [[INDEX]], [[OFFSET2]]
587; INTERLEAVE-NEXT:    [[TMP8:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP7]]
588; INTERLEAVE-NEXT:    [[TMP9:%.*]] = bitcast float* [[TMP8]] to <4 x float>*
589; INTERLEAVE-NEXT:    [[WIDE_LOAD9:%.*]] = load <4 x float>, <4 x float>* [[TMP9]], align 4, !alias.scope !7
590; INTERLEAVE-NEXT:    [[TMP10:%.*]] = getelementptr inbounds float, float* [[TMP8]], i64 4
591; INTERLEAVE-NEXT:    [[TMP11:%.*]] = bitcast float* [[TMP10]] to <4 x float>*
592; INTERLEAVE-NEXT:    [[WIDE_LOAD10:%.*]] = load <4 x float>, <4 x float>* [[TMP11]], align 4, !alias.scope !7
593; INTERLEAVE-NEXT:    [[TMP12:%.*]] = fmul fast <4 x float> [[BROADCAST_SPLAT]], [[WIDE_LOAD9]]
594; INTERLEAVE-NEXT:    [[TMP13:%.*]] = fmul fast <4 x float> [[BROADCAST_SPLAT12]], [[WIDE_LOAD10]]
595; INTERLEAVE-NEXT:    [[TMP14:%.*]] = fadd fast <4 x float> [[WIDE_LOAD]], [[TMP12]]
596; INTERLEAVE-NEXT:    [[TMP15:%.*]] = fadd fast <4 x float> [[WIDE_LOAD8]], [[TMP13]]
597; INTERLEAVE-NEXT:    [[TMP16:%.*]] = bitcast float* [[TMP3]] to <4 x float>*
598; INTERLEAVE-NEXT:    store <4 x float> [[TMP14]], <4 x float>* [[TMP16]], align 4, !alias.scope !4, !noalias !7
599; INTERLEAVE-NEXT:    [[TMP17:%.*]] = bitcast float* [[TMP5]] to <4 x float>*
600; INTERLEAVE-NEXT:    store <4 x float> [[TMP15]], <4 x float>* [[TMP17]], align 4, !alias.scope !4, !noalias !7
601; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
602; INTERLEAVE-NEXT:    [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
603; INTERLEAVE-NEXT:    br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
604; INTERLEAVE:       middle.block:
605; INTERLEAVE-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[N]]
606; INTERLEAVE-NEXT:    br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]]
607; INTERLEAVE:       scalar.ph:
608; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_MEMCHECK]] ]
609; INTERLEAVE-NEXT:    br label [[FOR_BODY:%.*]]
610; INTERLEAVE:       for.body:
611; INTERLEAVE-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ]
612; INTERLEAVE-NEXT:    [[IND_SUM:%.*]] = add i64 [[IV]], [[OFFSET]]
613; INTERLEAVE-NEXT:    [[ARR_IDX:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[IND_SUM]]
614; INTERLEAVE-NEXT:    [[L1:%.*]] = load float, float* [[ARR_IDX]], align 4
615; INTERLEAVE-NEXT:    [[IND_SUM2:%.*]] = add i64 [[IV]], [[OFFSET2]]
616; INTERLEAVE-NEXT:    [[ARR_IDX2:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[IND_SUM2]]
617; INTERLEAVE-NEXT:    [[L2:%.*]] = load float, float* [[ARR_IDX2]], align 4
618; INTERLEAVE-NEXT:    [[M:%.*]] = fmul fast float [[L2]], [[B]]
619; INTERLEAVE-NEXT:    [[AD:%.*]] = fadd fast float [[L1]], [[M]]
620; INTERLEAVE-NEXT:    store float [[AD]], float* [[ARR_IDX]], align 4
621; INTERLEAVE-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
622; INTERLEAVE-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
623; INTERLEAVE-NEXT:    br i1 [[EXITCOND]], label [[LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
624; INTERLEAVE:       loopexit:
625; INTERLEAVE-NEXT:    ret void
626;
627entry:
628  br label %for.body
629
630for.body:
631  %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
632  %ind.sum = add i64 %iv, %offset
633  %arr.idx = getelementptr inbounds float, float* %a, i64 %ind.sum
634  %l1 = load float, float* %arr.idx, align 4
635  %ind.sum2 = add i64 %iv, %offset2
636  %arr.idx2 = getelementptr inbounds float, float* %a, i64 %ind.sum2
637  %l2 = load float, float* %arr.idx2, align 4
638  %m = fmul fast float %b, %l2
639  %ad = fadd fast float %l1, %m
640  store float %ad, float* %arr.idx, align 4
641  %iv.next = add nuw nsw i64 %iv, 1
642  %exitcond = icmp eq i64 %iv.next, %n
643  br i1 %exitcond, label %loopexit, label %for.body
644
645loopexit:
646  ret void
647}
648
649; Make sure we don't create a vector induction phi node that is unused.
650; Scalarize the step vectors instead.
651;
652; for (int i = 0; i < n; ++i)
653;   sum += a[i];
654;
655;
656;
657;
658
659define i64 @scalarize_induction_variable_01(i64 *%a, i64 %n) {
660; CHECK-LABEL: @scalarize_induction_variable_01(
661; CHECK-NEXT:  entry:
662; CHECK-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)
663; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 2
664; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
665; CHECK:       vector.ph:
666; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[SMAX]], 2
667; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[SMAX]], [[N_MOD_VF]]
668; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
669; CHECK:       vector.body:
670; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
671; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP4:%.*]], [[VECTOR_BODY]] ]
672; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
673; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i64, i64* [[A:%.*]], i64 [[TMP0]]
674; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i64, i64* [[TMP1]], i32 0
675; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i64* [[TMP2]] to <2 x i64>*
676; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x i64>, <2 x i64>* [[TMP3]], align 8
677; CHECK-NEXT:    [[TMP4]] = add <2 x i64> [[WIDE_LOAD]], [[VEC_PHI]]
678; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
679; CHECK-NEXT:    [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
680; CHECK-NEXT:    br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]]
681; CHECK:       middle.block:
682; CHECK-NEXT:    [[TMP6:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[TMP4]])
683; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
684; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
685; CHECK:       scalar.ph:
686; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
687; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[TMP6]], [[MIDDLE_BLOCK]] ]
688; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
689; CHECK:       for.body:
690; CHECK-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
691; CHECK-NEXT:    [[SUM:%.*]] = phi i64 [ [[TMP9:%.*]], [[FOR_BODY]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ]
692; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i64, i64* [[A]], i64 [[I]]
693; CHECK-NEXT:    [[TMP8:%.*]] = load i64, i64* [[TMP7]], align 8
694; CHECK-NEXT:    [[TMP9]] = add i64 [[TMP8]], [[SUM]]
695; CHECK-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
696; CHECK-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
697; CHECK-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP12:![0-9]+]]
698; CHECK:       for.end:
699; CHECK-NEXT:    [[TMP10:%.*]] = phi i64 [ [[TMP9]], [[FOR_BODY]] ], [ [[TMP6]], [[MIDDLE_BLOCK]] ]
700; CHECK-NEXT:    ret i64 [[TMP10]]
701;
702; IND-LABEL: @scalarize_induction_variable_01(
703; IND-NEXT:  entry:
704; IND-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)
705; IND-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 2
706; IND-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
707; IND:       vector.ph:
708; IND-NEXT:    [[N_VEC:%.*]] = and i64 [[SMAX]], 9223372036854775806
709; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
710; IND:       vector.body:
711; IND-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
712; IND-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP2:%.*]], [[VECTOR_BODY]] ]
713; IND-NEXT:    [[TMP0:%.*]] = getelementptr inbounds i64, i64* [[A:%.*]], i64 [[INDEX]]
714; IND-NEXT:    [[TMP1:%.*]] = bitcast i64* [[TMP0]] to <2 x i64>*
715; IND-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x i64>, <2 x i64>* [[TMP1]], align 8
716; IND-NEXT:    [[TMP2]] = add <2 x i64> [[WIDE_LOAD]], [[VEC_PHI]]
717; IND-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
718; IND-NEXT:    [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
719; IND-NEXT:    br i1 [[TMP3]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]]
720; IND:       middle.block:
721; IND-NEXT:    [[TMP4:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[TMP2]])
722; IND-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
723; IND-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
724; IND:       scalar.ph:
725; IND-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
726; IND-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP4]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
727; IND-NEXT:    br label [[FOR_BODY:%.*]]
728; IND:       for.body:
729; IND-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
730; IND-NEXT:    [[SUM:%.*]] = phi i64 [ [[TMP7:%.*]], [[FOR_BODY]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ]
731; IND-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i64, i64* [[A]], i64 [[I]]
732; IND-NEXT:    [[TMP6:%.*]] = load i64, i64* [[TMP5]], align 8
733; IND-NEXT:    [[TMP7]] = add i64 [[TMP6]], [[SUM]]
734; IND-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
735; IND-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
736; IND-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP12:![0-9]+]]
737; IND:       for.end:
738; IND-NEXT:    [[TMP8:%.*]] = phi i64 [ [[TMP7]], [[FOR_BODY]] ], [ [[TMP4]], [[MIDDLE_BLOCK]] ]
739; IND-NEXT:    ret i64 [[TMP8]]
740;
741; UNROLL-LABEL: @scalarize_induction_variable_01(
742; UNROLL-NEXT:  entry:
743; UNROLL-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)
744; UNROLL-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 4
745; UNROLL-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
746; UNROLL:       vector.ph:
747; UNROLL-NEXT:    [[N_VEC:%.*]] = and i64 [[SMAX]], 9223372036854775804
748; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
749; UNROLL:       vector.body:
750; UNROLL-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
751; UNROLL-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP4:%.*]], [[VECTOR_BODY]] ]
752; UNROLL-NEXT:    [[VEC_PHI1:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP5:%.*]], [[VECTOR_BODY]] ]
753; UNROLL-NEXT:    [[TMP0:%.*]] = getelementptr inbounds i64, i64* [[A:%.*]], i64 [[INDEX]]
754; UNROLL-NEXT:    [[TMP1:%.*]] = bitcast i64* [[TMP0]] to <2 x i64>*
755; UNROLL-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x i64>, <2 x i64>* [[TMP1]], align 8
756; UNROLL-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i64, i64* [[TMP0]], i64 2
757; UNROLL-NEXT:    [[TMP3:%.*]] = bitcast i64* [[TMP2]] to <2 x i64>*
758; UNROLL-NEXT:    [[WIDE_LOAD2:%.*]] = load <2 x i64>, <2 x i64>* [[TMP3]], align 8
759; UNROLL-NEXT:    [[TMP4]] = add <2 x i64> [[WIDE_LOAD]], [[VEC_PHI]]
760; UNROLL-NEXT:    [[TMP5]] = add <2 x i64> [[WIDE_LOAD2]], [[VEC_PHI1]]
761; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
762; UNROLL-NEXT:    [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
763; UNROLL-NEXT:    br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]]
764; UNROLL:       middle.block:
765; UNROLL-NEXT:    [[BIN_RDX:%.*]] = add <2 x i64> [[TMP5]], [[TMP4]]
766; UNROLL-NEXT:    [[TMP7:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[BIN_RDX]])
767; UNROLL-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
768; UNROLL-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
769; UNROLL:       scalar.ph:
770; UNROLL-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
771; UNROLL-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP7]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
772; UNROLL-NEXT:    br label [[FOR_BODY:%.*]]
773; UNROLL:       for.body:
774; UNROLL-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
775; UNROLL-NEXT:    [[SUM:%.*]] = phi i64 [ [[TMP10:%.*]], [[FOR_BODY]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ]
776; UNROLL-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i64, i64* [[A]], i64 [[I]]
777; UNROLL-NEXT:    [[TMP9:%.*]] = load i64, i64* [[TMP8]], align 8
778; UNROLL-NEXT:    [[TMP10]] = add i64 [[TMP9]], [[SUM]]
779; UNROLL-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
780; UNROLL-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
781; UNROLL-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP12:![0-9]+]]
782; UNROLL:       for.end:
783; UNROLL-NEXT:    [[TMP11:%.*]] = phi i64 [ [[TMP10]], [[FOR_BODY]] ], [ [[TMP7]], [[MIDDLE_BLOCK]] ]
784; UNROLL-NEXT:    ret i64 [[TMP11]]
785;
786; UNROLL-NO-IC-LABEL: @scalarize_induction_variable_01(
787; UNROLL-NO-IC-NEXT:  entry:
788; UNROLL-NO-IC-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)
789; UNROLL-NO-IC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 4
790; UNROLL-NO-IC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
791; UNROLL-NO-IC:       vector.ph:
792; UNROLL-NO-IC-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[SMAX]], 4
793; UNROLL-NO-IC-NEXT:    [[N_VEC:%.*]] = sub i64 [[SMAX]], [[N_MOD_VF]]
794; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
795; UNROLL-NO-IC:       vector.body:
796; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
797; UNROLL-NO-IC-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP8:%.*]], [[VECTOR_BODY]] ]
798; UNROLL-NO-IC-NEXT:    [[VEC_PHI1:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP9:%.*]], [[VECTOR_BODY]] ]
799; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
800; UNROLL-NO-IC-NEXT:    [[TMP1:%.*]] = add i64 [[INDEX]], 2
801; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i64, i64* [[A:%.*]], i64 [[TMP0]]
802; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i64, i64* [[A]], i64 [[TMP1]]
803; UNROLL-NO-IC-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i64, i64* [[TMP2]], i32 0
804; UNROLL-NO-IC-NEXT:    [[TMP5:%.*]] = bitcast i64* [[TMP4]] to <2 x i64>*
805; UNROLL-NO-IC-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x i64>, <2 x i64>* [[TMP5]], align 8
806; UNROLL-NO-IC-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i64, i64* [[TMP2]], i32 2
807; UNROLL-NO-IC-NEXT:    [[TMP7:%.*]] = bitcast i64* [[TMP6]] to <2 x i64>*
808; UNROLL-NO-IC-NEXT:    [[WIDE_LOAD2:%.*]] = load <2 x i64>, <2 x i64>* [[TMP7]], align 8
809; UNROLL-NO-IC-NEXT:    [[TMP8]] = add <2 x i64> [[WIDE_LOAD]], [[VEC_PHI]]
810; UNROLL-NO-IC-NEXT:    [[TMP9]] = add <2 x i64> [[WIDE_LOAD2]], [[VEC_PHI1]]
811; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
812; UNROLL-NO-IC-NEXT:    [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
813; UNROLL-NO-IC-NEXT:    br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]]
814; UNROLL-NO-IC:       middle.block:
815; UNROLL-NO-IC-NEXT:    [[BIN_RDX:%.*]] = add <2 x i64> [[TMP9]], [[TMP8]]
816; UNROLL-NO-IC-NEXT:    [[TMP11:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[BIN_RDX]])
817; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
818; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
819; UNROLL-NO-IC:       scalar.ph:
820; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
821; UNROLL-NO-IC-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[TMP11]], [[MIDDLE_BLOCK]] ]
822; UNROLL-NO-IC-NEXT:    br label [[FOR_BODY:%.*]]
823; UNROLL-NO-IC:       for.body:
824; UNROLL-NO-IC-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
825; UNROLL-NO-IC-NEXT:    [[SUM:%.*]] = phi i64 [ [[TMP14:%.*]], [[FOR_BODY]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ]
826; UNROLL-NO-IC-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i64, i64* [[A]], i64 [[I]]
827; UNROLL-NO-IC-NEXT:    [[TMP13:%.*]] = load i64, i64* [[TMP12]], align 8
828; UNROLL-NO-IC-NEXT:    [[TMP14]] = add i64 [[TMP13]], [[SUM]]
829; UNROLL-NO-IC-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
830; UNROLL-NO-IC-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
831; UNROLL-NO-IC-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP12:![0-9]+]]
832; UNROLL-NO-IC:       for.end:
833; UNROLL-NO-IC-NEXT:    [[TMP15:%.*]] = phi i64 [ [[TMP14]], [[FOR_BODY]] ], [ [[TMP11]], [[MIDDLE_BLOCK]] ]
834; UNROLL-NO-IC-NEXT:    ret i64 [[TMP15]]
835;
836; INTERLEAVE-LABEL: @scalarize_induction_variable_01(
837; INTERLEAVE-NEXT:  entry:
838; INTERLEAVE-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)
839; INTERLEAVE-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 8
840; INTERLEAVE-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
841; INTERLEAVE:       vector.ph:
842; INTERLEAVE-NEXT:    [[N_VEC:%.*]] = and i64 [[SMAX]], 9223372036854775800
843; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
844; INTERLEAVE:       vector.body:
845; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
846; INTERLEAVE-NEXT:    [[VEC_PHI:%.*]] = phi <4 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP4:%.*]], [[VECTOR_BODY]] ]
847; INTERLEAVE-NEXT:    [[VEC_PHI1:%.*]] = phi <4 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP5:%.*]], [[VECTOR_BODY]] ]
848; INTERLEAVE-NEXT:    [[TMP0:%.*]] = getelementptr inbounds i64, i64* [[A:%.*]], i64 [[INDEX]]
849; INTERLEAVE-NEXT:    [[TMP1:%.*]] = bitcast i64* [[TMP0]] to <4 x i64>*
850; INTERLEAVE-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i64>, <4 x i64>* [[TMP1]], align 8
851; INTERLEAVE-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i64, i64* [[TMP0]], i64 4
852; INTERLEAVE-NEXT:    [[TMP3:%.*]] = bitcast i64* [[TMP2]] to <4 x i64>*
853; INTERLEAVE-NEXT:    [[WIDE_LOAD2:%.*]] = load <4 x i64>, <4 x i64>* [[TMP3]], align 8
854; INTERLEAVE-NEXT:    [[TMP4]] = add <4 x i64> [[WIDE_LOAD]], [[VEC_PHI]]
855; INTERLEAVE-NEXT:    [[TMP5]] = add <4 x i64> [[WIDE_LOAD2]], [[VEC_PHI1]]
856; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
857; INTERLEAVE-NEXT:    [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
858; INTERLEAVE-NEXT:    br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]]
859; INTERLEAVE:       middle.block:
860; INTERLEAVE-NEXT:    [[BIN_RDX:%.*]] = add <4 x i64> [[TMP5]], [[TMP4]]
861; INTERLEAVE-NEXT:    [[TMP7:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[BIN_RDX]])
862; INTERLEAVE-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
863; INTERLEAVE-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
864; INTERLEAVE:       scalar.ph:
865; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
866; INTERLEAVE-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP7]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
867; INTERLEAVE-NEXT:    br label [[FOR_BODY:%.*]]
868; INTERLEAVE:       for.body:
869; INTERLEAVE-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
870; INTERLEAVE-NEXT:    [[SUM:%.*]] = phi i64 [ [[TMP10:%.*]], [[FOR_BODY]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ]
871; INTERLEAVE-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i64, i64* [[A]], i64 [[I]]
872; INTERLEAVE-NEXT:    [[TMP9:%.*]] = load i64, i64* [[TMP8]], align 8
873; INTERLEAVE-NEXT:    [[TMP10]] = add i64 [[TMP9]], [[SUM]]
874; INTERLEAVE-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
875; INTERLEAVE-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
876; INTERLEAVE-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP12:![0-9]+]]
877; INTERLEAVE:       for.end:
878; INTERLEAVE-NEXT:    [[TMP11:%.*]] = phi i64 [ [[TMP10]], [[FOR_BODY]] ], [ [[TMP7]], [[MIDDLE_BLOCK]] ]
879; INTERLEAVE-NEXT:    ret i64 [[TMP11]]
880;
881entry:
882  br label %for.body
883
884for.body:
885  %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ]
886  %sum = phi i64 [ %2, %for.body ], [ 0, %entry ]
887  %0 = getelementptr inbounds i64, i64* %a, i64 %i
888  %1 = load i64, i64* %0, align 8
889  %2 = add i64 %1, %sum
890  %i.next = add nuw nsw i64 %i, 1
891  %cond = icmp slt i64 %i.next, %n
892  br i1 %cond, label %for.body, label %for.end
893
894for.end:
895  %3  = phi i64 [ %2, %for.body ]
896  ret i64 %3
897}
898
899; Make sure we scalarize the step vectors used for the pointer arithmetic. We
900; can't easily simplify vectorized step vectors.
901;
902; float s = 0;
903; for (int i ; 0; i < n; i += 8)
904;   s += (a[i] + b[i] + 1.0f);
905;
906;
907;
908;
909
910define float @scalarize_induction_variable_02(float* %a, float* %b, i64 %n) {
911; CHECK-LABEL: @scalarize_induction_variable_02(
912; CHECK-NEXT:  entry:
913; CHECK-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 8)
914; CHECK-NEXT:    [[TMP0:%.*]] = add nsw i64 [[SMAX]], -1
915; CHECK-NEXT:    [[TMP1:%.*]] = lshr i64 [[TMP0]], 3
916; CHECK-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
917; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 2
918; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
919; CHECK:       vector.ph:
920; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 2
921; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]]
922; CHECK-NEXT:    [[IND_END:%.*]] = mul i64 [[N_VEC]], 8
923; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
924; CHECK:       vector.body:
925; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
926; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <2 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP19:%.*]], [[VECTOR_BODY]] ]
927; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 8
928; CHECK-NEXT:    [[TMP3:%.*]] = add i64 [[OFFSET_IDX]], 0
929; CHECK-NEXT:    [[TMP4:%.*]] = add i64 [[OFFSET_IDX]], 8
930; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[TMP3]]
931; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP4]]
932; CHECK-NEXT:    [[TMP7:%.*]] = load float, float* [[TMP5]], align 4
933; CHECK-NEXT:    [[TMP8:%.*]] = load float, float* [[TMP6]], align 4
934; CHECK-NEXT:    [[TMP9:%.*]] = insertelement <2 x float> poison, float [[TMP7]], i32 0
935; CHECK-NEXT:    [[TMP10:%.*]] = insertelement <2 x float> [[TMP9]], float [[TMP8]], i32 1
936; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds float, float* [[B:%.*]], i64 [[TMP3]]
937; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[TMP4]]
938; CHECK-NEXT:    [[TMP13:%.*]] = load float, float* [[TMP11]], align 4
939; CHECK-NEXT:    [[TMP14:%.*]] = load float, float* [[TMP12]], align 4
940; CHECK-NEXT:    [[TMP15:%.*]] = insertelement <2 x float> poison, float [[TMP13]], i32 0
941; CHECK-NEXT:    [[TMP16:%.*]] = insertelement <2 x float> [[TMP15]], float [[TMP14]], i32 1
942; CHECK-NEXT:    [[TMP17:%.*]] = fadd fast <2 x float> [[VEC_PHI]], <float 1.000000e+00, float 1.000000e+00>
943; CHECK-NEXT:    [[TMP18:%.*]] = fadd fast <2 x float> [[TMP17]], [[TMP10]]
944; CHECK-NEXT:    [[TMP19]] = fadd fast <2 x float> [[TMP18]], [[TMP16]]
945; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
946; CHECK-NEXT:    [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
947; CHECK-NEXT:    br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]]
948; CHECK:       middle.block:
949; CHECK-NEXT:    [[TMP21:%.*]] = call fast float @llvm.vector.reduce.fadd.v2f32(float -0.000000e+00, <2 x float> [[TMP19]])
950; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
951; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
952; CHECK:       scalar.ph:
953; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
954; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi float [ 0.000000e+00, [[ENTRY]] ], [ [[TMP21]], [[MIDDLE_BLOCK]] ]
955; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
956; CHECK:       for.body:
957; CHECK-NEXT:    [[I:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[FOR_BODY]] ]
958; CHECK-NEXT:    [[S:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[TMP28:%.*]], [[FOR_BODY]] ]
959; CHECK-NEXT:    [[TMP22:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[I]]
960; CHECK-NEXT:    [[TMP23:%.*]] = load float, float* [[TMP22]], align 4
961; CHECK-NEXT:    [[TMP24:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[I]]
962; CHECK-NEXT:    [[TMP25:%.*]] = load float, float* [[TMP24]], align 4
963; CHECK-NEXT:    [[TMP26:%.*]] = fadd fast float [[S]], 1.000000e+00
964; CHECK-NEXT:    [[TMP27:%.*]] = fadd fast float [[TMP26]], [[TMP23]]
965; CHECK-NEXT:    [[TMP28]] = fadd fast float [[TMP27]], [[TMP25]]
966; CHECK-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 8
967; CHECK-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
968; CHECK-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP14:![0-9]+]]
969; CHECK:       for.end:
970; CHECK-NEXT:    [[S_LCSSA:%.*]] = phi float [ [[TMP28]], [[FOR_BODY]] ], [ [[TMP21]], [[MIDDLE_BLOCK]] ]
971; CHECK-NEXT:    ret float [[S_LCSSA]]
972;
973; IND-LABEL: @scalarize_induction_variable_02(
974; IND-NEXT:  entry:
975; IND-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 8)
976; IND-NEXT:    [[TMP0:%.*]] = add nsw i64 [[SMAX]], -1
977; IND-NEXT:    [[TMP1:%.*]] = lshr i64 [[TMP0]], 3
978; IND-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
979; IND-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 8
980; IND-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
981; IND:       vector.ph:
982; IND-NEXT:    [[N_VEC:%.*]] = and i64 [[TMP2]], 4611686018427387902
983; IND-NEXT:    [[IND_END:%.*]] = shl i64 [[N_VEC]], 3
984; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
985; IND:       vector.body:
986; IND-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
987; IND-NEXT:    [[VEC_PHI:%.*]] = phi <2 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP18:%.*]], [[VECTOR_BODY]] ]
988; IND-NEXT:    [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 3
989; IND-NEXT:    [[TMP3:%.*]] = or i64 [[OFFSET_IDX]], 8
990; IND-NEXT:    [[TMP4:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[OFFSET_IDX]]
991; IND-NEXT:    [[TMP5:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP3]]
992; IND-NEXT:    [[TMP6:%.*]] = load float, float* [[TMP4]], align 4
993; IND-NEXT:    [[TMP7:%.*]] = load float, float* [[TMP5]], align 4
994; IND-NEXT:    [[TMP8:%.*]] = insertelement <2 x float> poison, float [[TMP6]], i64 0
995; IND-NEXT:    [[TMP9:%.*]] = insertelement <2 x float> [[TMP8]], float [[TMP7]], i64 1
996; IND-NEXT:    [[TMP10:%.*]] = getelementptr inbounds float, float* [[B:%.*]], i64 [[OFFSET_IDX]]
997; IND-NEXT:    [[TMP11:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[TMP3]]
998; IND-NEXT:    [[TMP12:%.*]] = load float, float* [[TMP10]], align 4
999; IND-NEXT:    [[TMP13:%.*]] = load float, float* [[TMP11]], align 4
1000; IND-NEXT:    [[TMP14:%.*]] = insertelement <2 x float> poison, float [[TMP12]], i64 0
1001; IND-NEXT:    [[TMP15:%.*]] = insertelement <2 x float> [[TMP14]], float [[TMP13]], i64 1
1002; IND-NEXT:    [[TMP16:%.*]] = fadd fast <2 x float> [[VEC_PHI]], <float 1.000000e+00, float 1.000000e+00>
1003; IND-NEXT:    [[TMP17:%.*]] = fadd fast <2 x float> [[TMP16]], [[TMP9]]
1004; IND-NEXT:    [[TMP18]] = fadd fast <2 x float> [[TMP17]], [[TMP15]]
1005; IND-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
1006; IND-NEXT:    [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1007; IND-NEXT:    br i1 [[TMP19]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]]
1008; IND:       middle.block:
1009; IND-NEXT:    [[TMP20:%.*]] = call fast float @llvm.vector.reduce.fadd.v2f32(float -0.000000e+00, <2 x float> [[TMP18]])
1010; IND-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
1011; IND-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
1012; IND:       scalar.ph:
1013; IND-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
1014; IND-NEXT:    [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP20]], [[MIDDLE_BLOCK]] ], [ 0.000000e+00, [[ENTRY]] ]
1015; IND-NEXT:    br label [[FOR_BODY:%.*]]
1016; IND:       for.body:
1017; IND-NEXT:    [[I:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[FOR_BODY]] ]
1018; IND-NEXT:    [[S:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[TMP27:%.*]], [[FOR_BODY]] ]
1019; IND-NEXT:    [[TMP21:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[I]]
1020; IND-NEXT:    [[TMP22:%.*]] = load float, float* [[TMP21]], align 4
1021; IND-NEXT:    [[TMP23:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[I]]
1022; IND-NEXT:    [[TMP24:%.*]] = load float, float* [[TMP23]], align 4
1023; IND-NEXT:    [[TMP25:%.*]] = fadd fast float [[S]], 1.000000e+00
1024; IND-NEXT:    [[TMP26:%.*]] = fadd fast float [[TMP25]], [[TMP22]]
1025; IND-NEXT:    [[TMP27]] = fadd fast float [[TMP26]], [[TMP24]]
1026; IND-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 8
1027; IND-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
1028; IND-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP14:![0-9]+]]
1029; IND:       for.end:
1030; IND-NEXT:    [[S_LCSSA:%.*]] = phi float [ [[TMP27]], [[FOR_BODY]] ], [ [[TMP20]], [[MIDDLE_BLOCK]] ]
1031; IND-NEXT:    ret float [[S_LCSSA]]
1032;
1033; UNROLL-LABEL: @scalarize_induction_variable_02(
1034; UNROLL-NEXT:  entry:
1035; UNROLL-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 8)
1036; UNROLL-NEXT:    [[TMP0:%.*]] = add nsw i64 [[SMAX]], -1
1037; UNROLL-NEXT:    [[TMP1:%.*]] = lshr i64 [[TMP0]], 3
1038; UNROLL-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
1039; UNROLL-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 24
1040; UNROLL-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
1041; UNROLL:       vector.ph:
1042; UNROLL-NEXT:    [[N_VEC:%.*]] = and i64 [[TMP2]], 4611686018427387900
1043; UNROLL-NEXT:    [[IND_END:%.*]] = shl i64 [[N_VEC]], 3
1044; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
1045; UNROLL:       vector.body:
1046; UNROLL-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
1047; UNROLL-NEXT:    [[VEC_PHI:%.*]] = phi <2 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP34:%.*]], [[VECTOR_BODY]] ]
1048; UNROLL-NEXT:    [[VEC_PHI1:%.*]] = phi <2 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP35:%.*]], [[VECTOR_BODY]] ]
1049; UNROLL-NEXT:    [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 3
1050; UNROLL-NEXT:    [[TMP3:%.*]] = or i64 [[OFFSET_IDX]], 8
1051; UNROLL-NEXT:    [[TMP4:%.*]] = or i64 [[OFFSET_IDX]], 16
1052; UNROLL-NEXT:    [[TMP5:%.*]] = or i64 [[OFFSET_IDX]], 24
1053; UNROLL-NEXT:    [[TMP6:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[OFFSET_IDX]]
1054; UNROLL-NEXT:    [[TMP7:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP3]]
1055; UNROLL-NEXT:    [[TMP8:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP4]]
1056; UNROLL-NEXT:    [[TMP9:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP5]]
1057; UNROLL-NEXT:    [[TMP10:%.*]] = load float, float* [[TMP6]], align 4
1058; UNROLL-NEXT:    [[TMP11:%.*]] = load float, float* [[TMP7]], align 4
1059; UNROLL-NEXT:    [[TMP12:%.*]] = insertelement <2 x float> poison, float [[TMP10]], i64 0
1060; UNROLL-NEXT:    [[TMP13:%.*]] = insertelement <2 x float> [[TMP12]], float [[TMP11]], i64 1
1061; UNROLL-NEXT:    [[TMP14:%.*]] = load float, float* [[TMP8]], align 4
1062; UNROLL-NEXT:    [[TMP15:%.*]] = load float, float* [[TMP9]], align 4
1063; UNROLL-NEXT:    [[TMP16:%.*]] = insertelement <2 x float> poison, float [[TMP14]], i64 0
1064; UNROLL-NEXT:    [[TMP17:%.*]] = insertelement <2 x float> [[TMP16]], float [[TMP15]], i64 1
1065; UNROLL-NEXT:    [[TMP18:%.*]] = getelementptr inbounds float, float* [[B:%.*]], i64 [[OFFSET_IDX]]
1066; UNROLL-NEXT:    [[TMP19:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[TMP3]]
1067; UNROLL-NEXT:    [[TMP20:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[TMP4]]
1068; UNROLL-NEXT:    [[TMP21:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[TMP5]]
1069; UNROLL-NEXT:    [[TMP22:%.*]] = load float, float* [[TMP18]], align 4
1070; UNROLL-NEXT:    [[TMP23:%.*]] = load float, float* [[TMP19]], align 4
1071; UNROLL-NEXT:    [[TMP24:%.*]] = insertelement <2 x float> poison, float [[TMP22]], i64 0
1072; UNROLL-NEXT:    [[TMP25:%.*]] = insertelement <2 x float> [[TMP24]], float [[TMP23]], i64 1
1073; UNROLL-NEXT:    [[TMP26:%.*]] = load float, float* [[TMP20]], align 4
1074; UNROLL-NEXT:    [[TMP27:%.*]] = load float, float* [[TMP21]], align 4
1075; UNROLL-NEXT:    [[TMP28:%.*]] = insertelement <2 x float> poison, float [[TMP26]], i64 0
1076; UNROLL-NEXT:    [[TMP29:%.*]] = insertelement <2 x float> [[TMP28]], float [[TMP27]], i64 1
1077; UNROLL-NEXT:    [[TMP30:%.*]] = fadd fast <2 x float> [[VEC_PHI]], <float 1.000000e+00, float 1.000000e+00>
1078; UNROLL-NEXT:    [[TMP31:%.*]] = fadd fast <2 x float> [[VEC_PHI1]], <float 1.000000e+00, float 1.000000e+00>
1079; UNROLL-NEXT:    [[TMP32:%.*]] = fadd fast <2 x float> [[TMP30]], [[TMP13]]
1080; UNROLL-NEXT:    [[TMP33:%.*]] = fadd fast <2 x float> [[TMP31]], [[TMP17]]
1081; UNROLL-NEXT:    [[TMP34]] = fadd fast <2 x float> [[TMP32]], [[TMP25]]
1082; UNROLL-NEXT:    [[TMP35]] = fadd fast <2 x float> [[TMP33]], [[TMP29]]
1083; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
1084; UNROLL-NEXT:    [[TMP36:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1085; UNROLL-NEXT:    br i1 [[TMP36]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]]
1086; UNROLL:       middle.block:
1087; UNROLL-NEXT:    [[BIN_RDX:%.*]] = fadd fast <2 x float> [[TMP35]], [[TMP34]]
1088; UNROLL-NEXT:    [[TMP37:%.*]] = call fast float @llvm.vector.reduce.fadd.v2f32(float -0.000000e+00, <2 x float> [[BIN_RDX]])
1089; UNROLL-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
1090; UNROLL-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
1091; UNROLL:       scalar.ph:
1092; UNROLL-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
1093; UNROLL-NEXT:    [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP37]], [[MIDDLE_BLOCK]] ], [ 0.000000e+00, [[ENTRY]] ]
1094; UNROLL-NEXT:    br label [[FOR_BODY:%.*]]
1095; UNROLL:       for.body:
1096; UNROLL-NEXT:    [[I:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[FOR_BODY]] ]
1097; UNROLL-NEXT:    [[S:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[TMP44:%.*]], [[FOR_BODY]] ]
1098; UNROLL-NEXT:    [[TMP38:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[I]]
1099; UNROLL-NEXT:    [[TMP39:%.*]] = load float, float* [[TMP38]], align 4
1100; UNROLL-NEXT:    [[TMP40:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[I]]
1101; UNROLL-NEXT:    [[TMP41:%.*]] = load float, float* [[TMP40]], align 4
1102; UNROLL-NEXT:    [[TMP42:%.*]] = fadd fast float [[S]], 1.000000e+00
1103; UNROLL-NEXT:    [[TMP43:%.*]] = fadd fast float [[TMP42]], [[TMP39]]
1104; UNROLL-NEXT:    [[TMP44]] = fadd fast float [[TMP43]], [[TMP41]]
1105; UNROLL-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 8
1106; UNROLL-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
1107; UNROLL-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP14:![0-9]+]]
1108; UNROLL:       for.end:
1109; UNROLL-NEXT:    [[S_LCSSA:%.*]] = phi float [ [[TMP44]], [[FOR_BODY]] ], [ [[TMP37]], [[MIDDLE_BLOCK]] ]
1110; UNROLL-NEXT:    ret float [[S_LCSSA]]
1111;
1112; UNROLL-NO-IC-LABEL: @scalarize_induction_variable_02(
1113; UNROLL-NO-IC-NEXT:  entry:
1114; UNROLL-NO-IC-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 8)
1115; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = add nsw i64 [[SMAX]], -1
1116; UNROLL-NO-IC-NEXT:    [[TMP1:%.*]] = lshr i64 [[TMP0]], 3
1117; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
1118; UNROLL-NO-IC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 4
1119; UNROLL-NO-IC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
1120; UNROLL-NO-IC:       vector.ph:
1121; UNROLL-NO-IC-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 4
1122; UNROLL-NO-IC-NEXT:    [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]]
1123; UNROLL-NO-IC-NEXT:    [[IND_END:%.*]] = mul i64 [[N_VEC]], 8
1124; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
1125; UNROLL-NO-IC:       vector.body:
1126; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
1127; UNROLL-NO-IC-NEXT:    [[VEC_PHI:%.*]] = phi <2 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP35:%.*]], [[VECTOR_BODY]] ]
1128; UNROLL-NO-IC-NEXT:    [[VEC_PHI1:%.*]] = phi <2 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP36:%.*]], [[VECTOR_BODY]] ]
1129; UNROLL-NO-IC-NEXT:    [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 8
1130; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = add i64 [[OFFSET_IDX]], 0
1131; UNROLL-NO-IC-NEXT:    [[TMP4:%.*]] = add i64 [[OFFSET_IDX]], 8
1132; UNROLL-NO-IC-NEXT:    [[TMP5:%.*]] = add i64 [[OFFSET_IDX]], 16
1133; UNROLL-NO-IC-NEXT:    [[TMP6:%.*]] = add i64 [[OFFSET_IDX]], 24
1134; UNROLL-NO-IC-NEXT:    [[TMP7:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[TMP3]]
1135; UNROLL-NO-IC-NEXT:    [[TMP8:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP4]]
1136; UNROLL-NO-IC-NEXT:    [[TMP9:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP5]]
1137; UNROLL-NO-IC-NEXT:    [[TMP10:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP6]]
1138; UNROLL-NO-IC-NEXT:    [[TMP11:%.*]] = load float, float* [[TMP7]], align 4
1139; UNROLL-NO-IC-NEXT:    [[TMP12:%.*]] = load float, float* [[TMP8]], align 4
1140; UNROLL-NO-IC-NEXT:    [[TMP13:%.*]] = insertelement <2 x float> poison, float [[TMP11]], i32 0
1141; UNROLL-NO-IC-NEXT:    [[TMP14:%.*]] = insertelement <2 x float> [[TMP13]], float [[TMP12]], i32 1
1142; UNROLL-NO-IC-NEXT:    [[TMP15:%.*]] = load float, float* [[TMP9]], align 4
1143; UNROLL-NO-IC-NEXT:    [[TMP16:%.*]] = load float, float* [[TMP10]], align 4
1144; UNROLL-NO-IC-NEXT:    [[TMP17:%.*]] = insertelement <2 x float> poison, float [[TMP15]], i32 0
1145; UNROLL-NO-IC-NEXT:    [[TMP18:%.*]] = insertelement <2 x float> [[TMP17]], float [[TMP16]], i32 1
1146; UNROLL-NO-IC-NEXT:    [[TMP19:%.*]] = getelementptr inbounds float, float* [[B:%.*]], i64 [[TMP3]]
1147; UNROLL-NO-IC-NEXT:    [[TMP20:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[TMP4]]
1148; UNROLL-NO-IC-NEXT:    [[TMP21:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[TMP5]]
1149; UNROLL-NO-IC-NEXT:    [[TMP22:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[TMP6]]
1150; UNROLL-NO-IC-NEXT:    [[TMP23:%.*]] = load float, float* [[TMP19]], align 4
1151; UNROLL-NO-IC-NEXT:    [[TMP24:%.*]] = load float, float* [[TMP20]], align 4
1152; UNROLL-NO-IC-NEXT:    [[TMP25:%.*]] = insertelement <2 x float> poison, float [[TMP23]], i32 0
1153; UNROLL-NO-IC-NEXT:    [[TMP26:%.*]] = insertelement <2 x float> [[TMP25]], float [[TMP24]], i32 1
1154; UNROLL-NO-IC-NEXT:    [[TMP27:%.*]] = load float, float* [[TMP21]], align 4
1155; UNROLL-NO-IC-NEXT:    [[TMP28:%.*]] = load float, float* [[TMP22]], align 4
1156; UNROLL-NO-IC-NEXT:    [[TMP29:%.*]] = insertelement <2 x float> poison, float [[TMP27]], i32 0
1157; UNROLL-NO-IC-NEXT:    [[TMP30:%.*]] = insertelement <2 x float> [[TMP29]], float [[TMP28]], i32 1
1158; UNROLL-NO-IC-NEXT:    [[TMP31:%.*]] = fadd fast <2 x float> [[VEC_PHI]], <float 1.000000e+00, float 1.000000e+00>
1159; UNROLL-NO-IC-NEXT:    [[TMP32:%.*]] = fadd fast <2 x float> [[VEC_PHI1]], <float 1.000000e+00, float 1.000000e+00>
1160; UNROLL-NO-IC-NEXT:    [[TMP33:%.*]] = fadd fast <2 x float> [[TMP31]], [[TMP14]]
1161; UNROLL-NO-IC-NEXT:    [[TMP34:%.*]] = fadd fast <2 x float> [[TMP32]], [[TMP18]]
1162; UNROLL-NO-IC-NEXT:    [[TMP35]] = fadd fast <2 x float> [[TMP33]], [[TMP26]]
1163; UNROLL-NO-IC-NEXT:    [[TMP36]] = fadd fast <2 x float> [[TMP34]], [[TMP30]]
1164; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
1165; UNROLL-NO-IC-NEXT:    [[TMP37:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1166; UNROLL-NO-IC-NEXT:    br i1 [[TMP37]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]]
1167; UNROLL-NO-IC:       middle.block:
1168; UNROLL-NO-IC-NEXT:    [[BIN_RDX:%.*]] = fadd fast <2 x float> [[TMP36]], [[TMP35]]
1169; UNROLL-NO-IC-NEXT:    [[TMP38:%.*]] = call fast float @llvm.vector.reduce.fadd.v2f32(float -0.000000e+00, <2 x float> [[BIN_RDX]])
1170; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
1171; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
1172; UNROLL-NO-IC:       scalar.ph:
1173; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
1174; UNROLL-NO-IC-NEXT:    [[BC_MERGE_RDX:%.*]] = phi float [ 0.000000e+00, [[ENTRY]] ], [ [[TMP38]], [[MIDDLE_BLOCK]] ]
1175; UNROLL-NO-IC-NEXT:    br label [[FOR_BODY:%.*]]
1176; UNROLL-NO-IC:       for.body:
1177; UNROLL-NO-IC-NEXT:    [[I:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[FOR_BODY]] ]
1178; UNROLL-NO-IC-NEXT:    [[S:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[TMP45:%.*]], [[FOR_BODY]] ]
1179; UNROLL-NO-IC-NEXT:    [[TMP39:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[I]]
1180; UNROLL-NO-IC-NEXT:    [[TMP40:%.*]] = load float, float* [[TMP39]], align 4
1181; UNROLL-NO-IC-NEXT:    [[TMP41:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[I]]
1182; UNROLL-NO-IC-NEXT:    [[TMP42:%.*]] = load float, float* [[TMP41]], align 4
1183; UNROLL-NO-IC-NEXT:    [[TMP43:%.*]] = fadd fast float [[S]], 1.000000e+00
1184; UNROLL-NO-IC-NEXT:    [[TMP44:%.*]] = fadd fast float [[TMP43]], [[TMP40]]
1185; UNROLL-NO-IC-NEXT:    [[TMP45]] = fadd fast float [[TMP44]], [[TMP42]]
1186; UNROLL-NO-IC-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 8
1187; UNROLL-NO-IC-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
1188; UNROLL-NO-IC-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP14:![0-9]+]]
1189; UNROLL-NO-IC:       for.end:
1190; UNROLL-NO-IC-NEXT:    [[S_LCSSA:%.*]] = phi float [ [[TMP45]], [[FOR_BODY]] ], [ [[TMP38]], [[MIDDLE_BLOCK]] ]
1191; UNROLL-NO-IC-NEXT:    ret float [[S_LCSSA]]
1192;
1193; INTERLEAVE-LABEL: @scalarize_induction_variable_02(
1194; INTERLEAVE-NEXT:  entry:
1195; INTERLEAVE-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 8)
1196; INTERLEAVE-NEXT:    [[TMP0:%.*]] = add nsw i64 [[SMAX]], -1
1197; INTERLEAVE-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 64
1198; INTERLEAVE-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
1199; INTERLEAVE:       vector.ph:
1200; INTERLEAVE-NEXT:    [[TMP1:%.*]] = lshr i64 [[TMP0]], 3
1201; INTERLEAVE-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
1202; INTERLEAVE-NEXT:    [[N_MOD_VF:%.*]] = and i64 [[TMP2]], 7
1203; INTERLEAVE-NEXT:    [[TMP3:%.*]] = icmp eq i64 [[N_MOD_VF]], 0
1204; INTERLEAVE-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], i64 8, i64 [[N_MOD_VF]]
1205; INTERLEAVE-NEXT:    [[N_VEC:%.*]] = sub nsw i64 [[TMP2]], [[TMP4]]
1206; INTERLEAVE-NEXT:    [[IND_END:%.*]] = shl i64 [[N_VEC]], 3
1207; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
1208; INTERLEAVE:       vector.body:
1209; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
1210; INTERLEAVE-NEXT:    [[VEC_PHI:%.*]] = phi <4 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP18:%.*]], [[VECTOR_BODY]] ]
1211; INTERLEAVE-NEXT:    [[VEC_PHI1:%.*]] = phi <4 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP19:%.*]], [[VECTOR_BODY]] ]
1212; INTERLEAVE-NEXT:    [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 3
1213; INTERLEAVE-NEXT:    [[TMP5:%.*]] = or i64 [[OFFSET_IDX]], 32
1214; INTERLEAVE-NEXT:    [[TMP6:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[OFFSET_IDX]]
1215; INTERLEAVE-NEXT:    [[TMP7:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[TMP5]]
1216; INTERLEAVE-NEXT:    [[TMP8:%.*]] = bitcast float* [[TMP6]] to <32 x float>*
1217; INTERLEAVE-NEXT:    [[TMP9:%.*]] = bitcast float* [[TMP7]] to <32 x float>*
1218; INTERLEAVE-NEXT:    [[WIDE_VEC:%.*]] = load <32 x float>, <32 x float>* [[TMP8]], align 4
1219; INTERLEAVE-NEXT:    [[WIDE_VEC2:%.*]] = load <32 x float>, <32 x float>* [[TMP9]], align 4
1220; INTERLEAVE-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <32 x float> [[WIDE_VEC]], <32 x float> poison, <4 x i32> <i32 0, i32 8, i32 16, i32 24>
1221; INTERLEAVE-NEXT:    [[STRIDED_VEC3:%.*]] = shufflevector <32 x float> [[WIDE_VEC2]], <32 x float> poison, <4 x i32> <i32 0, i32 8, i32 16, i32 24>
1222; INTERLEAVE-NEXT:    [[TMP10:%.*]] = getelementptr inbounds float, float* [[B:%.*]], i64 [[OFFSET_IDX]]
1223; INTERLEAVE-NEXT:    [[TMP11:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[TMP5]]
1224; INTERLEAVE-NEXT:    [[TMP12:%.*]] = bitcast float* [[TMP10]] to <32 x float>*
1225; INTERLEAVE-NEXT:    [[TMP13:%.*]] = bitcast float* [[TMP11]] to <32 x float>*
1226; INTERLEAVE-NEXT:    [[WIDE_VEC4:%.*]] = load <32 x float>, <32 x float>* [[TMP12]], align 4
1227; INTERLEAVE-NEXT:    [[WIDE_VEC5:%.*]] = load <32 x float>, <32 x float>* [[TMP13]], align 4
1228; INTERLEAVE-NEXT:    [[STRIDED_VEC6:%.*]] = shufflevector <32 x float> [[WIDE_VEC4]], <32 x float> poison, <4 x i32> <i32 0, i32 8, i32 16, i32 24>
1229; INTERLEAVE-NEXT:    [[STRIDED_VEC7:%.*]] = shufflevector <32 x float> [[WIDE_VEC5]], <32 x float> poison, <4 x i32> <i32 0, i32 8, i32 16, i32 24>
1230; INTERLEAVE-NEXT:    [[TMP14:%.*]] = fadd fast <4 x float> [[VEC_PHI]], <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
1231; INTERLEAVE-NEXT:    [[TMP15:%.*]] = fadd fast <4 x float> [[VEC_PHI1]], <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
1232; INTERLEAVE-NEXT:    [[TMP16:%.*]] = fadd fast <4 x float> [[TMP14]], [[STRIDED_VEC]]
1233; INTERLEAVE-NEXT:    [[TMP17:%.*]] = fadd fast <4 x float> [[TMP15]], [[STRIDED_VEC3]]
1234; INTERLEAVE-NEXT:    [[TMP18]] = fadd fast <4 x float> [[TMP16]], [[STRIDED_VEC6]]
1235; INTERLEAVE-NEXT:    [[TMP19]] = fadd fast <4 x float> [[TMP17]], [[STRIDED_VEC7]]
1236; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
1237; INTERLEAVE-NEXT:    [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1238; INTERLEAVE-NEXT:    br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]]
1239; INTERLEAVE:       middle.block:
1240; INTERLEAVE-NEXT:    [[BIN_RDX:%.*]] = fadd fast <4 x float> [[TMP19]], [[TMP18]]
1241; INTERLEAVE-NEXT:    [[TMP21:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float -0.000000e+00, <4 x float> [[BIN_RDX]])
1242; INTERLEAVE-NEXT:    br label [[SCALAR_PH]]
1243; INTERLEAVE:       scalar.ph:
1244; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
1245; INTERLEAVE-NEXT:    [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP21]], [[MIDDLE_BLOCK]] ], [ 0.000000e+00, [[ENTRY]] ]
1246; INTERLEAVE-NEXT:    br label [[FOR_BODY:%.*]]
1247; INTERLEAVE:       for.body:
1248; INTERLEAVE-NEXT:    [[I:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[FOR_BODY]] ]
1249; INTERLEAVE-NEXT:    [[S:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[TMP28:%.*]], [[FOR_BODY]] ]
1250; INTERLEAVE-NEXT:    [[TMP22:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[I]]
1251; INTERLEAVE-NEXT:    [[TMP23:%.*]] = load float, float* [[TMP22]], align 4
1252; INTERLEAVE-NEXT:    [[TMP24:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[I]]
1253; INTERLEAVE-NEXT:    [[TMP25:%.*]] = load float, float* [[TMP24]], align 4
1254; INTERLEAVE-NEXT:    [[TMP26:%.*]] = fadd fast float [[S]], 1.000000e+00
1255; INTERLEAVE-NEXT:    [[TMP27:%.*]] = fadd fast float [[TMP26]], [[TMP23]]
1256; INTERLEAVE-NEXT:    [[TMP28]] = fadd fast float [[TMP27]], [[TMP25]]
1257; INTERLEAVE-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 8
1258; INTERLEAVE-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
1259; INTERLEAVE-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END:%.*]], !llvm.loop [[LOOP14:![0-9]+]]
1260; INTERLEAVE:       for.end:
1261; INTERLEAVE-NEXT:    ret float [[TMP28]]
1262;
1263entry:
1264  br label %for.body
1265
1266for.body:
1267  %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
1268  %s = phi float [ 0.0, %entry ], [ %6, %for.body ]
1269  %0 = getelementptr inbounds float, float* %a, i64 %i
1270  %1 = load float, float* %0, align 4
1271  %2 = getelementptr inbounds float, float* %b, i64 %i
1272  %3 = load float, float* %2, align 4
1273  %4 = fadd fast float %s, 1.0
1274  %5 = fadd fast float %4, %1
1275  %6 = fadd fast float %5, %3
1276  %i.next = add nuw nsw i64 %i, 8
1277  %cond = icmp slt i64 %i.next, %n
1278  br i1 %cond, label %for.body, label %for.end
1279
1280for.end:
1281  %s.lcssa = phi float [ %6, %for.body ]
1282  ret float %s.lcssa
1283}
1284
1285; Make sure we scalarize the step vectors used for the pointer arithmetic. We
1286; can't easily simplify vectorized step vectors. (Interleaved accesses.)
1287;
1288; for (int i = 0; i < n; ++i)
1289;   a[i].f ^= y;
1290;
1291
1292%pair.i32 = type { i32, i32 }
1293define void @scalarize_induction_variable_03(%pair.i32 *%p, i32 %y, i64 %n) {
1294; CHECK-LABEL: @scalarize_induction_variable_03(
1295; CHECK-NEXT:  entry:
1296; CHECK-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)
1297; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 2
1298; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
1299; CHECK:       vector.ph:
1300; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[SMAX]], 2
1301; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[SMAX]], [[N_MOD_VF]]
1302; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[Y:%.*]], i32 0
1303; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
1304; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
1305; CHECK:       vector.body:
1306; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
1307; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
1308; CHECK-NEXT:    [[TMP1:%.*]] = add i64 [[INDEX]], 1
1309; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], %pair.i32* [[P:%.*]], i64 [[TMP0]], i32 1
1310; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP1]], i32 1
1311; CHECK-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP2]], align 8
1312; CHECK-NEXT:    [[TMP5:%.*]] = load i32, i32* [[TMP3]], align 8
1313; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <2 x i32> poison, i32 [[TMP4]], i32 0
1314; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x i32> [[TMP6]], i32 [[TMP5]], i32 1
1315; CHECK-NEXT:    [[TMP8:%.*]] = xor <2 x i32> [[TMP7]], [[BROADCAST_SPLAT]]
1316; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <2 x i32> [[TMP8]], i32 0
1317; CHECK-NEXT:    store i32 [[TMP9]], i32* [[TMP2]], align 8
1318; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <2 x i32> [[TMP8]], i32 1
1319; CHECK-NEXT:    store i32 [[TMP10]], i32* [[TMP3]], align 8
1320; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
1321; CHECK-NEXT:    [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1322; CHECK-NEXT:    br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]
1323; CHECK:       middle.block:
1324; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
1325; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
1326; CHECK:       scalar.ph:
1327; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
1328; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
1329; CHECK:       for.body:
1330; CHECK-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
1331; CHECK-NEXT:    [[F:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[I]], i32 1
1332; CHECK-NEXT:    [[TMP12:%.*]] = load i32, i32* [[F]], align 8
1333; CHECK-NEXT:    [[TMP13:%.*]] = xor i32 [[TMP12]], [[Y]]
1334; CHECK-NEXT:    store i32 [[TMP13]], i32* [[F]], align 8
1335; CHECK-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
1336; CHECK-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
1337; CHECK-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP16:![0-9]+]]
1338; CHECK:       for.end:
1339; CHECK-NEXT:    ret void
1340;
1341; IND-LABEL: @scalarize_induction_variable_03(
1342; IND-NEXT:  entry:
1343; IND-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)
1344; IND-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 2
1345; IND-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
1346; IND:       vector.ph:
1347; IND-NEXT:    [[N_VEC:%.*]] = and i64 [[SMAX]], 9223372036854775806
1348; IND-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[Y:%.*]], i64 0
1349; IND-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
1350; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
1351; IND:       vector.body:
1352; IND-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
1353; IND-NEXT:    [[TMP0:%.*]] = or i64 [[INDEX]], 1
1354; IND-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], %pair.i32* [[P:%.*]], i64 [[INDEX]], i32 1
1355; IND-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP0]], i32 1
1356; IND-NEXT:    [[TMP3:%.*]] = load i32, i32* [[TMP1]], align 8
1357; IND-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP2]], align 8
1358; IND-NEXT:    [[TMP5:%.*]] = insertelement <2 x i32> poison, i32 [[TMP3]], i64 0
1359; IND-NEXT:    [[TMP6:%.*]] = insertelement <2 x i32> [[TMP5]], i32 [[TMP4]], i64 1
1360; IND-NEXT:    [[TMP7:%.*]] = xor <2 x i32> [[TMP6]], [[BROADCAST_SPLAT]]
1361; IND-NEXT:    [[TMP8:%.*]] = extractelement <2 x i32> [[TMP7]], i64 0
1362; IND-NEXT:    store i32 [[TMP8]], i32* [[TMP1]], align 8
1363; IND-NEXT:    [[TMP9:%.*]] = extractelement <2 x i32> [[TMP7]], i64 1
1364; IND-NEXT:    store i32 [[TMP9]], i32* [[TMP2]], align 8
1365; IND-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
1366; IND-NEXT:    [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1367; IND-NEXT:    br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]
1368; IND:       middle.block:
1369; IND-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
1370; IND-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
1371; IND:       scalar.ph:
1372; IND-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
1373; IND-NEXT:    br label [[FOR_BODY:%.*]]
1374; IND:       for.body:
1375; IND-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
1376; IND-NEXT:    [[F:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[I]], i32 1
1377; IND-NEXT:    [[TMP11:%.*]] = load i32, i32* [[F]], align 8
1378; IND-NEXT:    [[TMP12:%.*]] = xor i32 [[TMP11]], [[Y]]
1379; IND-NEXT:    store i32 [[TMP12]], i32* [[F]], align 8
1380; IND-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
1381; IND-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
1382; IND-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP16:![0-9]+]]
1383; IND:       for.end:
1384; IND-NEXT:    ret void
1385;
1386; UNROLL-LABEL: @scalarize_induction_variable_03(
1387; UNROLL-NEXT:  entry:
1388; UNROLL-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)
1389; UNROLL-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 4
1390; UNROLL-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
1391; UNROLL:       vector.ph:
1392; UNROLL-NEXT:    [[N_VEC:%.*]] = and i64 [[SMAX]], 9223372036854775804
1393; UNROLL-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[Y:%.*]], i64 0
1394; UNROLL-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
1395; UNROLL-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <2 x i32> poison, i32 [[Y]], i64 0
1396; UNROLL-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT1]], <2 x i32> poison, <2 x i32> zeroinitializer
1397; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
1398; UNROLL:       vector.body:
1399; UNROLL-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
1400; UNROLL-NEXT:    [[TMP0:%.*]] = or i64 [[INDEX]], 1
1401; UNROLL-NEXT:    [[TMP1:%.*]] = or i64 [[INDEX]], 2
1402; UNROLL-NEXT:    [[TMP2:%.*]] = or i64 [[INDEX]], 3
1403; UNROLL-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], %pair.i32* [[P:%.*]], i64 [[INDEX]], i32 1
1404; UNROLL-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP0]], i32 1
1405; UNROLL-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP1]], i32 1
1406; UNROLL-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP2]], i32 1
1407; UNROLL-NEXT:    [[TMP7:%.*]] = load i32, i32* [[TMP3]], align 8
1408; UNROLL-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP4]], align 8
1409; UNROLL-NEXT:    [[TMP9:%.*]] = insertelement <2 x i32> poison, i32 [[TMP7]], i64 0
1410; UNROLL-NEXT:    [[TMP10:%.*]] = insertelement <2 x i32> [[TMP9]], i32 [[TMP8]], i64 1
1411; UNROLL-NEXT:    [[TMP11:%.*]] = load i32, i32* [[TMP5]], align 8
1412; UNROLL-NEXT:    [[TMP12:%.*]] = load i32, i32* [[TMP6]], align 8
1413; UNROLL-NEXT:    [[TMP13:%.*]] = insertelement <2 x i32> poison, i32 [[TMP11]], i64 0
1414; UNROLL-NEXT:    [[TMP14:%.*]] = insertelement <2 x i32> [[TMP13]], i32 [[TMP12]], i64 1
1415; UNROLL-NEXT:    [[TMP15:%.*]] = xor <2 x i32> [[TMP10]], [[BROADCAST_SPLAT]]
1416; UNROLL-NEXT:    [[TMP16:%.*]] = xor <2 x i32> [[TMP14]], [[BROADCAST_SPLAT2]]
1417; UNROLL-NEXT:    [[TMP17:%.*]] = extractelement <2 x i32> [[TMP15]], i64 0
1418; UNROLL-NEXT:    store i32 [[TMP17]], i32* [[TMP3]], align 8
1419; UNROLL-NEXT:    [[TMP18:%.*]] = extractelement <2 x i32> [[TMP15]], i64 1
1420; UNROLL-NEXT:    store i32 [[TMP18]], i32* [[TMP4]], align 8
1421; UNROLL-NEXT:    [[TMP19:%.*]] = extractelement <2 x i32> [[TMP16]], i64 0
1422; UNROLL-NEXT:    store i32 [[TMP19]], i32* [[TMP5]], align 8
1423; UNROLL-NEXT:    [[TMP20:%.*]] = extractelement <2 x i32> [[TMP16]], i64 1
1424; UNROLL-NEXT:    store i32 [[TMP20]], i32* [[TMP6]], align 8
1425; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
1426; UNROLL-NEXT:    [[TMP21:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1427; UNROLL-NEXT:    br i1 [[TMP21]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]
1428; UNROLL:       middle.block:
1429; UNROLL-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
1430; UNROLL-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
1431; UNROLL:       scalar.ph:
1432; UNROLL-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
1433; UNROLL-NEXT:    br label [[FOR_BODY:%.*]]
1434; UNROLL:       for.body:
1435; UNROLL-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
1436; UNROLL-NEXT:    [[F:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[I]], i32 1
1437; UNROLL-NEXT:    [[TMP22:%.*]] = load i32, i32* [[F]], align 8
1438; UNROLL-NEXT:    [[TMP23:%.*]] = xor i32 [[TMP22]], [[Y]]
1439; UNROLL-NEXT:    store i32 [[TMP23]], i32* [[F]], align 8
1440; UNROLL-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
1441; UNROLL-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
1442; UNROLL-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP16:![0-9]+]]
1443; UNROLL:       for.end:
1444; UNROLL-NEXT:    ret void
1445;
1446; UNROLL-NO-IC-LABEL: @scalarize_induction_variable_03(
1447; UNROLL-NO-IC-NEXT:  entry:
1448; UNROLL-NO-IC-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)
1449; UNROLL-NO-IC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 4
1450; UNROLL-NO-IC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
1451; UNROLL-NO-IC:       vector.ph:
1452; UNROLL-NO-IC-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[SMAX]], 4
1453; UNROLL-NO-IC-NEXT:    [[N_VEC:%.*]] = sub i64 [[SMAX]], [[N_MOD_VF]]
1454; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[Y:%.*]], i32 0
1455; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
1456; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <2 x i32> poison, i32 [[Y]], i32 0
1457; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT1]], <2 x i32> poison, <2 x i32> zeroinitializer
1458; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
1459; UNROLL-NO-IC:       vector.body:
1460; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
1461; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
1462; UNROLL-NO-IC-NEXT:    [[TMP1:%.*]] = add i64 [[INDEX]], 1
1463; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = add i64 [[INDEX]], 2
1464; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = add i64 [[INDEX]], 3
1465; UNROLL-NO-IC-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], %pair.i32* [[P:%.*]], i64 [[TMP0]], i32 1
1466; UNROLL-NO-IC-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP1]], i32 1
1467; UNROLL-NO-IC-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP2]], i32 1
1468; UNROLL-NO-IC-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP3]], i32 1
1469; UNROLL-NO-IC-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP4]], align 8
1470; UNROLL-NO-IC-NEXT:    [[TMP9:%.*]] = load i32, i32* [[TMP5]], align 8
1471; UNROLL-NO-IC-NEXT:    [[TMP10:%.*]] = insertelement <2 x i32> poison, i32 [[TMP8]], i32 0
1472; UNROLL-NO-IC-NEXT:    [[TMP11:%.*]] = insertelement <2 x i32> [[TMP10]], i32 [[TMP9]], i32 1
1473; UNROLL-NO-IC-NEXT:    [[TMP12:%.*]] = load i32, i32* [[TMP6]], align 8
1474; UNROLL-NO-IC-NEXT:    [[TMP13:%.*]] = load i32, i32* [[TMP7]], align 8
1475; UNROLL-NO-IC-NEXT:    [[TMP14:%.*]] = insertelement <2 x i32> poison, i32 [[TMP12]], i32 0
1476; UNROLL-NO-IC-NEXT:    [[TMP15:%.*]] = insertelement <2 x i32> [[TMP14]], i32 [[TMP13]], i32 1
1477; UNROLL-NO-IC-NEXT:    [[TMP16:%.*]] = xor <2 x i32> [[TMP11]], [[BROADCAST_SPLAT]]
1478; UNROLL-NO-IC-NEXT:    [[TMP17:%.*]] = xor <2 x i32> [[TMP15]], [[BROADCAST_SPLAT2]]
1479; UNROLL-NO-IC-NEXT:    [[TMP18:%.*]] = extractelement <2 x i32> [[TMP16]], i32 0
1480; UNROLL-NO-IC-NEXT:    store i32 [[TMP18]], i32* [[TMP4]], align 8
1481; UNROLL-NO-IC-NEXT:    [[TMP19:%.*]] = extractelement <2 x i32> [[TMP16]], i32 1
1482; UNROLL-NO-IC-NEXT:    store i32 [[TMP19]], i32* [[TMP5]], align 8
1483; UNROLL-NO-IC-NEXT:    [[TMP20:%.*]] = extractelement <2 x i32> [[TMP17]], i32 0
1484; UNROLL-NO-IC-NEXT:    store i32 [[TMP20]], i32* [[TMP6]], align 8
1485; UNROLL-NO-IC-NEXT:    [[TMP21:%.*]] = extractelement <2 x i32> [[TMP17]], i32 1
1486; UNROLL-NO-IC-NEXT:    store i32 [[TMP21]], i32* [[TMP7]], align 8
1487; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
1488; UNROLL-NO-IC-NEXT:    [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1489; UNROLL-NO-IC-NEXT:    br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]
1490; UNROLL-NO-IC:       middle.block:
1491; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
1492; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
1493; UNROLL-NO-IC:       scalar.ph:
1494; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
1495; UNROLL-NO-IC-NEXT:    br label [[FOR_BODY:%.*]]
1496; UNROLL-NO-IC:       for.body:
1497; UNROLL-NO-IC-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
1498; UNROLL-NO-IC-NEXT:    [[F:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[I]], i32 1
1499; UNROLL-NO-IC-NEXT:    [[TMP23:%.*]] = load i32, i32* [[F]], align 8
1500; UNROLL-NO-IC-NEXT:    [[TMP24:%.*]] = xor i32 [[TMP23]], [[Y]]
1501; UNROLL-NO-IC-NEXT:    store i32 [[TMP24]], i32* [[F]], align 8
1502; UNROLL-NO-IC-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
1503; UNROLL-NO-IC-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
1504; UNROLL-NO-IC-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP16:![0-9]+]]
1505; UNROLL-NO-IC:       for.end:
1506; UNROLL-NO-IC-NEXT:    ret void
1507;
1508; INTERLEAVE-LABEL: @scalarize_induction_variable_03(
1509; INTERLEAVE-NEXT:  entry:
1510; INTERLEAVE-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)
1511; INTERLEAVE-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 9
1512; INTERLEAVE-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
1513; INTERLEAVE:       vector.ph:
1514; INTERLEAVE-NEXT:    [[N_MOD_VF:%.*]] = and i64 [[SMAX]], 7
1515; INTERLEAVE-NEXT:    [[TMP0:%.*]] = icmp eq i64 [[N_MOD_VF]], 0
1516; INTERLEAVE-NEXT:    [[TMP1:%.*]] = select i1 [[TMP0]], i64 8, i64 [[N_MOD_VF]]
1517; INTERLEAVE-NEXT:    [[N_VEC:%.*]] = sub nsw i64 [[SMAX]], [[TMP1]]
1518; INTERLEAVE-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[Y:%.*]], i64 0
1519; INTERLEAVE-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer
1520; INTERLEAVE-NEXT:    [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <4 x i32> poison, i32 [[Y]], i64 0
1521; INTERLEAVE-NEXT:    [[BROADCAST_SPLAT4:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT3]], <4 x i32> poison, <4 x i32> zeroinitializer
1522; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
1523; INTERLEAVE:       vector.body:
1524; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
1525; INTERLEAVE-NEXT:    [[TMP2:%.*]] = or i64 [[INDEX]], 1
1526; INTERLEAVE-NEXT:    [[TMP3:%.*]] = or i64 [[INDEX]], 2
1527; INTERLEAVE-NEXT:    [[TMP4:%.*]] = or i64 [[INDEX]], 3
1528; INTERLEAVE-NEXT:    [[TMP5:%.*]] = or i64 [[INDEX]], 4
1529; INTERLEAVE-NEXT:    [[TMP6:%.*]] = or i64 [[INDEX]], 5
1530; INTERLEAVE-NEXT:    [[TMP7:%.*]] = or i64 [[INDEX]], 6
1531; INTERLEAVE-NEXT:    [[TMP8:%.*]] = or i64 [[INDEX]], 7
1532; INTERLEAVE-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], %pair.i32* [[P:%.*]], i64 [[INDEX]], i32 1
1533; INTERLEAVE-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP2]], i32 1
1534; INTERLEAVE-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP3]], i32 1
1535; INTERLEAVE-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP4]], i32 1
1536; INTERLEAVE-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP5]], i32 1
1537; INTERLEAVE-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP6]], i32 1
1538; INTERLEAVE-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP7]], i32 1
1539; INTERLEAVE-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP8]], i32 1
1540; INTERLEAVE-NEXT:    [[TMP17:%.*]] = bitcast i32* [[TMP9]] to <8 x i32>*
1541; INTERLEAVE-NEXT:    [[TMP18:%.*]] = bitcast i32* [[TMP13]] to <8 x i32>*
1542; INTERLEAVE-NEXT:    [[WIDE_VEC:%.*]] = load <8 x i32>, <8 x i32>* [[TMP17]], align 8
1543; INTERLEAVE-NEXT:    [[WIDE_VEC1:%.*]] = load <8 x i32>, <8 x i32>* [[TMP18]], align 8
1544; INTERLEAVE-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
1545; INTERLEAVE-NEXT:    [[STRIDED_VEC2:%.*]] = shufflevector <8 x i32> [[WIDE_VEC1]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
1546; INTERLEAVE-NEXT:    [[TMP19:%.*]] = xor <4 x i32> [[STRIDED_VEC]], [[BROADCAST_SPLAT]]
1547; INTERLEAVE-NEXT:    [[TMP20:%.*]] = xor <4 x i32> [[STRIDED_VEC2]], [[BROADCAST_SPLAT4]]
1548; INTERLEAVE-NEXT:    [[TMP21:%.*]] = extractelement <4 x i32> [[TMP19]], i64 0
1549; INTERLEAVE-NEXT:    store i32 [[TMP21]], i32* [[TMP9]], align 8
1550; INTERLEAVE-NEXT:    [[TMP22:%.*]] = extractelement <4 x i32> [[TMP19]], i64 1
1551; INTERLEAVE-NEXT:    store i32 [[TMP22]], i32* [[TMP10]], align 8
1552; INTERLEAVE-NEXT:    [[TMP23:%.*]] = extractelement <4 x i32> [[TMP19]], i64 2
1553; INTERLEAVE-NEXT:    store i32 [[TMP23]], i32* [[TMP11]], align 8
1554; INTERLEAVE-NEXT:    [[TMP24:%.*]] = extractelement <4 x i32> [[TMP19]], i64 3
1555; INTERLEAVE-NEXT:    store i32 [[TMP24]], i32* [[TMP12]], align 8
1556; INTERLEAVE-NEXT:    [[TMP25:%.*]] = extractelement <4 x i32> [[TMP20]], i64 0
1557; INTERLEAVE-NEXT:    store i32 [[TMP25]], i32* [[TMP13]], align 8
1558; INTERLEAVE-NEXT:    [[TMP26:%.*]] = extractelement <4 x i32> [[TMP20]], i64 1
1559; INTERLEAVE-NEXT:    store i32 [[TMP26]], i32* [[TMP14]], align 8
1560; INTERLEAVE-NEXT:    [[TMP27:%.*]] = extractelement <4 x i32> [[TMP20]], i64 2
1561; INTERLEAVE-NEXT:    store i32 [[TMP27]], i32* [[TMP15]], align 8
1562; INTERLEAVE-NEXT:    [[TMP28:%.*]] = extractelement <4 x i32> [[TMP20]], i64 3
1563; INTERLEAVE-NEXT:    store i32 [[TMP28]], i32* [[TMP16]], align 8
1564; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
1565; INTERLEAVE-NEXT:    [[TMP29:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1566; INTERLEAVE-NEXT:    br i1 [[TMP29]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]
1567; INTERLEAVE:       middle.block:
1568; INTERLEAVE-NEXT:    br label [[SCALAR_PH]]
1569; INTERLEAVE:       scalar.ph:
1570; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
1571; INTERLEAVE-NEXT:    br label [[FOR_BODY:%.*]]
1572; INTERLEAVE:       for.body:
1573; INTERLEAVE-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
1574; INTERLEAVE-NEXT:    [[F:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[I]], i32 1
1575; INTERLEAVE-NEXT:    [[TMP30:%.*]] = load i32, i32* [[F]], align 8
1576; INTERLEAVE-NEXT:    [[TMP31:%.*]] = xor i32 [[TMP30]], [[Y]]
1577; INTERLEAVE-NEXT:    store i32 [[TMP31]], i32* [[F]], align 8
1578; INTERLEAVE-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
1579; INTERLEAVE-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
1580; INTERLEAVE-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END:%.*]], !llvm.loop [[LOOP16:![0-9]+]]
1581; INTERLEAVE:       for.end:
1582; INTERLEAVE-NEXT:    ret void
1583;
1584entry:
1585  br label %for.body
1586
1587for.body:
1588  %i  = phi i64 [ %i.next, %for.body ], [ 0, %entry ]
1589  %f = getelementptr inbounds %pair.i32, %pair.i32* %p, i64 %i, i32 1
1590  %0 = load i32, i32* %f, align 8
1591  %1 = xor i32 %0, %y
1592  store i32 %1, i32* %f, align 8
1593  %i.next = add nuw nsw i64 %i, 1
1594  %cond = icmp slt i64 %i.next, %n
1595  br i1 %cond, label %for.body, label %for.end
1596
1597for.end:
1598  ret void
1599}
1600
1601; Make sure we scalarize the step vectors used for the pointer arithmetic. We
1602; can't easily simplify vectorized step vectors. (Interleaved accesses.)
1603;
1604; for (int i = 0; i < n; ++i)
1605;   p[i].f = a[i * 4]
1606;
1607
1608define void @scalarize_induction_variable_04(i32* %a, %pair.i32* %p, i32 %n) {
1609; CHECK-LABEL: @scalarize_induction_variable_04(
1610; CHECK-NEXT:  entry:
1611; CHECK-NEXT:    [[A4:%.*]] = bitcast i32* [[A:%.*]] to i8*
1612; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[N:%.*]], -1
1613; CHECK-NEXT:    [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
1614; CHECK-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
1615; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 2
1616; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
1617; CHECK:       vector.memcheck:
1618; CHECK-NEXT:    [[SCEVGEP:%.*]] = getelementptr [[PAIR_I32:%.*]], %pair.i32* [[P:%.*]], i64 0, i32 1
1619; CHECK-NEXT:    [[SCEVGEP1:%.*]] = bitcast i32* [[SCEVGEP]] to i8*
1620; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[N]], -1
1621; CHECK-NEXT:    [[TMP4:%.*]] = zext i32 [[TMP3]] to i64
1622; CHECK-NEXT:    [[TMP5:%.*]] = add nuw nsw i64 [[TMP4]], 1
1623; CHECK-NEXT:    [[SCEVGEP2:%.*]] = getelementptr [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP5]]
1624; CHECK-NEXT:    [[SCEVGEP23:%.*]] = bitcast %pair.i32* [[SCEVGEP2]] to i8*
1625; CHECK-NEXT:    [[TMP6:%.*]] = shl nuw nsw i64 [[TMP4]], 2
1626; CHECK-NEXT:    [[TMP7:%.*]] = add nuw nsw i64 [[TMP6]], 1
1627; CHECK-NEXT:    [[SCEVGEP5:%.*]] = getelementptr i32, i32* [[A]], i64 [[TMP7]]
1628; CHECK-NEXT:    [[SCEVGEP56:%.*]] = bitcast i32* [[SCEVGEP5]] to i8*
1629; CHECK-NEXT:    [[BOUND0:%.*]] = icmp ult i8* [[SCEVGEP1]], [[SCEVGEP56]]
1630; CHECK-NEXT:    [[BOUND1:%.*]] = icmp ult i8* [[A4]], [[SCEVGEP23]]
1631; CHECK-NEXT:    [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
1632; CHECK-NEXT:    br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
1633; CHECK:       vector.ph:
1634; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 2
1635; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]]
1636; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
1637; CHECK:       vector.body:
1638; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
1639; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
1640; CHECK-NEXT:    [[TMP8:%.*]] = add i64 [[INDEX]], 0
1641; CHECK-NEXT:    [[TMP9:%.*]] = add i64 [[INDEX]], 1
1642; CHECK-NEXT:    [[TMP10:%.*]] = shl nsw <2 x i64> [[VEC_IND]], <i64 2, i64 2>
1643; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <2 x i64> [[TMP10]], i32 0
1644; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP11]]
1645; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <2 x i64> [[TMP10]], i32 1
1646; CHECK-NEXT:    [[TMP14:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP13]]
1647; CHECK-NEXT:    [[TMP15:%.*]] = load i32, i32* [[TMP12]], align 1, !alias.scope !17
1648; CHECK-NEXT:    [[TMP16:%.*]] = load i32, i32* [[TMP14]], align 1, !alias.scope !17
1649; CHECK-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP8]], i32 1
1650; CHECK-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP9]], i32 1
1651; CHECK-NEXT:    store i32 [[TMP15]], i32* [[TMP17]], align 1, !alias.scope !20, !noalias !17
1652; CHECK-NEXT:    store i32 [[TMP16]], i32* [[TMP18]], align 1, !alias.scope !20, !noalias !17
1653; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
1654; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 2, i64 2>
1655; CHECK-NEXT:    [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1656; CHECK-NEXT:    br i1 [[TMP19]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]]
1657; CHECK:       middle.block:
1658; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
1659; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
1660; CHECK:       scalar.ph:
1661; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_MEMCHECK]] ]
1662; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
1663; CHECK:       for.body:
1664; CHECK-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
1665; CHECK-NEXT:    [[TMP20:%.*]] = shl nsw i64 [[I]], 2
1666; CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP20]]
1667; CHECK-NEXT:    [[TMP22:%.*]] = load i32, i32* [[TMP21]], align 1
1668; CHECK-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[I]], i32 1
1669; CHECK-NEXT:    store i32 [[TMP22]], i32* [[TMP23]], align 1
1670; CHECK-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
1671; CHECK-NEXT:    [[TMP24:%.*]] = trunc i64 [[I_NEXT]] to i32
1672; CHECK-NEXT:    [[COND:%.*]] = icmp eq i32 [[TMP24]], [[N]]
1673; CHECK-NEXT:    br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP23:![0-9]+]]
1674; CHECK:       for.end:
1675; CHECK-NEXT:    ret void
1676;
1677; IND-LABEL: @scalarize_induction_variable_04(
1678; IND-NEXT:  entry:
1679; IND-NEXT:    [[TMP0:%.*]] = add i32 [[N:%.*]], -1
1680; IND-NEXT:    [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
1681; IND-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
1682; IND-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp eq i32 [[TMP0]], 0
1683; IND-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
1684; IND:       vector.memcheck:
1685; IND-NEXT:    [[SCEVGEP:%.*]] = getelementptr [[PAIR_I32:%.*]], %pair.i32* [[P:%.*]], i64 0, i32 1
1686; IND-NEXT:    [[TMP3:%.*]] = add i32 [[N]], -1
1687; IND-NEXT:    [[TMP4:%.*]] = zext i32 [[TMP3]] to i64
1688; IND-NEXT:    [[TMP5:%.*]] = add nuw nsw i64 [[TMP4]], 1
1689; IND-NEXT:    [[TMP6:%.*]] = shl nuw nsw i64 [[TMP4]], 2
1690; IND-NEXT:    [[TMP7:%.*]] = or i64 [[TMP6]], 1
1691; IND-NEXT:    [[SCEVGEP5:%.*]] = getelementptr i32, i32* [[A:%.*]], i64 [[TMP7]]
1692; IND-NEXT:    [[BOUND0:%.*]] = icmp ult i32* [[SCEVGEP]], [[SCEVGEP5]]
1693; IND-NEXT:    [[TMP8:%.*]] = getelementptr [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP5]], i32 0
1694; IND-NEXT:    [[BOUND1:%.*]] = icmp ugt i32* [[TMP8]], [[A]]
1695; IND-NEXT:    [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
1696; IND-NEXT:    br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
1697; IND:       vector.ph:
1698; IND-NEXT:    [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934590
1699; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
1700; IND:       vector.body:
1701; IND-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
1702; IND-NEXT:    [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
1703; IND-NEXT:    [[TMP9:%.*]] = or i64 [[INDEX]], 1
1704; IND-NEXT:    [[TMP10:%.*]] = shl nsw <2 x i64> [[VEC_IND]], <i64 2, i64 2>
1705; IND-NEXT:    [[TMP11:%.*]] = extractelement <2 x i64> [[TMP10]], i64 0
1706; IND-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP11]]
1707; IND-NEXT:    [[TMP13:%.*]] = extractelement <2 x i64> [[TMP10]], i64 1
1708; IND-NEXT:    [[TMP14:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP13]]
1709; IND-NEXT:    [[TMP15:%.*]] = load i32, i32* [[TMP12]], align 1, !alias.scope !17
1710; IND-NEXT:    [[TMP16:%.*]] = load i32, i32* [[TMP14]], align 1, !alias.scope !17
1711; IND-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[INDEX]], i32 1
1712; IND-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP9]], i32 1
1713; IND-NEXT:    store i32 [[TMP15]], i32* [[TMP17]], align 1, !alias.scope !20, !noalias !17
1714; IND-NEXT:    store i32 [[TMP16]], i32* [[TMP18]], align 1, !alias.scope !20, !noalias !17
1715; IND-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
1716; IND-NEXT:    [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 2, i64 2>
1717; IND-NEXT:    [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1718; IND-NEXT:    br i1 [[TMP19]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]]
1719; IND:       middle.block:
1720; IND-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
1721; IND-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
1722; IND:       scalar.ph:
1723; IND-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_MEMCHECK]] ]
1724; IND-NEXT:    br label [[FOR_BODY:%.*]]
1725; IND:       for.body:
1726; IND-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
1727; IND-NEXT:    [[TMP20:%.*]] = shl nsw i64 [[I]], 2
1728; IND-NEXT:    [[TMP21:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP20]]
1729; IND-NEXT:    [[TMP22:%.*]] = load i32, i32* [[TMP21]], align 1
1730; IND-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[I]], i32 1
1731; IND-NEXT:    store i32 [[TMP22]], i32* [[TMP23]], align 1
1732; IND-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
1733; IND-NEXT:    [[TMP24:%.*]] = trunc i64 [[I_NEXT]] to i32
1734; IND-NEXT:    [[COND:%.*]] = icmp eq i32 [[TMP24]], [[N]]
1735; IND-NEXT:    br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP23:![0-9]+]]
1736; IND:       for.end:
1737; IND-NEXT:    ret void
1738;
1739; UNROLL-LABEL: @scalarize_induction_variable_04(
1740; UNROLL-NEXT:  entry:
1741; UNROLL-NEXT:    [[TMP0:%.*]] = add i32 [[N:%.*]], -1
1742; UNROLL-NEXT:    [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
1743; UNROLL-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
1744; UNROLL-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 3
1745; UNROLL-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
1746; UNROLL:       vector.memcheck:
1747; UNROLL-NEXT:    [[SCEVGEP:%.*]] = getelementptr [[PAIR_I32:%.*]], %pair.i32* [[P:%.*]], i64 0, i32 1
1748; UNROLL-NEXT:    [[TMP3:%.*]] = add i32 [[N]], -1
1749; UNROLL-NEXT:    [[TMP4:%.*]] = zext i32 [[TMP3]] to i64
1750; UNROLL-NEXT:    [[TMP5:%.*]] = add nuw nsw i64 [[TMP4]], 1
1751; UNROLL-NEXT:    [[TMP6:%.*]] = shl nuw nsw i64 [[TMP4]], 2
1752; UNROLL-NEXT:    [[TMP7:%.*]] = or i64 [[TMP6]], 1
1753; UNROLL-NEXT:    [[SCEVGEP5:%.*]] = getelementptr i32, i32* [[A:%.*]], i64 [[TMP7]]
1754; UNROLL-NEXT:    [[BOUND0:%.*]] = icmp ult i32* [[SCEVGEP]], [[SCEVGEP5]]
1755; UNROLL-NEXT:    [[TMP8:%.*]] = getelementptr [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP5]], i32 0
1756; UNROLL-NEXT:    [[BOUND1:%.*]] = icmp ugt i32* [[TMP8]], [[A]]
1757; UNROLL-NEXT:    [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
1758; UNROLL-NEXT:    br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
1759; UNROLL:       vector.ph:
1760; UNROLL-NEXT:    [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934588
1761; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
1762; UNROLL:       vector.body:
1763; UNROLL-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
1764; UNROLL-NEXT:    [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
1765; UNROLL-NEXT:    [[TMP9:%.*]] = or i64 [[INDEX]], 1
1766; UNROLL-NEXT:    [[TMP10:%.*]] = or i64 [[INDEX]], 2
1767; UNROLL-NEXT:    [[TMP11:%.*]] = or i64 [[INDEX]], 3
1768; UNROLL-NEXT:    [[TMP12:%.*]] = shl nsw <2 x i64> [[VEC_IND]], <i64 2, i64 2>
1769; UNROLL-NEXT:    [[STEP_ADD:%.*]] = shl <2 x i64> [[VEC_IND]], <i64 2, i64 2>
1770; UNROLL-NEXT:    [[TMP13:%.*]] = add <2 x i64> [[STEP_ADD]], <i64 8, i64 8>
1771; UNROLL-NEXT:    [[TMP14:%.*]] = extractelement <2 x i64> [[TMP12]], i64 0
1772; UNROLL-NEXT:    [[TMP15:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP14]]
1773; UNROLL-NEXT:    [[TMP16:%.*]] = extractelement <2 x i64> [[TMP12]], i64 1
1774; UNROLL-NEXT:    [[TMP17:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP16]]
1775; UNROLL-NEXT:    [[TMP18:%.*]] = extractelement <2 x i64> [[TMP13]], i64 0
1776; UNROLL-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP18]]
1777; UNROLL-NEXT:    [[TMP20:%.*]] = extractelement <2 x i64> [[TMP13]], i64 1
1778; UNROLL-NEXT:    [[TMP21:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP20]]
1779; UNROLL-NEXT:    [[TMP22:%.*]] = load i32, i32* [[TMP15]], align 1, !alias.scope !17
1780; UNROLL-NEXT:    [[TMP23:%.*]] = load i32, i32* [[TMP17]], align 1, !alias.scope !17
1781; UNROLL-NEXT:    [[TMP24:%.*]] = load i32, i32* [[TMP19]], align 1, !alias.scope !17
1782; UNROLL-NEXT:    [[TMP25:%.*]] = load i32, i32* [[TMP21]], align 1, !alias.scope !17
1783; UNROLL-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[INDEX]], i32 1
1784; UNROLL-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP9]], i32 1
1785; UNROLL-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP10]], i32 1
1786; UNROLL-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP11]], i32 1
1787; UNROLL-NEXT:    store i32 [[TMP22]], i32* [[TMP26]], align 1, !alias.scope !20, !noalias !17
1788; UNROLL-NEXT:    store i32 [[TMP23]], i32* [[TMP27]], align 1, !alias.scope !20, !noalias !17
1789; UNROLL-NEXT:    store i32 [[TMP24]], i32* [[TMP28]], align 1, !alias.scope !20, !noalias !17
1790; UNROLL-NEXT:    store i32 [[TMP25]], i32* [[TMP29]], align 1, !alias.scope !20, !noalias !17
1791; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
1792; UNROLL-NEXT:    [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 4, i64 4>
1793; UNROLL-NEXT:    [[TMP30:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1794; UNROLL-NEXT:    br i1 [[TMP30]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]]
1795; UNROLL:       middle.block:
1796; UNROLL-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
1797; UNROLL-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
1798; UNROLL:       scalar.ph:
1799; UNROLL-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_MEMCHECK]] ]
1800; UNROLL-NEXT:    br label [[FOR_BODY:%.*]]
1801; UNROLL:       for.body:
1802; UNROLL-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
1803; UNROLL-NEXT:    [[TMP31:%.*]] = shl nsw i64 [[I]], 2
1804; UNROLL-NEXT:    [[TMP32:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP31]]
1805; UNROLL-NEXT:    [[TMP33:%.*]] = load i32, i32* [[TMP32]], align 1
1806; UNROLL-NEXT:    [[TMP34:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[I]], i32 1
1807; UNROLL-NEXT:    store i32 [[TMP33]], i32* [[TMP34]], align 1
1808; UNROLL-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
1809; UNROLL-NEXT:    [[TMP35:%.*]] = trunc i64 [[I_NEXT]] to i32
1810; UNROLL-NEXT:    [[COND:%.*]] = icmp eq i32 [[TMP35]], [[N]]
1811; UNROLL-NEXT:    br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP23:![0-9]+]]
1812; UNROLL:       for.end:
1813; UNROLL-NEXT:    ret void
1814;
1815; UNROLL-NO-IC-LABEL: @scalarize_induction_variable_04(
1816; UNROLL-NO-IC-NEXT:  entry:
1817; UNROLL-NO-IC-NEXT:    [[A4:%.*]] = bitcast i32* [[A:%.*]] to i8*
1818; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = add i32 [[N:%.*]], -1
1819; UNROLL-NO-IC-NEXT:    [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
1820; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
1821; UNROLL-NO-IC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 4
1822; UNROLL-NO-IC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
1823; UNROLL-NO-IC:       vector.memcheck:
1824; UNROLL-NO-IC-NEXT:    [[SCEVGEP:%.*]] = getelementptr [[PAIR_I32:%.*]], %pair.i32* [[P:%.*]], i64 0, i32 1
1825; UNROLL-NO-IC-NEXT:    [[SCEVGEP1:%.*]] = bitcast i32* [[SCEVGEP]] to i8*
1826; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = add i32 [[N]], -1
1827; UNROLL-NO-IC-NEXT:    [[TMP4:%.*]] = zext i32 [[TMP3]] to i64
1828; UNROLL-NO-IC-NEXT:    [[TMP5:%.*]] = add nuw nsw i64 [[TMP4]], 1
1829; UNROLL-NO-IC-NEXT:    [[SCEVGEP2:%.*]] = getelementptr [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP5]]
1830; UNROLL-NO-IC-NEXT:    [[SCEVGEP23:%.*]] = bitcast %pair.i32* [[SCEVGEP2]] to i8*
1831; UNROLL-NO-IC-NEXT:    [[TMP6:%.*]] = shl nuw nsw i64 [[TMP4]], 2
1832; UNROLL-NO-IC-NEXT:    [[TMP7:%.*]] = add nuw nsw i64 [[TMP6]], 1
1833; UNROLL-NO-IC-NEXT:    [[SCEVGEP5:%.*]] = getelementptr i32, i32* [[A]], i64 [[TMP7]]
1834; UNROLL-NO-IC-NEXT:    [[SCEVGEP56:%.*]] = bitcast i32* [[SCEVGEP5]] to i8*
1835; UNROLL-NO-IC-NEXT:    [[BOUND0:%.*]] = icmp ult i8* [[SCEVGEP1]], [[SCEVGEP56]]
1836; UNROLL-NO-IC-NEXT:    [[BOUND1:%.*]] = icmp ult i8* [[A4]], [[SCEVGEP23]]
1837; UNROLL-NO-IC-NEXT:    [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
1838; UNROLL-NO-IC-NEXT:    br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
1839; UNROLL-NO-IC:       vector.ph:
1840; UNROLL-NO-IC-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 4
1841; UNROLL-NO-IC-NEXT:    [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]]
1842; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
1843; UNROLL-NO-IC:       vector.body:
1844; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
1845; UNROLL-NO-IC-NEXT:    [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
1846; UNROLL-NO-IC-NEXT:    [[STEP_ADD:%.*]] = add <2 x i64> [[VEC_IND]], <i64 2, i64 2>
1847; UNROLL-NO-IC-NEXT:    [[TMP8:%.*]] = add i64 [[INDEX]], 0
1848; UNROLL-NO-IC-NEXT:    [[TMP9:%.*]] = add i64 [[INDEX]], 1
1849; UNROLL-NO-IC-NEXT:    [[TMP10:%.*]] = add i64 [[INDEX]], 2
1850; UNROLL-NO-IC-NEXT:    [[TMP11:%.*]] = add i64 [[INDEX]], 3
1851; UNROLL-NO-IC-NEXT:    [[TMP12:%.*]] = shl nsw <2 x i64> [[VEC_IND]], <i64 2, i64 2>
1852; UNROLL-NO-IC-NEXT:    [[TMP13:%.*]] = shl nsw <2 x i64> [[STEP_ADD]], <i64 2, i64 2>
1853; UNROLL-NO-IC-NEXT:    [[TMP14:%.*]] = extractelement <2 x i64> [[TMP12]], i32 0
1854; UNROLL-NO-IC-NEXT:    [[TMP15:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP14]]
1855; UNROLL-NO-IC-NEXT:    [[TMP16:%.*]] = extractelement <2 x i64> [[TMP12]], i32 1
1856; UNROLL-NO-IC-NEXT:    [[TMP17:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP16]]
1857; UNROLL-NO-IC-NEXT:    [[TMP18:%.*]] = extractelement <2 x i64> [[TMP13]], i32 0
1858; UNROLL-NO-IC-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP18]]
1859; UNROLL-NO-IC-NEXT:    [[TMP20:%.*]] = extractelement <2 x i64> [[TMP13]], i32 1
1860; UNROLL-NO-IC-NEXT:    [[TMP21:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP20]]
1861; UNROLL-NO-IC-NEXT:    [[TMP22:%.*]] = load i32, i32* [[TMP15]], align 1, !alias.scope !17
1862; UNROLL-NO-IC-NEXT:    [[TMP23:%.*]] = load i32, i32* [[TMP17]], align 1, !alias.scope !17
1863; UNROLL-NO-IC-NEXT:    [[TMP24:%.*]] = load i32, i32* [[TMP19]], align 1, !alias.scope !17
1864; UNROLL-NO-IC-NEXT:    [[TMP25:%.*]] = load i32, i32* [[TMP21]], align 1, !alias.scope !17
1865; UNROLL-NO-IC-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP8]], i32 1
1866; UNROLL-NO-IC-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP9]], i32 1
1867; UNROLL-NO-IC-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP10]], i32 1
1868; UNROLL-NO-IC-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP11]], i32 1
1869; UNROLL-NO-IC-NEXT:    store i32 [[TMP22]], i32* [[TMP26]], align 1, !alias.scope !20, !noalias !17
1870; UNROLL-NO-IC-NEXT:    store i32 [[TMP23]], i32* [[TMP27]], align 1, !alias.scope !20, !noalias !17
1871; UNROLL-NO-IC-NEXT:    store i32 [[TMP24]], i32* [[TMP28]], align 1, !alias.scope !20, !noalias !17
1872; UNROLL-NO-IC-NEXT:    store i32 [[TMP25]], i32* [[TMP29]], align 1, !alias.scope !20, !noalias !17
1873; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
1874; UNROLL-NO-IC-NEXT:    [[VEC_IND_NEXT]] = add <2 x i64> [[STEP_ADD]], <i64 2, i64 2>
1875; UNROLL-NO-IC-NEXT:    [[TMP30:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1876; UNROLL-NO-IC-NEXT:    br i1 [[TMP30]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]]
1877; UNROLL-NO-IC:       middle.block:
1878; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
1879; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
1880; UNROLL-NO-IC:       scalar.ph:
1881; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_MEMCHECK]] ]
1882; UNROLL-NO-IC-NEXT:    br label [[FOR_BODY:%.*]]
1883; UNROLL-NO-IC:       for.body:
1884; UNROLL-NO-IC-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
1885; UNROLL-NO-IC-NEXT:    [[TMP31:%.*]] = shl nsw i64 [[I]], 2
1886; UNROLL-NO-IC-NEXT:    [[TMP32:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP31]]
1887; UNROLL-NO-IC-NEXT:    [[TMP33:%.*]] = load i32, i32* [[TMP32]], align 1
1888; UNROLL-NO-IC-NEXT:    [[TMP34:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[I]], i32 1
1889; UNROLL-NO-IC-NEXT:    store i32 [[TMP33]], i32* [[TMP34]], align 1
1890; UNROLL-NO-IC-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
1891; UNROLL-NO-IC-NEXT:    [[TMP35:%.*]] = trunc i64 [[I_NEXT]] to i32
1892; UNROLL-NO-IC-NEXT:    [[COND:%.*]] = icmp eq i32 [[TMP35]], [[N]]
1893; UNROLL-NO-IC-NEXT:    br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP23:![0-9]+]]
1894; UNROLL-NO-IC:       for.end:
1895; UNROLL-NO-IC-NEXT:    ret void
1896;
1897; INTERLEAVE-LABEL: @scalarize_induction_variable_04(
1898; INTERLEAVE-NEXT:  entry:
1899; INTERLEAVE-NEXT:    [[TMP0:%.*]] = add i32 [[N:%.*]], -1
1900; INTERLEAVE-NEXT:    [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
1901; INTERLEAVE-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
1902; INTERLEAVE-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 8
1903; INTERLEAVE-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
1904; INTERLEAVE:       vector.memcheck:
1905; INTERLEAVE-NEXT:    [[SCEVGEP:%.*]] = getelementptr [[PAIR_I32:%.*]], %pair.i32* [[P:%.*]], i64 0, i32 1
1906; INTERLEAVE-NEXT:    [[TMP3:%.*]] = add i32 [[N]], -1
1907; INTERLEAVE-NEXT:    [[TMP4:%.*]] = zext i32 [[TMP3]] to i64
1908; INTERLEAVE-NEXT:    [[TMP5:%.*]] = add nuw nsw i64 [[TMP4]], 1
1909; INTERLEAVE-NEXT:    [[TMP6:%.*]] = shl nuw nsw i64 [[TMP4]], 2
1910; INTERLEAVE-NEXT:    [[TMP7:%.*]] = or i64 [[TMP6]], 1
1911; INTERLEAVE-NEXT:    [[SCEVGEP5:%.*]] = getelementptr i32, i32* [[A:%.*]], i64 [[TMP7]]
1912; INTERLEAVE-NEXT:    [[BOUND0:%.*]] = icmp ult i32* [[SCEVGEP]], [[SCEVGEP5]]
1913; INTERLEAVE-NEXT:    [[TMP8:%.*]] = getelementptr [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP5]], i32 0
1914; INTERLEAVE-NEXT:    [[BOUND1:%.*]] = icmp ugt i32* [[TMP8]], [[A]]
1915; INTERLEAVE-NEXT:    [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
1916; INTERLEAVE-NEXT:    br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
1917; INTERLEAVE:       vector.ph:
1918; INTERLEAVE-NEXT:    [[N_MOD_VF:%.*]] = and i64 [[TMP2]], 7
1919; INTERLEAVE-NEXT:    [[TMP9:%.*]] = icmp eq i64 [[N_MOD_VF]], 0
1920; INTERLEAVE-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], i64 8, i64 [[N_MOD_VF]]
1921; INTERLEAVE-NEXT:    [[N_VEC:%.*]] = sub nsw i64 [[TMP2]], [[TMP10]]
1922; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
1923; INTERLEAVE:       vector.body:
1924; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
1925; INTERLEAVE-NEXT:    [[TMP11:%.*]] = or i64 [[INDEX]], 1
1926; INTERLEAVE-NEXT:    [[TMP12:%.*]] = or i64 [[INDEX]], 2
1927; INTERLEAVE-NEXT:    [[TMP13:%.*]] = or i64 [[INDEX]], 3
1928; INTERLEAVE-NEXT:    [[TMP14:%.*]] = or i64 [[INDEX]], 4
1929; INTERLEAVE-NEXT:    [[TMP15:%.*]] = or i64 [[INDEX]], 5
1930; INTERLEAVE-NEXT:    [[TMP16:%.*]] = or i64 [[INDEX]], 6
1931; INTERLEAVE-NEXT:    [[TMP17:%.*]] = or i64 [[INDEX]], 7
1932; INTERLEAVE-NEXT:    [[TMP18:%.*]] = shl nsw i64 [[INDEX]], 2
1933; INTERLEAVE-NEXT:    [[TMP19:%.*]] = shl nsw i64 [[TMP14]], 2
1934; INTERLEAVE-NEXT:    [[TMP20:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP18]]
1935; INTERLEAVE-NEXT:    [[TMP21:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP19]]
1936; INTERLEAVE-NEXT:    [[TMP22:%.*]] = bitcast i32* [[TMP20]] to <16 x i32>*
1937; INTERLEAVE-NEXT:    [[TMP23:%.*]] = bitcast i32* [[TMP21]] to <16 x i32>*
1938; INTERLEAVE-NEXT:    [[WIDE_VEC:%.*]] = load <16 x i32>, <16 x i32>* [[TMP22]], align 1
1939; INTERLEAVE-NEXT:    [[WIDE_VEC7:%.*]] = load <16 x i32>, <16 x i32>* [[TMP23]], align 1
1940; INTERLEAVE-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[INDEX]], i32 1
1941; INTERLEAVE-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP11]], i32 1
1942; INTERLEAVE-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP12]], i32 1
1943; INTERLEAVE-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP13]], i32 1
1944; INTERLEAVE-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP14]], i32 1
1945; INTERLEAVE-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP15]], i32 1
1946; INTERLEAVE-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP16]], i32 1
1947; INTERLEAVE-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[TMP17]], i32 1
1948; INTERLEAVE-NEXT:    [[TMP32:%.*]] = extractelement <16 x i32> [[WIDE_VEC]], i64 0
1949; INTERLEAVE-NEXT:    store i32 [[TMP32]], i32* [[TMP24]], align 1, !alias.scope !17, !noalias !20
1950; INTERLEAVE-NEXT:    [[TMP33:%.*]] = extractelement <16 x i32> [[WIDE_VEC]], i64 4
1951; INTERLEAVE-NEXT:    store i32 [[TMP33]], i32* [[TMP25]], align 1, !alias.scope !17, !noalias !20
1952; INTERLEAVE-NEXT:    [[TMP34:%.*]] = extractelement <16 x i32> [[WIDE_VEC]], i64 8
1953; INTERLEAVE-NEXT:    store i32 [[TMP34]], i32* [[TMP26]], align 1, !alias.scope !17, !noalias !20
1954; INTERLEAVE-NEXT:    [[TMP35:%.*]] = extractelement <16 x i32> [[WIDE_VEC]], i64 12
1955; INTERLEAVE-NEXT:    store i32 [[TMP35]], i32* [[TMP27]], align 1, !alias.scope !17, !noalias !20
1956; INTERLEAVE-NEXT:    [[TMP36:%.*]] = extractelement <16 x i32> [[WIDE_VEC7]], i64 0
1957; INTERLEAVE-NEXT:    store i32 [[TMP36]], i32* [[TMP28]], align 1, !alias.scope !17, !noalias !20
1958; INTERLEAVE-NEXT:    [[TMP37:%.*]] = extractelement <16 x i32> [[WIDE_VEC7]], i64 4
1959; INTERLEAVE-NEXT:    store i32 [[TMP37]], i32* [[TMP29]], align 1, !alias.scope !17, !noalias !20
1960; INTERLEAVE-NEXT:    [[TMP38:%.*]] = extractelement <16 x i32> [[WIDE_VEC7]], i64 8
1961; INTERLEAVE-NEXT:    store i32 [[TMP38]], i32* [[TMP30]], align 1, !alias.scope !17, !noalias !20
1962; INTERLEAVE-NEXT:    [[TMP39:%.*]] = extractelement <16 x i32> [[WIDE_VEC7]], i64 12
1963; INTERLEAVE-NEXT:    store i32 [[TMP39]], i32* [[TMP31]], align 1, !alias.scope !17, !noalias !20
1964; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
1965; INTERLEAVE-NEXT:    [[TMP40:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1966; INTERLEAVE-NEXT:    br i1 [[TMP40]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]]
1967; INTERLEAVE:       middle.block:
1968; INTERLEAVE-NEXT:    br label [[SCALAR_PH]]
1969; INTERLEAVE:       scalar.ph:
1970; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_MEMCHECK]] ]
1971; INTERLEAVE-NEXT:    br label [[FOR_BODY:%.*]]
1972; INTERLEAVE:       for.body:
1973; INTERLEAVE-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
1974; INTERLEAVE-NEXT:    [[TMP41:%.*]] = shl nsw i64 [[I]], 2
1975; INTERLEAVE-NEXT:    [[TMP42:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP41]]
1976; INTERLEAVE-NEXT:    [[TMP43:%.*]] = load i32, i32* [[TMP42]], align 1
1977; INTERLEAVE-NEXT:    [[TMP44:%.*]] = getelementptr inbounds [[PAIR_I32]], %pair.i32* [[P]], i64 [[I]], i32 1
1978; INTERLEAVE-NEXT:    store i32 [[TMP43]], i32* [[TMP44]], align 1
1979; INTERLEAVE-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
1980; INTERLEAVE-NEXT:    [[TMP45:%.*]] = trunc i64 [[I_NEXT]] to i32
1981; INTERLEAVE-NEXT:    [[COND:%.*]] = icmp eq i32 [[TMP45]], [[N]]
1982; INTERLEAVE-NEXT:    br i1 [[COND]], label [[FOR_END:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP23:![0-9]+]]
1983; INTERLEAVE:       for.end:
1984; INTERLEAVE-NEXT:    ret void
1985;
1986entry:
1987  br label %for.body
1988
1989for.body:
1990  %i = phi i64 [ %i.next, %for.body ], [ 0, %entry]
1991  %0 = shl nsw i64 %i, 2
1992  %1 = getelementptr inbounds i32, i32* %a, i64 %0
1993  %2 = load i32, i32* %1, align 1
1994  %3 = getelementptr inbounds %pair.i32, %pair.i32* %p, i64 %i, i32 1
1995  store i32 %2, i32* %3, align 1
1996  %i.next = add nuw nsw i64 %i, 1
1997  %4 = trunc i64 %i.next to i32
1998  %cond = icmp eq i32 %4, %n
1999  br i1 %cond, label %for.end, label %for.body
2000
2001for.end:
2002  ret void
2003}
2004
2005; PR30542. Ensure we generate all the scalar steps for the induction variable.
2006; The scalar induction variable is used by a getelementptr instruction
2007; (uniform), and a udiv (non-uniform).
2008;
2009; int sum = 0;
2010; for (int i = 0; i < n; ++i) {
2011;   int x = a[i];
2012;   if (c)
2013;     x /= i;
2014;   sum += x;
2015; }
2016;
2017;
2018;
2019;
2020
2021define i32 @scalarize_induction_variable_05(i32* %a, i32 %x, i1 %c, i32 %n) {
2022; CHECK-LABEL: @scalarize_induction_variable_05(
2023; CHECK-NEXT:  entry:
2024; CHECK-NEXT:    [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[N:%.*]], i32 1)
2025; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[SMAX]], 2
2026; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
2027; CHECK:       vector.ph:
2028; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[SMAX]], 2
2029; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[SMAX]], [[N_MOD_VF]]
2030; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i1> poison, i1 [[C:%.*]], i32 0
2031; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i1> [[BROADCAST_SPLATINSERT]], <2 x i1> poison, <2 x i32> zeroinitializer
2032; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
2033; CHECK:       vector.body:
2034; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_UDIV_CONTINUE2:%.*]] ]
2035; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP16:%.*]], [[PRED_UDIV_CONTINUE2]] ]
2036; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[INDEX]], 0
2037; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 [[TMP0]]
2038; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[TMP1]], i32 0
2039; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <2 x i32>*
2040; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x i32>, <2 x i32>* [[TMP3]], align 4
2041; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x i1> [[BROADCAST_SPLAT]], i32 0
2042; CHECK-NEXT:    br i1 [[TMP4]], label [[PRED_UDIV_IF:%.*]], label [[PRED_UDIV_CONTINUE:%.*]]
2043; CHECK:       pred.udiv.if:
2044; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x i32> [[WIDE_LOAD]], i32 0
2045; CHECK-NEXT:    [[TMP6:%.*]] = udiv i32 [[TMP5]], [[TMP0]]
2046; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x i32> poison, i32 [[TMP6]], i32 0
2047; CHECK-NEXT:    br label [[PRED_UDIV_CONTINUE]]
2048; CHECK:       pred.udiv.continue:
2049; CHECK-NEXT:    [[TMP8:%.*]] = phi <2 x i32> [ poison, [[VECTOR_BODY]] ], [ [[TMP7]], [[PRED_UDIV_IF]] ]
2050; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <2 x i1> [[BROADCAST_SPLAT]], i32 1
2051; CHECK-NEXT:    br i1 [[TMP9]], label [[PRED_UDIV_IF1:%.*]], label [[PRED_UDIV_CONTINUE2]]
2052; CHECK:       pred.udiv.if1:
2053; CHECK-NEXT:    [[TMP10:%.*]] = add i32 [[INDEX]], 1
2054; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <2 x i32> [[WIDE_LOAD]], i32 1
2055; CHECK-NEXT:    [[TMP12:%.*]] = udiv i32 [[TMP11]], [[TMP10]]
2056; CHECK-NEXT:    [[TMP13:%.*]] = insertelement <2 x i32> [[TMP8]], i32 [[TMP12]], i32 1
2057; CHECK-NEXT:    br label [[PRED_UDIV_CONTINUE2]]
2058; CHECK:       pred.udiv.continue2:
2059; CHECK-NEXT:    [[TMP14:%.*]] = phi <2 x i32> [ [[TMP8]], [[PRED_UDIV_CONTINUE]] ], [ [[TMP13]], [[PRED_UDIV_IF1]] ]
2060; CHECK-NEXT:    [[TMP15:%.*]] = xor <2 x i1> [[BROADCAST_SPLAT]], <i1 true, i1 true>
2061; CHECK-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[TMP15]], <2 x i32> [[WIDE_LOAD]], <2 x i32> [[TMP14]]
2062; CHECK-NEXT:    [[TMP16]] = add <2 x i32> [[PREDPHI]], [[VEC_PHI]]
2063; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
2064; CHECK-NEXT:    [[TMP17:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
2065; CHECK-NEXT:    br i1 [[TMP17]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]]
2066; CHECK:       middle.block:
2067; CHECK-NEXT:    [[TMP18:%.*]] = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> [[TMP16]])
2068; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[SMAX]], [[N_VEC]]
2069; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
2070; CHECK:       scalar.ph:
2071; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
2072; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[TMP18]], [[MIDDLE_BLOCK]] ]
2073; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
2074; CHECK:       for.body:
2075; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[IF_END:%.*]] ]
2076; CHECK-NEXT:    [[SUM:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[VAR4:%.*]], [[IF_END]] ]
2077; CHECK-NEXT:    [[VAR0:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[I]]
2078; CHECK-NEXT:    [[VAR1:%.*]] = load i32, i32* [[VAR0]], align 4
2079; CHECK-NEXT:    br i1 [[C]], label [[IF_THEN:%.*]], label [[IF_END]]
2080; CHECK:       if.then:
2081; CHECK-NEXT:    [[VAR2:%.*]] = udiv i32 [[VAR1]], [[I]]
2082; CHECK-NEXT:    br label [[IF_END]]
2083; CHECK:       if.end:
2084; CHECK-NEXT:    [[VAR3:%.*]] = phi i32 [ [[VAR2]], [[IF_THEN]] ], [ [[VAR1]], [[FOR_BODY]] ]
2085; CHECK-NEXT:    [[VAR4]] = add i32 [[VAR3]], [[SUM]]
2086; CHECK-NEXT:    [[I_NEXT]] = add nuw nsw i32 [[I]], 1
2087; CHECK-NEXT:    [[COND:%.*]] = icmp slt i32 [[I_NEXT]], [[N]]
2088; CHECK-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP25:![0-9]+]]
2089; CHECK:       for.end:
2090; CHECK-NEXT:    [[VAR5:%.*]] = phi i32 [ [[VAR4]], [[IF_END]] ], [ [[TMP18]], [[MIDDLE_BLOCK]] ]
2091; CHECK-NEXT:    ret i32 [[VAR5]]
2092;
2093; IND-LABEL: @scalarize_induction_variable_05(
2094; IND-NEXT:  entry:
2095; IND-NEXT:    [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[N:%.*]], i32 1)
2096; IND-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[SMAX]], 2
2097; IND-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
2098; IND:       vector.ph:
2099; IND-NEXT:    [[N_VEC:%.*]] = and i32 [[SMAX]], 2147483646
2100; IND-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i1> poison, i1 [[C:%.*]], i64 0
2101; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
2102; IND:       vector.body:
2103; IND-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_UDIV_CONTINUE2:%.*]] ]
2104; IND-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP14:%.*]], [[PRED_UDIV_CONTINUE2]] ]
2105; IND-NEXT:    [[TMP0:%.*]] = sext i32 [[INDEX]] to i64
2106; IND-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP0]]
2107; IND-NEXT:    [[TMP2:%.*]] = bitcast i32* [[TMP1]] to <2 x i32>*
2108; IND-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x i32>, <2 x i32>* [[TMP2]], align 4
2109; IND-NEXT:    br i1 [[C]], label [[PRED_UDIV_IF:%.*]], label [[PRED_UDIV_CONTINUE:%.*]]
2110; IND:       pred.udiv.if:
2111; IND-NEXT:    [[TMP3:%.*]] = extractelement <2 x i32> [[WIDE_LOAD]], i64 0
2112; IND-NEXT:    [[TMP4:%.*]] = udiv i32 [[TMP3]], [[INDEX]]
2113; IND-NEXT:    [[TMP5:%.*]] = insertelement <2 x i32> poison, i32 [[TMP4]], i64 0
2114; IND-NEXT:    br label [[PRED_UDIV_CONTINUE]]
2115; IND:       pred.udiv.continue:
2116; IND-NEXT:    [[TMP6:%.*]] = phi <2 x i32> [ poison, [[VECTOR_BODY]] ], [ [[TMP5]], [[PRED_UDIV_IF]] ]
2117; IND-NEXT:    br i1 [[C]], label [[PRED_UDIV_IF1:%.*]], label [[PRED_UDIV_CONTINUE2]]
2118; IND:       pred.udiv.if1:
2119; IND-NEXT:    [[TMP7:%.*]] = or i32 [[INDEX]], 1
2120; IND-NEXT:    [[TMP8:%.*]] = extractelement <2 x i32> [[WIDE_LOAD]], i64 1
2121; IND-NEXT:    [[TMP9:%.*]] = udiv i32 [[TMP8]], [[TMP7]]
2122; IND-NEXT:    [[TMP10:%.*]] = insertelement <2 x i32> [[TMP6]], i32 [[TMP9]], i64 1
2123; IND-NEXT:    br label [[PRED_UDIV_CONTINUE2]]
2124; IND:       pred.udiv.continue2:
2125; IND-NEXT:    [[TMP11:%.*]] = phi <2 x i32> [ [[TMP6]], [[PRED_UDIV_CONTINUE]] ], [ [[TMP10]], [[PRED_UDIV_IF1]] ]
2126; IND-NEXT:    [[TMP12:%.*]] = xor <2 x i1> [[BROADCAST_SPLATINSERT]], <i1 true, i1 poison>
2127; IND-NEXT:    [[TMP13:%.*]] = shufflevector <2 x i1> [[TMP12]], <2 x i1> poison, <2 x i32> zeroinitializer
2128; IND-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[TMP13]], <2 x i32> [[WIDE_LOAD]], <2 x i32> [[TMP11]]
2129; IND-NEXT:    [[TMP14]] = add <2 x i32> [[PREDPHI]], [[VEC_PHI]]
2130; IND-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
2131; IND-NEXT:    [[TMP15:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
2132; IND-NEXT:    br i1 [[TMP15]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]]
2133; IND:       middle.block:
2134; IND-NEXT:    [[TMP16:%.*]] = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> [[TMP14]])
2135; IND-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[SMAX]], [[N_VEC]]
2136; IND-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
2137; IND:       scalar.ph:
2138; IND-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
2139; IND-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP16]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
2140; IND-NEXT:    br label [[FOR_BODY:%.*]]
2141; IND:       for.body:
2142; IND-NEXT:    [[I:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[IF_END:%.*]] ]
2143; IND-NEXT:    [[SUM:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[VAR4:%.*]], [[IF_END]] ]
2144; IND-NEXT:    [[TMP17:%.*]] = zext i32 [[I]] to i64
2145; IND-NEXT:    [[VAR0:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP17]]
2146; IND-NEXT:    [[VAR1:%.*]] = load i32, i32* [[VAR0]], align 4
2147; IND-NEXT:    br i1 [[C]], label [[IF_THEN:%.*]], label [[IF_END]]
2148; IND:       if.then:
2149; IND-NEXT:    [[VAR2:%.*]] = udiv i32 [[VAR1]], [[I]]
2150; IND-NEXT:    br label [[IF_END]]
2151; IND:       if.end:
2152; IND-NEXT:    [[VAR3:%.*]] = phi i32 [ [[VAR2]], [[IF_THEN]] ], [ [[VAR1]], [[FOR_BODY]] ]
2153; IND-NEXT:    [[VAR4]] = add i32 [[VAR3]], [[SUM]]
2154; IND-NEXT:    [[I_NEXT]] = add nuw nsw i32 [[I]], 1
2155; IND-NEXT:    [[COND:%.*]] = icmp slt i32 [[I_NEXT]], [[N]]
2156; IND-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP25:![0-9]+]]
2157; IND:       for.end:
2158; IND-NEXT:    [[VAR5:%.*]] = phi i32 [ [[VAR4]], [[IF_END]] ], [ [[TMP16]], [[MIDDLE_BLOCK]] ]
2159; IND-NEXT:    ret i32 [[VAR5]]
2160;
2161; UNROLL-LABEL: @scalarize_induction_variable_05(
2162; UNROLL-NEXT:  entry:
2163; UNROLL-NEXT:    [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[N:%.*]], i32 1)
2164; UNROLL-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[SMAX]], 4
2165; UNROLL-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
2166; UNROLL:       vector.ph:
2167; UNROLL-NEXT:    [[N_VEC:%.*]] = and i32 [[SMAX]], 2147483644
2168; UNROLL-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i1> poison, i1 [[C:%.*]], i64 0
2169; UNROLL-NEXT:    [[BROADCAST_SPLATINSERT5:%.*]] = insertelement <2 x i1> poison, i1 [[C]], i64 0
2170; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
2171; UNROLL:       vector.body:
2172; UNROLL-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_UDIV_CONTINUE10:%.*]] ]
2173; UNROLL-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP28:%.*]], [[PRED_UDIV_CONTINUE10]] ]
2174; UNROLL-NEXT:    [[VEC_PHI1:%.*]] = phi <2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP29:%.*]], [[PRED_UDIV_CONTINUE10]] ]
2175; UNROLL-NEXT:    [[TMP0:%.*]] = or i32 [[INDEX]], 2
2176; UNROLL-NEXT:    [[TMP1:%.*]] = sext i32 [[INDEX]] to i64
2177; UNROLL-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP1]]
2178; UNROLL-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <2 x i32>*
2179; UNROLL-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x i32>, <2 x i32>* [[TMP3]], align 4
2180; UNROLL-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[TMP2]], i64 2
2181; UNROLL-NEXT:    [[TMP5:%.*]] = bitcast i32* [[TMP4]] to <2 x i32>*
2182; UNROLL-NEXT:    [[WIDE_LOAD2:%.*]] = load <2 x i32>, <2 x i32>* [[TMP5]], align 4
2183; UNROLL-NEXT:    br i1 [[C]], label [[PRED_UDIV_IF:%.*]], label [[PRED_UDIV_CONTINUE:%.*]]
2184; UNROLL:       pred.udiv.if:
2185; UNROLL-NEXT:    [[TMP6:%.*]] = extractelement <2 x i32> [[WIDE_LOAD]], i64 0
2186; UNROLL-NEXT:    [[TMP7:%.*]] = udiv i32 [[TMP6]], [[INDEX]]
2187; UNROLL-NEXT:    [[TMP8:%.*]] = insertelement <2 x i32> poison, i32 [[TMP7]], i64 0
2188; UNROLL-NEXT:    br label [[PRED_UDIV_CONTINUE]]
2189; UNROLL:       pred.udiv.continue:
2190; UNROLL-NEXT:    [[TMP9:%.*]] = phi <2 x i32> [ poison, [[VECTOR_BODY]] ], [ [[TMP8]], [[PRED_UDIV_IF]] ]
2191; UNROLL-NEXT:    br i1 [[C]], label [[PRED_UDIV_IF3:%.*]], label [[PRED_UDIV_CONTINUE4:%.*]]
2192; UNROLL:       pred.udiv.if3:
2193; UNROLL-NEXT:    [[TMP10:%.*]] = or i32 [[INDEX]], 1
2194; UNROLL-NEXT:    [[TMP11:%.*]] = extractelement <2 x i32> [[WIDE_LOAD]], i64 1
2195; UNROLL-NEXT:    [[TMP12:%.*]] = udiv i32 [[TMP11]], [[TMP10]]
2196; UNROLL-NEXT:    [[TMP13:%.*]] = insertelement <2 x i32> [[TMP9]], i32 [[TMP12]], i64 1
2197; UNROLL-NEXT:    br label [[PRED_UDIV_CONTINUE4]]
2198; UNROLL:       pred.udiv.continue4:
2199; UNROLL-NEXT:    [[TMP14:%.*]] = phi <2 x i32> [ [[TMP9]], [[PRED_UDIV_CONTINUE]] ], [ [[TMP13]], [[PRED_UDIV_IF3]] ]
2200; UNROLL-NEXT:    br i1 [[C]], label [[PRED_UDIV_IF7:%.*]], label [[PRED_UDIV_CONTINUE8:%.*]]
2201; UNROLL:       pred.udiv.if7:
2202; UNROLL-NEXT:    [[TMP15:%.*]] = extractelement <2 x i32> [[WIDE_LOAD2]], i64 0
2203; UNROLL-NEXT:    [[TMP16:%.*]] = udiv i32 [[TMP15]], [[TMP0]]
2204; UNROLL-NEXT:    [[TMP17:%.*]] = insertelement <2 x i32> poison, i32 [[TMP16]], i64 0
2205; UNROLL-NEXT:    br label [[PRED_UDIV_CONTINUE8]]
2206; UNROLL:       pred.udiv.continue8:
2207; UNROLL-NEXT:    [[TMP18:%.*]] = phi <2 x i32> [ poison, [[PRED_UDIV_CONTINUE4]] ], [ [[TMP17]], [[PRED_UDIV_IF7]] ]
2208; UNROLL-NEXT:    br i1 [[C]], label [[PRED_UDIV_IF9:%.*]], label [[PRED_UDIV_CONTINUE10]]
2209; UNROLL:       pred.udiv.if9:
2210; UNROLL-NEXT:    [[TMP19:%.*]] = or i32 [[INDEX]], 3
2211; UNROLL-NEXT:    [[TMP20:%.*]] = extractelement <2 x i32> [[WIDE_LOAD2]], i64 1
2212; UNROLL-NEXT:    [[TMP21:%.*]] = udiv i32 [[TMP20]], [[TMP19]]
2213; UNROLL-NEXT:    [[TMP22:%.*]] = insertelement <2 x i32> [[TMP18]], i32 [[TMP21]], i64 1
2214; UNROLL-NEXT:    br label [[PRED_UDIV_CONTINUE10]]
2215; UNROLL:       pred.udiv.continue10:
2216; UNROLL-NEXT:    [[TMP23:%.*]] = phi <2 x i32> [ [[TMP18]], [[PRED_UDIV_CONTINUE8]] ], [ [[TMP22]], [[PRED_UDIV_IF9]] ]
2217; UNROLL-NEXT:    [[TMP24:%.*]] = xor <2 x i1> [[BROADCAST_SPLATINSERT]], <i1 true, i1 poison>
2218; UNROLL-NEXT:    [[TMP25:%.*]] = shufflevector <2 x i1> [[TMP24]], <2 x i1> poison, <2 x i32> zeroinitializer
2219; UNROLL-NEXT:    [[TMP26:%.*]] = xor <2 x i1> [[BROADCAST_SPLATINSERT5]], <i1 true, i1 poison>
2220; UNROLL-NEXT:    [[TMP27:%.*]] = shufflevector <2 x i1> [[TMP26]], <2 x i1> poison, <2 x i32> zeroinitializer
2221; UNROLL-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[TMP25]], <2 x i32> [[WIDE_LOAD]], <2 x i32> [[TMP14]]
2222; UNROLL-NEXT:    [[PREDPHI11:%.*]] = select <2 x i1> [[TMP27]], <2 x i32> [[WIDE_LOAD2]], <2 x i32> [[TMP23]]
2223; UNROLL-NEXT:    [[TMP28]] = add <2 x i32> [[PREDPHI]], [[VEC_PHI]]
2224; UNROLL-NEXT:    [[TMP29]] = add <2 x i32> [[PREDPHI11]], [[VEC_PHI1]]
2225; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
2226; UNROLL-NEXT:    [[TMP30:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
2227; UNROLL-NEXT:    br i1 [[TMP30]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]]
2228; UNROLL:       middle.block:
2229; UNROLL-NEXT:    [[BIN_RDX:%.*]] = add <2 x i32> [[TMP29]], [[TMP28]]
2230; UNROLL-NEXT:    [[TMP31:%.*]] = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> [[BIN_RDX]])
2231; UNROLL-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[SMAX]], [[N_VEC]]
2232; UNROLL-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
2233; UNROLL:       scalar.ph:
2234; UNROLL-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
2235; UNROLL-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP31]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
2236; UNROLL-NEXT:    br label [[FOR_BODY:%.*]]
2237; UNROLL:       for.body:
2238; UNROLL-NEXT:    [[I:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[IF_END:%.*]] ]
2239; UNROLL-NEXT:    [[SUM:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[VAR4:%.*]], [[IF_END]] ]
2240; UNROLL-NEXT:    [[TMP32:%.*]] = zext i32 [[I]] to i64
2241; UNROLL-NEXT:    [[VAR0:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP32]]
2242; UNROLL-NEXT:    [[VAR1:%.*]] = load i32, i32* [[VAR0]], align 4
2243; UNROLL-NEXT:    br i1 [[C]], label [[IF_THEN:%.*]], label [[IF_END]]
2244; UNROLL:       if.then:
2245; UNROLL-NEXT:    [[VAR2:%.*]] = udiv i32 [[VAR1]], [[I]]
2246; UNROLL-NEXT:    br label [[IF_END]]
2247; UNROLL:       if.end:
2248; UNROLL-NEXT:    [[VAR3:%.*]] = phi i32 [ [[VAR2]], [[IF_THEN]] ], [ [[VAR1]], [[FOR_BODY]] ]
2249; UNROLL-NEXT:    [[VAR4]] = add i32 [[VAR3]], [[SUM]]
2250; UNROLL-NEXT:    [[I_NEXT]] = add nuw nsw i32 [[I]], 1
2251; UNROLL-NEXT:    [[COND:%.*]] = icmp slt i32 [[I_NEXT]], [[N]]
2252; UNROLL-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP25:![0-9]+]]
2253; UNROLL:       for.end:
2254; UNROLL-NEXT:    [[VAR5:%.*]] = phi i32 [ [[VAR4]], [[IF_END]] ], [ [[TMP31]], [[MIDDLE_BLOCK]] ]
2255; UNROLL-NEXT:    ret i32 [[VAR5]]
2256;
2257; UNROLL-NO-IC-LABEL: @scalarize_induction_variable_05(
2258; UNROLL-NO-IC-NEXT:  entry:
2259; UNROLL-NO-IC-NEXT:    [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[N:%.*]], i32 1)
2260; UNROLL-NO-IC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[SMAX]], 4
2261; UNROLL-NO-IC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
2262; UNROLL-NO-IC:       vector.ph:
2263; UNROLL-NO-IC-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[SMAX]], 4
2264; UNROLL-NO-IC-NEXT:    [[N_VEC:%.*]] = sub i32 [[SMAX]], [[N_MOD_VF]]
2265; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i1> poison, i1 [[C:%.*]], i32 0
2266; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i1> [[BROADCAST_SPLATINSERT]], <2 x i1> poison, <2 x i32> zeroinitializer
2267; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLATINSERT5:%.*]] = insertelement <2 x i1> poison, i1 [[C]], i32 0
2268; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLAT6:%.*]] = shufflevector <2 x i1> [[BROADCAST_SPLATINSERT5]], <2 x i1> poison, <2 x i32> zeroinitializer
2269; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
2270; UNROLL-NO-IC:       vector.body:
2271; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_UDIV_CONTINUE10:%.*]] ]
2272; UNROLL-NO-IC-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP32:%.*]], [[PRED_UDIV_CONTINUE10]] ]
2273; UNROLL-NO-IC-NEXT:    [[VEC_PHI1:%.*]] = phi <2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP33:%.*]], [[PRED_UDIV_CONTINUE10]] ]
2274; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = add i32 [[INDEX]], 0
2275; UNROLL-NO-IC-NEXT:    [[TMP1:%.*]] = add i32 [[INDEX]], 2
2276; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 [[TMP0]]
2277; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[TMP1]]
2278; UNROLL-NO-IC-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[TMP2]], i32 0
2279; UNROLL-NO-IC-NEXT:    [[TMP5:%.*]] = bitcast i32* [[TMP4]] to <2 x i32>*
2280; UNROLL-NO-IC-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x i32>, <2 x i32>* [[TMP5]], align 4
2281; UNROLL-NO-IC-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i32, i32* [[TMP2]], i32 2
2282; UNROLL-NO-IC-NEXT:    [[TMP7:%.*]] = bitcast i32* [[TMP6]] to <2 x i32>*
2283; UNROLL-NO-IC-NEXT:    [[WIDE_LOAD2:%.*]] = load <2 x i32>, <2 x i32>* [[TMP7]], align 4
2284; UNROLL-NO-IC-NEXT:    [[TMP8:%.*]] = extractelement <2 x i1> [[BROADCAST_SPLAT]], i32 0
2285; UNROLL-NO-IC-NEXT:    br i1 [[TMP8]], label [[PRED_UDIV_IF:%.*]], label [[PRED_UDIV_CONTINUE:%.*]]
2286; UNROLL-NO-IC:       pred.udiv.if:
2287; UNROLL-NO-IC-NEXT:    [[TMP9:%.*]] = extractelement <2 x i32> [[WIDE_LOAD]], i32 0
2288; UNROLL-NO-IC-NEXT:    [[TMP10:%.*]] = udiv i32 [[TMP9]], [[TMP0]]
2289; UNROLL-NO-IC-NEXT:    [[TMP11:%.*]] = insertelement <2 x i32> poison, i32 [[TMP10]], i32 0
2290; UNROLL-NO-IC-NEXT:    br label [[PRED_UDIV_CONTINUE]]
2291; UNROLL-NO-IC:       pred.udiv.continue:
2292; UNROLL-NO-IC-NEXT:    [[TMP12:%.*]] = phi <2 x i32> [ poison, [[VECTOR_BODY]] ], [ [[TMP11]], [[PRED_UDIV_IF]] ]
2293; UNROLL-NO-IC-NEXT:    [[TMP13:%.*]] = extractelement <2 x i1> [[BROADCAST_SPLAT]], i32 1
2294; UNROLL-NO-IC-NEXT:    br i1 [[TMP13]], label [[PRED_UDIV_IF3:%.*]], label [[PRED_UDIV_CONTINUE4:%.*]]
2295; UNROLL-NO-IC:       pred.udiv.if3:
2296; UNROLL-NO-IC-NEXT:    [[TMP14:%.*]] = add i32 [[INDEX]], 1
2297; UNROLL-NO-IC-NEXT:    [[TMP15:%.*]] = extractelement <2 x i32> [[WIDE_LOAD]], i32 1
2298; UNROLL-NO-IC-NEXT:    [[TMP16:%.*]] = udiv i32 [[TMP15]], [[TMP14]]
2299; UNROLL-NO-IC-NEXT:    [[TMP17:%.*]] = insertelement <2 x i32> [[TMP12]], i32 [[TMP16]], i32 1
2300; UNROLL-NO-IC-NEXT:    br label [[PRED_UDIV_CONTINUE4]]
2301; UNROLL-NO-IC:       pred.udiv.continue4:
2302; UNROLL-NO-IC-NEXT:    [[TMP18:%.*]] = phi <2 x i32> [ [[TMP12]], [[PRED_UDIV_CONTINUE]] ], [ [[TMP17]], [[PRED_UDIV_IF3]] ]
2303; UNROLL-NO-IC-NEXT:    [[TMP19:%.*]] = extractelement <2 x i1> [[BROADCAST_SPLAT6]], i32 0
2304; UNROLL-NO-IC-NEXT:    br i1 [[TMP19]], label [[PRED_UDIV_IF7:%.*]], label [[PRED_UDIV_CONTINUE8:%.*]]
2305; UNROLL-NO-IC:       pred.udiv.if7:
2306; UNROLL-NO-IC-NEXT:    [[TMP20:%.*]] = extractelement <2 x i32> [[WIDE_LOAD2]], i32 0
2307; UNROLL-NO-IC-NEXT:    [[TMP21:%.*]] = udiv i32 [[TMP20]], [[TMP1]]
2308; UNROLL-NO-IC-NEXT:    [[TMP22:%.*]] = insertelement <2 x i32> poison, i32 [[TMP21]], i32 0
2309; UNROLL-NO-IC-NEXT:    br label [[PRED_UDIV_CONTINUE8]]
2310; UNROLL-NO-IC:       pred.udiv.continue8:
2311; UNROLL-NO-IC-NEXT:    [[TMP23:%.*]] = phi <2 x i32> [ poison, [[PRED_UDIV_CONTINUE4]] ], [ [[TMP22]], [[PRED_UDIV_IF7]] ]
2312; UNROLL-NO-IC-NEXT:    [[TMP24:%.*]] = extractelement <2 x i1> [[BROADCAST_SPLAT6]], i32 1
2313; UNROLL-NO-IC-NEXT:    br i1 [[TMP24]], label [[PRED_UDIV_IF9:%.*]], label [[PRED_UDIV_CONTINUE10]]
2314; UNROLL-NO-IC:       pred.udiv.if9:
2315; UNROLL-NO-IC-NEXT:    [[TMP25:%.*]] = add i32 [[INDEX]], 3
2316; UNROLL-NO-IC-NEXT:    [[TMP26:%.*]] = extractelement <2 x i32> [[WIDE_LOAD2]], i32 1
2317; UNROLL-NO-IC-NEXT:    [[TMP27:%.*]] = udiv i32 [[TMP26]], [[TMP25]]
2318; UNROLL-NO-IC-NEXT:    [[TMP28:%.*]] = insertelement <2 x i32> [[TMP23]], i32 [[TMP27]], i32 1
2319; UNROLL-NO-IC-NEXT:    br label [[PRED_UDIV_CONTINUE10]]
2320; UNROLL-NO-IC:       pred.udiv.continue10:
2321; UNROLL-NO-IC-NEXT:    [[TMP29:%.*]] = phi <2 x i32> [ [[TMP23]], [[PRED_UDIV_CONTINUE8]] ], [ [[TMP28]], [[PRED_UDIV_IF9]] ]
2322; UNROLL-NO-IC-NEXT:    [[TMP30:%.*]] = xor <2 x i1> [[BROADCAST_SPLAT]], <i1 true, i1 true>
2323; UNROLL-NO-IC-NEXT:    [[TMP31:%.*]] = xor <2 x i1> [[BROADCAST_SPLAT6]], <i1 true, i1 true>
2324; UNROLL-NO-IC-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[TMP30]], <2 x i32> [[WIDE_LOAD]], <2 x i32> [[TMP18]]
2325; UNROLL-NO-IC-NEXT:    [[PREDPHI11:%.*]] = select <2 x i1> [[TMP31]], <2 x i32> [[WIDE_LOAD2]], <2 x i32> [[TMP29]]
2326; UNROLL-NO-IC-NEXT:    [[TMP32]] = add <2 x i32> [[PREDPHI]], [[VEC_PHI]]
2327; UNROLL-NO-IC-NEXT:    [[TMP33]] = add <2 x i32> [[PREDPHI11]], [[VEC_PHI1]]
2328; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
2329; UNROLL-NO-IC-NEXT:    [[TMP34:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
2330; UNROLL-NO-IC-NEXT:    br i1 [[TMP34]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]]
2331; UNROLL-NO-IC:       middle.block:
2332; UNROLL-NO-IC-NEXT:    [[BIN_RDX:%.*]] = add <2 x i32> [[TMP33]], [[TMP32]]
2333; UNROLL-NO-IC-NEXT:    [[TMP35:%.*]] = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> [[BIN_RDX]])
2334; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[SMAX]], [[N_VEC]]
2335; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
2336; UNROLL-NO-IC:       scalar.ph:
2337; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
2338; UNROLL-NO-IC-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[TMP35]], [[MIDDLE_BLOCK]] ]
2339; UNROLL-NO-IC-NEXT:    br label [[FOR_BODY:%.*]]
2340; UNROLL-NO-IC:       for.body:
2341; UNROLL-NO-IC-NEXT:    [[I:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[IF_END:%.*]] ]
2342; UNROLL-NO-IC-NEXT:    [[SUM:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[VAR4:%.*]], [[IF_END]] ]
2343; UNROLL-NO-IC-NEXT:    [[VAR0:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[I]]
2344; UNROLL-NO-IC-NEXT:    [[VAR1:%.*]] = load i32, i32* [[VAR0]], align 4
2345; UNROLL-NO-IC-NEXT:    br i1 [[C]], label [[IF_THEN:%.*]], label [[IF_END]]
2346; UNROLL-NO-IC:       if.then:
2347; UNROLL-NO-IC-NEXT:    [[VAR2:%.*]] = udiv i32 [[VAR1]], [[I]]
2348; UNROLL-NO-IC-NEXT:    br label [[IF_END]]
2349; UNROLL-NO-IC:       if.end:
2350; UNROLL-NO-IC-NEXT:    [[VAR3:%.*]] = phi i32 [ [[VAR2]], [[IF_THEN]] ], [ [[VAR1]], [[FOR_BODY]] ]
2351; UNROLL-NO-IC-NEXT:    [[VAR4]] = add i32 [[VAR3]], [[SUM]]
2352; UNROLL-NO-IC-NEXT:    [[I_NEXT]] = add nuw nsw i32 [[I]], 1
2353; UNROLL-NO-IC-NEXT:    [[COND:%.*]] = icmp slt i32 [[I_NEXT]], [[N]]
2354; UNROLL-NO-IC-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP25:![0-9]+]]
2355; UNROLL-NO-IC:       for.end:
2356; UNROLL-NO-IC-NEXT:    [[VAR5:%.*]] = phi i32 [ [[VAR4]], [[IF_END]] ], [ [[TMP35]], [[MIDDLE_BLOCK]] ]
2357; UNROLL-NO-IC-NEXT:    ret i32 [[VAR5]]
2358;
2359; INTERLEAVE-LABEL: @scalarize_induction_variable_05(
2360; INTERLEAVE-NEXT:  entry:
2361; INTERLEAVE-NEXT:    [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[N:%.*]], i32 1)
2362; INTERLEAVE-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[SMAX]], 8
2363; INTERLEAVE-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
2364; INTERLEAVE:       vector.ph:
2365; INTERLEAVE-NEXT:    [[N_VEC:%.*]] = and i32 [[SMAX]], 2147483640
2366; INTERLEAVE-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i1> poison, i1 [[C:%.*]], i64 0
2367; INTERLEAVE-NEXT:    [[BROADCAST_SPLATINSERT9:%.*]] = insertelement <4 x i1> poison, i1 [[C]], i64 0
2368; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
2369; INTERLEAVE:       vector.body:
2370; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_UDIV_CONTINUE18:%.*]] ]
2371; INTERLEAVE-NEXT:    [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP48:%.*]], [[PRED_UDIV_CONTINUE18]] ]
2372; INTERLEAVE-NEXT:    [[VEC_PHI1:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP49:%.*]], [[PRED_UDIV_CONTINUE18]] ]
2373; INTERLEAVE-NEXT:    [[TMP0:%.*]] = or i32 [[INDEX]], 4
2374; INTERLEAVE-NEXT:    [[TMP1:%.*]] = sext i32 [[INDEX]] to i64
2375; INTERLEAVE-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP1]]
2376; INTERLEAVE-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <4 x i32>*
2377; INTERLEAVE-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i32>, <4 x i32>* [[TMP3]], align 4
2378; INTERLEAVE-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[TMP2]], i64 4
2379; INTERLEAVE-NEXT:    [[TMP5:%.*]] = bitcast i32* [[TMP4]] to <4 x i32>*
2380; INTERLEAVE-NEXT:    [[WIDE_LOAD2:%.*]] = load <4 x i32>, <4 x i32>* [[TMP5]], align 4
2381; INTERLEAVE-NEXT:    br i1 [[C]], label [[PRED_UDIV_IF:%.*]], label [[PRED_UDIV_CONTINUE:%.*]]
2382; INTERLEAVE:       pred.udiv.if:
2383; INTERLEAVE-NEXT:    [[TMP6:%.*]] = extractelement <4 x i32> [[WIDE_LOAD]], i64 0
2384; INTERLEAVE-NEXT:    [[TMP7:%.*]] = udiv i32 [[TMP6]], [[INDEX]]
2385; INTERLEAVE-NEXT:    [[TMP8:%.*]] = insertelement <4 x i32> poison, i32 [[TMP7]], i64 0
2386; INTERLEAVE-NEXT:    br label [[PRED_UDIV_CONTINUE]]
2387; INTERLEAVE:       pred.udiv.continue:
2388; INTERLEAVE-NEXT:    [[TMP9:%.*]] = phi <4 x i32> [ poison, [[VECTOR_BODY]] ], [ [[TMP8]], [[PRED_UDIV_IF]] ]
2389; INTERLEAVE-NEXT:    br i1 [[C]], label [[PRED_UDIV_IF3:%.*]], label [[PRED_UDIV_CONTINUE4:%.*]]
2390; INTERLEAVE:       pred.udiv.if3:
2391; INTERLEAVE-NEXT:    [[TMP10:%.*]] = or i32 [[INDEX]], 1
2392; INTERLEAVE-NEXT:    [[TMP11:%.*]] = extractelement <4 x i32> [[WIDE_LOAD]], i64 1
2393; INTERLEAVE-NEXT:    [[TMP12:%.*]] = udiv i32 [[TMP11]], [[TMP10]]
2394; INTERLEAVE-NEXT:    [[TMP13:%.*]] = insertelement <4 x i32> [[TMP9]], i32 [[TMP12]], i64 1
2395; INTERLEAVE-NEXT:    br label [[PRED_UDIV_CONTINUE4]]
2396; INTERLEAVE:       pred.udiv.continue4:
2397; INTERLEAVE-NEXT:    [[TMP14:%.*]] = phi <4 x i32> [ [[TMP9]], [[PRED_UDIV_CONTINUE]] ], [ [[TMP13]], [[PRED_UDIV_IF3]] ]
2398; INTERLEAVE-NEXT:    br i1 [[C]], label [[PRED_UDIV_IF5:%.*]], label [[PRED_UDIV_CONTINUE6:%.*]]
2399; INTERLEAVE:       pred.udiv.if5:
2400; INTERLEAVE-NEXT:    [[TMP15:%.*]] = or i32 [[INDEX]], 2
2401; INTERLEAVE-NEXT:    [[TMP16:%.*]] = extractelement <4 x i32> [[WIDE_LOAD]], i64 2
2402; INTERLEAVE-NEXT:    [[TMP17:%.*]] = udiv i32 [[TMP16]], [[TMP15]]
2403; INTERLEAVE-NEXT:    [[TMP18:%.*]] = insertelement <4 x i32> [[TMP14]], i32 [[TMP17]], i64 2
2404; INTERLEAVE-NEXT:    br label [[PRED_UDIV_CONTINUE6]]
2405; INTERLEAVE:       pred.udiv.continue6:
2406; INTERLEAVE-NEXT:    [[TMP19:%.*]] = phi <4 x i32> [ [[TMP14]], [[PRED_UDIV_CONTINUE4]] ], [ [[TMP18]], [[PRED_UDIV_IF5]] ]
2407; INTERLEAVE-NEXT:    br i1 [[C]], label [[PRED_UDIV_IF7:%.*]], label [[PRED_UDIV_CONTINUE8:%.*]]
2408; INTERLEAVE:       pred.udiv.if7:
2409; INTERLEAVE-NEXT:    [[TMP20:%.*]] = or i32 [[INDEX]], 3
2410; INTERLEAVE-NEXT:    [[TMP21:%.*]] = extractelement <4 x i32> [[WIDE_LOAD]], i64 3
2411; INTERLEAVE-NEXT:    [[TMP22:%.*]] = udiv i32 [[TMP21]], [[TMP20]]
2412; INTERLEAVE-NEXT:    [[TMP23:%.*]] = insertelement <4 x i32> [[TMP19]], i32 [[TMP22]], i64 3
2413; INTERLEAVE-NEXT:    br label [[PRED_UDIV_CONTINUE8]]
2414; INTERLEAVE:       pred.udiv.continue8:
2415; INTERLEAVE-NEXT:    [[TMP24:%.*]] = phi <4 x i32> [ [[TMP19]], [[PRED_UDIV_CONTINUE6]] ], [ [[TMP23]], [[PRED_UDIV_IF7]] ]
2416; INTERLEAVE-NEXT:    br i1 [[C]], label [[PRED_UDIV_IF11:%.*]], label [[PRED_UDIV_CONTINUE12:%.*]]
2417; INTERLEAVE:       pred.udiv.if11:
2418; INTERLEAVE-NEXT:    [[TMP25:%.*]] = extractelement <4 x i32> [[WIDE_LOAD2]], i64 0
2419; INTERLEAVE-NEXT:    [[TMP26:%.*]] = udiv i32 [[TMP25]], [[TMP0]]
2420; INTERLEAVE-NEXT:    [[TMP27:%.*]] = insertelement <4 x i32> poison, i32 [[TMP26]], i64 0
2421; INTERLEAVE-NEXT:    br label [[PRED_UDIV_CONTINUE12]]
2422; INTERLEAVE:       pred.udiv.continue12:
2423; INTERLEAVE-NEXT:    [[TMP28:%.*]] = phi <4 x i32> [ poison, [[PRED_UDIV_CONTINUE8]] ], [ [[TMP27]], [[PRED_UDIV_IF11]] ]
2424; INTERLEAVE-NEXT:    br i1 [[C]], label [[PRED_UDIV_IF13:%.*]], label [[PRED_UDIV_CONTINUE14:%.*]]
2425; INTERLEAVE:       pred.udiv.if13:
2426; INTERLEAVE-NEXT:    [[TMP29:%.*]] = or i32 [[INDEX]], 5
2427; INTERLEAVE-NEXT:    [[TMP30:%.*]] = extractelement <4 x i32> [[WIDE_LOAD2]], i64 1
2428; INTERLEAVE-NEXT:    [[TMP31:%.*]] = udiv i32 [[TMP30]], [[TMP29]]
2429; INTERLEAVE-NEXT:    [[TMP32:%.*]] = insertelement <4 x i32> [[TMP28]], i32 [[TMP31]], i64 1
2430; INTERLEAVE-NEXT:    br label [[PRED_UDIV_CONTINUE14]]
2431; INTERLEAVE:       pred.udiv.continue14:
2432; INTERLEAVE-NEXT:    [[TMP33:%.*]] = phi <4 x i32> [ [[TMP28]], [[PRED_UDIV_CONTINUE12]] ], [ [[TMP32]], [[PRED_UDIV_IF13]] ]
2433; INTERLEAVE-NEXT:    br i1 [[C]], label [[PRED_UDIV_IF15:%.*]], label [[PRED_UDIV_CONTINUE16:%.*]]
2434; INTERLEAVE:       pred.udiv.if15:
2435; INTERLEAVE-NEXT:    [[TMP34:%.*]] = or i32 [[INDEX]], 6
2436; INTERLEAVE-NEXT:    [[TMP35:%.*]] = extractelement <4 x i32> [[WIDE_LOAD2]], i64 2
2437; INTERLEAVE-NEXT:    [[TMP36:%.*]] = udiv i32 [[TMP35]], [[TMP34]]
2438; INTERLEAVE-NEXT:    [[TMP37:%.*]] = insertelement <4 x i32> [[TMP33]], i32 [[TMP36]], i64 2
2439; INTERLEAVE-NEXT:    br label [[PRED_UDIV_CONTINUE16]]
2440; INTERLEAVE:       pred.udiv.continue16:
2441; INTERLEAVE-NEXT:    [[TMP38:%.*]] = phi <4 x i32> [ [[TMP33]], [[PRED_UDIV_CONTINUE14]] ], [ [[TMP37]], [[PRED_UDIV_IF15]] ]
2442; INTERLEAVE-NEXT:    br i1 [[C]], label [[PRED_UDIV_IF17:%.*]], label [[PRED_UDIV_CONTINUE18]]
2443; INTERLEAVE:       pred.udiv.if17:
2444; INTERLEAVE-NEXT:    [[TMP39:%.*]] = or i32 [[INDEX]], 7
2445; INTERLEAVE-NEXT:    [[TMP40:%.*]] = extractelement <4 x i32> [[WIDE_LOAD2]], i64 3
2446; INTERLEAVE-NEXT:    [[TMP41:%.*]] = udiv i32 [[TMP40]], [[TMP39]]
2447; INTERLEAVE-NEXT:    [[TMP42:%.*]] = insertelement <4 x i32> [[TMP38]], i32 [[TMP41]], i64 3
2448; INTERLEAVE-NEXT:    br label [[PRED_UDIV_CONTINUE18]]
2449; INTERLEAVE:       pred.udiv.continue18:
2450; INTERLEAVE-NEXT:    [[TMP43:%.*]] = phi <4 x i32> [ [[TMP38]], [[PRED_UDIV_CONTINUE16]] ], [ [[TMP42]], [[PRED_UDIV_IF17]] ]
2451; INTERLEAVE-NEXT:    [[TMP44:%.*]] = xor <4 x i1> [[BROADCAST_SPLATINSERT]], <i1 true, i1 poison, i1 poison, i1 poison>
2452; INTERLEAVE-NEXT:    [[TMP45:%.*]] = shufflevector <4 x i1> [[TMP44]], <4 x i1> poison, <4 x i32> zeroinitializer
2453; INTERLEAVE-NEXT:    [[TMP46:%.*]] = xor <4 x i1> [[BROADCAST_SPLATINSERT9]], <i1 true, i1 poison, i1 poison, i1 poison>
2454; INTERLEAVE-NEXT:    [[TMP47:%.*]] = shufflevector <4 x i1> [[TMP46]], <4 x i1> poison, <4 x i32> zeroinitializer
2455; INTERLEAVE-NEXT:    [[PREDPHI:%.*]] = select <4 x i1> [[TMP45]], <4 x i32> [[WIDE_LOAD]], <4 x i32> [[TMP24]]
2456; INTERLEAVE-NEXT:    [[PREDPHI19:%.*]] = select <4 x i1> [[TMP47]], <4 x i32> [[WIDE_LOAD2]], <4 x i32> [[TMP43]]
2457; INTERLEAVE-NEXT:    [[TMP48]] = add <4 x i32> [[PREDPHI]], [[VEC_PHI]]
2458; INTERLEAVE-NEXT:    [[TMP49]] = add <4 x i32> [[PREDPHI19]], [[VEC_PHI1]]
2459; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 8
2460; INTERLEAVE-NEXT:    [[TMP50:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
2461; INTERLEAVE-NEXT:    br i1 [[TMP50]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]]
2462; INTERLEAVE:       middle.block:
2463; INTERLEAVE-NEXT:    [[BIN_RDX:%.*]] = add <4 x i32> [[TMP49]], [[TMP48]]
2464; INTERLEAVE-NEXT:    [[TMP51:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[BIN_RDX]])
2465; INTERLEAVE-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[SMAX]], [[N_VEC]]
2466; INTERLEAVE-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
2467; INTERLEAVE:       scalar.ph:
2468; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
2469; INTERLEAVE-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP51]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
2470; INTERLEAVE-NEXT:    br label [[FOR_BODY:%.*]]
2471; INTERLEAVE:       for.body:
2472; INTERLEAVE-NEXT:    [[I:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[IF_END:%.*]] ]
2473; INTERLEAVE-NEXT:    [[SUM:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[VAR4:%.*]], [[IF_END]] ]
2474; INTERLEAVE-NEXT:    [[TMP52:%.*]] = zext i32 [[I]] to i64
2475; INTERLEAVE-NEXT:    [[VAR0:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP52]]
2476; INTERLEAVE-NEXT:    [[VAR1:%.*]] = load i32, i32* [[VAR0]], align 4
2477; INTERLEAVE-NEXT:    br i1 [[C]], label [[IF_THEN:%.*]], label [[IF_END]]
2478; INTERLEAVE:       if.then:
2479; INTERLEAVE-NEXT:    [[VAR2:%.*]] = udiv i32 [[VAR1]], [[I]]
2480; INTERLEAVE-NEXT:    br label [[IF_END]]
2481; INTERLEAVE:       if.end:
2482; INTERLEAVE-NEXT:    [[VAR3:%.*]] = phi i32 [ [[VAR2]], [[IF_THEN]] ], [ [[VAR1]], [[FOR_BODY]] ]
2483; INTERLEAVE-NEXT:    [[VAR4]] = add i32 [[VAR3]], [[SUM]]
2484; INTERLEAVE-NEXT:    [[I_NEXT]] = add nuw nsw i32 [[I]], 1
2485; INTERLEAVE-NEXT:    [[COND:%.*]] = icmp slt i32 [[I_NEXT]], [[N]]
2486; INTERLEAVE-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP25:![0-9]+]]
2487; INTERLEAVE:       for.end:
2488; INTERLEAVE-NEXT:    [[VAR5:%.*]] = phi i32 [ [[VAR4]], [[IF_END]] ], [ [[TMP51]], [[MIDDLE_BLOCK]] ]
2489; INTERLEAVE-NEXT:    ret i32 [[VAR5]]
2490;
2491entry:
2492  br label %for.body
2493
2494for.body:
2495  %i = phi i32 [ 0, %entry ], [ %i.next, %if.end ]
2496  %sum = phi i32 [ 0, %entry ], [ %var4, %if.end ]
2497  %var0 = getelementptr inbounds i32, i32* %a, i32 %i
2498  %var1 = load i32, i32* %var0, align 4
2499  br i1 %c, label %if.then, label %if.end
2500
2501if.then:
2502  %var2 = udiv i32 %var1, %i
2503  br label %if.end
2504
2505if.end:
2506  %var3 = phi i32 [ %var2, %if.then ], [ %var1, %for.body ]
2507  %var4 = add i32 %var3, %sum
2508  %i.next = add nuw nsw i32 %i, 1
2509  %cond = icmp slt i32 %i.next, %n
2510  br i1 %cond, label %for.body, label %for.end
2511
2512for.end:
2513  %var5  = phi i32 [ %var4, %if.end ]
2514  ret i32 %var5
2515}
2516
2517; Ensure we generate both a vector and a scalar induction variable. In this
2518; test, the induction variable is used by an instruction that will be
2519; vectorized (trunc) as well as an instruction that will remain in scalar form
2520; (gepelementptr).
2521;
2522;
2523;
2524
2525%pair.i16 = type { i16, i16 }
2526define void @iv_vector_and_scalar_users(%pair.i16* %p, i32 %a, i32 %n) {
2527; CHECK-LABEL: @iv_vector_and_scalar_users(
2528; CHECK-NEXT:  entry:
2529; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[N:%.*]], -1
2530; CHECK-NEXT:    [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
2531; CHECK-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
2532; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 2
2533; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
2534; CHECK:       vector.ph:
2535; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 2
2536; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]]
2537; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[A:%.*]], i32 0
2538; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
2539; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
2540; CHECK:       vector.body:
2541; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
2542; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
2543; CHECK-NEXT:    [[TMP3:%.*]] = add i64 [[INDEX]], 0
2544; CHECK-NEXT:    [[TMP4:%.*]] = add i64 [[INDEX]], 1
2545; CHECK-NEXT:    [[TMP5:%.*]] = add <2 x i32> [[BROADCAST_SPLAT]], [[VEC_IND]]
2546; CHECK-NEXT:    [[TMP6:%.*]] = trunc <2 x i32> [[TMP5]] to <2 x i16>
2547; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [[PAIR_I16:%.*]], %pair.i16* [[P:%.*]], i64 [[TMP3]], i32 1
2548; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[TMP4]], i32 1
2549; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <2 x i16> [[TMP6]], i32 0
2550; CHECK-NEXT:    store i16 [[TMP9]], i16* [[TMP7]], align 2
2551; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <2 x i16> [[TMP6]], i32 1
2552; CHECK-NEXT:    store i16 [[TMP10]], i16* [[TMP8]], align 2
2553; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
2554; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
2555; CHECK-NEXT:    [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
2556; CHECK-NEXT:    br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP26:![0-9]+]]
2557; CHECK:       middle.block:
2558; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
2559; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
2560; CHECK:       scalar.ph:
2561; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
2562; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
2563; CHECK:       for.body:
2564; CHECK-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
2565; CHECK-NEXT:    [[TMP12:%.*]] = trunc i64 [[I]] to i32
2566; CHECK-NEXT:    [[TMP13:%.*]] = add i32 [[A]], [[TMP12]]
2567; CHECK-NEXT:    [[TMP14:%.*]] = trunc i32 [[TMP13]] to i16
2568; CHECK-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[I]], i32 1
2569; CHECK-NEXT:    store i16 [[TMP14]], i16* [[TMP15]], align 2
2570; CHECK-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
2571; CHECK-NEXT:    [[TMP16:%.*]] = trunc i64 [[I_NEXT]] to i32
2572; CHECK-NEXT:    [[COND:%.*]] = icmp eq i32 [[TMP16]], [[N]]
2573; CHECK-NEXT:    br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP27:![0-9]+]]
2574; CHECK:       for.end:
2575; CHECK-NEXT:    ret void
2576;
2577; IND-LABEL: @iv_vector_and_scalar_users(
2578; IND-NEXT:  entry:
2579; IND-NEXT:    [[TMP0:%.*]] = add i32 [[N:%.*]], -1
2580; IND-NEXT:    [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
2581; IND-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
2582; IND-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp eq i32 [[TMP0]], 0
2583; IND-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
2584; IND:       vector.ph:
2585; IND-NEXT:    [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934590
2586; IND-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[A:%.*]], i64 0
2587; IND-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
2588; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
2589; IND:       vector.body:
2590; IND-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
2591; IND-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
2592; IND-NEXT:    [[TMP3:%.*]] = or i64 [[INDEX]], 1
2593; IND-NEXT:    [[TMP4:%.*]] = add <2 x i32> [[BROADCAST_SPLAT]], [[VEC_IND]]
2594; IND-NEXT:    [[TMP5:%.*]] = trunc <2 x i32> [[TMP4]] to <2 x i16>
2595; IND-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[PAIR_I16:%.*]], %pair.i16* [[P:%.*]], i64 [[INDEX]], i32 1
2596; IND-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[TMP3]], i32 1
2597; IND-NEXT:    [[TMP8:%.*]] = extractelement <2 x i16> [[TMP5]], i64 0
2598; IND-NEXT:    store i16 [[TMP8]], i16* [[TMP6]], align 2
2599; IND-NEXT:    [[TMP9:%.*]] = extractelement <2 x i16> [[TMP5]], i64 1
2600; IND-NEXT:    store i16 [[TMP9]], i16* [[TMP7]], align 2
2601; IND-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
2602; IND-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
2603; IND-NEXT:    [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
2604; IND-NEXT:    br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP26:![0-9]+]]
2605; IND:       middle.block:
2606; IND-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
2607; IND-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
2608; IND:       scalar.ph:
2609; IND-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
2610; IND-NEXT:    br label [[FOR_BODY:%.*]]
2611; IND:       for.body:
2612; IND-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
2613; IND-NEXT:    [[TMP11:%.*]] = trunc i64 [[I]] to i32
2614; IND-NEXT:    [[TMP12:%.*]] = add i32 [[TMP11]], [[A]]
2615; IND-NEXT:    [[TMP13:%.*]] = trunc i32 [[TMP12]] to i16
2616; IND-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[I]], i32 1
2617; IND-NEXT:    store i16 [[TMP13]], i16* [[TMP14]], align 2
2618; IND-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
2619; IND-NEXT:    [[TMP15:%.*]] = trunc i64 [[I_NEXT]] to i32
2620; IND-NEXT:    [[COND:%.*]] = icmp eq i32 [[TMP15]], [[N]]
2621; IND-NEXT:    br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP27:![0-9]+]]
2622; IND:       for.end:
2623; IND-NEXT:    ret void
2624;
2625; UNROLL-LABEL: @iv_vector_and_scalar_users(
2626; UNROLL-NEXT:  entry:
2627; UNROLL-NEXT:    [[TMP0:%.*]] = add i32 [[N:%.*]], -1
2628; UNROLL-NEXT:    [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
2629; UNROLL-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
2630; UNROLL-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 3
2631; UNROLL-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
2632; UNROLL:       vector.ph:
2633; UNROLL-NEXT:    [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934588
2634; UNROLL-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[A:%.*]], i64 0
2635; UNROLL-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
2636; UNROLL-NEXT:    [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <2 x i32> poison, i32 [[A]], i64 0
2637; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
2638; UNROLL:       vector.body:
2639; UNROLL-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
2640; UNROLL-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
2641; UNROLL-NEXT:    [[TMP3:%.*]] = or i64 [[INDEX]], 1
2642; UNROLL-NEXT:    [[TMP4:%.*]] = or i64 [[INDEX]], 2
2643; UNROLL-NEXT:    [[TMP5:%.*]] = or i64 [[INDEX]], 3
2644; UNROLL-NEXT:    [[TMP6:%.*]] = add <2 x i32> [[BROADCAST_SPLAT]], [[VEC_IND]]
2645; UNROLL-NEXT:    [[TMP7:%.*]] = add <2 x i32> [[BROADCAST_SPLATINSERT2]], <i32 2, i32 poison>
2646; UNROLL-NEXT:    [[TMP8:%.*]] = shufflevector <2 x i32> [[TMP7]], <2 x i32> poison, <2 x i32> zeroinitializer
2647; UNROLL-NEXT:    [[TMP9:%.*]] = add <2 x i32> [[TMP8]], [[VEC_IND]]
2648; UNROLL-NEXT:    [[TMP10:%.*]] = trunc <2 x i32> [[TMP6]] to <2 x i16>
2649; UNROLL-NEXT:    [[TMP11:%.*]] = trunc <2 x i32> [[TMP9]] to <2 x i16>
2650; UNROLL-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[PAIR_I16:%.*]], %pair.i16* [[P:%.*]], i64 [[INDEX]], i32 1
2651; UNROLL-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[TMP3]], i32 1
2652; UNROLL-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[TMP4]], i32 1
2653; UNROLL-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[TMP5]], i32 1
2654; UNROLL-NEXT:    [[TMP16:%.*]] = extractelement <2 x i16> [[TMP10]], i64 0
2655; UNROLL-NEXT:    store i16 [[TMP16]], i16* [[TMP12]], align 2
2656; UNROLL-NEXT:    [[TMP17:%.*]] = extractelement <2 x i16> [[TMP10]], i64 1
2657; UNROLL-NEXT:    store i16 [[TMP17]], i16* [[TMP13]], align 2
2658; UNROLL-NEXT:    [[TMP18:%.*]] = extractelement <2 x i16> [[TMP11]], i64 0
2659; UNROLL-NEXT:    store i16 [[TMP18]], i16* [[TMP14]], align 2
2660; UNROLL-NEXT:    [[TMP19:%.*]] = extractelement <2 x i16> [[TMP11]], i64 1
2661; UNROLL-NEXT:    store i16 [[TMP19]], i16* [[TMP15]], align 2
2662; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
2663; UNROLL-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 4, i32 4>
2664; UNROLL-NEXT:    [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
2665; UNROLL-NEXT:    br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP26:![0-9]+]]
2666; UNROLL:       middle.block:
2667; UNROLL-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
2668; UNROLL-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
2669; UNROLL:       scalar.ph:
2670; UNROLL-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
2671; UNROLL-NEXT:    br label [[FOR_BODY:%.*]]
2672; UNROLL:       for.body:
2673; UNROLL-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
2674; UNROLL-NEXT:    [[TMP21:%.*]] = trunc i64 [[I]] to i32
2675; UNROLL-NEXT:    [[TMP22:%.*]] = add i32 [[TMP21]], [[A]]
2676; UNROLL-NEXT:    [[TMP23:%.*]] = trunc i32 [[TMP22]] to i16
2677; UNROLL-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[I]], i32 1
2678; UNROLL-NEXT:    store i16 [[TMP23]], i16* [[TMP24]], align 2
2679; UNROLL-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
2680; UNROLL-NEXT:    [[TMP25:%.*]] = trunc i64 [[I_NEXT]] to i32
2681; UNROLL-NEXT:    [[COND:%.*]] = icmp eq i32 [[TMP25]], [[N]]
2682; UNROLL-NEXT:    br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP27:![0-9]+]]
2683; UNROLL:       for.end:
2684; UNROLL-NEXT:    ret void
2685;
2686; UNROLL-NO-IC-LABEL: @iv_vector_and_scalar_users(
2687; UNROLL-NO-IC-NEXT:  entry:
2688; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = add i32 [[N:%.*]], -1
2689; UNROLL-NO-IC-NEXT:    [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
2690; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
2691; UNROLL-NO-IC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 4
2692; UNROLL-NO-IC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
2693; UNROLL-NO-IC:       vector.ph:
2694; UNROLL-NO-IC-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 4
2695; UNROLL-NO-IC-NEXT:    [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]]
2696; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[A:%.*]], i32 0
2697; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
2698; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <2 x i32> poison, i32 [[A]], i32 0
2699; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLAT3:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT2]], <2 x i32> poison, <2 x i32> zeroinitializer
2700; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
2701; UNROLL-NO-IC:       vector.body:
2702; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
2703; UNROLL-NO-IC-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
2704; UNROLL-NO-IC-NEXT:    [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
2705; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = add i64 [[INDEX]], 0
2706; UNROLL-NO-IC-NEXT:    [[TMP4:%.*]] = add i64 [[INDEX]], 1
2707; UNROLL-NO-IC-NEXT:    [[TMP5:%.*]] = add i64 [[INDEX]], 2
2708; UNROLL-NO-IC-NEXT:    [[TMP6:%.*]] = add i64 [[INDEX]], 3
2709; UNROLL-NO-IC-NEXT:    [[TMP7:%.*]] = add <2 x i32> [[BROADCAST_SPLAT]], [[VEC_IND]]
2710; UNROLL-NO-IC-NEXT:    [[TMP8:%.*]] = add <2 x i32> [[BROADCAST_SPLAT3]], [[STEP_ADD]]
2711; UNROLL-NO-IC-NEXT:    [[TMP9:%.*]] = trunc <2 x i32> [[TMP7]] to <2 x i16>
2712; UNROLL-NO-IC-NEXT:    [[TMP10:%.*]] = trunc <2 x i32> [[TMP8]] to <2 x i16>
2713; UNROLL-NO-IC-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[PAIR_I16:%.*]], %pair.i16* [[P:%.*]], i64 [[TMP3]], i32 1
2714; UNROLL-NO-IC-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[TMP4]], i32 1
2715; UNROLL-NO-IC-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[TMP5]], i32 1
2716; UNROLL-NO-IC-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[TMP6]], i32 1
2717; UNROLL-NO-IC-NEXT:    [[TMP15:%.*]] = extractelement <2 x i16> [[TMP9]], i32 0
2718; UNROLL-NO-IC-NEXT:    store i16 [[TMP15]], i16* [[TMP11]], align 2
2719; UNROLL-NO-IC-NEXT:    [[TMP16:%.*]] = extractelement <2 x i16> [[TMP9]], i32 1
2720; UNROLL-NO-IC-NEXT:    store i16 [[TMP16]], i16* [[TMP12]], align 2
2721; UNROLL-NO-IC-NEXT:    [[TMP17:%.*]] = extractelement <2 x i16> [[TMP10]], i32 0
2722; UNROLL-NO-IC-NEXT:    store i16 [[TMP17]], i16* [[TMP13]], align 2
2723; UNROLL-NO-IC-NEXT:    [[TMP18:%.*]] = extractelement <2 x i16> [[TMP10]], i32 1
2724; UNROLL-NO-IC-NEXT:    store i16 [[TMP18]], i16* [[TMP14]], align 2
2725; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
2726; UNROLL-NO-IC-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[STEP_ADD]], <i32 2, i32 2>
2727; UNROLL-NO-IC-NEXT:    [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
2728; UNROLL-NO-IC-NEXT:    br i1 [[TMP19]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP26:![0-9]+]]
2729; UNROLL-NO-IC:       middle.block:
2730; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
2731; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
2732; UNROLL-NO-IC:       scalar.ph:
2733; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
2734; UNROLL-NO-IC-NEXT:    br label [[FOR_BODY:%.*]]
2735; UNROLL-NO-IC:       for.body:
2736; UNROLL-NO-IC-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
2737; UNROLL-NO-IC-NEXT:    [[TMP20:%.*]] = trunc i64 [[I]] to i32
2738; UNROLL-NO-IC-NEXT:    [[TMP21:%.*]] = add i32 [[A]], [[TMP20]]
2739; UNROLL-NO-IC-NEXT:    [[TMP22:%.*]] = trunc i32 [[TMP21]] to i16
2740; UNROLL-NO-IC-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[I]], i32 1
2741; UNROLL-NO-IC-NEXT:    store i16 [[TMP22]], i16* [[TMP23]], align 2
2742; UNROLL-NO-IC-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
2743; UNROLL-NO-IC-NEXT:    [[TMP24:%.*]] = trunc i64 [[I_NEXT]] to i32
2744; UNROLL-NO-IC-NEXT:    [[COND:%.*]] = icmp eq i32 [[TMP24]], [[N]]
2745; UNROLL-NO-IC-NEXT:    br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP27:![0-9]+]]
2746; UNROLL-NO-IC:       for.end:
2747; UNROLL-NO-IC-NEXT:    ret void
2748;
2749; INTERLEAVE-LABEL: @iv_vector_and_scalar_users(
2750; INTERLEAVE-NEXT:  entry:
2751; INTERLEAVE-NEXT:    [[TMP0:%.*]] = add i32 [[N:%.*]], -1
2752; INTERLEAVE-NEXT:    [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
2753; INTERLEAVE-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
2754; INTERLEAVE-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 7
2755; INTERLEAVE-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
2756; INTERLEAVE:       vector.ph:
2757; INTERLEAVE-NEXT:    [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934584
2758; INTERLEAVE-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[A:%.*]], i64 0
2759; INTERLEAVE-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer
2760; INTERLEAVE-NEXT:    [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <4 x i32> poison, i32 [[A]], i64 0
2761; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
2762; INTERLEAVE:       vector.body:
2763; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
2764; INTERLEAVE-NEXT:    [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
2765; INTERLEAVE-NEXT:    [[TMP3:%.*]] = or i64 [[INDEX]], 1
2766; INTERLEAVE-NEXT:    [[TMP4:%.*]] = or i64 [[INDEX]], 2
2767; INTERLEAVE-NEXT:    [[TMP5:%.*]] = or i64 [[INDEX]], 3
2768; INTERLEAVE-NEXT:    [[TMP6:%.*]] = or i64 [[INDEX]], 4
2769; INTERLEAVE-NEXT:    [[TMP7:%.*]] = or i64 [[INDEX]], 5
2770; INTERLEAVE-NEXT:    [[TMP8:%.*]] = or i64 [[INDEX]], 6
2771; INTERLEAVE-NEXT:    [[TMP9:%.*]] = or i64 [[INDEX]], 7
2772; INTERLEAVE-NEXT:    [[TMP10:%.*]] = add <4 x i32> [[BROADCAST_SPLAT]], [[VEC_IND]]
2773; INTERLEAVE-NEXT:    [[TMP11:%.*]] = add <4 x i32> [[BROADCAST_SPLATINSERT2]], <i32 4, i32 poison, i32 poison, i32 poison>
2774; INTERLEAVE-NEXT:    [[TMP12:%.*]] = shufflevector <4 x i32> [[TMP11]], <4 x i32> poison, <4 x i32> zeroinitializer
2775; INTERLEAVE-NEXT:    [[TMP13:%.*]] = add <4 x i32> [[TMP12]], [[VEC_IND]]
2776; INTERLEAVE-NEXT:    [[TMP14:%.*]] = trunc <4 x i32> [[TMP10]] to <4 x i16>
2777; INTERLEAVE-NEXT:    [[TMP15:%.*]] = trunc <4 x i32> [[TMP13]] to <4 x i16>
2778; INTERLEAVE-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [[PAIR_I16:%.*]], %pair.i16* [[P:%.*]], i64 [[INDEX]], i32 1
2779; INTERLEAVE-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[TMP3]], i32 1
2780; INTERLEAVE-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[TMP4]], i32 1
2781; INTERLEAVE-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[TMP5]], i32 1
2782; INTERLEAVE-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[TMP6]], i32 1
2783; INTERLEAVE-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[TMP7]], i32 1
2784; INTERLEAVE-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[TMP8]], i32 1
2785; INTERLEAVE-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[TMP9]], i32 1
2786; INTERLEAVE-NEXT:    [[TMP24:%.*]] = extractelement <4 x i16> [[TMP14]], i64 0
2787; INTERLEAVE-NEXT:    store i16 [[TMP24]], i16* [[TMP16]], align 2
2788; INTERLEAVE-NEXT:    [[TMP25:%.*]] = extractelement <4 x i16> [[TMP14]], i64 1
2789; INTERLEAVE-NEXT:    store i16 [[TMP25]], i16* [[TMP17]], align 2
2790; INTERLEAVE-NEXT:    [[TMP26:%.*]] = extractelement <4 x i16> [[TMP14]], i64 2
2791; INTERLEAVE-NEXT:    store i16 [[TMP26]], i16* [[TMP18]], align 2
2792; INTERLEAVE-NEXT:    [[TMP27:%.*]] = extractelement <4 x i16> [[TMP14]], i64 3
2793; INTERLEAVE-NEXT:    store i16 [[TMP27]], i16* [[TMP19]], align 2
2794; INTERLEAVE-NEXT:    [[TMP28:%.*]] = extractelement <4 x i16> [[TMP15]], i64 0
2795; INTERLEAVE-NEXT:    store i16 [[TMP28]], i16* [[TMP20]], align 2
2796; INTERLEAVE-NEXT:    [[TMP29:%.*]] = extractelement <4 x i16> [[TMP15]], i64 1
2797; INTERLEAVE-NEXT:    store i16 [[TMP29]], i16* [[TMP21]], align 2
2798; INTERLEAVE-NEXT:    [[TMP30:%.*]] = extractelement <4 x i16> [[TMP15]], i64 2
2799; INTERLEAVE-NEXT:    store i16 [[TMP30]], i16* [[TMP22]], align 2
2800; INTERLEAVE-NEXT:    [[TMP31:%.*]] = extractelement <4 x i16> [[TMP15]], i64 3
2801; INTERLEAVE-NEXT:    store i16 [[TMP31]], i16* [[TMP23]], align 2
2802; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
2803; INTERLEAVE-NEXT:    [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 8, i32 8, i32 8, i32 8>
2804; INTERLEAVE-NEXT:    [[TMP32:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
2805; INTERLEAVE-NEXT:    br i1 [[TMP32]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP26:![0-9]+]]
2806; INTERLEAVE:       middle.block:
2807; INTERLEAVE-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
2808; INTERLEAVE-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
2809; INTERLEAVE:       scalar.ph:
2810; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
2811; INTERLEAVE-NEXT:    br label [[FOR_BODY:%.*]]
2812; INTERLEAVE:       for.body:
2813; INTERLEAVE-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
2814; INTERLEAVE-NEXT:    [[TMP33:%.*]] = trunc i64 [[I]] to i32
2815; INTERLEAVE-NEXT:    [[TMP34:%.*]] = add i32 [[TMP33]], [[A]]
2816; INTERLEAVE-NEXT:    [[TMP35:%.*]] = trunc i32 [[TMP34]] to i16
2817; INTERLEAVE-NEXT:    [[TMP36:%.*]] = getelementptr inbounds [[PAIR_I16]], %pair.i16* [[P]], i64 [[I]], i32 1
2818; INTERLEAVE-NEXT:    store i16 [[TMP35]], i16* [[TMP36]], align 2
2819; INTERLEAVE-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
2820; INTERLEAVE-NEXT:    [[TMP37:%.*]] = trunc i64 [[I_NEXT]] to i32
2821; INTERLEAVE-NEXT:    [[COND:%.*]] = icmp eq i32 [[TMP37]], [[N]]
2822; INTERLEAVE-NEXT:    br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP27:![0-9]+]]
2823; INTERLEAVE:       for.end:
2824; INTERLEAVE-NEXT:    ret void
2825;
2826entry:
2827  br label %for.body
2828
2829for.body:
2830  %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ]
2831  %0 = trunc i64 %i to i32
2832  %1 = add i32 %a, %0
2833  %2 = trunc i32 %1 to i16
2834  %3 = getelementptr inbounds %pair.i16, %pair.i16* %p, i64 %i, i32 1
2835  store i16 %2, i16* %3, align 2
2836  %i.next = add nuw nsw i64 %i, 1
2837  %4 = trunc i64 %i.next to i32
2838  %cond = icmp eq i32 %4, %n
2839  br i1 %cond, label %for.end, label %for.body
2840
2841for.end:
2842  ret void
2843}
2844
2845; Make sure that the loop exit count computation does not overflow for i8 and
2846; i16. The exit count of these loops is i8/i16 max + 1. If we don't cast the
2847; induction variable to a bigger type the exit count computation will overflow
2848; to 0.
2849; PR17532
2850
2851define i32 @i8_loop() nounwind readnone ssp uwtable {
2852; CHECK-LABEL: @i8_loop(
2853; CHECK-NEXT:  entry:
2854; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
2855; CHECK:       vector.ph:
2856; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
2857; CHECK:       vector.body:
2858; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
2859; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ <i32 1, i32 -1>, [[VECTOR_PH]] ], [ [[TMP0:%.*]], [[VECTOR_BODY]] ]
2860; CHECK-NEXT:    [[TMP0]] = and <2 x i32> [[VEC_PHI]], <i32 4, i32 4>
2861; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
2862; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq i32 [[INDEX_NEXT]], 256
2863; CHECK-NEXT:    br i1 [[TMP1]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP28:![0-9]+]]
2864; CHECK:       middle.block:
2865; CHECK-NEXT:    [[TMP2:%.*]] = call i32 @llvm.vector.reduce.and.v2i32(<2 x i32> [[TMP0]])
2866; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 256, 256
2867; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
2868; CHECK:       scalar.ph:
2869; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i8 [ 0, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
2870; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ 1, [[ENTRY]] ], [ [[TMP2]], [[MIDDLE_BLOCK]] ]
2871; CHECK-NEXT:    br label [[LOOP:%.*]]
2872; CHECK:       loop:
2873; CHECK-NEXT:    [[A_0:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[A_0_AND:%.*]], [[LOOP]] ]
2874; CHECK-NEXT:    [[B_0:%.*]] = phi i8 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[B_NEXT:%.*]], [[LOOP]] ]
2875; CHECK-NEXT:    [[A_0_AND]] = and i32 [[A_0]], 4
2876; CHECK-NEXT:    [[B_NEXT]] = add i8 [[B_0]], -1
2877; CHECK-NEXT:    [[EC:%.*]] = icmp eq i8 [[B_NEXT]], 0
2878; CHECK-NEXT:    br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP29:![0-9]+]]
2879; CHECK:       exit:
2880; CHECK-NEXT:    [[A_0_AND_LCSSA:%.*]] = phi i32 [ [[A_0_AND]], [[LOOP]] ], [ [[TMP2]], [[MIDDLE_BLOCK]] ]
2881; CHECK-NEXT:    ret i32 [[A_0_AND_LCSSA]]
2882;
2883; IND-LABEL: @i8_loop(
2884; IND-NEXT:  entry:
2885; IND-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
2886; IND:       vector.ph:
2887; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
2888; IND:       vector.body:
2889; IND-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
2890; IND-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
2891; IND-NEXT:    [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], 256
2892; IND-NEXT:    br i1 [[TMP0]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP28:![0-9]+]]
2893; IND:       middle.block:
2894; IND-NEXT:    br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
2895; IND:       scalar.ph:
2896; IND-NEXT:    br label [[LOOP:%.*]]
2897; IND:       loop:
2898; IND-NEXT:    br i1 poison, label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP29:![0-9]+]]
2899; IND:       exit:
2900; IND-NEXT:    ret i32 0
2901;
2902; UNROLL-LABEL: @i8_loop(
2903; UNROLL-NEXT:  entry:
2904; UNROLL-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
2905; UNROLL:       vector.ph:
2906; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
2907; UNROLL:       vector.body:
2908; UNROLL-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
2909; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
2910; UNROLL-NEXT:    [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], 256
2911; UNROLL-NEXT:    br i1 [[TMP0]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP28:![0-9]+]]
2912; UNROLL:       middle.block:
2913; UNROLL-NEXT:    br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
2914; UNROLL:       scalar.ph:
2915; UNROLL-NEXT:    br label [[LOOP:%.*]]
2916; UNROLL:       loop:
2917; UNROLL-NEXT:    br i1 poison, label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP29:![0-9]+]]
2918; UNROLL:       exit:
2919; UNROLL-NEXT:    ret i32 0
2920;
2921; UNROLL-NO-IC-LABEL: @i8_loop(
2922; UNROLL-NO-IC-NEXT:  entry:
2923; UNROLL-NO-IC-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
2924; UNROLL-NO-IC:       vector.ph:
2925; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
2926; UNROLL-NO-IC:       vector.body:
2927; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
2928; UNROLL-NO-IC-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ <i32 1, i32 -1>, [[VECTOR_PH]] ], [ [[TMP0:%.*]], [[VECTOR_BODY]] ]
2929; UNROLL-NO-IC-NEXT:    [[VEC_PHI1:%.*]] = phi <2 x i32> [ <i32 -1, i32 -1>, [[VECTOR_PH]] ], [ [[TMP1:%.*]], [[VECTOR_BODY]] ]
2930; UNROLL-NO-IC-NEXT:    [[TMP0]] = and <2 x i32> [[VEC_PHI]], <i32 4, i32 4>
2931; UNROLL-NO-IC-NEXT:    [[TMP1]] = and <2 x i32> [[VEC_PHI1]], <i32 4, i32 4>
2932; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
2933; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = icmp eq i32 [[INDEX_NEXT]], 256
2934; UNROLL-NO-IC-NEXT:    br i1 [[TMP2]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP28:![0-9]+]]
2935; UNROLL-NO-IC:       middle.block:
2936; UNROLL-NO-IC-NEXT:    [[BIN_RDX:%.*]] = and <2 x i32> [[TMP1]], [[TMP0]]
2937; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.and.v2i32(<2 x i32> [[BIN_RDX]])
2938; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i32 256, 256
2939; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
2940; UNROLL-NO-IC:       scalar.ph:
2941; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i8 [ 0, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
2942; UNROLL-NO-IC-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ 1, [[ENTRY]] ], [ [[TMP3]], [[MIDDLE_BLOCK]] ]
2943; UNROLL-NO-IC-NEXT:    br label [[LOOP:%.*]]
2944; UNROLL-NO-IC:       loop:
2945; UNROLL-NO-IC-NEXT:    [[A_0:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[A_0_AND:%.*]], [[LOOP]] ]
2946; UNROLL-NO-IC-NEXT:    [[B_0:%.*]] = phi i8 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[B_NEXT:%.*]], [[LOOP]] ]
2947; UNROLL-NO-IC-NEXT:    [[A_0_AND]] = and i32 [[A_0]], 4
2948; UNROLL-NO-IC-NEXT:    [[B_NEXT]] = add i8 [[B_0]], -1
2949; UNROLL-NO-IC-NEXT:    [[EC:%.*]] = icmp eq i8 [[B_NEXT]], 0
2950; UNROLL-NO-IC-NEXT:    br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP29:![0-9]+]]
2951; UNROLL-NO-IC:       exit:
2952; UNROLL-NO-IC-NEXT:    [[A_0_AND_LCSSA:%.*]] = phi i32 [ [[A_0_AND]], [[LOOP]] ], [ [[TMP3]], [[MIDDLE_BLOCK]] ]
2953; UNROLL-NO-IC-NEXT:    ret i32 [[A_0_AND_LCSSA]]
2954;
2955; INTERLEAVE-LABEL: @i8_loop(
2956; INTERLEAVE-NEXT:  entry:
2957; INTERLEAVE-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
2958; INTERLEAVE:       vector.ph:
2959; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
2960; INTERLEAVE:       vector.body:
2961; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
2962; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 8
2963; INTERLEAVE-NEXT:    [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], 256
2964; INTERLEAVE-NEXT:    br i1 [[TMP0]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP28:![0-9]+]]
2965; INTERLEAVE:       middle.block:
2966; INTERLEAVE-NEXT:    br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
2967; INTERLEAVE:       scalar.ph:
2968; INTERLEAVE-NEXT:    br label [[LOOP:%.*]]
2969; INTERLEAVE:       loop:
2970; INTERLEAVE-NEXT:    br i1 poison, label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP29:![0-9]+]]
2971; INTERLEAVE:       exit:
2972; INTERLEAVE-NEXT:    ret i32 0
2973;
2974entry:
2975  br label %loop
2976
2977loop:
2978  %a.0 = phi i32 [ 1, %entry ], [ %a.0.and, %loop ]
2979  %b.0 = phi i8 [ 0, %entry ], [ %b.next, %loop ]
2980  %a.0.and = and i32 %a.0, 4
2981  %b.next = add i8 %b.0, -1
2982  %ec = icmp eq i8 %b.next, 0
2983  br i1 %ec, label %exit, label %loop
2984
2985exit:
2986  ret i32 %a.0.and
2987}
2988
2989
2990define i32 @i16_loop() nounwind readnone ssp uwtable {
2991; CHECK-LABEL: @i16_loop(
2992; CHECK-NEXT:  entry:
2993; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
2994; CHECK:       vector.ph:
2995; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
2996; CHECK:       vector.body:
2997; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
2998; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ <i32 1, i32 -1>, [[VECTOR_PH]] ], [ [[TMP0:%.*]], [[VECTOR_BODY]] ]
2999; CHECK-NEXT:    [[TMP0]] = and <2 x i32> [[VEC_PHI]], <i32 4, i32 4>
3000; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
3001; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq i32 [[INDEX_NEXT]], 65536
3002; CHECK-NEXT:    br i1 [[TMP1]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP30:![0-9]+]]
3003; CHECK:       middle.block:
3004; CHECK-NEXT:    [[TMP2:%.*]] = call i32 @llvm.vector.reduce.and.v2i32(<2 x i32> [[TMP0]])
3005; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 65536, 65536
3006; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
3007; CHECK:       scalar.ph:
3008; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i16 [ 0, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
3009; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ 1, [[ENTRY]] ], [ [[TMP2]], [[MIDDLE_BLOCK]] ]
3010; CHECK-NEXT:    br label [[LOOP:%.*]]
3011; CHECK:       loop:
3012; CHECK-NEXT:    [[A_0:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[A_0_AND:%.*]], [[LOOP]] ]
3013; CHECK-NEXT:    [[B_0:%.*]] = phi i16 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[B_0_NEXT:%.*]], [[LOOP]] ]
3014; CHECK-NEXT:    [[A_0_AND]] = and i32 [[A_0]], 4
3015; CHECK-NEXT:    [[B_0_NEXT]] = add i16 [[B_0]], -1
3016; CHECK-NEXT:    [[EC:%.*]] = icmp eq i16 [[B_0_NEXT]], 0
3017; CHECK-NEXT:    br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP31:![0-9]+]]
3018; CHECK:       exit:
3019; CHECK-NEXT:    [[A_0_AND_LCSSA:%.*]] = phi i32 [ [[A_0_AND]], [[LOOP]] ], [ [[TMP2]], [[MIDDLE_BLOCK]] ]
3020; CHECK-NEXT:    ret i32 [[A_0_AND_LCSSA]]
3021;
3022; IND-LABEL: @i16_loop(
3023; IND-NEXT:  entry:
3024; IND-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
3025; IND:       vector.ph:
3026; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
3027; IND:       vector.body:
3028; IND-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
3029; IND-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
3030; IND-NEXT:    [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], 65536
3031; IND-NEXT:    br i1 [[TMP0]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP30:![0-9]+]]
3032; IND:       middle.block:
3033; IND-NEXT:    br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
3034; IND:       scalar.ph:
3035; IND-NEXT:    br label [[LOOP:%.*]]
3036; IND:       loop:
3037; IND-NEXT:    br i1 poison, label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP31:![0-9]+]]
3038; IND:       exit:
3039; IND-NEXT:    ret i32 0
3040;
3041; UNROLL-LABEL: @i16_loop(
3042; UNROLL-NEXT:  entry:
3043; UNROLL-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
3044; UNROLL:       vector.ph:
3045; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
3046; UNROLL:       vector.body:
3047; UNROLL-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
3048; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
3049; UNROLL-NEXT:    [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], 65536
3050; UNROLL-NEXT:    br i1 [[TMP0]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP30:![0-9]+]]
3051; UNROLL:       middle.block:
3052; UNROLL-NEXT:    br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
3053; UNROLL:       scalar.ph:
3054; UNROLL-NEXT:    br label [[LOOP:%.*]]
3055; UNROLL:       loop:
3056; UNROLL-NEXT:    br i1 poison, label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP31:![0-9]+]]
3057; UNROLL:       exit:
3058; UNROLL-NEXT:    ret i32 0
3059;
3060; UNROLL-NO-IC-LABEL: @i16_loop(
3061; UNROLL-NO-IC-NEXT:  entry:
3062; UNROLL-NO-IC-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
3063; UNROLL-NO-IC:       vector.ph:
3064; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
3065; UNROLL-NO-IC:       vector.body:
3066; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
3067; UNROLL-NO-IC-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ <i32 1, i32 -1>, [[VECTOR_PH]] ], [ [[TMP0:%.*]], [[VECTOR_BODY]] ]
3068; UNROLL-NO-IC-NEXT:    [[VEC_PHI1:%.*]] = phi <2 x i32> [ <i32 -1, i32 -1>, [[VECTOR_PH]] ], [ [[TMP1:%.*]], [[VECTOR_BODY]] ]
3069; UNROLL-NO-IC-NEXT:    [[TMP0]] = and <2 x i32> [[VEC_PHI]], <i32 4, i32 4>
3070; UNROLL-NO-IC-NEXT:    [[TMP1]] = and <2 x i32> [[VEC_PHI1]], <i32 4, i32 4>
3071; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
3072; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = icmp eq i32 [[INDEX_NEXT]], 65536
3073; UNROLL-NO-IC-NEXT:    br i1 [[TMP2]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP30:![0-9]+]]
3074; UNROLL-NO-IC:       middle.block:
3075; UNROLL-NO-IC-NEXT:    [[BIN_RDX:%.*]] = and <2 x i32> [[TMP1]], [[TMP0]]
3076; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.and.v2i32(<2 x i32> [[BIN_RDX]])
3077; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i32 65536, 65536
3078; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
3079; UNROLL-NO-IC:       scalar.ph:
3080; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i16 [ 0, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
3081; UNROLL-NO-IC-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ 1, [[ENTRY]] ], [ [[TMP3]], [[MIDDLE_BLOCK]] ]
3082; UNROLL-NO-IC-NEXT:    br label [[LOOP:%.*]]
3083; UNROLL-NO-IC:       loop:
3084; UNROLL-NO-IC-NEXT:    [[A_0:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[A_0_AND:%.*]], [[LOOP]] ]
3085; UNROLL-NO-IC-NEXT:    [[B_0:%.*]] = phi i16 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[B_0_NEXT:%.*]], [[LOOP]] ]
3086; UNROLL-NO-IC-NEXT:    [[A_0_AND]] = and i32 [[A_0]], 4
3087; UNROLL-NO-IC-NEXT:    [[B_0_NEXT]] = add i16 [[B_0]], -1
3088; UNROLL-NO-IC-NEXT:    [[EC:%.*]] = icmp eq i16 [[B_0_NEXT]], 0
3089; UNROLL-NO-IC-NEXT:    br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP31:![0-9]+]]
3090; UNROLL-NO-IC:       exit:
3091; UNROLL-NO-IC-NEXT:    [[A_0_AND_LCSSA:%.*]] = phi i32 [ [[A_0_AND]], [[LOOP]] ], [ [[TMP3]], [[MIDDLE_BLOCK]] ]
3092; UNROLL-NO-IC-NEXT:    ret i32 [[A_0_AND_LCSSA]]
3093;
3094; INTERLEAVE-LABEL: @i16_loop(
3095; INTERLEAVE-NEXT:  entry:
3096; INTERLEAVE-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
3097; INTERLEAVE:       vector.ph:
3098; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
3099; INTERLEAVE:       vector.body:
3100; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
3101; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 8
3102; INTERLEAVE-NEXT:    [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], 65536
3103; INTERLEAVE-NEXT:    br i1 [[TMP0]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP30:![0-9]+]]
3104; INTERLEAVE:       middle.block:
3105; INTERLEAVE-NEXT:    br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
3106; INTERLEAVE:       scalar.ph:
3107; INTERLEAVE-NEXT:    br label [[LOOP:%.*]]
3108; INTERLEAVE:       loop:
3109; INTERLEAVE-NEXT:    br i1 poison, label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP31:![0-9]+]]
3110; INTERLEAVE:       exit:
3111; INTERLEAVE-NEXT:    ret i32 0
3112;
3113entry:
3114  br label %loop
3115
3116loop:
3117  %a.0 = phi i32 [ 1, %entry ], [ %a.0.and, %loop ]
3118  %b.0 = phi i16 [ 0, %entry ], [ %b.0.next, %loop ]
3119  %a.0.and = and i32 %a.0, 4
3120  %b.0.next = add i16 %b.0, -1
3121  %ec = icmp eq i16 %b.0.next, 0
3122  br i1 %ec, label %exit, label %loop
3123
3124exit:
3125  ret i32 %a.0.and
3126}
3127
3128; This loop has a backedge taken count of i32_max. We need to check for this
3129; condition and branch directly to the scalar loop.
3130
3131
3132
3133define i32 @max_i32_backedgetaken() nounwind readnone ssp uwtable {
3134; CHECK-LABEL: @max_i32_backedgetaken(
3135; CHECK-NEXT:  entry:
3136; CHECK-NEXT:    br i1 true, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
3137; CHECK:       vector.ph:
3138; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
3139; CHECK:       vector.body:
3140; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
3141; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ <i32 1, i32 -1>, [[VECTOR_PH]] ], [ [[TMP0:%.*]], [[VECTOR_BODY]] ]
3142; CHECK-NEXT:    [[TMP0]] = and <2 x i32> [[VEC_PHI]], <i32 4, i32 4>
3143; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
3144; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq i32 [[INDEX_NEXT]], 0
3145; CHECK-NEXT:    br i1 [[TMP1]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP32:![0-9]+]]
3146; CHECK:       middle.block:
3147; CHECK-NEXT:    [[TMP2:%.*]] = call i32 @llvm.vector.reduce.and.v2i32(<2 x i32> [[TMP0]])
3148; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 0, 0
3149; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
3150; CHECK:       scalar.ph:
3151; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ 0, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
3152; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ 1, [[ENTRY]] ], [ [[TMP2]], [[MIDDLE_BLOCK]] ]
3153; CHECK-NEXT:    br label [[LOOP:%.*]]
3154; CHECK:       loop:
3155; CHECK-NEXT:    [[A_0:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[A_0_AND:%.*]], [[LOOP]] ]
3156; CHECK-NEXT:    [[B_0:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[B_NEXT:%.*]], [[LOOP]] ]
3157; CHECK-NEXT:    [[A_0_AND]] = and i32 [[A_0]], 4
3158; CHECK-NEXT:    [[B_NEXT]] = add i32 [[B_0]], -1
3159; CHECK-NEXT:    [[EC:%.*]] = icmp eq i32 [[B_NEXT]], 0
3160; CHECK-NEXT:    br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP33:![0-9]+]]
3161; CHECK:       exit:
3162; CHECK-NEXT:    [[A_0_AND_LCSSA:%.*]] = phi i32 [ [[A_0_AND]], [[LOOP]] ], [ [[TMP2]], [[MIDDLE_BLOCK]] ]
3163; CHECK-NEXT:    ret i32 [[A_0_AND_LCSSA]]
3164;
3165; IND-LABEL: @max_i32_backedgetaken(
3166; IND-NEXT:  entry:
3167; IND-NEXT:    br i1 true, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
3168; IND:       vector.ph:
3169; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
3170; IND:       vector.body:
3171; IND-NEXT:    br i1 poison, label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP32:![0-9]+]]
3172; IND:       middle.block:
3173; IND-NEXT:    br i1 poison, label [[EXIT:%.*]], label [[SCALAR_PH]]
3174; IND:       scalar.ph:
3175; IND-NEXT:    br label [[LOOP:%.*]]
3176; IND:       loop:
3177; IND-NEXT:    [[B_0:%.*]] = phi i32 [ 0, [[SCALAR_PH]] ], [ [[B_NEXT:%.*]], [[LOOP]] ]
3178; IND-NEXT:    [[B_NEXT]] = add i32 [[B_0]], -1
3179; IND-NEXT:    [[EC:%.*]] = icmp eq i32 [[B_NEXT]], 0
3180; IND-NEXT:    br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP33:![0-9]+]]
3181; IND:       exit:
3182; IND-NEXT:    ret i32 0
3183;
3184; UNROLL-LABEL: @max_i32_backedgetaken(
3185; UNROLL-NEXT:  entry:
3186; UNROLL-NEXT:    br i1 true, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
3187; UNROLL:       vector.ph:
3188; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
3189; UNROLL:       vector.body:
3190; UNROLL-NEXT:    br i1 poison, label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP32:![0-9]+]]
3191; UNROLL:       middle.block:
3192; UNROLL-NEXT:    br i1 poison, label [[EXIT:%.*]], label [[SCALAR_PH]]
3193; UNROLL:       scalar.ph:
3194; UNROLL-NEXT:    br label [[LOOP:%.*]]
3195; UNROLL:       loop:
3196; UNROLL-NEXT:    [[B_0:%.*]] = phi i32 [ 0, [[SCALAR_PH]] ], [ [[B_NEXT:%.*]], [[LOOP]] ]
3197; UNROLL-NEXT:    [[B_NEXT]] = add i32 [[B_0]], -1
3198; UNROLL-NEXT:    [[EC:%.*]] = icmp eq i32 [[B_NEXT]], 0
3199; UNROLL-NEXT:    br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP33:![0-9]+]]
3200; UNROLL:       exit:
3201; UNROLL-NEXT:    ret i32 0
3202;
3203; UNROLL-NO-IC-LABEL: @max_i32_backedgetaken(
3204; UNROLL-NO-IC-NEXT:  entry:
3205; UNROLL-NO-IC-NEXT:    br i1 true, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
3206; UNROLL-NO-IC:       vector.ph:
3207; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
3208; UNROLL-NO-IC:       vector.body:
3209; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
3210; UNROLL-NO-IC-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ <i32 1, i32 -1>, [[VECTOR_PH]] ], [ [[TMP0:%.*]], [[VECTOR_BODY]] ]
3211; UNROLL-NO-IC-NEXT:    [[VEC_PHI1:%.*]] = phi <2 x i32> [ <i32 -1, i32 -1>, [[VECTOR_PH]] ], [ [[TMP1:%.*]], [[VECTOR_BODY]] ]
3212; UNROLL-NO-IC-NEXT:    [[TMP0]] = and <2 x i32> [[VEC_PHI]], <i32 4, i32 4>
3213; UNROLL-NO-IC-NEXT:    [[TMP1]] = and <2 x i32> [[VEC_PHI1]], <i32 4, i32 4>
3214; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
3215; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = icmp eq i32 [[INDEX_NEXT]], 0
3216; UNROLL-NO-IC-NEXT:    br i1 [[TMP2]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP32:![0-9]+]]
3217; UNROLL-NO-IC:       middle.block:
3218; UNROLL-NO-IC-NEXT:    [[BIN_RDX:%.*]] = and <2 x i32> [[TMP1]], [[TMP0]]
3219; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.and.v2i32(<2 x i32> [[BIN_RDX]])
3220; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i32 0, 0
3221; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
3222; UNROLL-NO-IC:       scalar.ph:
3223; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ 0, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
3224; UNROLL-NO-IC-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ 1, [[ENTRY]] ], [ [[TMP3]], [[MIDDLE_BLOCK]] ]
3225; UNROLL-NO-IC-NEXT:    br label [[LOOP:%.*]]
3226; UNROLL-NO-IC:       loop:
3227; UNROLL-NO-IC-NEXT:    [[A_0:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[A_0_AND:%.*]], [[LOOP]] ]
3228; UNROLL-NO-IC-NEXT:    [[B_0:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[B_NEXT:%.*]], [[LOOP]] ]
3229; UNROLL-NO-IC-NEXT:    [[A_0_AND]] = and i32 [[A_0]], 4
3230; UNROLL-NO-IC-NEXT:    [[B_NEXT]] = add i32 [[B_0]], -1
3231; UNROLL-NO-IC-NEXT:    [[EC:%.*]] = icmp eq i32 [[B_NEXT]], 0
3232; UNROLL-NO-IC-NEXT:    br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP33:![0-9]+]]
3233; UNROLL-NO-IC:       exit:
3234; UNROLL-NO-IC-NEXT:    [[A_0_AND_LCSSA:%.*]] = phi i32 [ [[A_0_AND]], [[LOOP]] ], [ [[TMP3]], [[MIDDLE_BLOCK]] ]
3235; UNROLL-NO-IC-NEXT:    ret i32 [[A_0_AND_LCSSA]]
3236;
3237; INTERLEAVE-LABEL: @max_i32_backedgetaken(
3238; INTERLEAVE-NEXT:  entry:
3239; INTERLEAVE-NEXT:    br i1 true, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
3240; INTERLEAVE:       vector.ph:
3241; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
3242; INTERLEAVE:       vector.body:
3243; INTERLEAVE-NEXT:    br i1 poison, label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP32:![0-9]+]]
3244; INTERLEAVE:       middle.block:
3245; INTERLEAVE-NEXT:    br i1 poison, label [[EXIT:%.*]], label [[SCALAR_PH]]
3246; INTERLEAVE:       scalar.ph:
3247; INTERLEAVE-NEXT:    br label [[LOOP:%.*]]
3248; INTERLEAVE:       loop:
3249; INTERLEAVE-NEXT:    [[B_0:%.*]] = phi i32 [ 0, [[SCALAR_PH]] ], [ [[B_NEXT:%.*]], [[LOOP]] ]
3250; INTERLEAVE-NEXT:    [[B_NEXT]] = add i32 [[B_0]], -1
3251; INTERLEAVE-NEXT:    [[EC:%.*]] = icmp eq i32 [[B_NEXT]], 0
3252; INTERLEAVE-NEXT:    br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP33:![0-9]+]]
3253; INTERLEAVE:       exit:
3254; INTERLEAVE-NEXT:    ret i32 0
3255;
3256entry:
3257  br label %loop
3258
3259loop:
3260  %a.0 = phi i32 [ 1, %entry ], [ %a.0.and, %loop ]
3261  %b.0 = phi i32 [ 0, %entry ], [ %b.next, %loop ]
3262  %a.0.and = and i32 %a.0, 4
3263  %b.next = add i32 %b.0, -1
3264  %ec = icmp eq i32 %b.next, 0
3265  br i1 %ec, label %exit, label %loop
3266
3267exit:
3268  ret i32 %a.0.and
3269}
3270
3271; When generating the overflow check we must sure that the induction start value
3272; is defined before the branch to the scalar preheader.
3273
3274
3275
3276@e = global i8 1, align 1
3277@d = common global i32 0, align 4
3278@c = common global i32 0, align 4
3279define i32 @testoverflowcheck() {
3280; CHECK-LABEL: @testoverflowcheck(
3281; CHECK-NEXT:  entry:
3282; CHECK-NEXT:    [[DOTPR_I:%.*]] = load i8, i8* @e, align 1
3283; CHECK-NEXT:    [[TMP0:%.*]] = load i32, i32* @d, align 4
3284; CHECK-NEXT:    [[C_PROMOTED_I:%.*]] = load i32, i32* @c, align 4
3285; CHECK-NEXT:    [[TMP1:%.*]] = sub i8 -1, [[DOTPR_I]]
3286; CHECK-NEXT:    [[TMP2:%.*]] = zext i8 [[TMP1]] to i32
3287; CHECK-NEXT:    [[TMP3:%.*]] = add nuw nsw i32 [[TMP2]], 1
3288; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP3]], 2
3289; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
3290; CHECK:       vector.ph:
3291; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[TMP3]], 2
3292; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[TMP3]], [[N_MOD_VF]]
3293; CHECK-NEXT:    [[CAST_VTC:%.*]] = trunc i32 [[N_VEC]] to i8
3294; CHECK-NEXT:    [[IND_END:%.*]] = add i8 [[DOTPR_I]], [[CAST_VTC]]
3295; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <2 x i32> <i32 -1, i32 -1>, i32 [[C_PROMOTED_I]], i32 0
3296; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[TMP0]], i32 0
3297; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
3298; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
3299; CHECK:       vector.body:
3300; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
3301; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ [[TMP4]], [[VECTOR_PH]] ], [ [[TMP5:%.*]], [[VECTOR_BODY]] ]
3302; CHECK-NEXT:    [[TMP5]] = and <2 x i32> [[BROADCAST_SPLAT]], [[VEC_PHI]]
3303; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
3304; CHECK-NEXT:    [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
3305; CHECK-NEXT:    br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP34:![0-9]+]]
3306; CHECK:       middle.block:
3307; CHECK-NEXT:    [[TMP7:%.*]] = call i32 @llvm.vector.reduce.and.v2i32(<2 x i32> [[TMP5]])
3308; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP3]], [[N_VEC]]
3309; CHECK-NEXT:    br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]]
3310; CHECK:       scalar.ph:
3311; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[DOTPR_I]], [[ENTRY:%.*]] ]
3312; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ [[C_PROMOTED_I]], [[ENTRY]] ], [ [[TMP7]], [[MIDDLE_BLOCK]] ]
3313; CHECK-NEXT:    br label [[COND_END_I:%.*]]
3314; CHECK:       cond.end.i:
3315; CHECK-NEXT:    [[INC4_I:%.*]] = phi i8 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC_I:%.*]], [[COND_END_I]] ]
3316; CHECK-NEXT:    [[AND3_I:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[AND_I:%.*]], [[COND_END_I]] ]
3317; CHECK-NEXT:    [[AND_I]] = and i32 [[TMP0]], [[AND3_I]]
3318; CHECK-NEXT:    [[INC_I]] = add i8 [[INC4_I]], 1
3319; CHECK-NEXT:    [[TOBOOL_I:%.*]] = icmp eq i8 [[INC_I]], 0
3320; CHECK-NEXT:    br i1 [[TOBOOL_I]], label [[LOOPEXIT]], label [[COND_END_I]], !llvm.loop [[LOOP35:![0-9]+]]
3321; CHECK:       loopexit:
3322; CHECK-NEXT:    [[AND_I_LCSSA:%.*]] = phi i32 [ [[AND_I]], [[COND_END_I]] ], [ [[TMP7]], [[MIDDLE_BLOCK]] ]
3323; CHECK-NEXT:    ret i32 [[AND_I_LCSSA]]
3324;
3325; IND-LABEL: @testoverflowcheck(
3326; IND-NEXT:  entry:
3327; IND-NEXT:    [[DOTPR_I:%.*]] = load i8, i8* @e, align 1
3328; IND-NEXT:    [[TMP0:%.*]] = load i32, i32* @d, align 4
3329; IND-NEXT:    [[C_PROMOTED_I:%.*]] = load i32, i32* @c, align 4
3330; IND-NEXT:    [[TMP1:%.*]] = xor i8 [[DOTPR_I]], -1
3331; IND-NEXT:    [[TMP2:%.*]] = zext i8 [[TMP1]] to i32
3332; IND-NEXT:    [[TMP3:%.*]] = add nuw nsw i32 [[TMP2]], 1
3333; IND-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp eq i8 [[DOTPR_I]], -1
3334; IND-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
3335; IND:       vector.ph:
3336; IND-NEXT:    [[N_VEC:%.*]] = and i32 [[TMP3]], 510
3337; IND-NEXT:    [[CAST_VTC:%.*]] = trunc i32 [[N_VEC]] to i8
3338; IND-NEXT:    [[IND_END:%.*]] = add i8 [[DOTPR_I]], [[CAST_VTC]]
3339; IND-NEXT:    [[TMP4:%.*]] = insertelement <2 x i32> <i32 poison, i32 -1>, i32 [[C_PROMOTED_I]], i64 0
3340; IND-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[TMP0]], i64 0
3341; IND-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
3342; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
3343; IND:       vector.body:
3344; IND-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
3345; IND-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
3346; IND-NEXT:    [[TMP5:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
3347; IND-NEXT:    br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP34:![0-9]+]]
3348; IND:       middle.block:
3349; IND-NEXT:    [[TMP6:%.*]] = and <2 x i32> [[TMP4]], [[BROADCAST_SPLAT]]
3350; IND-NEXT:    [[TMP7:%.*]] = call i32 @llvm.vector.reduce.and.v2i32(<2 x i32> [[TMP6]])
3351; IND-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP3]], [[N_VEC]]
3352; IND-NEXT:    br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]]
3353; IND:       scalar.ph:
3354; IND-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[DOTPR_I]], [[ENTRY:%.*]] ]
3355; IND-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP7]], [[MIDDLE_BLOCK]] ], [ [[C_PROMOTED_I]], [[ENTRY]] ]
3356; IND-NEXT:    br label [[COND_END_I:%.*]]
3357; IND:       cond.end.i:
3358; IND-NEXT:    [[INC4_I:%.*]] = phi i8 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC_I:%.*]], [[COND_END_I]] ]
3359; IND-NEXT:    [[TMP8:%.*]] = and i32 [[BC_MERGE_RDX]], [[TMP0]]
3360; IND-NEXT:    [[INC_I]] = add i8 [[INC4_I]], 1
3361; IND-NEXT:    [[TOBOOL_I:%.*]] = icmp eq i8 [[INC_I]], 0
3362; IND-NEXT:    br i1 [[TOBOOL_I]], label [[LOOPEXIT]], label [[COND_END_I]], !llvm.loop [[LOOP35:![0-9]+]]
3363; IND:       loopexit:
3364; IND-NEXT:    [[AND_I_LCSSA:%.*]] = phi i32 [ [[TMP8]], [[COND_END_I]] ], [ [[TMP7]], [[MIDDLE_BLOCK]] ]
3365; IND-NEXT:    ret i32 [[AND_I_LCSSA]]
3366;
3367; UNROLL-LABEL: @testoverflowcheck(
3368; UNROLL-NEXT:  entry:
3369; UNROLL-NEXT:    [[DOTPR_I:%.*]] = load i8, i8* @e, align 1
3370; UNROLL-NEXT:    [[TMP0:%.*]] = load i32, i32* @d, align 4
3371; UNROLL-NEXT:    [[C_PROMOTED_I:%.*]] = load i32, i32* @c, align 4
3372; UNROLL-NEXT:    [[TMP1:%.*]] = xor i8 [[DOTPR_I]], -1
3373; UNROLL-NEXT:    [[TMP2:%.*]] = zext i8 [[TMP1]] to i32
3374; UNROLL-NEXT:    [[TMP3:%.*]] = add nuw nsw i32 [[TMP2]], 1
3375; UNROLL-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ugt i8 [[DOTPR_I]], -4
3376; UNROLL-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
3377; UNROLL:       vector.ph:
3378; UNROLL-NEXT:    [[N_VEC:%.*]] = and i32 [[TMP3]], 508
3379; UNROLL-NEXT:    [[CAST_VTC:%.*]] = trunc i32 [[N_VEC]] to i8
3380; UNROLL-NEXT:    [[IND_END:%.*]] = add i8 [[DOTPR_I]], [[CAST_VTC]]
3381; UNROLL-NEXT:    [[TMP4:%.*]] = insertelement <2 x i32> <i32 poison, i32 -1>, i32 [[C_PROMOTED_I]], i64 0
3382; UNROLL-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[TMP0]], i64 0
3383; UNROLL-NEXT:    [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <2 x i32> poison, i32 [[TMP0]], i64 0
3384; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
3385; UNROLL:       vector.body:
3386; UNROLL-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
3387; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
3388; UNROLL-NEXT:    [[TMP5:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
3389; UNROLL-NEXT:    br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP34:![0-9]+]]
3390; UNROLL:       middle.block:
3391; UNROLL-NEXT:    [[TMP6:%.*]] = and <2 x i32> [[BROADCAST_SPLATINSERT2]], [[BROADCAST_SPLATINSERT]]
3392; UNROLL-NEXT:    [[TMP7:%.*]] = shufflevector <2 x i32> [[TMP6]], <2 x i32> poison, <2 x i32> zeroinitializer
3393; UNROLL-NEXT:    [[BIN_RDX:%.*]] = and <2 x i32> [[TMP7]], [[TMP4]]
3394; UNROLL-NEXT:    [[TMP8:%.*]] = call i32 @llvm.vector.reduce.and.v2i32(<2 x i32> [[BIN_RDX]])
3395; UNROLL-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP3]], [[N_VEC]]
3396; UNROLL-NEXT:    br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]]
3397; UNROLL:       scalar.ph:
3398; UNROLL-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[DOTPR_I]], [[ENTRY:%.*]] ]
3399; UNROLL-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP8]], [[MIDDLE_BLOCK]] ], [ [[C_PROMOTED_I]], [[ENTRY]] ]
3400; UNROLL-NEXT:    br label [[COND_END_I:%.*]]
3401; UNROLL:       cond.end.i:
3402; UNROLL-NEXT:    [[INC4_I:%.*]] = phi i8 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC_I:%.*]], [[COND_END_I]] ]
3403; UNROLL-NEXT:    [[TMP9:%.*]] = and i32 [[BC_MERGE_RDX]], [[TMP0]]
3404; UNROLL-NEXT:    [[INC_I]] = add i8 [[INC4_I]], 1
3405; UNROLL-NEXT:    [[TOBOOL_I:%.*]] = icmp eq i8 [[INC_I]], 0
3406; UNROLL-NEXT:    br i1 [[TOBOOL_I]], label [[LOOPEXIT]], label [[COND_END_I]], !llvm.loop [[LOOP35:![0-9]+]]
3407; UNROLL:       loopexit:
3408; UNROLL-NEXT:    [[AND_I_LCSSA:%.*]] = phi i32 [ [[TMP9]], [[COND_END_I]] ], [ [[TMP8]], [[MIDDLE_BLOCK]] ]
3409; UNROLL-NEXT:    ret i32 [[AND_I_LCSSA]]
3410;
3411; UNROLL-NO-IC-LABEL: @testoverflowcheck(
3412; UNROLL-NO-IC-NEXT:  entry:
3413; UNROLL-NO-IC-NEXT:    [[DOTPR_I:%.*]] = load i8, i8* @e, align 1
3414; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = load i32, i32* @d, align 4
3415; UNROLL-NO-IC-NEXT:    [[C_PROMOTED_I:%.*]] = load i32, i32* @c, align 4
3416; UNROLL-NO-IC-NEXT:    [[TMP1:%.*]] = sub i8 -1, [[DOTPR_I]]
3417; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = zext i8 [[TMP1]] to i32
3418; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = add nuw nsw i32 [[TMP2]], 1
3419; UNROLL-NO-IC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP3]], 4
3420; UNROLL-NO-IC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
3421; UNROLL-NO-IC:       vector.ph:
3422; UNROLL-NO-IC-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[TMP3]], 4
3423; UNROLL-NO-IC-NEXT:    [[N_VEC:%.*]] = sub i32 [[TMP3]], [[N_MOD_VF]]
3424; UNROLL-NO-IC-NEXT:    [[CAST_VTC:%.*]] = trunc i32 [[N_VEC]] to i8
3425; UNROLL-NO-IC-NEXT:    [[IND_END:%.*]] = add i8 [[DOTPR_I]], [[CAST_VTC]]
3426; UNROLL-NO-IC-NEXT:    [[TMP4:%.*]] = insertelement <2 x i32> <i32 -1, i32 -1>, i32 [[C_PROMOTED_I]], i32 0
3427; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[TMP0]], i32 0
3428; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
3429; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <2 x i32> poison, i32 [[TMP0]], i32 0
3430; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLAT3:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT2]], <2 x i32> poison, <2 x i32> zeroinitializer
3431; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
3432; UNROLL-NO-IC:       vector.body:
3433; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
3434; UNROLL-NO-IC-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ [[TMP4]], [[VECTOR_PH]] ], [ [[TMP5:%.*]], [[VECTOR_BODY]] ]
3435; UNROLL-NO-IC-NEXT:    [[VEC_PHI1:%.*]] = phi <2 x i32> [ <i32 -1, i32 -1>, [[VECTOR_PH]] ], [ [[TMP6:%.*]], [[VECTOR_BODY]] ]
3436; UNROLL-NO-IC-NEXT:    [[TMP5]] = and <2 x i32> [[BROADCAST_SPLAT]], [[VEC_PHI]]
3437; UNROLL-NO-IC-NEXT:    [[TMP6]] = and <2 x i32> [[BROADCAST_SPLAT3]], [[VEC_PHI1]]
3438; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
3439; UNROLL-NO-IC-NEXT:    [[TMP7:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
3440; UNROLL-NO-IC-NEXT:    br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP34:![0-9]+]]
3441; UNROLL-NO-IC:       middle.block:
3442; UNROLL-NO-IC-NEXT:    [[BIN_RDX:%.*]] = and <2 x i32> [[TMP6]], [[TMP5]]
3443; UNROLL-NO-IC-NEXT:    [[TMP8:%.*]] = call i32 @llvm.vector.reduce.and.v2i32(<2 x i32> [[BIN_RDX]])
3444; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP3]], [[N_VEC]]
3445; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]]
3446; UNROLL-NO-IC:       scalar.ph:
3447; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[DOTPR_I]], [[ENTRY:%.*]] ]
3448; UNROLL-NO-IC-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ [[C_PROMOTED_I]], [[ENTRY]] ], [ [[TMP8]], [[MIDDLE_BLOCK]] ]
3449; UNROLL-NO-IC-NEXT:    br label [[COND_END_I:%.*]]
3450; UNROLL-NO-IC:       cond.end.i:
3451; UNROLL-NO-IC-NEXT:    [[INC4_I:%.*]] = phi i8 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC_I:%.*]], [[COND_END_I]] ]
3452; UNROLL-NO-IC-NEXT:    [[AND3_I:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[AND_I:%.*]], [[COND_END_I]] ]
3453; UNROLL-NO-IC-NEXT:    [[AND_I]] = and i32 [[TMP0]], [[AND3_I]]
3454; UNROLL-NO-IC-NEXT:    [[INC_I]] = add i8 [[INC4_I]], 1
3455; UNROLL-NO-IC-NEXT:    [[TOBOOL_I:%.*]] = icmp eq i8 [[INC_I]], 0
3456; UNROLL-NO-IC-NEXT:    br i1 [[TOBOOL_I]], label [[LOOPEXIT]], label [[COND_END_I]], !llvm.loop [[LOOP35:![0-9]+]]
3457; UNROLL-NO-IC:       loopexit:
3458; UNROLL-NO-IC-NEXT:    [[AND_I_LCSSA:%.*]] = phi i32 [ [[AND_I]], [[COND_END_I]] ], [ [[TMP8]], [[MIDDLE_BLOCK]] ]
3459; UNROLL-NO-IC-NEXT:    ret i32 [[AND_I_LCSSA]]
3460;
3461; INTERLEAVE-LABEL: @testoverflowcheck(
3462; INTERLEAVE-NEXT:  entry:
3463; INTERLEAVE-NEXT:    [[DOTPR_I:%.*]] = load i8, i8* @e, align 1
3464; INTERLEAVE-NEXT:    [[TMP0:%.*]] = load i32, i32* @d, align 4
3465; INTERLEAVE-NEXT:    [[C_PROMOTED_I:%.*]] = load i32, i32* @c, align 4
3466; INTERLEAVE-NEXT:    [[TMP1:%.*]] = xor i8 [[DOTPR_I]], -1
3467; INTERLEAVE-NEXT:    [[TMP2:%.*]] = zext i8 [[TMP1]] to i32
3468; INTERLEAVE-NEXT:    [[TMP3:%.*]] = add nuw nsw i32 [[TMP2]], 1
3469; INTERLEAVE-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ugt i8 [[DOTPR_I]], -8
3470; INTERLEAVE-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
3471; INTERLEAVE:       vector.ph:
3472; INTERLEAVE-NEXT:    [[N_VEC:%.*]] = and i32 [[TMP3]], 504
3473; INTERLEAVE-NEXT:    [[CAST_VTC:%.*]] = trunc i32 [[N_VEC]] to i8
3474; INTERLEAVE-NEXT:    [[IND_END:%.*]] = add i8 [[DOTPR_I]], [[CAST_VTC]]
3475; INTERLEAVE-NEXT:    [[TMP4:%.*]] = insertelement <4 x i32> <i32 poison, i32 -1, i32 -1, i32 -1>, i32 [[C_PROMOTED_I]], i64 0
3476; INTERLEAVE-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[TMP0]], i64 0
3477; INTERLEAVE-NEXT:    [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <4 x i32> poison, i32 [[TMP0]], i64 0
3478; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
3479; INTERLEAVE:       vector.body:
3480; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
3481; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 8
3482; INTERLEAVE-NEXT:    [[TMP5:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
3483; INTERLEAVE-NEXT:    br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP34:![0-9]+]]
3484; INTERLEAVE:       middle.block:
3485; INTERLEAVE-NEXT:    [[TMP6:%.*]] = and <4 x i32> [[BROADCAST_SPLATINSERT2]], [[BROADCAST_SPLATINSERT]]
3486; INTERLEAVE-NEXT:    [[TMP7:%.*]] = shufflevector <4 x i32> [[TMP6]], <4 x i32> poison, <4 x i32> zeroinitializer
3487; INTERLEAVE-NEXT:    [[BIN_RDX:%.*]] = and <4 x i32> [[TMP7]], [[TMP4]]
3488; INTERLEAVE-NEXT:    [[TMP8:%.*]] = call i32 @llvm.vector.reduce.and.v4i32(<4 x i32> [[BIN_RDX]])
3489; INTERLEAVE-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP3]], [[N_VEC]]
3490; INTERLEAVE-NEXT:    br i1 [[CMP_N]], label [[LOOPEXIT:%.*]], label [[SCALAR_PH]]
3491; INTERLEAVE:       scalar.ph:
3492; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[DOTPR_I]], [[ENTRY:%.*]] ]
3493; INTERLEAVE-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP8]], [[MIDDLE_BLOCK]] ], [ [[C_PROMOTED_I]], [[ENTRY]] ]
3494; INTERLEAVE-NEXT:    br label [[COND_END_I:%.*]]
3495; INTERLEAVE:       cond.end.i:
3496; INTERLEAVE-NEXT:    [[INC4_I:%.*]] = phi i8 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC_I:%.*]], [[COND_END_I]] ]
3497; INTERLEAVE-NEXT:    [[TMP9:%.*]] = and i32 [[BC_MERGE_RDX]], [[TMP0]]
3498; INTERLEAVE-NEXT:    [[INC_I]] = add i8 [[INC4_I]], 1
3499; INTERLEAVE-NEXT:    [[TOBOOL_I:%.*]] = icmp eq i8 [[INC_I]], 0
3500; INTERLEAVE-NEXT:    br i1 [[TOBOOL_I]], label [[LOOPEXIT]], label [[COND_END_I]], !llvm.loop [[LOOP35:![0-9]+]]
3501; INTERLEAVE:       loopexit:
3502; INTERLEAVE-NEXT:    [[AND_I_LCSSA:%.*]] = phi i32 [ [[TMP9]], [[COND_END_I]] ], [ [[TMP8]], [[MIDDLE_BLOCK]] ]
3503; INTERLEAVE-NEXT:    ret i32 [[AND_I_LCSSA]]
3504;
3505entry:
3506  %.pr.i = load i8, i8* @e, align 1
3507  %0 = load i32, i32* @d, align 4
3508  %c.promoted.i = load i32, i32* @c, align 4
3509  br label %cond.end.i
3510
3511cond.end.i:
3512  %inc4.i = phi i8 [ %.pr.i, %entry ], [ %inc.i, %cond.end.i ]
3513  %and3.i = phi i32 [ %c.promoted.i, %entry ], [ %and.i, %cond.end.i ]
3514  %and.i = and i32 %0, %and3.i
3515  %inc.i = add i8 %inc4.i, 1
3516  %tobool.i = icmp eq i8 %inc.i, 0
3517  br i1 %tobool.i, label %loopexit, label %cond.end.i
3518
3519loopexit:
3520  ret i32 %and.i
3521}
3522
3523; The SCEV expression of %sphi is (zext i8 {%t,+,1}<%loop> to i32)
3524; In order to recognize %sphi as an induction PHI and vectorize this loop,
3525; we need to convert the SCEV expression into an AddRecExpr.
3526; The expression gets converted to {zext i8 %t to i32,+,1}.
3527
3528define void @wrappingindvars1(i8 %t, i32 %len, i32 *%A) {
3529; CHECK-LABEL: @wrappingindvars1(
3530; CHECK-NEXT:  entry:
3531; CHECK-NEXT:    [[ST:%.*]] = zext i8 [[T:%.*]] to i16
3532; CHECK-NEXT:    [[EXT:%.*]] = zext i8 [[T]] to i32
3533; CHECK-NEXT:    [[ECMP:%.*]] = icmp ult i16 [[ST]], 42
3534; CHECK-NEXT:    br i1 [[ECMP]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]]
3535; CHECK:       loop.preheader:
3536; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[LEN:%.*]], 1
3537; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 2
3538; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
3539; CHECK:       vector.scevcheck:
3540; CHECK-NEXT:    [[TMP1:%.*]] = trunc i32 [[LEN]] to i8
3541; CHECK-NEXT:    [[TMP2:%.*]] = add i8 [[T]], [[TMP1]]
3542; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult i8 [[TMP2]], [[T]]
3543; CHECK-NEXT:    [[TMP4:%.*]] = icmp ugt i32 [[LEN]], 255
3544; CHECK-NEXT:    [[TMP5:%.*]] = or i1 [[TMP3]], [[TMP4]]
3545; CHECK-NEXT:    [[TMP6:%.*]] = trunc i32 [[LEN]] to i8
3546; CHECK-NEXT:    [[TMP7:%.*]] = add i8 [[T]], [[TMP6]]
3547; CHECK-NEXT:    [[TMP8:%.*]] = icmp slt i8 [[TMP7]], [[T]]
3548; CHECK-NEXT:    [[TMP9:%.*]] = icmp ugt i32 [[LEN]], 255
3549; CHECK-NEXT:    [[TMP10:%.*]] = or i1 [[TMP8]], [[TMP9]]
3550; CHECK-NEXT:    [[TMP11:%.*]] = or i1 [[TMP5]], [[TMP10]]
3551; CHECK-NEXT:    br i1 [[TMP11]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
3552; CHECK:       vector.ph:
3553; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[TMP0]], 2
3554; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[TMP0]], [[N_MOD_VF]]
3555; CHECK-NEXT:    [[CAST_VTC:%.*]] = trunc i32 [[N_VEC]] to i8
3556; CHECK-NEXT:    [[IND_END:%.*]] = add i8 [[T]], [[CAST_VTC]]
3557; CHECK-NEXT:    [[IND_END2:%.*]] = add i32 [[EXT]], [[N_VEC]]
3558; CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[EXT]], i32 0
3559; CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
3560; CHECK-NEXT:    [[INDUCTION:%.*]] = add <2 x i32> [[DOTSPLAT]], <i32 0, i32 1>
3561; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
3562; CHECK:       vector.body:
3563; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
3564; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
3565; CHECK-NEXT:    [[TMP12:%.*]] = trunc i32 [[INDEX]] to i8
3566; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = add i8 [[T]], [[TMP12]]
3567; CHECK-NEXT:    [[TMP13:%.*]] = add i8 [[OFFSET_IDX]], 0
3568; CHECK-NEXT:    [[TMP14:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i8 [[TMP13]]
3569; CHECK-NEXT:    [[TMP15:%.*]] = getelementptr inbounds i32, i32* [[TMP14]], i32 0
3570; CHECK-NEXT:    [[TMP16:%.*]] = bitcast i32* [[TMP15]] to <2 x i32>*
3571; CHECK-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP16]], align 4
3572; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
3573; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
3574; CHECK-NEXT:    [[TMP17:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
3575; CHECK-NEXT:    br i1 [[TMP17]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP36:![0-9]+]]
3576; CHECK:       middle.block:
3577; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
3578; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]]
3579; CHECK:       scalar.ph:
3580; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[T]], [[LOOP_PREHEADER]] ], [ [[T]], [[VECTOR_SCEVCHECK]] ]
3581; CHECK-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[LOOP_PREHEADER]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
3582; CHECK-NEXT:    [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[IND_END2]], [[MIDDLE_BLOCK]] ], [ [[EXT]], [[LOOP_PREHEADER]] ], [ [[EXT]], [[VECTOR_SCEVCHECK]] ]
3583; CHECK-NEXT:    br label [[LOOP:%.*]]
3584; CHECK:       loop:
3585; CHECK-NEXT:    [[IDX:%.*]] = phi i8 [ [[IDX_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
3586; CHECK-NEXT:    [[IDX_B:%.*]] = phi i32 [ [[IDX_B_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]
3587; CHECK-NEXT:    [[SPHI:%.*]] = phi i32 [ [[IDX_INC_EXT:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ]
3588; CHECK-NEXT:    [[PTR:%.*]] = getelementptr inbounds i32, i32* [[A]], i8 [[IDX]]
3589; CHECK-NEXT:    store i32 [[SPHI]], i32* [[PTR]], align 4
3590; CHECK-NEXT:    [[IDX_INC]] = add i8 [[IDX]], 1
3591; CHECK-NEXT:    [[IDX_INC_EXT]] = zext i8 [[IDX_INC]] to i32
3592; CHECK-NEXT:    [[IDX_B_INC]] = add nuw nsw i32 [[IDX_B]], 1
3593; CHECK-NEXT:    [[C:%.*]] = icmp ult i32 [[IDX_B]], [[LEN]]
3594; CHECK-NEXT:    br i1 [[C]], label [[LOOP]], label [[EXIT_LOOPEXIT]], !llvm.loop [[LOOP37:![0-9]+]]
3595; CHECK:       exit.loopexit:
3596; CHECK-NEXT:    br label [[EXIT]]
3597; CHECK:       exit:
3598; CHECK-NEXT:    ret void
3599;
3600; IND-LABEL: @wrappingindvars1(
3601; IND-NEXT:  entry:
3602; IND-NEXT:    [[EXT:%.*]] = zext i8 [[T:%.*]] to i32
3603; IND-NEXT:    [[ECMP:%.*]] = icmp ult i8 [[T]], 42
3604; IND-NEXT:    br i1 [[ECMP]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]]
3605; IND:       loop.preheader:
3606; IND-NEXT:    [[TMP0:%.*]] = add i32 [[LEN:%.*]], 1
3607; IND-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 2
3608; IND-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
3609; IND:       vector.scevcheck:
3610; IND-NEXT:    [[TMP1:%.*]] = trunc i32 [[LEN]] to i8
3611; IND-NEXT:    [[TMP2:%.*]] = xor i8 [[T]], -1
3612; IND-NEXT:    [[TMP3:%.*]] = icmp ult i8 [[TMP2]], [[TMP1]]
3613; IND-NEXT:    [[TMP4:%.*]] = trunc i32 [[LEN]] to i8
3614; IND-NEXT:    [[TMP5:%.*]] = add i8 [[TMP4]], [[T]]
3615; IND-NEXT:    [[TMP6:%.*]] = icmp slt i8 [[TMP5]], [[T]]
3616; IND-NEXT:    [[TMP7:%.*]] = icmp ugt i32 [[LEN]], 255
3617; IND-NEXT:    [[TMP8:%.*]] = or i1 [[TMP6]], [[TMP7]]
3618; IND-NEXT:    [[TMP9:%.*]] = or i1 [[TMP3]], [[TMP8]]
3619; IND-NEXT:    br i1 [[TMP9]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
3620; IND:       vector.ph:
3621; IND-NEXT:    [[N_VEC:%.*]] = and i32 [[TMP0]], -2
3622; IND-NEXT:    [[CAST_VTC:%.*]] = trunc i32 [[N_VEC]] to i8
3623; IND-NEXT:    [[IND_END:%.*]] = add i8 [[CAST_VTC]], [[T]]
3624; IND-NEXT:    [[IND_END2:%.*]] = add i32 [[N_VEC]], [[EXT]]
3625; IND-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[EXT]], i64 0
3626; IND-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
3627; IND-NEXT:    [[INDUCTION:%.*]] = add nuw nsw <2 x i32> [[DOTSPLAT]], <i32 0, i32 1>
3628; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
3629; IND:       vector.body:
3630; IND-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
3631; IND-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
3632; IND-NEXT:    [[TMP10:%.*]] = trunc i32 [[INDEX]] to i8
3633; IND-NEXT:    [[OFFSET_IDX:%.*]] = add i8 [[TMP10]], [[T]]
3634; IND-NEXT:    [[TMP11:%.*]] = sext i8 [[OFFSET_IDX]] to i64
3635; IND-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP11]]
3636; IND-NEXT:    [[TMP13:%.*]] = bitcast i32* [[TMP12]] to <2 x i32>*
3637; IND-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP13]], align 4
3638; IND-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
3639; IND-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
3640; IND-NEXT:    [[TMP14:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
3641; IND-NEXT:    br i1 [[TMP14]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP36:![0-9]+]]
3642; IND:       middle.block:
3643; IND-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
3644; IND-NEXT:    br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]]
3645; IND:       scalar.ph:
3646; IND-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[T]], [[LOOP_PREHEADER]] ], [ [[T]], [[VECTOR_SCEVCHECK]] ]
3647; IND-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[LOOP_PREHEADER]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
3648; IND-NEXT:    [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[IND_END2]], [[MIDDLE_BLOCK]] ], [ [[EXT]], [[LOOP_PREHEADER]] ], [ [[EXT]], [[VECTOR_SCEVCHECK]] ]
3649; IND-NEXT:    br label [[LOOP:%.*]]
3650; IND:       loop:
3651; IND-NEXT:    [[IDX:%.*]] = phi i8 [ [[IDX_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
3652; IND-NEXT:    [[IDX_B:%.*]] = phi i32 [ [[IDX_B_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]
3653; IND-NEXT:    [[SPHI:%.*]] = phi i32 [ [[IDX_INC_EXT:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ]
3654; IND-NEXT:    [[TMP15:%.*]] = sext i8 [[IDX]] to i64
3655; IND-NEXT:    [[PTR:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP15]]
3656; IND-NEXT:    store i32 [[SPHI]], i32* [[PTR]], align 4
3657; IND-NEXT:    [[IDX_INC]] = add i8 [[IDX]], 1
3658; IND-NEXT:    [[IDX_INC_EXT]] = zext i8 [[IDX_INC]] to i32
3659; IND-NEXT:    [[IDX_B_INC]] = add nuw nsw i32 [[IDX_B]], 1
3660; IND-NEXT:    [[C:%.*]] = icmp ult i32 [[IDX_B]], [[LEN]]
3661; IND-NEXT:    br i1 [[C]], label [[LOOP]], label [[EXIT_LOOPEXIT]], !llvm.loop [[LOOP37:![0-9]+]]
3662; IND:       exit.loopexit:
3663; IND-NEXT:    br label [[EXIT]]
3664; IND:       exit:
3665; IND-NEXT:    ret void
3666;
3667; UNROLL-LABEL: @wrappingindvars1(
3668; UNROLL-NEXT:  entry:
3669; UNROLL-NEXT:    [[EXT:%.*]] = zext i8 [[T:%.*]] to i32
3670; UNROLL-NEXT:    [[ECMP:%.*]] = icmp ult i8 [[T]], 42
3671; UNROLL-NEXT:    br i1 [[ECMP]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]]
3672; UNROLL:       loop.preheader:
3673; UNROLL-NEXT:    [[TMP0:%.*]] = add i32 [[LEN:%.*]], 1
3674; UNROLL-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 4
3675; UNROLL-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
3676; UNROLL:       vector.scevcheck:
3677; UNROLL-NEXT:    [[TMP1:%.*]] = trunc i32 [[LEN]] to i8
3678; UNROLL-NEXT:    [[TMP2:%.*]] = xor i8 [[T]], -1
3679; UNROLL-NEXT:    [[TMP3:%.*]] = icmp ult i8 [[TMP2]], [[TMP1]]
3680; UNROLL-NEXT:    [[TMP4:%.*]] = trunc i32 [[LEN]] to i8
3681; UNROLL-NEXT:    [[TMP5:%.*]] = add i8 [[TMP4]], [[T]]
3682; UNROLL-NEXT:    [[TMP6:%.*]] = icmp slt i8 [[TMP5]], [[T]]
3683; UNROLL-NEXT:    [[TMP7:%.*]] = icmp ugt i32 [[LEN]], 255
3684; UNROLL-NEXT:    [[TMP8:%.*]] = or i1 [[TMP6]], [[TMP7]]
3685; UNROLL-NEXT:    [[TMP9:%.*]] = or i1 [[TMP3]], [[TMP8]]
3686; UNROLL-NEXT:    br i1 [[TMP9]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
3687; UNROLL:       vector.ph:
3688; UNROLL-NEXT:    [[N_VEC:%.*]] = and i32 [[TMP0]], -4
3689; UNROLL-NEXT:    [[CAST_VTC:%.*]] = trunc i32 [[N_VEC]] to i8
3690; UNROLL-NEXT:    [[IND_END:%.*]] = add i8 [[CAST_VTC]], [[T]]
3691; UNROLL-NEXT:    [[IND_END2:%.*]] = add i32 [[N_VEC]], [[EXT]]
3692; UNROLL-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[EXT]], i64 0
3693; UNROLL-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
3694; UNROLL-NEXT:    [[INDUCTION:%.*]] = add nuw nsw <2 x i32> [[DOTSPLAT]], <i32 0, i32 1>
3695; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
3696; UNROLL:       vector.body:
3697; UNROLL-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
3698; UNROLL-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
3699; UNROLL-NEXT:    [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
3700; UNROLL-NEXT:    [[TMP10:%.*]] = trunc i32 [[INDEX]] to i8
3701; UNROLL-NEXT:    [[OFFSET_IDX:%.*]] = add i8 [[TMP10]], [[T]]
3702; UNROLL-NEXT:    [[TMP11:%.*]] = sext i8 [[OFFSET_IDX]] to i64
3703; UNROLL-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP11]]
3704; UNROLL-NEXT:    [[TMP13:%.*]] = bitcast i32* [[TMP12]] to <2 x i32>*
3705; UNROLL-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP13]], align 4
3706; UNROLL-NEXT:    [[TMP14:%.*]] = getelementptr inbounds i32, i32* [[TMP12]], i64 2
3707; UNROLL-NEXT:    [[TMP15:%.*]] = bitcast i32* [[TMP14]] to <2 x i32>*
3708; UNROLL-NEXT:    store <2 x i32> [[STEP_ADD]], <2 x i32>* [[TMP15]], align 4
3709; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
3710; UNROLL-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 4, i32 4>
3711; UNROLL-NEXT:    [[TMP16:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
3712; UNROLL-NEXT:    br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP36:![0-9]+]]
3713; UNROLL:       middle.block:
3714; UNROLL-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
3715; UNROLL-NEXT:    br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]]
3716; UNROLL:       scalar.ph:
3717; UNROLL-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[T]], [[LOOP_PREHEADER]] ], [ [[T]], [[VECTOR_SCEVCHECK]] ]
3718; UNROLL-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[LOOP_PREHEADER]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
3719; UNROLL-NEXT:    [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[IND_END2]], [[MIDDLE_BLOCK]] ], [ [[EXT]], [[LOOP_PREHEADER]] ], [ [[EXT]], [[VECTOR_SCEVCHECK]] ]
3720; UNROLL-NEXT:    br label [[LOOP:%.*]]
3721; UNROLL:       loop:
3722; UNROLL-NEXT:    [[IDX:%.*]] = phi i8 [ [[IDX_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
3723; UNROLL-NEXT:    [[IDX_B:%.*]] = phi i32 [ [[IDX_B_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]
3724; UNROLL-NEXT:    [[SPHI:%.*]] = phi i32 [ [[IDX_INC_EXT:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ]
3725; UNROLL-NEXT:    [[TMP17:%.*]] = sext i8 [[IDX]] to i64
3726; UNROLL-NEXT:    [[PTR:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP17]]
3727; UNROLL-NEXT:    store i32 [[SPHI]], i32* [[PTR]], align 4
3728; UNROLL-NEXT:    [[IDX_INC]] = add i8 [[IDX]], 1
3729; UNROLL-NEXT:    [[IDX_INC_EXT]] = zext i8 [[IDX_INC]] to i32
3730; UNROLL-NEXT:    [[IDX_B_INC]] = add nuw nsw i32 [[IDX_B]], 1
3731; UNROLL-NEXT:    [[C:%.*]] = icmp ult i32 [[IDX_B]], [[LEN]]
3732; UNROLL-NEXT:    br i1 [[C]], label [[LOOP]], label [[EXIT_LOOPEXIT]], !llvm.loop [[LOOP37:![0-9]+]]
3733; UNROLL:       exit.loopexit:
3734; UNROLL-NEXT:    br label [[EXIT]]
3735; UNROLL:       exit:
3736; UNROLL-NEXT:    ret void
3737;
3738; UNROLL-NO-IC-LABEL: @wrappingindvars1(
3739; UNROLL-NO-IC-NEXT:  entry:
3740; UNROLL-NO-IC-NEXT:    [[ST:%.*]] = zext i8 [[T:%.*]] to i16
3741; UNROLL-NO-IC-NEXT:    [[EXT:%.*]] = zext i8 [[T]] to i32
3742; UNROLL-NO-IC-NEXT:    [[ECMP:%.*]] = icmp ult i16 [[ST]], 42
3743; UNROLL-NO-IC-NEXT:    br i1 [[ECMP]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]]
3744; UNROLL-NO-IC:       loop.preheader:
3745; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = add i32 [[LEN:%.*]], 1
3746; UNROLL-NO-IC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 4
3747; UNROLL-NO-IC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
3748; UNROLL-NO-IC:       vector.scevcheck:
3749; UNROLL-NO-IC-NEXT:    [[TMP1:%.*]] = trunc i32 [[LEN]] to i8
3750; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = add i8 [[T]], [[TMP1]]
3751; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = icmp ult i8 [[TMP2]], [[T]]
3752; UNROLL-NO-IC-NEXT:    [[TMP4:%.*]] = icmp ugt i32 [[LEN]], 255
3753; UNROLL-NO-IC-NEXT:    [[TMP5:%.*]] = or i1 [[TMP3]], [[TMP4]]
3754; UNROLL-NO-IC-NEXT:    [[TMP6:%.*]] = trunc i32 [[LEN]] to i8
3755; UNROLL-NO-IC-NEXT:    [[TMP7:%.*]] = add i8 [[T]], [[TMP6]]
3756; UNROLL-NO-IC-NEXT:    [[TMP8:%.*]] = icmp slt i8 [[TMP7]], [[T]]
3757; UNROLL-NO-IC-NEXT:    [[TMP9:%.*]] = icmp ugt i32 [[LEN]], 255
3758; UNROLL-NO-IC-NEXT:    [[TMP10:%.*]] = or i1 [[TMP8]], [[TMP9]]
3759; UNROLL-NO-IC-NEXT:    [[TMP11:%.*]] = or i1 [[TMP5]], [[TMP10]]
3760; UNROLL-NO-IC-NEXT:    br i1 [[TMP11]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
3761; UNROLL-NO-IC:       vector.ph:
3762; UNROLL-NO-IC-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[TMP0]], 4
3763; UNROLL-NO-IC-NEXT:    [[N_VEC:%.*]] = sub i32 [[TMP0]], [[N_MOD_VF]]
3764; UNROLL-NO-IC-NEXT:    [[CAST_VTC:%.*]] = trunc i32 [[N_VEC]] to i8
3765; UNROLL-NO-IC-NEXT:    [[IND_END:%.*]] = add i8 [[T]], [[CAST_VTC]]
3766; UNROLL-NO-IC-NEXT:    [[IND_END2:%.*]] = add i32 [[EXT]], [[N_VEC]]
3767; UNROLL-NO-IC-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[EXT]], i32 0
3768; UNROLL-NO-IC-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
3769; UNROLL-NO-IC-NEXT:    [[INDUCTION:%.*]] = add <2 x i32> [[DOTSPLAT]], <i32 0, i32 1>
3770; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
3771; UNROLL-NO-IC:       vector.body:
3772; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
3773; UNROLL-NO-IC-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
3774; UNROLL-NO-IC-NEXT:    [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
3775; UNROLL-NO-IC-NEXT:    [[TMP12:%.*]] = trunc i32 [[INDEX]] to i8
3776; UNROLL-NO-IC-NEXT:    [[OFFSET_IDX:%.*]] = add i8 [[T]], [[TMP12]]
3777; UNROLL-NO-IC-NEXT:    [[TMP13:%.*]] = add i8 [[OFFSET_IDX]], 0
3778; UNROLL-NO-IC-NEXT:    [[TMP14:%.*]] = add i8 [[OFFSET_IDX]], 2
3779; UNROLL-NO-IC-NEXT:    [[TMP15:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i8 [[TMP13]]
3780; UNROLL-NO-IC-NEXT:    [[TMP16:%.*]] = getelementptr inbounds i32, i32* [[A]], i8 [[TMP14]]
3781; UNROLL-NO-IC-NEXT:    [[TMP17:%.*]] = getelementptr inbounds i32, i32* [[TMP15]], i32 0
3782; UNROLL-NO-IC-NEXT:    [[TMP18:%.*]] = bitcast i32* [[TMP17]] to <2 x i32>*
3783; UNROLL-NO-IC-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP18]], align 4
3784; UNROLL-NO-IC-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i32, i32* [[TMP15]], i32 2
3785; UNROLL-NO-IC-NEXT:    [[TMP20:%.*]] = bitcast i32* [[TMP19]] to <2 x i32>*
3786; UNROLL-NO-IC-NEXT:    store <2 x i32> [[STEP_ADD]], <2 x i32>* [[TMP20]], align 4
3787; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
3788; UNROLL-NO-IC-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[STEP_ADD]], <i32 2, i32 2>
3789; UNROLL-NO-IC-NEXT:    [[TMP21:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
3790; UNROLL-NO-IC-NEXT:    br i1 [[TMP21]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP36:![0-9]+]]
3791; UNROLL-NO-IC:       middle.block:
3792; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
3793; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]]
3794; UNROLL-NO-IC:       scalar.ph:
3795; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[T]], [[LOOP_PREHEADER]] ], [ [[T]], [[VECTOR_SCEVCHECK]] ]
3796; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[LOOP_PREHEADER]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
3797; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[IND_END2]], [[MIDDLE_BLOCK]] ], [ [[EXT]], [[LOOP_PREHEADER]] ], [ [[EXT]], [[VECTOR_SCEVCHECK]] ]
3798; UNROLL-NO-IC-NEXT:    br label [[LOOP:%.*]]
3799; UNROLL-NO-IC:       loop:
3800; UNROLL-NO-IC-NEXT:    [[IDX:%.*]] = phi i8 [ [[IDX_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
3801; UNROLL-NO-IC-NEXT:    [[IDX_B:%.*]] = phi i32 [ [[IDX_B_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]
3802; UNROLL-NO-IC-NEXT:    [[SPHI:%.*]] = phi i32 [ [[IDX_INC_EXT:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ]
3803; UNROLL-NO-IC-NEXT:    [[PTR:%.*]] = getelementptr inbounds i32, i32* [[A]], i8 [[IDX]]
3804; UNROLL-NO-IC-NEXT:    store i32 [[SPHI]], i32* [[PTR]], align 4
3805; UNROLL-NO-IC-NEXT:    [[IDX_INC]] = add i8 [[IDX]], 1
3806; UNROLL-NO-IC-NEXT:    [[IDX_INC_EXT]] = zext i8 [[IDX_INC]] to i32
3807; UNROLL-NO-IC-NEXT:    [[IDX_B_INC]] = add nuw nsw i32 [[IDX_B]], 1
3808; UNROLL-NO-IC-NEXT:    [[C:%.*]] = icmp ult i32 [[IDX_B]], [[LEN]]
3809; UNROLL-NO-IC-NEXT:    br i1 [[C]], label [[LOOP]], label [[EXIT_LOOPEXIT]], !llvm.loop [[LOOP37:![0-9]+]]
3810; UNROLL-NO-IC:       exit.loopexit:
3811; UNROLL-NO-IC-NEXT:    br label [[EXIT]]
3812; UNROLL-NO-IC:       exit:
3813; UNROLL-NO-IC-NEXT:    ret void
3814;
3815; INTERLEAVE-LABEL: @wrappingindvars1(
3816; INTERLEAVE-NEXT:  entry:
3817; INTERLEAVE-NEXT:    [[EXT:%.*]] = zext i8 [[T:%.*]] to i32
3818; INTERLEAVE-NEXT:    [[ECMP:%.*]] = icmp ult i8 [[T]], 42
3819; INTERLEAVE-NEXT:    br i1 [[ECMP]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]]
3820; INTERLEAVE:       loop.preheader:
3821; INTERLEAVE-NEXT:    [[TMP0:%.*]] = add i32 [[LEN:%.*]], 1
3822; INTERLEAVE-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 8
3823; INTERLEAVE-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
3824; INTERLEAVE:       vector.scevcheck:
3825; INTERLEAVE-NEXT:    [[TMP1:%.*]] = trunc i32 [[LEN]] to i8
3826; INTERLEAVE-NEXT:    [[TMP2:%.*]] = xor i8 [[T]], -1
3827; INTERLEAVE-NEXT:    [[TMP3:%.*]] = icmp ult i8 [[TMP2]], [[TMP1]]
3828; INTERLEAVE-NEXT:    [[TMP4:%.*]] = trunc i32 [[LEN]] to i8
3829; INTERLEAVE-NEXT:    [[TMP5:%.*]] = add i8 [[TMP4]], [[T]]
3830; INTERLEAVE-NEXT:    [[TMP6:%.*]] = icmp slt i8 [[TMP5]], [[T]]
3831; INTERLEAVE-NEXT:    [[TMP7:%.*]] = icmp ugt i32 [[LEN]], 255
3832; INTERLEAVE-NEXT:    [[TMP8:%.*]] = or i1 [[TMP6]], [[TMP7]]
3833; INTERLEAVE-NEXT:    [[TMP9:%.*]] = or i1 [[TMP3]], [[TMP8]]
3834; INTERLEAVE-NEXT:    br i1 [[TMP9]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
3835; INTERLEAVE:       vector.ph:
3836; INTERLEAVE-NEXT:    [[N_VEC:%.*]] = and i32 [[TMP0]], -8
3837; INTERLEAVE-NEXT:    [[CAST_VTC:%.*]] = trunc i32 [[N_VEC]] to i8
3838; INTERLEAVE-NEXT:    [[IND_END:%.*]] = add i8 [[CAST_VTC]], [[T]]
3839; INTERLEAVE-NEXT:    [[IND_END2:%.*]] = add i32 [[N_VEC]], [[EXT]]
3840; INTERLEAVE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[EXT]], i64 0
3841; INTERLEAVE-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <4 x i32> [[DOTSPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer
3842; INTERLEAVE-NEXT:    [[INDUCTION:%.*]] = add nuw nsw <4 x i32> [[DOTSPLAT]], <i32 0, i32 1, i32 2, i32 3>
3843; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
3844; INTERLEAVE:       vector.body:
3845; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
3846; INTERLEAVE-NEXT:    [[VEC_IND:%.*]] = phi <4 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
3847; INTERLEAVE-NEXT:    [[STEP_ADD:%.*]] = add <4 x i32> [[VEC_IND]], <i32 4, i32 4, i32 4, i32 4>
3848; INTERLEAVE-NEXT:    [[TMP10:%.*]] = trunc i32 [[INDEX]] to i8
3849; INTERLEAVE-NEXT:    [[OFFSET_IDX:%.*]] = add i8 [[TMP10]], [[T]]
3850; INTERLEAVE-NEXT:    [[TMP11:%.*]] = sext i8 [[OFFSET_IDX]] to i64
3851; INTERLEAVE-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP11]]
3852; INTERLEAVE-NEXT:    [[TMP13:%.*]] = bitcast i32* [[TMP12]] to <4 x i32>*
3853; INTERLEAVE-NEXT:    store <4 x i32> [[VEC_IND]], <4 x i32>* [[TMP13]], align 4
3854; INTERLEAVE-NEXT:    [[TMP14:%.*]] = getelementptr inbounds i32, i32* [[TMP12]], i64 4
3855; INTERLEAVE-NEXT:    [[TMP15:%.*]] = bitcast i32* [[TMP14]] to <4 x i32>*
3856; INTERLEAVE-NEXT:    store <4 x i32> [[STEP_ADD]], <4 x i32>* [[TMP15]], align 4
3857; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 8
3858; INTERLEAVE-NEXT:    [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 8, i32 8, i32 8, i32 8>
3859; INTERLEAVE-NEXT:    [[TMP16:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
3860; INTERLEAVE-NEXT:    br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP36:![0-9]+]]
3861; INTERLEAVE:       middle.block:
3862; INTERLEAVE-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
3863; INTERLEAVE-NEXT:    br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]]
3864; INTERLEAVE:       scalar.ph:
3865; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[T]], [[LOOP_PREHEADER]] ], [ [[T]], [[VECTOR_SCEVCHECK]] ]
3866; INTERLEAVE-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[LOOP_PREHEADER]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
3867; INTERLEAVE-NEXT:    [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[IND_END2]], [[MIDDLE_BLOCK]] ], [ [[EXT]], [[LOOP_PREHEADER]] ], [ [[EXT]], [[VECTOR_SCEVCHECK]] ]
3868; INTERLEAVE-NEXT:    br label [[LOOP:%.*]]
3869; INTERLEAVE:       loop:
3870; INTERLEAVE-NEXT:    [[IDX:%.*]] = phi i8 [ [[IDX_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
3871; INTERLEAVE-NEXT:    [[IDX_B:%.*]] = phi i32 [ [[IDX_B_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]
3872; INTERLEAVE-NEXT:    [[SPHI:%.*]] = phi i32 [ [[IDX_INC_EXT:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ]
3873; INTERLEAVE-NEXT:    [[TMP17:%.*]] = sext i8 [[IDX]] to i64
3874; INTERLEAVE-NEXT:    [[PTR:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP17]]
3875; INTERLEAVE-NEXT:    store i32 [[SPHI]], i32* [[PTR]], align 4
3876; INTERLEAVE-NEXT:    [[IDX_INC]] = add i8 [[IDX]], 1
3877; INTERLEAVE-NEXT:    [[IDX_INC_EXT]] = zext i8 [[IDX_INC]] to i32
3878; INTERLEAVE-NEXT:    [[IDX_B_INC]] = add nuw nsw i32 [[IDX_B]], 1
3879; INTERLEAVE-NEXT:    [[C:%.*]] = icmp ult i32 [[IDX_B]], [[LEN]]
3880; INTERLEAVE-NEXT:    br i1 [[C]], label [[LOOP]], label [[EXIT_LOOPEXIT]], !llvm.loop [[LOOP37:![0-9]+]]
3881; INTERLEAVE:       exit.loopexit:
3882; INTERLEAVE-NEXT:    br label [[EXIT]]
3883; INTERLEAVE:       exit:
3884; INTERLEAVE-NEXT:    ret void
3885;
3886  entry:
3887  %st = zext i8 %t to i16
3888  %ext = zext i8 %t to i32
3889  %ecmp = icmp ult i16 %st, 42
3890  br i1 %ecmp, label %loop, label %exit
3891
3892  loop:
3893
3894  %idx = phi i8 [ %t, %entry ], [ %idx.inc, %loop ]
3895  %idx.b = phi i32 [ 0, %entry ], [ %idx.b.inc, %loop ]
3896  %sphi = phi i32 [ %ext, %entry ], [%idx.inc.ext, %loop]
3897
3898  %ptr = getelementptr inbounds i32, i32* %A, i8 %idx
3899  store i32 %sphi, i32* %ptr
3900
3901  %idx.inc = add i8 %idx, 1
3902  %idx.inc.ext = zext i8 %idx.inc to i32
3903  %idx.b.inc = add nuw nsw i32 %idx.b, 1
3904
3905  %c = icmp ult i32 %idx.b, %len
3906  br i1 %c, label %loop, label %exit
3907
3908  exit:
3909  ret void
3910}
3911
3912; The SCEV expression of %sphi is (4 * (zext i8 {%t,+,1}<%loop> to i32))
3913; In order to recognize %sphi as an induction PHI and vectorize this loop,
3914; we need to convert the SCEV expression into an AddRecExpr.
3915; The expression gets converted to ({4 * (zext %t to i32),+,4}).
3916define void @wrappingindvars2(i8 %t, i32 %len, i32 *%A) {
3917; CHECK-LABEL: @wrappingindvars2(
3918; CHECK-NEXT:  entry:
3919; CHECK-NEXT:    [[ST:%.*]] = zext i8 [[T:%.*]] to i16
3920; CHECK-NEXT:    [[EXT:%.*]] = zext i8 [[T]] to i32
3921; CHECK-NEXT:    [[EXT_MUL:%.*]] = mul i32 [[EXT]], 4
3922; CHECK-NEXT:    [[ECMP:%.*]] = icmp ult i16 [[ST]], 42
3923; CHECK-NEXT:    br i1 [[ECMP]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]]
3924; CHECK:       loop.preheader:
3925; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[LEN:%.*]], 1
3926; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 2
3927; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
3928; CHECK:       vector.scevcheck:
3929; CHECK-NEXT:    [[TMP1:%.*]] = trunc i32 [[LEN]] to i8
3930; CHECK-NEXT:    [[TMP2:%.*]] = add i8 [[T]], [[TMP1]]
3931; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult i8 [[TMP2]], [[T]]
3932; CHECK-NEXT:    [[TMP4:%.*]] = icmp ugt i32 [[LEN]], 255
3933; CHECK-NEXT:    [[TMP5:%.*]] = or i1 [[TMP3]], [[TMP4]]
3934; CHECK-NEXT:    [[TMP6:%.*]] = trunc i32 [[LEN]] to i8
3935; CHECK-NEXT:    [[TMP7:%.*]] = add i8 [[T]], [[TMP6]]
3936; CHECK-NEXT:    [[TMP8:%.*]] = icmp slt i8 [[TMP7]], [[T]]
3937; CHECK-NEXT:    [[TMP9:%.*]] = icmp ugt i32 [[LEN]], 255
3938; CHECK-NEXT:    [[TMP10:%.*]] = or i1 [[TMP8]], [[TMP9]]
3939; CHECK-NEXT:    [[TMP11:%.*]] = or i1 [[TMP5]], [[TMP10]]
3940; CHECK-NEXT:    br i1 [[TMP11]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
3941; CHECK:       vector.ph:
3942; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[TMP0]], 2
3943; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[TMP0]], [[N_MOD_VF]]
3944; CHECK-NEXT:    [[CAST_VTC:%.*]] = trunc i32 [[N_VEC]] to i8
3945; CHECK-NEXT:    [[IND_END:%.*]] = add i8 [[T]], [[CAST_VTC]]
3946; CHECK-NEXT:    [[TMP12:%.*]] = mul i32 [[N_VEC]], 4
3947; CHECK-NEXT:    [[IND_END1:%.*]] = add i32 [[EXT_MUL]], [[TMP12]]
3948; CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[EXT_MUL]], i32 0
3949; CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
3950; CHECK-NEXT:    [[INDUCTION:%.*]] = add <2 x i32> [[DOTSPLAT]], <i32 0, i32 4>
3951; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
3952; CHECK:       vector.body:
3953; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
3954; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
3955; CHECK-NEXT:    [[TMP13:%.*]] = trunc i32 [[INDEX]] to i8
3956; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = add i8 [[T]], [[TMP13]]
3957; CHECK-NEXT:    [[TMP14:%.*]] = add i8 [[OFFSET_IDX]], 0
3958; CHECK-NEXT:    [[TMP15:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i8 [[TMP14]]
3959; CHECK-NEXT:    [[TMP16:%.*]] = getelementptr inbounds i32, i32* [[TMP15]], i32 0
3960; CHECK-NEXT:    [[TMP17:%.*]] = bitcast i32* [[TMP16]] to <2 x i32>*
3961; CHECK-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP17]], align 4
3962; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
3963; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 8, i32 8>
3964; CHECK-NEXT:    [[TMP18:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
3965; CHECK-NEXT:    br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP38:![0-9]+]]
3966; CHECK:       middle.block:
3967; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
3968; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]]
3969; CHECK:       scalar.ph:
3970; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[T]], [[LOOP_PREHEADER]] ], [ [[T]], [[VECTOR_SCEVCHECK]] ]
3971; CHECK-NEXT:    [[BC_RESUME_VAL2:%.*]] = phi i32 [ [[IND_END1]], [[MIDDLE_BLOCK]] ], [ [[EXT_MUL]], [[LOOP_PREHEADER]] ], [ [[EXT_MUL]], [[VECTOR_SCEVCHECK]] ]
3972; CHECK-NEXT:    [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[LOOP_PREHEADER]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
3973; CHECK-NEXT:    br label [[LOOP:%.*]]
3974; CHECK:       loop:
3975; CHECK-NEXT:    [[IDX:%.*]] = phi i8 [ [[IDX_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
3976; CHECK-NEXT:    [[SPHI:%.*]] = phi i32 [ [[MUL:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL2]], [[SCALAR_PH]] ]
3977; CHECK-NEXT:    [[IDX_B:%.*]] = phi i32 [ [[IDX_B_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ]
3978; CHECK-NEXT:    [[PTR:%.*]] = getelementptr inbounds i32, i32* [[A]], i8 [[IDX]]
3979; CHECK-NEXT:    store i32 [[SPHI]], i32* [[PTR]], align 4
3980; CHECK-NEXT:    [[IDX_INC]] = add i8 [[IDX]], 1
3981; CHECK-NEXT:    [[IDX_INC_EXT:%.*]] = zext i8 [[IDX_INC]] to i32
3982; CHECK-NEXT:    [[MUL]] = mul i32 [[IDX_INC_EXT]], 4
3983; CHECK-NEXT:    [[IDX_B_INC]] = add nuw nsw i32 [[IDX_B]], 1
3984; CHECK-NEXT:    [[C:%.*]] = icmp ult i32 [[IDX_B]], [[LEN]]
3985; CHECK-NEXT:    br i1 [[C]], label [[LOOP]], label [[EXIT_LOOPEXIT]], !llvm.loop [[LOOP39:![0-9]+]]
3986; CHECK:       exit.loopexit:
3987; CHECK-NEXT:    br label [[EXIT]]
3988; CHECK:       exit:
3989; CHECK-NEXT:    ret void
3990;
3991; IND-LABEL: @wrappingindvars2(
3992; IND-NEXT:  entry:
3993; IND-NEXT:    [[EXT:%.*]] = zext i8 [[T:%.*]] to i32
3994; IND-NEXT:    [[EXT_MUL:%.*]] = shl nuw nsw i32 [[EXT]], 2
3995; IND-NEXT:    [[ECMP:%.*]] = icmp ult i8 [[T]], 42
3996; IND-NEXT:    br i1 [[ECMP]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]]
3997; IND:       loop.preheader:
3998; IND-NEXT:    [[TMP0:%.*]] = add i32 [[LEN:%.*]], 1
3999; IND-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 2
4000; IND-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
4001; IND:       vector.scevcheck:
4002; IND-NEXT:    [[TMP1:%.*]] = trunc i32 [[LEN]] to i8
4003; IND-NEXT:    [[TMP2:%.*]] = xor i8 [[T]], -1
4004; IND-NEXT:    [[TMP3:%.*]] = icmp ult i8 [[TMP2]], [[TMP1]]
4005; IND-NEXT:    [[TMP4:%.*]] = trunc i32 [[LEN]] to i8
4006; IND-NEXT:    [[TMP5:%.*]] = add i8 [[TMP4]], [[T]]
4007; IND-NEXT:    [[TMP6:%.*]] = icmp slt i8 [[TMP5]], [[T]]
4008; IND-NEXT:    [[TMP7:%.*]] = icmp ugt i32 [[LEN]], 255
4009; IND-NEXT:    [[TMP8:%.*]] = or i1 [[TMP6]], [[TMP7]]
4010; IND-NEXT:    [[TMP9:%.*]] = or i1 [[TMP3]], [[TMP8]]
4011; IND-NEXT:    br i1 [[TMP9]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
4012; IND:       vector.ph:
4013; IND-NEXT:    [[N_VEC:%.*]] = and i32 [[TMP0]], -2
4014; IND-NEXT:    [[CAST_VTC:%.*]] = trunc i32 [[N_VEC]] to i8
4015; IND-NEXT:    [[IND_END:%.*]] = add i8 [[CAST_VTC]], [[T]]
4016; IND-NEXT:    [[TMP10:%.*]] = add i32 [[N_VEC]], [[EXT]]
4017; IND-NEXT:    [[IND_END1:%.*]] = shl i32 [[TMP10]], 2
4018; IND-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[EXT_MUL]], i64 0
4019; IND-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
4020; IND-NEXT:    [[INDUCTION:%.*]] = add nuw nsw <2 x i32> [[DOTSPLAT]], <i32 0, i32 4>
4021; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
4022; IND:       vector.body:
4023; IND-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4024; IND-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
4025; IND-NEXT:    [[TMP11:%.*]] = trunc i32 [[INDEX]] to i8
4026; IND-NEXT:    [[OFFSET_IDX:%.*]] = add i8 [[TMP11]], [[T]]
4027; IND-NEXT:    [[TMP12:%.*]] = sext i8 [[OFFSET_IDX]] to i64
4028; IND-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP12]]
4029; IND-NEXT:    [[TMP14:%.*]] = bitcast i32* [[TMP13]] to <2 x i32>*
4030; IND-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP14]], align 4
4031; IND-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
4032; IND-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 8, i32 8>
4033; IND-NEXT:    [[TMP15:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
4034; IND-NEXT:    br i1 [[TMP15]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP38:![0-9]+]]
4035; IND:       middle.block:
4036; IND-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
4037; IND-NEXT:    br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]]
4038; IND:       scalar.ph:
4039; IND-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[T]], [[LOOP_PREHEADER]] ], [ [[T]], [[VECTOR_SCEVCHECK]] ]
4040; IND-NEXT:    [[BC_RESUME_VAL2:%.*]] = phi i32 [ [[IND_END1]], [[MIDDLE_BLOCK]] ], [ [[EXT_MUL]], [[LOOP_PREHEADER]] ], [ [[EXT_MUL]], [[VECTOR_SCEVCHECK]] ]
4041; IND-NEXT:    [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[LOOP_PREHEADER]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
4042; IND-NEXT:    br label [[LOOP:%.*]]
4043; IND:       loop:
4044; IND-NEXT:    [[IDX:%.*]] = phi i8 [ [[IDX_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
4045; IND-NEXT:    [[SPHI:%.*]] = phi i32 [ [[MUL:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL2]], [[SCALAR_PH]] ]
4046; IND-NEXT:    [[IDX_B:%.*]] = phi i32 [ [[IDX_B_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ]
4047; IND-NEXT:    [[TMP16:%.*]] = sext i8 [[IDX]] to i64
4048; IND-NEXT:    [[PTR:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP16]]
4049; IND-NEXT:    store i32 [[SPHI]], i32* [[PTR]], align 4
4050; IND-NEXT:    [[IDX_INC]] = add i8 [[IDX]], 1
4051; IND-NEXT:    [[IDX_INC_EXT:%.*]] = zext i8 [[IDX_INC]] to i32
4052; IND-NEXT:    [[MUL]] = shl nuw nsw i32 [[IDX_INC_EXT]], 2
4053; IND-NEXT:    [[IDX_B_INC]] = add nuw nsw i32 [[IDX_B]], 1
4054; IND-NEXT:    [[C:%.*]] = icmp ult i32 [[IDX_B]], [[LEN]]
4055; IND-NEXT:    br i1 [[C]], label [[LOOP]], label [[EXIT_LOOPEXIT]], !llvm.loop [[LOOP39:![0-9]+]]
4056; IND:       exit.loopexit:
4057; IND-NEXT:    br label [[EXIT]]
4058; IND:       exit:
4059; IND-NEXT:    ret void
4060;
4061; UNROLL-LABEL: @wrappingindvars2(
4062; UNROLL-NEXT:  entry:
4063; UNROLL-NEXT:    [[EXT:%.*]] = zext i8 [[T:%.*]] to i32
4064; UNROLL-NEXT:    [[EXT_MUL:%.*]] = shl nuw nsw i32 [[EXT]], 2
4065; UNROLL-NEXT:    [[ECMP:%.*]] = icmp ult i8 [[T]], 42
4066; UNROLL-NEXT:    br i1 [[ECMP]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]]
4067; UNROLL:       loop.preheader:
4068; UNROLL-NEXT:    [[TMP0:%.*]] = add i32 [[LEN:%.*]], 1
4069; UNROLL-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 4
4070; UNROLL-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
4071; UNROLL:       vector.scevcheck:
4072; UNROLL-NEXT:    [[TMP1:%.*]] = trunc i32 [[LEN]] to i8
4073; UNROLL-NEXT:    [[TMP2:%.*]] = xor i8 [[T]], -1
4074; UNROLL-NEXT:    [[TMP3:%.*]] = icmp ult i8 [[TMP2]], [[TMP1]]
4075; UNROLL-NEXT:    [[TMP4:%.*]] = trunc i32 [[LEN]] to i8
4076; UNROLL-NEXT:    [[TMP5:%.*]] = add i8 [[TMP4]], [[T]]
4077; UNROLL-NEXT:    [[TMP6:%.*]] = icmp slt i8 [[TMP5]], [[T]]
4078; UNROLL-NEXT:    [[TMP7:%.*]] = icmp ugt i32 [[LEN]], 255
4079; UNROLL-NEXT:    [[TMP8:%.*]] = or i1 [[TMP6]], [[TMP7]]
4080; UNROLL-NEXT:    [[TMP9:%.*]] = or i1 [[TMP3]], [[TMP8]]
4081; UNROLL-NEXT:    br i1 [[TMP9]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
4082; UNROLL:       vector.ph:
4083; UNROLL-NEXT:    [[N_VEC:%.*]] = and i32 [[TMP0]], -4
4084; UNROLL-NEXT:    [[CAST_VTC:%.*]] = trunc i32 [[N_VEC]] to i8
4085; UNROLL-NEXT:    [[IND_END:%.*]] = add i8 [[CAST_VTC]], [[T]]
4086; UNROLL-NEXT:    [[TMP10:%.*]] = add i32 [[N_VEC]], [[EXT]]
4087; UNROLL-NEXT:    [[IND_END1:%.*]] = shl i32 [[TMP10]], 2
4088; UNROLL-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[EXT_MUL]], i64 0
4089; UNROLL-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
4090; UNROLL-NEXT:    [[INDUCTION:%.*]] = add nuw nsw <2 x i32> [[DOTSPLAT]], <i32 0, i32 4>
4091; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
4092; UNROLL:       vector.body:
4093; UNROLL-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4094; UNROLL-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
4095; UNROLL-NEXT:    [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], <i32 8, i32 8>
4096; UNROLL-NEXT:    [[TMP11:%.*]] = trunc i32 [[INDEX]] to i8
4097; UNROLL-NEXT:    [[OFFSET_IDX:%.*]] = add i8 [[TMP11]], [[T]]
4098; UNROLL-NEXT:    [[TMP12:%.*]] = sext i8 [[OFFSET_IDX]] to i64
4099; UNROLL-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP12]]
4100; UNROLL-NEXT:    [[TMP14:%.*]] = bitcast i32* [[TMP13]] to <2 x i32>*
4101; UNROLL-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP14]], align 4
4102; UNROLL-NEXT:    [[TMP15:%.*]] = getelementptr inbounds i32, i32* [[TMP13]], i64 2
4103; UNROLL-NEXT:    [[TMP16:%.*]] = bitcast i32* [[TMP15]] to <2 x i32>*
4104; UNROLL-NEXT:    store <2 x i32> [[STEP_ADD]], <2 x i32>* [[TMP16]], align 4
4105; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
4106; UNROLL-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 16, i32 16>
4107; UNROLL-NEXT:    [[TMP17:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
4108; UNROLL-NEXT:    br i1 [[TMP17]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP38:![0-9]+]]
4109; UNROLL:       middle.block:
4110; UNROLL-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
4111; UNROLL-NEXT:    br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]]
4112; UNROLL:       scalar.ph:
4113; UNROLL-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[T]], [[LOOP_PREHEADER]] ], [ [[T]], [[VECTOR_SCEVCHECK]] ]
4114; UNROLL-NEXT:    [[BC_RESUME_VAL2:%.*]] = phi i32 [ [[IND_END1]], [[MIDDLE_BLOCK]] ], [ [[EXT_MUL]], [[LOOP_PREHEADER]] ], [ [[EXT_MUL]], [[VECTOR_SCEVCHECK]] ]
4115; UNROLL-NEXT:    [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[LOOP_PREHEADER]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
4116; UNROLL-NEXT:    br label [[LOOP:%.*]]
4117; UNROLL:       loop:
4118; UNROLL-NEXT:    [[IDX:%.*]] = phi i8 [ [[IDX_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
4119; UNROLL-NEXT:    [[SPHI:%.*]] = phi i32 [ [[MUL:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL2]], [[SCALAR_PH]] ]
4120; UNROLL-NEXT:    [[IDX_B:%.*]] = phi i32 [ [[IDX_B_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ]
4121; UNROLL-NEXT:    [[TMP18:%.*]] = sext i8 [[IDX]] to i64
4122; UNROLL-NEXT:    [[PTR:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP18]]
4123; UNROLL-NEXT:    store i32 [[SPHI]], i32* [[PTR]], align 4
4124; UNROLL-NEXT:    [[IDX_INC]] = add i8 [[IDX]], 1
4125; UNROLL-NEXT:    [[IDX_INC_EXT:%.*]] = zext i8 [[IDX_INC]] to i32
4126; UNROLL-NEXT:    [[MUL]] = shl nuw nsw i32 [[IDX_INC_EXT]], 2
4127; UNROLL-NEXT:    [[IDX_B_INC]] = add nuw nsw i32 [[IDX_B]], 1
4128; UNROLL-NEXT:    [[C:%.*]] = icmp ult i32 [[IDX_B]], [[LEN]]
4129; UNROLL-NEXT:    br i1 [[C]], label [[LOOP]], label [[EXIT_LOOPEXIT]], !llvm.loop [[LOOP39:![0-9]+]]
4130; UNROLL:       exit.loopexit:
4131; UNROLL-NEXT:    br label [[EXIT]]
4132; UNROLL:       exit:
4133; UNROLL-NEXT:    ret void
4134;
4135; UNROLL-NO-IC-LABEL: @wrappingindvars2(
4136; UNROLL-NO-IC-NEXT:  entry:
4137; UNROLL-NO-IC-NEXT:    [[ST:%.*]] = zext i8 [[T:%.*]] to i16
4138; UNROLL-NO-IC-NEXT:    [[EXT:%.*]] = zext i8 [[T]] to i32
4139; UNROLL-NO-IC-NEXT:    [[EXT_MUL:%.*]] = mul i32 [[EXT]], 4
4140; UNROLL-NO-IC-NEXT:    [[ECMP:%.*]] = icmp ult i16 [[ST]], 42
4141; UNROLL-NO-IC-NEXT:    br i1 [[ECMP]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]]
4142; UNROLL-NO-IC:       loop.preheader:
4143; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = add i32 [[LEN:%.*]], 1
4144; UNROLL-NO-IC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 4
4145; UNROLL-NO-IC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
4146; UNROLL-NO-IC:       vector.scevcheck:
4147; UNROLL-NO-IC-NEXT:    [[TMP1:%.*]] = trunc i32 [[LEN]] to i8
4148; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = add i8 [[T]], [[TMP1]]
4149; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = icmp ult i8 [[TMP2]], [[T]]
4150; UNROLL-NO-IC-NEXT:    [[TMP4:%.*]] = icmp ugt i32 [[LEN]], 255
4151; UNROLL-NO-IC-NEXT:    [[TMP5:%.*]] = or i1 [[TMP3]], [[TMP4]]
4152; UNROLL-NO-IC-NEXT:    [[TMP6:%.*]] = trunc i32 [[LEN]] to i8
4153; UNROLL-NO-IC-NEXT:    [[TMP7:%.*]] = add i8 [[T]], [[TMP6]]
4154; UNROLL-NO-IC-NEXT:    [[TMP8:%.*]] = icmp slt i8 [[TMP7]], [[T]]
4155; UNROLL-NO-IC-NEXT:    [[TMP9:%.*]] = icmp ugt i32 [[LEN]], 255
4156; UNROLL-NO-IC-NEXT:    [[TMP10:%.*]] = or i1 [[TMP8]], [[TMP9]]
4157; UNROLL-NO-IC-NEXT:    [[TMP11:%.*]] = or i1 [[TMP5]], [[TMP10]]
4158; UNROLL-NO-IC-NEXT:    br i1 [[TMP11]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
4159; UNROLL-NO-IC:       vector.ph:
4160; UNROLL-NO-IC-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[TMP0]], 4
4161; UNROLL-NO-IC-NEXT:    [[N_VEC:%.*]] = sub i32 [[TMP0]], [[N_MOD_VF]]
4162; UNROLL-NO-IC-NEXT:    [[CAST_VTC:%.*]] = trunc i32 [[N_VEC]] to i8
4163; UNROLL-NO-IC-NEXT:    [[IND_END:%.*]] = add i8 [[T]], [[CAST_VTC]]
4164; UNROLL-NO-IC-NEXT:    [[TMP12:%.*]] = mul i32 [[N_VEC]], 4
4165; UNROLL-NO-IC-NEXT:    [[IND_END1:%.*]] = add i32 [[EXT_MUL]], [[TMP12]]
4166; UNROLL-NO-IC-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[EXT_MUL]], i32 0
4167; UNROLL-NO-IC-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
4168; UNROLL-NO-IC-NEXT:    [[INDUCTION:%.*]] = add <2 x i32> [[DOTSPLAT]], <i32 0, i32 4>
4169; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
4170; UNROLL-NO-IC:       vector.body:
4171; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4172; UNROLL-NO-IC-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
4173; UNROLL-NO-IC-NEXT:    [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], <i32 8, i32 8>
4174; UNROLL-NO-IC-NEXT:    [[TMP13:%.*]] = trunc i32 [[INDEX]] to i8
4175; UNROLL-NO-IC-NEXT:    [[OFFSET_IDX:%.*]] = add i8 [[T]], [[TMP13]]
4176; UNROLL-NO-IC-NEXT:    [[TMP14:%.*]] = add i8 [[OFFSET_IDX]], 0
4177; UNROLL-NO-IC-NEXT:    [[TMP15:%.*]] = add i8 [[OFFSET_IDX]], 2
4178; UNROLL-NO-IC-NEXT:    [[TMP16:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i8 [[TMP14]]
4179; UNROLL-NO-IC-NEXT:    [[TMP17:%.*]] = getelementptr inbounds i32, i32* [[A]], i8 [[TMP15]]
4180; UNROLL-NO-IC-NEXT:    [[TMP18:%.*]] = getelementptr inbounds i32, i32* [[TMP16]], i32 0
4181; UNROLL-NO-IC-NEXT:    [[TMP19:%.*]] = bitcast i32* [[TMP18]] to <2 x i32>*
4182; UNROLL-NO-IC-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP19]], align 4
4183; UNROLL-NO-IC-NEXT:    [[TMP20:%.*]] = getelementptr inbounds i32, i32* [[TMP16]], i32 2
4184; UNROLL-NO-IC-NEXT:    [[TMP21:%.*]] = bitcast i32* [[TMP20]] to <2 x i32>*
4185; UNROLL-NO-IC-NEXT:    store <2 x i32> [[STEP_ADD]], <2 x i32>* [[TMP21]], align 4
4186; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
4187; UNROLL-NO-IC-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[STEP_ADD]], <i32 8, i32 8>
4188; UNROLL-NO-IC-NEXT:    [[TMP22:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
4189; UNROLL-NO-IC-NEXT:    br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP38:![0-9]+]]
4190; UNROLL-NO-IC:       middle.block:
4191; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
4192; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]]
4193; UNROLL-NO-IC:       scalar.ph:
4194; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[T]], [[LOOP_PREHEADER]] ], [ [[T]], [[VECTOR_SCEVCHECK]] ]
4195; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL2:%.*]] = phi i32 [ [[IND_END1]], [[MIDDLE_BLOCK]] ], [ [[EXT_MUL]], [[LOOP_PREHEADER]] ], [ [[EXT_MUL]], [[VECTOR_SCEVCHECK]] ]
4196; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[LOOP_PREHEADER]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
4197; UNROLL-NO-IC-NEXT:    br label [[LOOP:%.*]]
4198; UNROLL-NO-IC:       loop:
4199; UNROLL-NO-IC-NEXT:    [[IDX:%.*]] = phi i8 [ [[IDX_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
4200; UNROLL-NO-IC-NEXT:    [[SPHI:%.*]] = phi i32 [ [[MUL:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL2]], [[SCALAR_PH]] ]
4201; UNROLL-NO-IC-NEXT:    [[IDX_B:%.*]] = phi i32 [ [[IDX_B_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ]
4202; UNROLL-NO-IC-NEXT:    [[PTR:%.*]] = getelementptr inbounds i32, i32* [[A]], i8 [[IDX]]
4203; UNROLL-NO-IC-NEXT:    store i32 [[SPHI]], i32* [[PTR]], align 4
4204; UNROLL-NO-IC-NEXT:    [[IDX_INC]] = add i8 [[IDX]], 1
4205; UNROLL-NO-IC-NEXT:    [[IDX_INC_EXT:%.*]] = zext i8 [[IDX_INC]] to i32
4206; UNROLL-NO-IC-NEXT:    [[MUL]] = mul i32 [[IDX_INC_EXT]], 4
4207; UNROLL-NO-IC-NEXT:    [[IDX_B_INC]] = add nuw nsw i32 [[IDX_B]], 1
4208; UNROLL-NO-IC-NEXT:    [[C:%.*]] = icmp ult i32 [[IDX_B]], [[LEN]]
4209; UNROLL-NO-IC-NEXT:    br i1 [[C]], label [[LOOP]], label [[EXIT_LOOPEXIT]], !llvm.loop [[LOOP39:![0-9]+]]
4210; UNROLL-NO-IC:       exit.loopexit:
4211; UNROLL-NO-IC-NEXT:    br label [[EXIT]]
4212; UNROLL-NO-IC:       exit:
4213; UNROLL-NO-IC-NEXT:    ret void
4214;
4215; INTERLEAVE-LABEL: @wrappingindvars2(
4216; INTERLEAVE-NEXT:  entry:
4217; INTERLEAVE-NEXT:    [[EXT:%.*]] = zext i8 [[T:%.*]] to i32
4218; INTERLEAVE-NEXT:    [[EXT_MUL:%.*]] = shl nuw nsw i32 [[EXT]], 2
4219; INTERLEAVE-NEXT:    [[ECMP:%.*]] = icmp ult i8 [[T]], 42
4220; INTERLEAVE-NEXT:    br i1 [[ECMP]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]]
4221; INTERLEAVE:       loop.preheader:
4222; INTERLEAVE-NEXT:    [[TMP0:%.*]] = add i32 [[LEN:%.*]], 1
4223; INTERLEAVE-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 8
4224; INTERLEAVE-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
4225; INTERLEAVE:       vector.scevcheck:
4226; INTERLEAVE-NEXT:    [[TMP1:%.*]] = trunc i32 [[LEN]] to i8
4227; INTERLEAVE-NEXT:    [[TMP2:%.*]] = xor i8 [[T]], -1
4228; INTERLEAVE-NEXT:    [[TMP3:%.*]] = icmp ult i8 [[TMP2]], [[TMP1]]
4229; INTERLEAVE-NEXT:    [[TMP4:%.*]] = trunc i32 [[LEN]] to i8
4230; INTERLEAVE-NEXT:    [[TMP5:%.*]] = add i8 [[TMP4]], [[T]]
4231; INTERLEAVE-NEXT:    [[TMP6:%.*]] = icmp slt i8 [[TMP5]], [[T]]
4232; INTERLEAVE-NEXT:    [[TMP7:%.*]] = icmp ugt i32 [[LEN]], 255
4233; INTERLEAVE-NEXT:    [[TMP8:%.*]] = or i1 [[TMP6]], [[TMP7]]
4234; INTERLEAVE-NEXT:    [[TMP9:%.*]] = or i1 [[TMP3]], [[TMP8]]
4235; INTERLEAVE-NEXT:    br i1 [[TMP9]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
4236; INTERLEAVE:       vector.ph:
4237; INTERLEAVE-NEXT:    [[N_VEC:%.*]] = and i32 [[TMP0]], -8
4238; INTERLEAVE-NEXT:    [[CAST_VTC:%.*]] = trunc i32 [[N_VEC]] to i8
4239; INTERLEAVE-NEXT:    [[IND_END:%.*]] = add i8 [[CAST_VTC]], [[T]]
4240; INTERLEAVE-NEXT:    [[TMP10:%.*]] = add i32 [[N_VEC]], [[EXT]]
4241; INTERLEAVE-NEXT:    [[IND_END1:%.*]] = shl i32 [[TMP10]], 2
4242; INTERLEAVE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[EXT_MUL]], i64 0
4243; INTERLEAVE-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <4 x i32> [[DOTSPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer
4244; INTERLEAVE-NEXT:    [[INDUCTION:%.*]] = add nuw nsw <4 x i32> [[DOTSPLAT]], <i32 0, i32 4, i32 8, i32 12>
4245; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
4246; INTERLEAVE:       vector.body:
4247; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4248; INTERLEAVE-NEXT:    [[VEC_IND:%.*]] = phi <4 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
4249; INTERLEAVE-NEXT:    [[STEP_ADD:%.*]] = add <4 x i32> [[VEC_IND]], <i32 16, i32 16, i32 16, i32 16>
4250; INTERLEAVE-NEXT:    [[TMP11:%.*]] = trunc i32 [[INDEX]] to i8
4251; INTERLEAVE-NEXT:    [[OFFSET_IDX:%.*]] = add i8 [[TMP11]], [[T]]
4252; INTERLEAVE-NEXT:    [[TMP12:%.*]] = sext i8 [[OFFSET_IDX]] to i64
4253; INTERLEAVE-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP12]]
4254; INTERLEAVE-NEXT:    [[TMP14:%.*]] = bitcast i32* [[TMP13]] to <4 x i32>*
4255; INTERLEAVE-NEXT:    store <4 x i32> [[VEC_IND]], <4 x i32>* [[TMP14]], align 4
4256; INTERLEAVE-NEXT:    [[TMP15:%.*]] = getelementptr inbounds i32, i32* [[TMP13]], i64 4
4257; INTERLEAVE-NEXT:    [[TMP16:%.*]] = bitcast i32* [[TMP15]] to <4 x i32>*
4258; INTERLEAVE-NEXT:    store <4 x i32> [[STEP_ADD]], <4 x i32>* [[TMP16]], align 4
4259; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 8
4260; INTERLEAVE-NEXT:    [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 32, i32 32, i32 32, i32 32>
4261; INTERLEAVE-NEXT:    [[TMP17:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
4262; INTERLEAVE-NEXT:    br i1 [[TMP17]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP38:![0-9]+]]
4263; INTERLEAVE:       middle.block:
4264; INTERLEAVE-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
4265; INTERLEAVE-NEXT:    br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]]
4266; INTERLEAVE:       scalar.ph:
4267; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i8 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[T]], [[LOOP_PREHEADER]] ], [ [[T]], [[VECTOR_SCEVCHECK]] ]
4268; INTERLEAVE-NEXT:    [[BC_RESUME_VAL2:%.*]] = phi i32 [ [[IND_END1]], [[MIDDLE_BLOCK]] ], [ [[EXT_MUL]], [[LOOP_PREHEADER]] ], [ [[EXT_MUL]], [[VECTOR_SCEVCHECK]] ]
4269; INTERLEAVE-NEXT:    [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[LOOP_PREHEADER]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
4270; INTERLEAVE-NEXT:    br label [[LOOP:%.*]]
4271; INTERLEAVE:       loop:
4272; INTERLEAVE-NEXT:    [[IDX:%.*]] = phi i8 [ [[IDX_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
4273; INTERLEAVE-NEXT:    [[SPHI:%.*]] = phi i32 [ [[MUL:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL2]], [[SCALAR_PH]] ]
4274; INTERLEAVE-NEXT:    [[IDX_B:%.*]] = phi i32 [ [[IDX_B_INC:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ]
4275; INTERLEAVE-NEXT:    [[TMP18:%.*]] = sext i8 [[IDX]] to i64
4276; INTERLEAVE-NEXT:    [[PTR:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP18]]
4277; INTERLEAVE-NEXT:    store i32 [[SPHI]], i32* [[PTR]], align 4
4278; INTERLEAVE-NEXT:    [[IDX_INC]] = add i8 [[IDX]], 1
4279; INTERLEAVE-NEXT:    [[IDX_INC_EXT:%.*]] = zext i8 [[IDX_INC]] to i32
4280; INTERLEAVE-NEXT:    [[MUL]] = shl nuw nsw i32 [[IDX_INC_EXT]], 2
4281; INTERLEAVE-NEXT:    [[IDX_B_INC]] = add nuw nsw i32 [[IDX_B]], 1
4282; INTERLEAVE-NEXT:    [[C:%.*]] = icmp ult i32 [[IDX_B]], [[LEN]]
4283; INTERLEAVE-NEXT:    br i1 [[C]], label [[LOOP]], label [[EXIT_LOOPEXIT]], !llvm.loop [[LOOP39:![0-9]+]]
4284; INTERLEAVE:       exit.loopexit:
4285; INTERLEAVE-NEXT:    br label [[EXIT]]
4286; INTERLEAVE:       exit:
4287; INTERLEAVE-NEXT:    ret void
4288;
4289entry:
4290  %st = zext i8 %t to i16
4291  %ext = zext i8 %t to i32
4292  %ext.mul = mul i32 %ext, 4
4293
4294  %ecmp = icmp ult i16 %st, 42
4295  br i1 %ecmp, label %loop, label %exit
4296
4297  loop:
4298
4299  %idx = phi i8 [ %t, %entry ], [ %idx.inc, %loop ]
4300  %sphi = phi i32 [ %ext.mul, %entry ], [%mul, %loop]
4301  %idx.b = phi i32 [ 0, %entry ], [ %idx.b.inc, %loop ]
4302
4303  %ptr = getelementptr inbounds i32, i32* %A, i8 %idx
4304  store i32 %sphi, i32* %ptr
4305
4306  %idx.inc = add i8 %idx, 1
4307  %idx.inc.ext = zext i8 %idx.inc to i32
4308  %mul = mul i32 %idx.inc.ext, 4
4309  %idx.b.inc = add nuw nsw i32 %idx.b, 1
4310
4311  %c = icmp ult i32 %idx.b, %len
4312  br i1 %c, label %loop, label %exit
4313
4314  exit:
4315  ret void
4316}
4317
4318; Check that we generate vectorized IVs in the pre-header
4319; instead of widening the scalar IV inside the loop, when
4320; we know how to do that.
4321define void @veciv(i32* nocapture %a, i32 %start, i32 %k) {
4322; CHECK-LABEL: @veciv(
4323; CHECK-NEXT:  for.body.preheader:
4324; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[K:%.*]], 2
4325; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
4326; CHECK:       vector.ph:
4327; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[K]], 2
4328; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[K]], [[N_MOD_VF]]
4329; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
4330; CHECK:       vector.body:
4331; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4332; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
4333; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[INDEX]], 0
4334; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 [[TMP0]]
4335; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[TMP1]], i32 0
4336; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <2 x i32>*
4337; CHECK-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP3]], align 4
4338; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
4339; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
4340; CHECK-NEXT:    [[TMP4:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
4341; CHECK-NEXT:    br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP40:![0-9]+]]
4342; CHECK:       middle.block:
4343; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[K]], [[N_VEC]]
4344; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
4345; CHECK:       scalar.ph:
4346; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER:%.*]] ]
4347; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
4348; CHECK:       for.body:
4349; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i32 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
4350; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[INDVARS_IV]]
4351; CHECK-NEXT:    store i32 [[INDVARS_IV]], i32* [[ARRAYIDX]], align 4
4352; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i32 [[INDVARS_IV]], 1
4353; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[INDVARS_IV_NEXT]], [[K]]
4354; CHECK-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP41:![0-9]+]]
4355; CHECK:       exit:
4356; CHECK-NEXT:    ret void
4357;
4358; IND-LABEL: @veciv(
4359; IND-NEXT:  for.body.preheader:
4360; IND-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[K:%.*]], 2
4361; IND-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
4362; IND:       vector.ph:
4363; IND-NEXT:    [[N_VEC:%.*]] = and i32 [[K]], -2
4364; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
4365; IND:       vector.body:
4366; IND-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4367; IND-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
4368; IND-NEXT:    [[TMP0:%.*]] = sext i32 [[INDEX]] to i64
4369; IND-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP0]]
4370; IND-NEXT:    [[TMP2:%.*]] = bitcast i32* [[TMP1]] to <2 x i32>*
4371; IND-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP2]], align 4
4372; IND-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
4373; IND-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
4374; IND-NEXT:    [[TMP3:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
4375; IND-NEXT:    br i1 [[TMP3]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP40:![0-9]+]]
4376; IND:       middle.block:
4377; IND-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[N_VEC]], [[K]]
4378; IND-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
4379; IND:       scalar.ph:
4380; IND-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER:%.*]] ]
4381; IND-NEXT:    br label [[FOR_BODY:%.*]]
4382; IND:       for.body:
4383; IND-NEXT:    [[INDVARS_IV:%.*]] = phi i32 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
4384; IND-NEXT:    [[TMP4:%.*]] = sext i32 [[INDVARS_IV]] to i64
4385; IND-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP4]]
4386; IND-NEXT:    store i32 [[INDVARS_IV]], i32* [[ARRAYIDX]], align 4
4387; IND-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i32 [[INDVARS_IV]], 1
4388; IND-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[INDVARS_IV_NEXT]], [[K]]
4389; IND-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP41:![0-9]+]]
4390; IND:       exit:
4391; IND-NEXT:    ret void
4392;
4393; UNROLL-LABEL: @veciv(
4394; UNROLL-NEXT:  for.body.preheader:
4395; UNROLL-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[K:%.*]], 4
4396; UNROLL-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
4397; UNROLL:       vector.ph:
4398; UNROLL-NEXT:    [[N_VEC:%.*]] = and i32 [[K]], -4
4399; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
4400; UNROLL:       vector.body:
4401; UNROLL-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4402; UNROLL-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
4403; UNROLL-NEXT:    [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
4404; UNROLL-NEXT:    [[TMP0:%.*]] = sext i32 [[INDEX]] to i64
4405; UNROLL-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP0]]
4406; UNROLL-NEXT:    [[TMP2:%.*]] = bitcast i32* [[TMP1]] to <2 x i32>*
4407; UNROLL-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP2]], align 4
4408; UNROLL-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[TMP1]], i64 2
4409; UNROLL-NEXT:    [[TMP4:%.*]] = bitcast i32* [[TMP3]] to <2 x i32>*
4410; UNROLL-NEXT:    store <2 x i32> [[STEP_ADD]], <2 x i32>* [[TMP4]], align 4
4411; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
4412; UNROLL-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 4, i32 4>
4413; UNROLL-NEXT:    [[TMP5:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
4414; UNROLL-NEXT:    br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP40:![0-9]+]]
4415; UNROLL:       middle.block:
4416; UNROLL-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[N_VEC]], [[K]]
4417; UNROLL-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
4418; UNROLL:       scalar.ph:
4419; UNROLL-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER:%.*]] ]
4420; UNROLL-NEXT:    br label [[FOR_BODY:%.*]]
4421; UNROLL:       for.body:
4422; UNROLL-NEXT:    [[INDVARS_IV:%.*]] = phi i32 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
4423; UNROLL-NEXT:    [[TMP6:%.*]] = sext i32 [[INDVARS_IV]] to i64
4424; UNROLL-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP6]]
4425; UNROLL-NEXT:    store i32 [[INDVARS_IV]], i32* [[ARRAYIDX]], align 4
4426; UNROLL-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i32 [[INDVARS_IV]], 1
4427; UNROLL-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[INDVARS_IV_NEXT]], [[K]]
4428; UNROLL-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP41:![0-9]+]]
4429; UNROLL:       exit:
4430; UNROLL-NEXT:    ret void
4431;
4432; UNROLL-NO-IC-LABEL: @veciv(
4433; UNROLL-NO-IC-NEXT:  for.body.preheader:
4434; UNROLL-NO-IC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[K:%.*]], 4
4435; UNROLL-NO-IC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
4436; UNROLL-NO-IC:       vector.ph:
4437; UNROLL-NO-IC-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[K]], 4
4438; UNROLL-NO-IC-NEXT:    [[N_VEC:%.*]] = sub i32 [[K]], [[N_MOD_VF]]
4439; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
4440; UNROLL-NO-IC:       vector.body:
4441; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4442; UNROLL-NO-IC-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
4443; UNROLL-NO-IC-NEXT:    [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
4444; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = add i32 [[INDEX]], 0
4445; UNROLL-NO-IC-NEXT:    [[TMP1:%.*]] = add i32 [[INDEX]], 2
4446; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 [[TMP0]]
4447; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[TMP1]]
4448; UNROLL-NO-IC-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[TMP2]], i32 0
4449; UNROLL-NO-IC-NEXT:    [[TMP5:%.*]] = bitcast i32* [[TMP4]] to <2 x i32>*
4450; UNROLL-NO-IC-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP5]], align 4
4451; UNROLL-NO-IC-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i32, i32* [[TMP2]], i32 2
4452; UNROLL-NO-IC-NEXT:    [[TMP7:%.*]] = bitcast i32* [[TMP6]] to <2 x i32>*
4453; UNROLL-NO-IC-NEXT:    store <2 x i32> [[STEP_ADD]], <2 x i32>* [[TMP7]], align 4
4454; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
4455; UNROLL-NO-IC-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[STEP_ADD]], <i32 2, i32 2>
4456; UNROLL-NO-IC-NEXT:    [[TMP8:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
4457; UNROLL-NO-IC-NEXT:    br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP40:![0-9]+]]
4458; UNROLL-NO-IC:       middle.block:
4459; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[K]], [[N_VEC]]
4460; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
4461; UNROLL-NO-IC:       scalar.ph:
4462; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER:%.*]] ]
4463; UNROLL-NO-IC-NEXT:    br label [[FOR_BODY:%.*]]
4464; UNROLL-NO-IC:       for.body:
4465; UNROLL-NO-IC-NEXT:    [[INDVARS_IV:%.*]] = phi i32 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
4466; UNROLL-NO-IC-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[INDVARS_IV]]
4467; UNROLL-NO-IC-NEXT:    store i32 [[INDVARS_IV]], i32* [[ARRAYIDX]], align 4
4468; UNROLL-NO-IC-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i32 [[INDVARS_IV]], 1
4469; UNROLL-NO-IC-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[INDVARS_IV_NEXT]], [[K]]
4470; UNROLL-NO-IC-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP41:![0-9]+]]
4471; UNROLL-NO-IC:       exit:
4472; UNROLL-NO-IC-NEXT:    ret void
4473;
4474; INTERLEAVE-LABEL: @veciv(
4475; INTERLEAVE-NEXT:  for.body.preheader:
4476; INTERLEAVE-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[K:%.*]], 8
4477; INTERLEAVE-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
4478; INTERLEAVE:       vector.ph:
4479; INTERLEAVE-NEXT:    [[N_VEC:%.*]] = and i32 [[K]], -8
4480; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
4481; INTERLEAVE:       vector.body:
4482; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4483; INTERLEAVE-NEXT:    [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
4484; INTERLEAVE-NEXT:    [[STEP_ADD:%.*]] = add <4 x i32> [[VEC_IND]], <i32 4, i32 4, i32 4, i32 4>
4485; INTERLEAVE-NEXT:    [[TMP0:%.*]] = sext i32 [[INDEX]] to i64
4486; INTERLEAVE-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP0]]
4487; INTERLEAVE-NEXT:    [[TMP2:%.*]] = bitcast i32* [[TMP1]] to <4 x i32>*
4488; INTERLEAVE-NEXT:    store <4 x i32> [[VEC_IND]], <4 x i32>* [[TMP2]], align 4
4489; INTERLEAVE-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[TMP1]], i64 4
4490; INTERLEAVE-NEXT:    [[TMP4:%.*]] = bitcast i32* [[TMP3]] to <4 x i32>*
4491; INTERLEAVE-NEXT:    store <4 x i32> [[STEP_ADD]], <4 x i32>* [[TMP4]], align 4
4492; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 8
4493; INTERLEAVE-NEXT:    [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 8, i32 8, i32 8, i32 8>
4494; INTERLEAVE-NEXT:    [[TMP5:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
4495; INTERLEAVE-NEXT:    br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP40:![0-9]+]]
4496; INTERLEAVE:       middle.block:
4497; INTERLEAVE-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[N_VEC]], [[K]]
4498; INTERLEAVE-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
4499; INTERLEAVE:       scalar.ph:
4500; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER:%.*]] ]
4501; INTERLEAVE-NEXT:    br label [[FOR_BODY:%.*]]
4502; INTERLEAVE:       for.body:
4503; INTERLEAVE-NEXT:    [[INDVARS_IV:%.*]] = phi i32 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
4504; INTERLEAVE-NEXT:    [[TMP6:%.*]] = sext i32 [[INDVARS_IV]] to i64
4505; INTERLEAVE-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP6]]
4506; INTERLEAVE-NEXT:    store i32 [[INDVARS_IV]], i32* [[ARRAYIDX]], align 4
4507; INTERLEAVE-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i32 [[INDVARS_IV]], 1
4508; INTERLEAVE-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[INDVARS_IV_NEXT]], [[K]]
4509; INTERLEAVE-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP41:![0-9]+]]
4510; INTERLEAVE:       exit:
4511; INTERLEAVE-NEXT:    ret void
4512;
4513for.body.preheader:
4514  br label %for.body
4515
4516for.body:
4517  %indvars.iv = phi i32 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]
4518  %arrayidx = getelementptr inbounds i32, i32* %a, i32 %indvars.iv
4519  store i32 %indvars.iv, i32* %arrayidx, align 4
4520  %indvars.iv.next = add nuw nsw i32 %indvars.iv, 1
4521  %exitcond = icmp eq i32 %indvars.iv.next, %k
4522  br i1 %exitcond, label %exit, label %for.body
4523
4524exit:
4525  ret void
4526}
4527
4528define void @trunciv(i32* nocapture %a, i32 %start, i64 %k) {
4529; CHECK-LABEL: @trunciv(
4530; CHECK-NEXT:  for.body.preheader:
4531; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[K:%.*]], 2
4532; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
4533; CHECK:       vector.scevcheck:
4534; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[K]], -1
4535; CHECK-NEXT:    [[TMP1:%.*]] = trunc i64 [[TMP0]] to i32
4536; CHECK-NEXT:    [[TMP2:%.*]] = icmp slt i32 [[TMP1]], 0
4537; CHECK-NEXT:    [[TMP3:%.*]] = icmp ugt i64 [[TMP0]], 4294967295
4538; CHECK-NEXT:    [[TMP4:%.*]] = or i1 [[TMP2]], [[TMP3]]
4539; CHECK-NEXT:    br i1 [[TMP4]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
4540; CHECK:       vector.ph:
4541; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[K]], 2
4542; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[K]], [[N_MOD_VF]]
4543; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
4544; CHECK:       vector.body:
4545; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4546; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
4547; CHECK-NEXT:    [[TMP5:%.*]] = trunc i64 [[INDEX]] to i32
4548; CHECK-NEXT:    [[TMP6:%.*]] = add i32 [[TMP5]], 0
4549; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 [[TMP6]]
4550; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i32, i32* [[TMP7]], i32 0
4551; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i32* [[TMP8]] to <2 x i32>*
4552; CHECK-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP9]], align 4
4553; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
4554; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
4555; CHECK-NEXT:    [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
4556; CHECK-NEXT:    br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP42:![0-9]+]]
4557; CHECK:       middle.block:
4558; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[K]], [[N_VEC]]
4559; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
4560; CHECK:       scalar.ph:
4561; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER:%.*]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
4562; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
4563; CHECK:       for.body:
4564; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
4565; CHECK-NEXT:    [[TRUNC_IV:%.*]] = trunc i64 [[INDVARS_IV]] to i32
4566; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[TRUNC_IV]]
4567; CHECK-NEXT:    store i32 [[TRUNC_IV]], i32* [[ARRAYIDX]], align 4
4568; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
4569; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[K]]
4570; CHECK-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP43:![0-9]+]]
4571; CHECK:       exit:
4572; CHECK-NEXT:    ret void
4573;
4574; IND-LABEL: @trunciv(
4575; IND-NEXT:  for.body.preheader:
4576; IND-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[K:%.*]], 2
4577; IND-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
4578; IND:       vector.scevcheck:
4579; IND-NEXT:    [[TMP0:%.*]] = add i64 [[K]], -1
4580; IND-NEXT:    [[DOTNOT:%.*]] = icmp ult i64 [[TMP0]], 2147483648
4581; IND-NEXT:    br i1 [[DOTNOT]], label [[VECTOR_PH:%.*]], label [[SCALAR_PH]]
4582; IND:       vector.ph:
4583; IND-NEXT:    [[N_VEC:%.*]] = and i64 [[K]], -2
4584; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
4585; IND:       vector.body:
4586; IND-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4587; IND-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
4588; IND-NEXT:    [[SEXT:%.*]] = shl i64 [[INDEX]], 32
4589; IND-NEXT:    [[TMP1:%.*]] = ashr exact i64 [[SEXT]], 32
4590; IND-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP1]]
4591; IND-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <2 x i32>*
4592; IND-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP3]], align 4
4593; IND-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
4594; IND-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
4595; IND-NEXT:    [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
4596; IND-NEXT:    br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP42:![0-9]+]]
4597; IND:       middle.block:
4598; IND-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[K]]
4599; IND-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
4600; IND:       scalar.ph:
4601; IND-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER:%.*]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
4602; IND-NEXT:    br label [[FOR_BODY:%.*]]
4603; IND:       for.body:
4604; IND-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
4605; IND-NEXT:    [[TRUNC_IV:%.*]] = trunc i64 [[INDVARS_IV]] to i32
4606; IND-NEXT:    [[SEXT1:%.*]] = shl i64 [[INDVARS_IV]], 32
4607; IND-NEXT:    [[TMP5:%.*]] = ashr exact i64 [[SEXT1]], 32
4608; IND-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP5]]
4609; IND-NEXT:    store i32 [[TRUNC_IV]], i32* [[ARRAYIDX]], align 4
4610; IND-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
4611; IND-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[K]]
4612; IND-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP43:![0-9]+]]
4613; IND:       exit:
4614; IND-NEXT:    ret void
4615;
4616; UNROLL-LABEL: @trunciv(
4617; UNROLL-NEXT:  for.body.preheader:
4618; UNROLL-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[K:%.*]], 4
4619; UNROLL-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
4620; UNROLL:       vector.scevcheck:
4621; UNROLL-NEXT:    [[TMP0:%.*]] = add i64 [[K]], -1
4622; UNROLL-NEXT:    [[DOTNOT:%.*]] = icmp ult i64 [[TMP0]], 2147483648
4623; UNROLL-NEXT:    br i1 [[DOTNOT]], label [[VECTOR_PH:%.*]], label [[SCALAR_PH]]
4624; UNROLL:       vector.ph:
4625; UNROLL-NEXT:    [[N_VEC:%.*]] = and i64 [[K]], -4
4626; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
4627; UNROLL:       vector.body:
4628; UNROLL-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4629; UNROLL-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
4630; UNROLL-NEXT:    [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
4631; UNROLL-NEXT:    [[SEXT:%.*]] = shl i64 [[INDEX]], 32
4632; UNROLL-NEXT:    [[TMP1:%.*]] = ashr exact i64 [[SEXT]], 32
4633; UNROLL-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP1]]
4634; UNROLL-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <2 x i32>*
4635; UNROLL-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP3]], align 4
4636; UNROLL-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[TMP2]], i64 2
4637; UNROLL-NEXT:    [[TMP5:%.*]] = bitcast i32* [[TMP4]] to <2 x i32>*
4638; UNROLL-NEXT:    store <2 x i32> [[STEP_ADD]], <2 x i32>* [[TMP5]], align 4
4639; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
4640; UNROLL-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 4, i32 4>
4641; UNROLL-NEXT:    [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
4642; UNROLL-NEXT:    br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP42:![0-9]+]]
4643; UNROLL:       middle.block:
4644; UNROLL-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[K]]
4645; UNROLL-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
4646; UNROLL:       scalar.ph:
4647; UNROLL-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER:%.*]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
4648; UNROLL-NEXT:    br label [[FOR_BODY:%.*]]
4649; UNROLL:       for.body:
4650; UNROLL-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
4651; UNROLL-NEXT:    [[TRUNC_IV:%.*]] = trunc i64 [[INDVARS_IV]] to i32
4652; UNROLL-NEXT:    [[SEXT2:%.*]] = shl i64 [[INDVARS_IV]], 32
4653; UNROLL-NEXT:    [[TMP7:%.*]] = ashr exact i64 [[SEXT2]], 32
4654; UNROLL-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP7]]
4655; UNROLL-NEXT:    store i32 [[TRUNC_IV]], i32* [[ARRAYIDX]], align 4
4656; UNROLL-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
4657; UNROLL-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[K]]
4658; UNROLL-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP43:![0-9]+]]
4659; UNROLL:       exit:
4660; UNROLL-NEXT:    ret void
4661;
4662; UNROLL-NO-IC-LABEL: @trunciv(
4663; UNROLL-NO-IC-NEXT:  for.body.preheader:
4664; UNROLL-NO-IC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[K:%.*]], 4
4665; UNROLL-NO-IC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
4666; UNROLL-NO-IC:       vector.scevcheck:
4667; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = add i64 [[K]], -1
4668; UNROLL-NO-IC-NEXT:    [[TMP1:%.*]] = trunc i64 [[TMP0]] to i32
4669; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = icmp slt i32 [[TMP1]], 0
4670; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = icmp ugt i64 [[TMP0]], 4294967295
4671; UNROLL-NO-IC-NEXT:    [[TMP4:%.*]] = or i1 [[TMP2]], [[TMP3]]
4672; UNROLL-NO-IC-NEXT:    br i1 [[TMP4]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
4673; UNROLL-NO-IC:       vector.ph:
4674; UNROLL-NO-IC-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[K]], 4
4675; UNROLL-NO-IC-NEXT:    [[N_VEC:%.*]] = sub i64 [[K]], [[N_MOD_VF]]
4676; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
4677; UNROLL-NO-IC:       vector.body:
4678; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4679; UNROLL-NO-IC-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
4680; UNROLL-NO-IC-NEXT:    [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
4681; UNROLL-NO-IC-NEXT:    [[TMP5:%.*]] = trunc i64 [[INDEX]] to i32
4682; UNROLL-NO-IC-NEXT:    [[TMP6:%.*]] = add i32 [[TMP5]], 0
4683; UNROLL-NO-IC-NEXT:    [[TMP7:%.*]] = add i32 [[TMP5]], 2
4684; UNROLL-NO-IC-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 [[TMP6]]
4685; UNROLL-NO-IC-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[TMP7]]
4686; UNROLL-NO-IC-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i32, i32* [[TMP8]], i32 0
4687; UNROLL-NO-IC-NEXT:    [[TMP11:%.*]] = bitcast i32* [[TMP10]] to <2 x i32>*
4688; UNROLL-NO-IC-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP11]], align 4
4689; UNROLL-NO-IC-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i32, i32* [[TMP8]], i32 2
4690; UNROLL-NO-IC-NEXT:    [[TMP13:%.*]] = bitcast i32* [[TMP12]] to <2 x i32>*
4691; UNROLL-NO-IC-NEXT:    store <2 x i32> [[STEP_ADD]], <2 x i32>* [[TMP13]], align 4
4692; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
4693; UNROLL-NO-IC-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[STEP_ADD]], <i32 2, i32 2>
4694; UNROLL-NO-IC-NEXT:    [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
4695; UNROLL-NO-IC-NEXT:    br i1 [[TMP14]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP42:![0-9]+]]
4696; UNROLL-NO-IC:       middle.block:
4697; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[K]], [[N_VEC]]
4698; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
4699; UNROLL-NO-IC:       scalar.ph:
4700; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER:%.*]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
4701; UNROLL-NO-IC-NEXT:    br label [[FOR_BODY:%.*]]
4702; UNROLL-NO-IC:       for.body:
4703; UNROLL-NO-IC-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
4704; UNROLL-NO-IC-NEXT:    [[TRUNC_IV:%.*]] = trunc i64 [[INDVARS_IV]] to i32
4705; UNROLL-NO-IC-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[TRUNC_IV]]
4706; UNROLL-NO-IC-NEXT:    store i32 [[TRUNC_IV]], i32* [[ARRAYIDX]], align 4
4707; UNROLL-NO-IC-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
4708; UNROLL-NO-IC-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[K]]
4709; UNROLL-NO-IC-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP43:![0-9]+]]
4710; UNROLL-NO-IC:       exit:
4711; UNROLL-NO-IC-NEXT:    ret void
4712;
4713; INTERLEAVE-LABEL: @trunciv(
4714; INTERLEAVE-NEXT:  for.body.preheader:
4715; INTERLEAVE-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[K:%.*]], 8
4716; INTERLEAVE-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
4717; INTERLEAVE:       vector.scevcheck:
4718; INTERLEAVE-NEXT:    [[TMP0:%.*]] = add i64 [[K]], -1
4719; INTERLEAVE-NEXT:    [[DOTNOT:%.*]] = icmp ult i64 [[TMP0]], 2147483648
4720; INTERLEAVE-NEXT:    br i1 [[DOTNOT]], label [[VECTOR_PH:%.*]], label [[SCALAR_PH]]
4721; INTERLEAVE:       vector.ph:
4722; INTERLEAVE-NEXT:    [[N_VEC:%.*]] = and i64 [[K]], -8
4723; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
4724; INTERLEAVE:       vector.body:
4725; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4726; INTERLEAVE-NEXT:    [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
4727; INTERLEAVE-NEXT:    [[STEP_ADD:%.*]] = add <4 x i32> [[VEC_IND]], <i32 4, i32 4, i32 4, i32 4>
4728; INTERLEAVE-NEXT:    [[SEXT:%.*]] = shl i64 [[INDEX]], 32
4729; INTERLEAVE-NEXT:    [[TMP1:%.*]] = ashr exact i64 [[SEXT]], 32
4730; INTERLEAVE-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP1]]
4731; INTERLEAVE-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <4 x i32>*
4732; INTERLEAVE-NEXT:    store <4 x i32> [[VEC_IND]], <4 x i32>* [[TMP3]], align 4
4733; INTERLEAVE-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[TMP2]], i64 4
4734; INTERLEAVE-NEXT:    [[TMP5:%.*]] = bitcast i32* [[TMP4]] to <4 x i32>*
4735; INTERLEAVE-NEXT:    store <4 x i32> [[STEP_ADD]], <4 x i32>* [[TMP5]], align 4
4736; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
4737; INTERLEAVE-NEXT:    [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 8, i32 8, i32 8, i32 8>
4738; INTERLEAVE-NEXT:    [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
4739; INTERLEAVE-NEXT:    br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP42:![0-9]+]]
4740; INTERLEAVE:       middle.block:
4741; INTERLEAVE-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[K]]
4742; INTERLEAVE-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
4743; INTERLEAVE:       scalar.ph:
4744; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER:%.*]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
4745; INTERLEAVE-NEXT:    br label [[FOR_BODY:%.*]]
4746; INTERLEAVE:       for.body:
4747; INTERLEAVE-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
4748; INTERLEAVE-NEXT:    [[TRUNC_IV:%.*]] = trunc i64 [[INDVARS_IV]] to i32
4749; INTERLEAVE-NEXT:    [[SEXT2:%.*]] = shl i64 [[INDVARS_IV]], 32
4750; INTERLEAVE-NEXT:    [[TMP7:%.*]] = ashr exact i64 [[SEXT2]], 32
4751; INTERLEAVE-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP7]]
4752; INTERLEAVE-NEXT:    store i32 [[TRUNC_IV]], i32* [[ARRAYIDX]], align 4
4753; INTERLEAVE-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
4754; INTERLEAVE-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[K]]
4755; INTERLEAVE-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP43:![0-9]+]]
4756; INTERLEAVE:       exit:
4757; INTERLEAVE-NEXT:    ret void
4758;
4759for.body.preheader:
4760  br label %for.body
4761
4762for.body:
4763  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]
4764  %trunc.iv = trunc i64 %indvars.iv to i32
4765  %arrayidx = getelementptr inbounds i32, i32* %a, i32 %trunc.iv
4766  store i32 %trunc.iv, i32* %arrayidx, align 4
4767  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
4768  %exitcond = icmp eq i64 %indvars.iv.next, %k
4769  br i1 %exitcond, label %exit, label %for.body
4770
4771exit:
4772  ret void
4773}
4774
4775;
4776;
4777define void @nonprimary(i32* nocapture %a, i32 %start, i32 %i, i32 %k) {
4778; CHECK-LABEL: @nonprimary(
4779; CHECK-NEXT:  for.body.preheader:
4780; CHECK-NEXT:    [[TMP0:%.*]] = sub i32 [[K:%.*]], [[I:%.*]]
4781; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 2
4782; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
4783; CHECK:       vector.ph:
4784; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[TMP0]], 2
4785; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[TMP0]], [[N_MOD_VF]]
4786; CHECK-NEXT:    [[IND_END:%.*]] = add i32 [[I]], [[N_VEC]]
4787; CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[I]], i32 0
4788; CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
4789; CHECK-NEXT:    [[INDUCTION:%.*]] = add <2 x i32> [[DOTSPLAT]], <i32 0, i32 1>
4790; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
4791; CHECK:       vector.body:
4792; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4793; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
4794; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = add i32 [[I]], [[INDEX]]
4795; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[OFFSET_IDX]], 0
4796; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 [[TMP1]]
4797; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[TMP2]], i32 0
4798; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i32* [[TMP3]] to <2 x i32>*
4799; CHECK-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP4]], align 4
4800; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
4801; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
4802; CHECK-NEXT:    [[TMP5:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
4803; CHECK-NEXT:    br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP44:![0-9]+]]
4804; CHECK:       middle.block:
4805; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
4806; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
4807; CHECK:       scalar.ph:
4808; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[I]], [[FOR_BODY_PREHEADER:%.*]] ]
4809; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
4810; CHECK:       for.body:
4811; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i32 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
4812; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[INDVARS_IV]]
4813; CHECK-NEXT:    store i32 [[INDVARS_IV]], i32* [[ARRAYIDX]], align 4
4814; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i32 [[INDVARS_IV]], 1
4815; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[INDVARS_IV_NEXT]], [[K]]
4816; CHECK-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP45:![0-9]+]]
4817; CHECK:       exit:
4818; CHECK-NEXT:    ret void
4819;
4820; IND-LABEL: @nonprimary(
4821; IND-NEXT:  for.body.preheader:
4822; IND-NEXT:    [[TMP0:%.*]] = sub i32 [[K:%.*]], [[I:%.*]]
4823; IND-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 2
4824; IND-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
4825; IND:       vector.ph:
4826; IND-NEXT:    [[N_VEC:%.*]] = and i32 [[TMP0]], -2
4827; IND-NEXT:    [[IND_END:%.*]] = add i32 [[N_VEC]], [[I]]
4828; IND-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[I]], i64 0
4829; IND-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
4830; IND-NEXT:    [[INDUCTION:%.*]] = add <2 x i32> [[DOTSPLAT]], <i32 0, i32 1>
4831; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
4832; IND:       vector.body:
4833; IND-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4834; IND-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
4835; IND-NEXT:    [[OFFSET_IDX:%.*]] = add i32 [[INDEX]], [[I]]
4836; IND-NEXT:    [[TMP1:%.*]] = sext i32 [[OFFSET_IDX]] to i64
4837; IND-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP1]]
4838; IND-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <2 x i32>*
4839; IND-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP3]], align 4
4840; IND-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
4841; IND-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
4842; IND-NEXT:    [[TMP4:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
4843; IND-NEXT:    br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP44:![0-9]+]]
4844; IND:       middle.block:
4845; IND-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
4846; IND-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
4847; IND:       scalar.ph:
4848; IND-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[I]], [[FOR_BODY_PREHEADER:%.*]] ]
4849; IND-NEXT:    br label [[FOR_BODY:%.*]]
4850; IND:       for.body:
4851; IND-NEXT:    [[INDVARS_IV:%.*]] = phi i32 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
4852; IND-NEXT:    [[TMP5:%.*]] = sext i32 [[INDVARS_IV]] to i64
4853; IND-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP5]]
4854; IND-NEXT:    store i32 [[INDVARS_IV]], i32* [[ARRAYIDX]], align 4
4855; IND-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i32 [[INDVARS_IV]], 1
4856; IND-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[INDVARS_IV_NEXT]], [[K]]
4857; IND-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP45:![0-9]+]]
4858; IND:       exit:
4859; IND-NEXT:    ret void
4860;
4861; UNROLL-LABEL: @nonprimary(
4862; UNROLL-NEXT:  for.body.preheader:
4863; UNROLL-NEXT:    [[TMP0:%.*]] = sub i32 [[K:%.*]], [[I:%.*]]
4864; UNROLL-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 4
4865; UNROLL-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
4866; UNROLL:       vector.ph:
4867; UNROLL-NEXT:    [[N_VEC:%.*]] = and i32 [[TMP0]], -4
4868; UNROLL-NEXT:    [[IND_END:%.*]] = add i32 [[N_VEC]], [[I]]
4869; UNROLL-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[I]], i64 0
4870; UNROLL-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
4871; UNROLL-NEXT:    [[INDUCTION:%.*]] = add <2 x i32> [[DOTSPLAT]], <i32 0, i32 1>
4872; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
4873; UNROLL:       vector.body:
4874; UNROLL-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4875; UNROLL-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
4876; UNROLL-NEXT:    [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
4877; UNROLL-NEXT:    [[OFFSET_IDX:%.*]] = add i32 [[INDEX]], [[I]]
4878; UNROLL-NEXT:    [[TMP1:%.*]] = sext i32 [[OFFSET_IDX]] to i64
4879; UNROLL-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP1]]
4880; UNROLL-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <2 x i32>*
4881; UNROLL-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP3]], align 4
4882; UNROLL-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[TMP2]], i64 2
4883; UNROLL-NEXT:    [[TMP5:%.*]] = bitcast i32* [[TMP4]] to <2 x i32>*
4884; UNROLL-NEXT:    store <2 x i32> [[STEP_ADD]], <2 x i32>* [[TMP5]], align 4
4885; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
4886; UNROLL-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 4, i32 4>
4887; UNROLL-NEXT:    [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
4888; UNROLL-NEXT:    br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP44:![0-9]+]]
4889; UNROLL:       middle.block:
4890; UNROLL-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
4891; UNROLL-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
4892; UNROLL:       scalar.ph:
4893; UNROLL-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[I]], [[FOR_BODY_PREHEADER:%.*]] ]
4894; UNROLL-NEXT:    br label [[FOR_BODY:%.*]]
4895; UNROLL:       for.body:
4896; UNROLL-NEXT:    [[INDVARS_IV:%.*]] = phi i32 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
4897; UNROLL-NEXT:    [[TMP7:%.*]] = sext i32 [[INDVARS_IV]] to i64
4898; UNROLL-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP7]]
4899; UNROLL-NEXT:    store i32 [[INDVARS_IV]], i32* [[ARRAYIDX]], align 4
4900; UNROLL-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i32 [[INDVARS_IV]], 1
4901; UNROLL-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[INDVARS_IV_NEXT]], [[K]]
4902; UNROLL-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP45:![0-9]+]]
4903; UNROLL:       exit:
4904; UNROLL-NEXT:    ret void
4905;
4906; UNROLL-NO-IC-LABEL: @nonprimary(
4907; UNROLL-NO-IC-NEXT:  for.body.preheader:
4908; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = sub i32 [[K:%.*]], [[I:%.*]]
4909; UNROLL-NO-IC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 4
4910; UNROLL-NO-IC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
4911; UNROLL-NO-IC:       vector.ph:
4912; UNROLL-NO-IC-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[TMP0]], 4
4913; UNROLL-NO-IC-NEXT:    [[N_VEC:%.*]] = sub i32 [[TMP0]], [[N_MOD_VF]]
4914; UNROLL-NO-IC-NEXT:    [[IND_END:%.*]] = add i32 [[I]], [[N_VEC]]
4915; UNROLL-NO-IC-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[I]], i32 0
4916; UNROLL-NO-IC-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
4917; UNROLL-NO-IC-NEXT:    [[INDUCTION:%.*]] = add <2 x i32> [[DOTSPLAT]], <i32 0, i32 1>
4918; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
4919; UNROLL-NO-IC:       vector.body:
4920; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4921; UNROLL-NO-IC-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
4922; UNROLL-NO-IC-NEXT:    [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
4923; UNROLL-NO-IC-NEXT:    [[OFFSET_IDX:%.*]] = add i32 [[I]], [[INDEX]]
4924; UNROLL-NO-IC-NEXT:    [[TMP1:%.*]] = add i32 [[OFFSET_IDX]], 0
4925; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = add i32 [[OFFSET_IDX]], 2
4926; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i32 [[TMP1]]
4927; UNROLL-NO-IC-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[TMP2]]
4928; UNROLL-NO-IC-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[TMP3]], i32 0
4929; UNROLL-NO-IC-NEXT:    [[TMP6:%.*]] = bitcast i32* [[TMP5]] to <2 x i32>*
4930; UNROLL-NO-IC-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP6]], align 4
4931; UNROLL-NO-IC-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i32, i32* [[TMP3]], i32 2
4932; UNROLL-NO-IC-NEXT:    [[TMP8:%.*]] = bitcast i32* [[TMP7]] to <2 x i32>*
4933; UNROLL-NO-IC-NEXT:    store <2 x i32> [[STEP_ADD]], <2 x i32>* [[TMP8]], align 4
4934; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
4935; UNROLL-NO-IC-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[STEP_ADD]], <i32 2, i32 2>
4936; UNROLL-NO-IC-NEXT:    [[TMP9:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
4937; UNROLL-NO-IC-NEXT:    br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP44:![0-9]+]]
4938; UNROLL-NO-IC:       middle.block:
4939; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
4940; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
4941; UNROLL-NO-IC:       scalar.ph:
4942; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[I]], [[FOR_BODY_PREHEADER:%.*]] ]
4943; UNROLL-NO-IC-NEXT:    br label [[FOR_BODY:%.*]]
4944; UNROLL-NO-IC:       for.body:
4945; UNROLL-NO-IC-NEXT:    [[INDVARS_IV:%.*]] = phi i32 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
4946; UNROLL-NO-IC-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[INDVARS_IV]]
4947; UNROLL-NO-IC-NEXT:    store i32 [[INDVARS_IV]], i32* [[ARRAYIDX]], align 4
4948; UNROLL-NO-IC-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i32 [[INDVARS_IV]], 1
4949; UNROLL-NO-IC-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[INDVARS_IV_NEXT]], [[K]]
4950; UNROLL-NO-IC-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP45:![0-9]+]]
4951; UNROLL-NO-IC:       exit:
4952; UNROLL-NO-IC-NEXT:    ret void
4953;
4954; INTERLEAVE-LABEL: @nonprimary(
4955; INTERLEAVE-NEXT:  for.body.preheader:
4956; INTERLEAVE-NEXT:    [[TMP0:%.*]] = sub i32 [[K:%.*]], [[I:%.*]]
4957; INTERLEAVE-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 8
4958; INTERLEAVE-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
4959; INTERLEAVE:       vector.ph:
4960; INTERLEAVE-NEXT:    [[N_VEC:%.*]] = and i32 [[TMP0]], -8
4961; INTERLEAVE-NEXT:    [[IND_END:%.*]] = add i32 [[N_VEC]], [[I]]
4962; INTERLEAVE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[I]], i64 0
4963; INTERLEAVE-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <4 x i32> [[DOTSPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer
4964; INTERLEAVE-NEXT:    [[INDUCTION:%.*]] = add <4 x i32> [[DOTSPLAT]], <i32 0, i32 1, i32 2, i32 3>
4965; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
4966; INTERLEAVE:       vector.body:
4967; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
4968; INTERLEAVE-NEXT:    [[VEC_IND:%.*]] = phi <4 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
4969; INTERLEAVE-NEXT:    [[STEP_ADD:%.*]] = add <4 x i32> [[VEC_IND]], <i32 4, i32 4, i32 4, i32 4>
4970; INTERLEAVE-NEXT:    [[OFFSET_IDX:%.*]] = add i32 [[INDEX]], [[I]]
4971; INTERLEAVE-NEXT:    [[TMP1:%.*]] = sext i32 [[OFFSET_IDX]] to i64
4972; INTERLEAVE-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP1]]
4973; INTERLEAVE-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <4 x i32>*
4974; INTERLEAVE-NEXT:    store <4 x i32> [[VEC_IND]], <4 x i32>* [[TMP3]], align 4
4975; INTERLEAVE-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[TMP2]], i64 4
4976; INTERLEAVE-NEXT:    [[TMP5:%.*]] = bitcast i32* [[TMP4]] to <4 x i32>*
4977; INTERLEAVE-NEXT:    store <4 x i32> [[STEP_ADD]], <4 x i32>* [[TMP5]], align 4
4978; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 8
4979; INTERLEAVE-NEXT:    [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 8, i32 8, i32 8, i32 8>
4980; INTERLEAVE-NEXT:    [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
4981; INTERLEAVE-NEXT:    br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP44:![0-9]+]]
4982; INTERLEAVE:       middle.block:
4983; INTERLEAVE-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
4984; INTERLEAVE-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
4985; INTERLEAVE:       scalar.ph:
4986; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[I]], [[FOR_BODY_PREHEADER:%.*]] ]
4987; INTERLEAVE-NEXT:    br label [[FOR_BODY:%.*]]
4988; INTERLEAVE:       for.body:
4989; INTERLEAVE-NEXT:    [[INDVARS_IV:%.*]] = phi i32 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
4990; INTERLEAVE-NEXT:    [[TMP7:%.*]] = sext i32 [[INDVARS_IV]] to i64
4991; INTERLEAVE-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP7]]
4992; INTERLEAVE-NEXT:    store i32 [[INDVARS_IV]], i32* [[ARRAYIDX]], align 4
4993; INTERLEAVE-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i32 [[INDVARS_IV]], 1
4994; INTERLEAVE-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[INDVARS_IV_NEXT]], [[K]]
4995; INTERLEAVE-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP45:![0-9]+]]
4996; INTERLEAVE:       exit:
4997; INTERLEAVE-NEXT:    ret void
4998;
4999for.body.preheader:
5000  br label %for.body
5001
5002for.body:
5003  %indvars.iv = phi i32 [ %indvars.iv.next, %for.body ], [ %i, %for.body.preheader ]
5004  %arrayidx = getelementptr inbounds i32, i32* %a, i32 %indvars.iv
5005  store i32 %indvars.iv, i32* %arrayidx, align 4
5006  %indvars.iv.next = add nuw nsw i32 %indvars.iv, 1
5007  %exitcond = icmp eq i32 %indvars.iv.next, %k
5008  br i1 %exitcond, label %exit, label %for.body
5009
5010exit:
5011  ret void
5012}
5013
5014define void @non_primary_iv_trunc(i32* %a, i64 %n) {
5015; CHECK-LABEL: @non_primary_iv_trunc(
5016; CHECK-NEXT:  entry:
5017; CHECK-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)
5018; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 2
5019; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
5020; CHECK:       vector.ph:
5021; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[SMAX]], 2
5022; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[SMAX]], [[N_MOD_VF]]
5023; CHECK-NEXT:    [[IND_END:%.*]] = mul i64 [[N_VEC]], 2
5024; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
5025; CHECK:       vector.body:
5026; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
5027; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 2>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
5028; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
5029; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP0]]
5030; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[TMP1]], i32 0
5031; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <2 x i32>*
5032; CHECK-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP3]], align 4
5033; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
5034; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 4, i32 4>
5035; CHECK-NEXT:    [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
5036; CHECK-NEXT:    br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP46:![0-9]+]]
5037; CHECK:       middle.block:
5038; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
5039; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
5040; CHECK:       scalar.ph:
5041; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
5042; CHECK-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
5043; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
5044; CHECK:       for.body:
5045; CHECK-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
5046; CHECK-NEXT:    [[J:%.*]] = phi i64 [ [[J_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]
5047; CHECK-NEXT:    [[VAR0:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[I]]
5048; CHECK-NEXT:    [[VAR1:%.*]] = trunc i64 [[J]] to i32
5049; CHECK-NEXT:    store i32 [[VAR1]], i32* [[VAR0]], align 4
5050; CHECK-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
5051; CHECK-NEXT:    [[J_NEXT]] = add nuw nsw i64 [[J]], 2
5052; CHECK-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
5053; CHECK-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP47:![0-9]+]]
5054; CHECK:       for.end:
5055; CHECK-NEXT:    ret void
5056;
5057; IND-LABEL: @non_primary_iv_trunc(
5058; IND-NEXT:  entry:
5059; IND-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)
5060; IND-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 2
5061; IND-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
5062; IND:       vector.ph:
5063; IND-NEXT:    [[N_VEC:%.*]] = and i64 [[SMAX]], 9223372036854775806
5064; IND-NEXT:    [[IND_END:%.*]] = shl nuw i64 [[N_VEC]], 1
5065; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
5066; IND:       vector.body:
5067; IND-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
5068; IND-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 2>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
5069; IND-NEXT:    [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[INDEX]]
5070; IND-NEXT:    [[TMP1:%.*]] = bitcast i32* [[TMP0]] to <2 x i32>*
5071; IND-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP1]], align 4
5072; IND-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
5073; IND-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 4, i32 4>
5074; IND-NEXT:    [[TMP2:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
5075; IND-NEXT:    br i1 [[TMP2]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP46:![0-9]+]]
5076; IND:       middle.block:
5077; IND-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
5078; IND-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
5079; IND:       scalar.ph:
5080; IND-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
5081; IND-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
5082; IND-NEXT:    br label [[FOR_BODY:%.*]]
5083; IND:       for.body:
5084; IND-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
5085; IND-NEXT:    [[J:%.*]] = phi i64 [ [[J_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]
5086; IND-NEXT:    [[VAR0:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[I]]
5087; IND-NEXT:    [[VAR1:%.*]] = trunc i64 [[J]] to i32
5088; IND-NEXT:    store i32 [[VAR1]], i32* [[VAR0]], align 4
5089; IND-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
5090; IND-NEXT:    [[J_NEXT]] = add nuw nsw i64 [[J]], 2
5091; IND-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
5092; IND-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP47:![0-9]+]]
5093; IND:       for.end:
5094; IND-NEXT:    ret void
5095;
5096; UNROLL-LABEL: @non_primary_iv_trunc(
5097; UNROLL-NEXT:  entry:
5098; UNROLL-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)
5099; UNROLL-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 4
5100; UNROLL-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
5101; UNROLL:       vector.ph:
5102; UNROLL-NEXT:    [[N_VEC:%.*]] = and i64 [[SMAX]], 9223372036854775804
5103; UNROLL-NEXT:    [[IND_END:%.*]] = shl nuw i64 [[N_VEC]], 1
5104; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
5105; UNROLL:       vector.body:
5106; UNROLL-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
5107; UNROLL-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 2>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
5108; UNROLL-NEXT:    [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], <i32 4, i32 4>
5109; UNROLL-NEXT:    [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[INDEX]]
5110; UNROLL-NEXT:    [[TMP1:%.*]] = bitcast i32* [[TMP0]] to <2 x i32>*
5111; UNROLL-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP1]], align 4
5112; UNROLL-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[TMP0]], i64 2
5113; UNROLL-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <2 x i32>*
5114; UNROLL-NEXT:    store <2 x i32> [[STEP_ADD]], <2 x i32>* [[TMP3]], align 4
5115; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
5116; UNROLL-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 8, i32 8>
5117; UNROLL-NEXT:    [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
5118; UNROLL-NEXT:    br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP46:![0-9]+]]
5119; UNROLL:       middle.block:
5120; UNROLL-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
5121; UNROLL-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
5122; UNROLL:       scalar.ph:
5123; UNROLL-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
5124; UNROLL-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
5125; UNROLL-NEXT:    br label [[FOR_BODY:%.*]]
5126; UNROLL:       for.body:
5127; UNROLL-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
5128; UNROLL-NEXT:    [[J:%.*]] = phi i64 [ [[J_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]
5129; UNROLL-NEXT:    [[VAR0:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[I]]
5130; UNROLL-NEXT:    [[VAR1:%.*]] = trunc i64 [[J]] to i32
5131; UNROLL-NEXT:    store i32 [[VAR1]], i32* [[VAR0]], align 4
5132; UNROLL-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
5133; UNROLL-NEXT:    [[J_NEXT]] = add nuw nsw i64 [[J]], 2
5134; UNROLL-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
5135; UNROLL-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP47:![0-9]+]]
5136; UNROLL:       for.end:
5137; UNROLL-NEXT:    ret void
5138;
5139; UNROLL-NO-IC-LABEL: @non_primary_iv_trunc(
5140; UNROLL-NO-IC-NEXT:  entry:
5141; UNROLL-NO-IC-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)
5142; UNROLL-NO-IC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 4
5143; UNROLL-NO-IC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
5144; UNROLL-NO-IC:       vector.ph:
5145; UNROLL-NO-IC-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[SMAX]], 4
5146; UNROLL-NO-IC-NEXT:    [[N_VEC:%.*]] = sub i64 [[SMAX]], [[N_MOD_VF]]
5147; UNROLL-NO-IC-NEXT:    [[IND_END:%.*]] = mul i64 [[N_VEC]], 2
5148; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
5149; UNROLL-NO-IC:       vector.body:
5150; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
5151; UNROLL-NO-IC-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 2>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
5152; UNROLL-NO-IC-NEXT:    [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], <i32 4, i32 4>
5153; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
5154; UNROLL-NO-IC-NEXT:    [[TMP1:%.*]] = add i64 [[INDEX]], 2
5155; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP0]]
5156; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP1]]
5157; UNROLL-NO-IC-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[TMP2]], i32 0
5158; UNROLL-NO-IC-NEXT:    [[TMP5:%.*]] = bitcast i32* [[TMP4]] to <2 x i32>*
5159; UNROLL-NO-IC-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP5]], align 4
5160; UNROLL-NO-IC-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i32, i32* [[TMP2]], i32 2
5161; UNROLL-NO-IC-NEXT:    [[TMP7:%.*]] = bitcast i32* [[TMP6]] to <2 x i32>*
5162; UNROLL-NO-IC-NEXT:    store <2 x i32> [[STEP_ADD]], <2 x i32>* [[TMP7]], align 4
5163; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
5164; UNROLL-NO-IC-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[STEP_ADD]], <i32 4, i32 4>
5165; UNROLL-NO-IC-NEXT:    [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
5166; UNROLL-NO-IC-NEXT:    br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP46:![0-9]+]]
5167; UNROLL-NO-IC:       middle.block:
5168; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
5169; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
5170; UNROLL-NO-IC:       scalar.ph:
5171; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
5172; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
5173; UNROLL-NO-IC-NEXT:    br label [[FOR_BODY:%.*]]
5174; UNROLL-NO-IC:       for.body:
5175; UNROLL-NO-IC-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
5176; UNROLL-NO-IC-NEXT:    [[J:%.*]] = phi i64 [ [[J_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]
5177; UNROLL-NO-IC-NEXT:    [[VAR0:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[I]]
5178; UNROLL-NO-IC-NEXT:    [[VAR1:%.*]] = trunc i64 [[J]] to i32
5179; UNROLL-NO-IC-NEXT:    store i32 [[VAR1]], i32* [[VAR0]], align 4
5180; UNROLL-NO-IC-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
5181; UNROLL-NO-IC-NEXT:    [[J_NEXT]] = add nuw nsw i64 [[J]], 2
5182; UNROLL-NO-IC-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
5183; UNROLL-NO-IC-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP47:![0-9]+]]
5184; UNROLL-NO-IC:       for.end:
5185; UNROLL-NO-IC-NEXT:    ret void
5186;
5187; INTERLEAVE-LABEL: @non_primary_iv_trunc(
5188; INTERLEAVE-NEXT:  entry:
5189; INTERLEAVE-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)
5190; INTERLEAVE-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 8
5191; INTERLEAVE-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
5192; INTERLEAVE:       vector.ph:
5193; INTERLEAVE-NEXT:    [[N_VEC:%.*]] = and i64 [[SMAX]], 9223372036854775800
5194; INTERLEAVE-NEXT:    [[IND_END:%.*]] = shl nuw i64 [[N_VEC]], 1
5195; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
5196; INTERLEAVE:       vector.body:
5197; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
5198; INTERLEAVE-NEXT:    [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 2, i32 4, i32 6>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
5199; INTERLEAVE-NEXT:    [[STEP_ADD:%.*]] = add <4 x i32> [[VEC_IND]], <i32 8, i32 8, i32 8, i32 8>
5200; INTERLEAVE-NEXT:    [[TMP0:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[INDEX]]
5201; INTERLEAVE-NEXT:    [[TMP1:%.*]] = bitcast i32* [[TMP0]] to <4 x i32>*
5202; INTERLEAVE-NEXT:    store <4 x i32> [[VEC_IND]], <4 x i32>* [[TMP1]], align 4
5203; INTERLEAVE-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[TMP0]], i64 4
5204; INTERLEAVE-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <4 x i32>*
5205; INTERLEAVE-NEXT:    store <4 x i32> [[STEP_ADD]], <4 x i32>* [[TMP3]], align 4
5206; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
5207; INTERLEAVE-NEXT:    [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 16, i32 16, i32 16, i32 16>
5208; INTERLEAVE-NEXT:    [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
5209; INTERLEAVE-NEXT:    br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP46:![0-9]+]]
5210; INTERLEAVE:       middle.block:
5211; INTERLEAVE-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
5212; INTERLEAVE-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
5213; INTERLEAVE:       scalar.ph:
5214; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
5215; INTERLEAVE-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
5216; INTERLEAVE-NEXT:    br label [[FOR_BODY:%.*]]
5217; INTERLEAVE:       for.body:
5218; INTERLEAVE-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
5219; INTERLEAVE-NEXT:    [[J:%.*]] = phi i64 [ [[J_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]
5220; INTERLEAVE-NEXT:    [[VAR0:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[I]]
5221; INTERLEAVE-NEXT:    [[VAR1:%.*]] = trunc i64 [[J]] to i32
5222; INTERLEAVE-NEXT:    store i32 [[VAR1]], i32* [[VAR0]], align 4
5223; INTERLEAVE-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
5224; INTERLEAVE-NEXT:    [[J_NEXT]] = add nuw nsw i64 [[J]], 2
5225; INTERLEAVE-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
5226; INTERLEAVE-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP47:![0-9]+]]
5227; INTERLEAVE:       for.end:
5228; INTERLEAVE-NEXT:    ret void
5229;
5230entry:
5231  br label %for.body
5232
5233for.body:
5234  %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ]
5235  %j = phi i64 [ %j.next, %for.body ], [ 0, %entry ]
5236  %var0 = getelementptr inbounds i32, i32* %a, i64 %i
5237  %var1 = trunc i64 %j to i32
5238  store i32 %var1, i32* %var0, align 4
5239  %i.next = add nuw nsw i64 %i, 1
5240  %j.next = add nuw nsw i64 %j, 2
5241  %cond = icmp slt i64 %i.next, %n
5242  br i1 %cond, label %for.body, label %for.end
5243
5244for.end:
5245  ret void
5246}
5247
5248; PR32419. Ensure we transform truncated non-primary induction variables. In
5249; the test case below we replace %var1 with a new induction variable. Because
5250; the truncated value is non-primary, we must compute an offset from the
5251; primary induction variable.
5252;
5253;
5254define i32 @PR32419(i32 %a, i16 %b) {
5255; CHECK-LABEL: @PR32419(
5256; CHECK-NEXT:  entry:
5257; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
5258; CHECK:       vector.ph:
5259; CHECK-NEXT:    [[TMP0:%.*]] = insertelement <2 x i32> zeroinitializer, i32 [[A:%.*]], i32 0
5260; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
5261; CHECK:       vector.body:
5262; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_UREM_CONTINUE2:%.*]] ]
5263; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ [[TMP0]], [[VECTOR_PH]] ], [ [[TMP15:%.*]], [[PRED_UREM_CONTINUE2]] ]
5264; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i16> [ <i16 -20, i16 -19>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_UREM_CONTINUE2]] ]
5265; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = add i32 -20, [[INDEX]]
5266; CHECK-NEXT:    [[TMP1:%.*]] = trunc i32 [[OFFSET_IDX]] to i16
5267; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq <2 x i16> [[VEC_IND]], zeroinitializer
5268; CHECK-NEXT:    [[TMP3:%.*]] = xor <2 x i1> [[TMP2]], <i1 true, i1 true>
5269; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x i1> [[TMP3]], i32 0
5270; CHECK-NEXT:    br i1 [[TMP4]], label [[PRED_UREM_IF:%.*]], label [[PRED_UREM_CONTINUE:%.*]]
5271; CHECK:       pred.urem.if:
5272; CHECK-NEXT:    [[TMP5:%.*]] = add i16 [[TMP1]], 0
5273; CHECK-NEXT:    [[TMP6:%.*]] = urem i16 [[B:%.*]], [[TMP5]]
5274; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x i16> poison, i16 [[TMP6]], i32 0
5275; CHECK-NEXT:    br label [[PRED_UREM_CONTINUE]]
5276; CHECK:       pred.urem.continue:
5277; CHECK-NEXT:    [[TMP8:%.*]] = phi <2 x i16> [ poison, [[VECTOR_BODY]] ], [ [[TMP7]], [[PRED_UREM_IF]] ]
5278; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <2 x i1> [[TMP3]], i32 1
5279; CHECK-NEXT:    br i1 [[TMP9]], label [[PRED_UREM_IF1:%.*]], label [[PRED_UREM_CONTINUE2]]
5280; CHECK:       pred.urem.if1:
5281; CHECK-NEXT:    [[TMP10:%.*]] = add i16 [[TMP1]], 1
5282; CHECK-NEXT:    [[TMP11:%.*]] = urem i16 [[B]], [[TMP10]]
5283; CHECK-NEXT:    [[TMP12:%.*]] = insertelement <2 x i16> [[TMP8]], i16 [[TMP11]], i32 1
5284; CHECK-NEXT:    br label [[PRED_UREM_CONTINUE2]]
5285; CHECK:       pred.urem.continue2:
5286; CHECK-NEXT:    [[TMP13:%.*]] = phi <2 x i16> [ [[TMP8]], [[PRED_UREM_CONTINUE]] ], [ [[TMP12]], [[PRED_UREM_IF1]] ]
5287; CHECK-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[TMP2]], <2 x i16> zeroinitializer, <2 x i16> [[TMP13]]
5288; CHECK-NEXT:    [[TMP14:%.*]] = sext <2 x i16> [[PREDPHI]] to <2 x i32>
5289; CHECK-NEXT:    [[TMP15]] = or <2 x i32> [[VEC_PHI]], [[TMP14]]
5290; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
5291; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i16> [[VEC_IND]], <i16 2, i16 2>
5292; CHECK-NEXT:    [[TMP16:%.*]] = icmp eq i32 [[INDEX_NEXT]], 20
5293; CHECK-NEXT:    br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP48:![0-9]+]]
5294; CHECK:       middle.block:
5295; CHECK-NEXT:    [[TMP17:%.*]] = call i32 @llvm.vector.reduce.or.v2i32(<2 x i32> [[TMP15]])
5296; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 20, 20
5297; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
5298; CHECK:       scalar.ph:
5299; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ 0, [[MIDDLE_BLOCK]] ], [ -20, [[ENTRY:%.*]] ]
5300; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ [[A]], [[ENTRY]] ], [ [[TMP17]], [[MIDDLE_BLOCK]] ]
5301; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
5302; CHECK:       for.body:
5303; CHECK-NEXT:    [[I:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[FOR_INC:%.*]] ]
5304; CHECK-NEXT:    [[VAR0:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[VAR6:%.*]], [[FOR_INC]] ]
5305; CHECK-NEXT:    [[VAR1:%.*]] = trunc i32 [[I]] to i16
5306; CHECK-NEXT:    [[VAR2:%.*]] = icmp eq i16 [[VAR1]], 0
5307; CHECK-NEXT:    br i1 [[VAR2]], label [[FOR_INC]], label [[FOR_COND:%.*]]
5308; CHECK:       for.cond:
5309; CHECK-NEXT:    [[VAR3:%.*]] = urem i16 [[B]], [[VAR1]]
5310; CHECK-NEXT:    br label [[FOR_INC]]
5311; CHECK:       for.inc:
5312; CHECK-NEXT:    [[VAR4:%.*]] = phi i16 [ [[VAR3]], [[FOR_COND]] ], [ 0, [[FOR_BODY]] ]
5313; CHECK-NEXT:    [[VAR5:%.*]] = sext i16 [[VAR4]] to i32
5314; CHECK-NEXT:    [[VAR6]] = or i32 [[VAR0]], [[VAR5]]
5315; CHECK-NEXT:    [[I_NEXT]] = add nsw i32 [[I]], 1
5316; CHECK-NEXT:    [[COND:%.*]] = icmp eq i32 [[I_NEXT]], 0
5317; CHECK-NEXT:    br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP49:![0-9]+]]
5318; CHECK:       for.end:
5319; CHECK-NEXT:    [[VAR7:%.*]] = phi i32 [ [[VAR6]], [[FOR_INC]] ], [ [[TMP17]], [[MIDDLE_BLOCK]] ]
5320; CHECK-NEXT:    ret i32 [[VAR7]]
5321;
5322; IND-LABEL: @PR32419(
5323; IND-NEXT:  entry:
5324; IND-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
5325; IND:       vector.ph:
5326; IND-NEXT:    [[TMP0:%.*]] = insertelement <2 x i32> <i32 poison, i32 0>, i32 [[A:%.*]], i64 0
5327; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
5328; IND:       vector.body:
5329; IND-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_UREM_CONTINUE2:%.*]] ]
5330; IND-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ [[TMP0]], [[VECTOR_PH]] ], [ [[TMP15:%.*]], [[PRED_UREM_CONTINUE2]] ]
5331; IND-NEXT:    [[VEC_IND:%.*]] = phi <2 x i16> [ <i16 -20, i16 -19>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_UREM_CONTINUE2]] ]
5332; IND-NEXT:    [[TMP1:%.*]] = trunc i32 [[INDEX]] to i16
5333; IND-NEXT:    [[TMP2:%.*]] = icmp eq <2 x i16> [[VEC_IND]], zeroinitializer
5334; IND-NEXT:    [[TMP3:%.*]] = xor <2 x i1> [[TMP2]], <i1 true, i1 true>
5335; IND-NEXT:    [[TMP4:%.*]] = extractelement <2 x i1> [[TMP3]], i64 0
5336; IND-NEXT:    br i1 [[TMP4]], label [[PRED_UREM_IF:%.*]], label [[PRED_UREM_CONTINUE:%.*]]
5337; IND:       pred.urem.if:
5338; IND-NEXT:    [[TMP5:%.*]] = add i16 [[TMP1]], -20
5339; IND-NEXT:    [[TMP6:%.*]] = urem i16 [[B:%.*]], [[TMP5]]
5340; IND-NEXT:    [[TMP7:%.*]] = insertelement <2 x i16> poison, i16 [[TMP6]], i64 0
5341; IND-NEXT:    br label [[PRED_UREM_CONTINUE]]
5342; IND:       pred.urem.continue:
5343; IND-NEXT:    [[TMP8:%.*]] = phi <2 x i16> [ poison, [[VECTOR_BODY]] ], [ [[TMP7]], [[PRED_UREM_IF]] ]
5344; IND-NEXT:    [[TMP9:%.*]] = extractelement <2 x i1> [[TMP3]], i64 1
5345; IND-NEXT:    br i1 [[TMP9]], label [[PRED_UREM_IF1:%.*]], label [[PRED_UREM_CONTINUE2]]
5346; IND:       pred.urem.if1:
5347; IND-NEXT:    [[TMP10:%.*]] = add i16 [[TMP1]], -19
5348; IND-NEXT:    [[TMP11:%.*]] = urem i16 [[B]], [[TMP10]]
5349; IND-NEXT:    [[TMP12:%.*]] = insertelement <2 x i16> [[TMP8]], i16 [[TMP11]], i64 1
5350; IND-NEXT:    br label [[PRED_UREM_CONTINUE2]]
5351; IND:       pred.urem.continue2:
5352; IND-NEXT:    [[TMP13:%.*]] = phi <2 x i16> [ [[TMP8]], [[PRED_UREM_CONTINUE]] ], [ [[TMP12]], [[PRED_UREM_IF1]] ]
5353; IND-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[TMP2]], <2 x i16> zeroinitializer, <2 x i16> [[TMP13]]
5354; IND-NEXT:    [[TMP14:%.*]] = sext <2 x i16> [[PREDPHI]] to <2 x i32>
5355; IND-NEXT:    [[TMP15]] = or <2 x i32> [[VEC_PHI]], [[TMP14]]
5356; IND-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
5357; IND-NEXT:    [[VEC_IND_NEXT]] = add <2 x i16> [[VEC_IND]], <i16 2, i16 2>
5358; IND-NEXT:    [[TMP16:%.*]] = icmp eq i32 [[INDEX_NEXT]], 20
5359; IND-NEXT:    br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP48:![0-9]+]]
5360; IND:       middle.block:
5361; IND-NEXT:    [[TMP17:%.*]] = call i32 @llvm.vector.reduce.or.v2i32(<2 x i32> [[TMP15]])
5362; IND-NEXT:    br i1 true, label [[FOR_END:%.*]], label [[SCALAR_PH]]
5363; IND:       scalar.ph:
5364; IND-NEXT:    br label [[FOR_BODY:%.*]]
5365; IND:       for.body:
5366; IND-NEXT:    br i1 poison, label [[FOR_INC:%.*]], label [[FOR_COND:%.*]]
5367; IND:       for.cond:
5368; IND-NEXT:    br label [[FOR_INC]]
5369; IND:       for.inc:
5370; IND-NEXT:    br i1 poison, label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP49:![0-9]+]]
5371; IND:       for.end:
5372; IND-NEXT:    [[VAR7:%.*]] = phi i32 [ poison, [[FOR_INC]] ], [ [[TMP17]], [[MIDDLE_BLOCK]] ]
5373; IND-NEXT:    ret i32 [[VAR7]]
5374;
5375; UNROLL-LABEL: @PR32419(
5376; UNROLL-NEXT:  entry:
5377; UNROLL-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
5378; UNROLL:       vector.ph:
5379; UNROLL-NEXT:    [[TMP0:%.*]] = insertelement <2 x i32> <i32 poison, i32 0>, i32 [[A:%.*]], i64 0
5380; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
5381; UNROLL:       vector.body:
5382; UNROLL-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_UREM_CONTINUE8:%.*]] ]
5383; UNROLL-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ [[TMP0]], [[VECTOR_PH]] ], [ [[TMP28:%.*]], [[PRED_UREM_CONTINUE8]] ]
5384; UNROLL-NEXT:    [[VEC_PHI1:%.*]] = phi <2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP29:%.*]], [[PRED_UREM_CONTINUE8]] ]
5385; UNROLL-NEXT:    [[VEC_IND:%.*]] = phi <2 x i16> [ <i16 -20, i16 -19>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_UREM_CONTINUE8]] ]
5386; UNROLL-NEXT:    [[TMP1:%.*]] = trunc i32 [[INDEX]] to i16
5387; UNROLL-NEXT:    [[TMP2:%.*]] = icmp eq <2 x i16> [[VEC_IND]], zeroinitializer
5388; UNROLL-NEXT:    [[TMP3:%.*]] = icmp eq <2 x i16> [[VEC_IND]], <i16 -2, i16 -2>
5389; UNROLL-NEXT:    [[TMP4:%.*]] = xor <2 x i1> [[TMP2]], <i1 true, i1 true>
5390; UNROLL-NEXT:    [[TMP5:%.*]] = xor <2 x i1> [[TMP3]], <i1 true, i1 true>
5391; UNROLL-NEXT:    [[TMP6:%.*]] = extractelement <2 x i1> [[TMP4]], i64 0
5392; UNROLL-NEXT:    br i1 [[TMP6]], label [[PRED_UREM_IF:%.*]], label [[PRED_UREM_CONTINUE:%.*]]
5393; UNROLL:       pred.urem.if:
5394; UNROLL-NEXT:    [[TMP7:%.*]] = add i16 [[TMP1]], -20
5395; UNROLL-NEXT:    [[TMP8:%.*]] = urem i16 [[B:%.*]], [[TMP7]]
5396; UNROLL-NEXT:    [[TMP9:%.*]] = insertelement <2 x i16> poison, i16 [[TMP8]], i64 0
5397; UNROLL-NEXT:    br label [[PRED_UREM_CONTINUE]]
5398; UNROLL:       pred.urem.continue:
5399; UNROLL-NEXT:    [[TMP10:%.*]] = phi <2 x i16> [ poison, [[VECTOR_BODY]] ], [ [[TMP9]], [[PRED_UREM_IF]] ]
5400; UNROLL-NEXT:    [[TMP11:%.*]] = extractelement <2 x i1> [[TMP4]], i64 1
5401; UNROLL-NEXT:    br i1 [[TMP11]], label [[PRED_UREM_IF3:%.*]], label [[PRED_UREM_CONTINUE4:%.*]]
5402; UNROLL:       pred.urem.if3:
5403; UNROLL-NEXT:    [[TMP12:%.*]] = add i16 [[TMP1]], -19
5404; UNROLL-NEXT:    [[TMP13:%.*]] = urem i16 [[B]], [[TMP12]]
5405; UNROLL-NEXT:    [[TMP14:%.*]] = insertelement <2 x i16> [[TMP10]], i16 [[TMP13]], i64 1
5406; UNROLL-NEXT:    br label [[PRED_UREM_CONTINUE4]]
5407; UNROLL:       pred.urem.continue4:
5408; UNROLL-NEXT:    [[TMP15:%.*]] = phi <2 x i16> [ [[TMP10]], [[PRED_UREM_CONTINUE]] ], [ [[TMP14]], [[PRED_UREM_IF3]] ]
5409; UNROLL-NEXT:    [[TMP16:%.*]] = extractelement <2 x i1> [[TMP5]], i64 0
5410; UNROLL-NEXT:    br i1 [[TMP16]], label [[PRED_UREM_IF5:%.*]], label [[PRED_UREM_CONTINUE6:%.*]]
5411; UNROLL:       pred.urem.if5:
5412; UNROLL-NEXT:    [[TMP17:%.*]] = add i16 [[TMP1]], -18
5413; UNROLL-NEXT:    [[TMP18:%.*]] = urem i16 [[B]], [[TMP17]]
5414; UNROLL-NEXT:    [[TMP19:%.*]] = insertelement <2 x i16> poison, i16 [[TMP18]], i64 0
5415; UNROLL-NEXT:    br label [[PRED_UREM_CONTINUE6]]
5416; UNROLL:       pred.urem.continue6:
5417; UNROLL-NEXT:    [[TMP20:%.*]] = phi <2 x i16> [ poison, [[PRED_UREM_CONTINUE4]] ], [ [[TMP19]], [[PRED_UREM_IF5]] ]
5418; UNROLL-NEXT:    [[TMP21:%.*]] = extractelement <2 x i1> [[TMP5]], i64 1
5419; UNROLL-NEXT:    br i1 [[TMP21]], label [[PRED_UREM_IF7:%.*]], label [[PRED_UREM_CONTINUE8]]
5420; UNROLL:       pred.urem.if7:
5421; UNROLL-NEXT:    [[TMP22:%.*]] = add i16 [[TMP1]], -17
5422; UNROLL-NEXT:    [[TMP23:%.*]] = urem i16 [[B]], [[TMP22]]
5423; UNROLL-NEXT:    [[TMP24:%.*]] = insertelement <2 x i16> [[TMP20]], i16 [[TMP23]], i64 1
5424; UNROLL-NEXT:    br label [[PRED_UREM_CONTINUE8]]
5425; UNROLL:       pred.urem.continue8:
5426; UNROLL-NEXT:    [[TMP25:%.*]] = phi <2 x i16> [ [[TMP20]], [[PRED_UREM_CONTINUE6]] ], [ [[TMP24]], [[PRED_UREM_IF7]] ]
5427; UNROLL-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[TMP2]], <2 x i16> zeroinitializer, <2 x i16> [[TMP15]]
5428; UNROLL-NEXT:    [[PREDPHI9:%.*]] = select <2 x i1> [[TMP3]], <2 x i16> zeroinitializer, <2 x i16> [[TMP25]]
5429; UNROLL-NEXT:    [[TMP26:%.*]] = sext <2 x i16> [[PREDPHI]] to <2 x i32>
5430; UNROLL-NEXT:    [[TMP27:%.*]] = sext <2 x i16> [[PREDPHI9]] to <2 x i32>
5431; UNROLL-NEXT:    [[TMP28]] = or <2 x i32> [[VEC_PHI]], [[TMP26]]
5432; UNROLL-NEXT:    [[TMP29]] = or <2 x i32> [[VEC_PHI1]], [[TMP27]]
5433; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
5434; UNROLL-NEXT:    [[VEC_IND_NEXT]] = add <2 x i16> [[VEC_IND]], <i16 4, i16 4>
5435; UNROLL-NEXT:    [[TMP30:%.*]] = icmp eq i32 [[INDEX_NEXT]], 20
5436; UNROLL-NEXT:    br i1 [[TMP30]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP48:![0-9]+]]
5437; UNROLL:       middle.block:
5438; UNROLL-NEXT:    [[BIN_RDX:%.*]] = or <2 x i32> [[TMP29]], [[TMP28]]
5439; UNROLL-NEXT:    [[TMP31:%.*]] = call i32 @llvm.vector.reduce.or.v2i32(<2 x i32> [[BIN_RDX]])
5440; UNROLL-NEXT:    br i1 true, label [[FOR_END:%.*]], label [[SCALAR_PH]]
5441; UNROLL:       scalar.ph:
5442; UNROLL-NEXT:    br label [[FOR_BODY:%.*]]
5443; UNROLL:       for.body:
5444; UNROLL-NEXT:    br i1 poison, label [[FOR_INC:%.*]], label [[FOR_COND:%.*]]
5445; UNROLL:       for.cond:
5446; UNROLL-NEXT:    br label [[FOR_INC]]
5447; UNROLL:       for.inc:
5448; UNROLL-NEXT:    br i1 poison, label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP49:![0-9]+]]
5449; UNROLL:       for.end:
5450; UNROLL-NEXT:    [[VAR7:%.*]] = phi i32 [ poison, [[FOR_INC]] ], [ [[TMP31]], [[MIDDLE_BLOCK]] ]
5451; UNROLL-NEXT:    ret i32 [[VAR7]]
5452;
5453; UNROLL-NO-IC-LABEL: @PR32419(
5454; UNROLL-NO-IC-NEXT:  entry:
5455; UNROLL-NO-IC-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
5456; UNROLL-NO-IC:       vector.ph:
5457; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = insertelement <2 x i32> zeroinitializer, i32 [[A:%.*]], i32 0
5458; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
5459; UNROLL-NO-IC:       vector.body:
5460; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_UREM_CONTINUE8:%.*]] ]
5461; UNROLL-NO-IC-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ [[TMP0]], [[VECTOR_PH]] ], [ [[TMP28:%.*]], [[PRED_UREM_CONTINUE8]] ]
5462; UNROLL-NO-IC-NEXT:    [[VEC_PHI1:%.*]] = phi <2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP29:%.*]], [[PRED_UREM_CONTINUE8]] ]
5463; UNROLL-NO-IC-NEXT:    [[VEC_IND:%.*]] = phi <2 x i16> [ <i16 -20, i16 -19>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_UREM_CONTINUE8]] ]
5464; UNROLL-NO-IC-NEXT:    [[STEP_ADD:%.*]] = add <2 x i16> [[VEC_IND]], <i16 2, i16 2>
5465; UNROLL-NO-IC-NEXT:    [[OFFSET_IDX:%.*]] = add i32 -20, [[INDEX]]
5466; UNROLL-NO-IC-NEXT:    [[TMP1:%.*]] = trunc i32 [[OFFSET_IDX]] to i16
5467; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = icmp eq <2 x i16> [[VEC_IND]], zeroinitializer
5468; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = icmp eq <2 x i16> [[STEP_ADD]], zeroinitializer
5469; UNROLL-NO-IC-NEXT:    [[TMP4:%.*]] = xor <2 x i1> [[TMP2]], <i1 true, i1 true>
5470; UNROLL-NO-IC-NEXT:    [[TMP5:%.*]] = xor <2 x i1> [[TMP3]], <i1 true, i1 true>
5471; UNROLL-NO-IC-NEXT:    [[TMP6:%.*]] = extractelement <2 x i1> [[TMP4]], i32 0
5472; UNROLL-NO-IC-NEXT:    br i1 [[TMP6]], label [[PRED_UREM_IF:%.*]], label [[PRED_UREM_CONTINUE:%.*]]
5473; UNROLL-NO-IC:       pred.urem.if:
5474; UNROLL-NO-IC-NEXT:    [[TMP7:%.*]] = add i16 [[TMP1]], 0
5475; UNROLL-NO-IC-NEXT:    [[TMP8:%.*]] = urem i16 [[B:%.*]], [[TMP7]]
5476; UNROLL-NO-IC-NEXT:    [[TMP9:%.*]] = insertelement <2 x i16> poison, i16 [[TMP8]], i32 0
5477; UNROLL-NO-IC-NEXT:    br label [[PRED_UREM_CONTINUE]]
5478; UNROLL-NO-IC:       pred.urem.continue:
5479; UNROLL-NO-IC-NEXT:    [[TMP10:%.*]] = phi <2 x i16> [ poison, [[VECTOR_BODY]] ], [ [[TMP9]], [[PRED_UREM_IF]] ]
5480; UNROLL-NO-IC-NEXT:    [[TMP11:%.*]] = extractelement <2 x i1> [[TMP4]], i32 1
5481; UNROLL-NO-IC-NEXT:    br i1 [[TMP11]], label [[PRED_UREM_IF3:%.*]], label [[PRED_UREM_CONTINUE4:%.*]]
5482; UNROLL-NO-IC:       pred.urem.if3:
5483; UNROLL-NO-IC-NEXT:    [[TMP12:%.*]] = add i16 [[TMP1]], 1
5484; UNROLL-NO-IC-NEXT:    [[TMP13:%.*]] = urem i16 [[B]], [[TMP12]]
5485; UNROLL-NO-IC-NEXT:    [[TMP14:%.*]] = insertelement <2 x i16> [[TMP10]], i16 [[TMP13]], i32 1
5486; UNROLL-NO-IC-NEXT:    br label [[PRED_UREM_CONTINUE4]]
5487; UNROLL-NO-IC:       pred.urem.continue4:
5488; UNROLL-NO-IC-NEXT:    [[TMP15:%.*]] = phi <2 x i16> [ [[TMP10]], [[PRED_UREM_CONTINUE]] ], [ [[TMP14]], [[PRED_UREM_IF3]] ]
5489; UNROLL-NO-IC-NEXT:    [[TMP16:%.*]] = extractelement <2 x i1> [[TMP5]], i32 0
5490; UNROLL-NO-IC-NEXT:    br i1 [[TMP16]], label [[PRED_UREM_IF5:%.*]], label [[PRED_UREM_CONTINUE6:%.*]]
5491; UNROLL-NO-IC:       pred.urem.if5:
5492; UNROLL-NO-IC-NEXT:    [[TMP17:%.*]] = add i16 [[TMP1]], 2
5493; UNROLL-NO-IC-NEXT:    [[TMP18:%.*]] = urem i16 [[B]], [[TMP17]]
5494; UNROLL-NO-IC-NEXT:    [[TMP19:%.*]] = insertelement <2 x i16> poison, i16 [[TMP18]], i32 0
5495; UNROLL-NO-IC-NEXT:    br label [[PRED_UREM_CONTINUE6]]
5496; UNROLL-NO-IC:       pred.urem.continue6:
5497; UNROLL-NO-IC-NEXT:    [[TMP20:%.*]] = phi <2 x i16> [ poison, [[PRED_UREM_CONTINUE4]] ], [ [[TMP19]], [[PRED_UREM_IF5]] ]
5498; UNROLL-NO-IC-NEXT:    [[TMP21:%.*]] = extractelement <2 x i1> [[TMP5]], i32 1
5499; UNROLL-NO-IC-NEXT:    br i1 [[TMP21]], label [[PRED_UREM_IF7:%.*]], label [[PRED_UREM_CONTINUE8]]
5500; UNROLL-NO-IC:       pred.urem.if7:
5501; UNROLL-NO-IC-NEXT:    [[TMP22:%.*]] = add i16 [[TMP1]], 3
5502; UNROLL-NO-IC-NEXT:    [[TMP23:%.*]] = urem i16 [[B]], [[TMP22]]
5503; UNROLL-NO-IC-NEXT:    [[TMP24:%.*]] = insertelement <2 x i16> [[TMP20]], i16 [[TMP23]], i32 1
5504; UNROLL-NO-IC-NEXT:    br label [[PRED_UREM_CONTINUE8]]
5505; UNROLL-NO-IC:       pred.urem.continue8:
5506; UNROLL-NO-IC-NEXT:    [[TMP25:%.*]] = phi <2 x i16> [ [[TMP20]], [[PRED_UREM_CONTINUE6]] ], [ [[TMP24]], [[PRED_UREM_IF7]] ]
5507; UNROLL-NO-IC-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[TMP2]], <2 x i16> zeroinitializer, <2 x i16> [[TMP15]]
5508; UNROLL-NO-IC-NEXT:    [[PREDPHI9:%.*]] = select <2 x i1> [[TMP3]], <2 x i16> zeroinitializer, <2 x i16> [[TMP25]]
5509; UNROLL-NO-IC-NEXT:    [[TMP26:%.*]] = sext <2 x i16> [[PREDPHI]] to <2 x i32>
5510; UNROLL-NO-IC-NEXT:    [[TMP27:%.*]] = sext <2 x i16> [[PREDPHI9]] to <2 x i32>
5511; UNROLL-NO-IC-NEXT:    [[TMP28]] = or <2 x i32> [[VEC_PHI]], [[TMP26]]
5512; UNROLL-NO-IC-NEXT:    [[TMP29]] = or <2 x i32> [[VEC_PHI1]], [[TMP27]]
5513; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
5514; UNROLL-NO-IC-NEXT:    [[VEC_IND_NEXT]] = add <2 x i16> [[STEP_ADD]], <i16 2, i16 2>
5515; UNROLL-NO-IC-NEXT:    [[TMP30:%.*]] = icmp eq i32 [[INDEX_NEXT]], 20
5516; UNROLL-NO-IC-NEXT:    br i1 [[TMP30]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP48:![0-9]+]]
5517; UNROLL-NO-IC:       middle.block:
5518; UNROLL-NO-IC-NEXT:    [[BIN_RDX:%.*]] = or <2 x i32> [[TMP29]], [[TMP28]]
5519; UNROLL-NO-IC-NEXT:    [[TMP31:%.*]] = call i32 @llvm.vector.reduce.or.v2i32(<2 x i32> [[BIN_RDX]])
5520; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i32 20, 20
5521; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
5522; UNROLL-NO-IC:       scalar.ph:
5523; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ 0, [[MIDDLE_BLOCK]] ], [ -20, [[ENTRY:%.*]] ]
5524; UNROLL-NO-IC-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ [[A]], [[ENTRY]] ], [ [[TMP31]], [[MIDDLE_BLOCK]] ]
5525; UNROLL-NO-IC-NEXT:    br label [[FOR_BODY:%.*]]
5526; UNROLL-NO-IC:       for.body:
5527; UNROLL-NO-IC-NEXT:    [[I:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[FOR_INC:%.*]] ]
5528; UNROLL-NO-IC-NEXT:    [[VAR0:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[VAR6:%.*]], [[FOR_INC]] ]
5529; UNROLL-NO-IC-NEXT:    [[VAR1:%.*]] = trunc i32 [[I]] to i16
5530; UNROLL-NO-IC-NEXT:    [[VAR2:%.*]] = icmp eq i16 [[VAR1]], 0
5531; UNROLL-NO-IC-NEXT:    br i1 [[VAR2]], label [[FOR_INC]], label [[FOR_COND:%.*]]
5532; UNROLL-NO-IC:       for.cond:
5533; UNROLL-NO-IC-NEXT:    [[VAR3:%.*]] = urem i16 [[B]], [[VAR1]]
5534; UNROLL-NO-IC-NEXT:    br label [[FOR_INC]]
5535; UNROLL-NO-IC:       for.inc:
5536; UNROLL-NO-IC-NEXT:    [[VAR4:%.*]] = phi i16 [ [[VAR3]], [[FOR_COND]] ], [ 0, [[FOR_BODY]] ]
5537; UNROLL-NO-IC-NEXT:    [[VAR5:%.*]] = sext i16 [[VAR4]] to i32
5538; UNROLL-NO-IC-NEXT:    [[VAR6]] = or i32 [[VAR0]], [[VAR5]]
5539; UNROLL-NO-IC-NEXT:    [[I_NEXT]] = add nsw i32 [[I]], 1
5540; UNROLL-NO-IC-NEXT:    [[COND:%.*]] = icmp eq i32 [[I_NEXT]], 0
5541; UNROLL-NO-IC-NEXT:    br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP49:![0-9]+]]
5542; UNROLL-NO-IC:       for.end:
5543; UNROLL-NO-IC-NEXT:    [[VAR7:%.*]] = phi i32 [ [[VAR6]], [[FOR_INC]] ], [ [[TMP31]], [[MIDDLE_BLOCK]] ]
5544; UNROLL-NO-IC-NEXT:    ret i32 [[VAR7]]
5545;
5546; INTERLEAVE-LABEL: @PR32419(
5547; INTERLEAVE-NEXT:  entry:
5548; INTERLEAVE-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
5549; INTERLEAVE:       vector.ph:
5550; INTERLEAVE-NEXT:    [[TMP0:%.*]] = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 [[A:%.*]], i64 0
5551; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
5552; INTERLEAVE:       vector.body:
5553; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_UREM_CONTINUE16:%.*]] ]
5554; INTERLEAVE-NEXT:    [[VEC_PHI:%.*]] = phi <4 x i32> [ [[TMP0]], [[VECTOR_PH]] ], [ [[TMP48:%.*]], [[PRED_UREM_CONTINUE16]] ]
5555; INTERLEAVE-NEXT:    [[VEC_PHI1:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP49:%.*]], [[PRED_UREM_CONTINUE16]] ]
5556; INTERLEAVE-NEXT:    [[VEC_IND:%.*]] = phi <4 x i16> [ <i16 -20, i16 -19, i16 -18, i16 -17>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_UREM_CONTINUE16]] ]
5557; INTERLEAVE-NEXT:    [[TMP1:%.*]] = trunc i32 [[INDEX]] to i16
5558; INTERLEAVE-NEXT:    [[TMP2:%.*]] = icmp eq <4 x i16> [[VEC_IND]], zeroinitializer
5559; INTERLEAVE-NEXT:    [[TMP3:%.*]] = icmp eq <4 x i16> [[VEC_IND]], <i16 -4, i16 -4, i16 -4, i16 -4>
5560; INTERLEAVE-NEXT:    [[TMP4:%.*]] = xor <4 x i1> [[TMP2]], <i1 true, i1 true, i1 true, i1 true>
5561; INTERLEAVE-NEXT:    [[TMP5:%.*]] = xor <4 x i1> [[TMP3]], <i1 true, i1 true, i1 true, i1 true>
5562; INTERLEAVE-NEXT:    [[TMP6:%.*]] = extractelement <4 x i1> [[TMP4]], i64 0
5563; INTERLEAVE-NEXT:    br i1 [[TMP6]], label [[PRED_UREM_IF:%.*]], label [[PRED_UREM_CONTINUE:%.*]]
5564; INTERLEAVE:       pred.urem.if:
5565; INTERLEAVE-NEXT:    [[TMP7:%.*]] = add i16 [[TMP1]], -20
5566; INTERLEAVE-NEXT:    [[TMP8:%.*]] = urem i16 [[B:%.*]], [[TMP7]]
5567; INTERLEAVE-NEXT:    [[TMP9:%.*]] = insertelement <4 x i16> poison, i16 [[TMP8]], i64 0
5568; INTERLEAVE-NEXT:    br label [[PRED_UREM_CONTINUE]]
5569; INTERLEAVE:       pred.urem.continue:
5570; INTERLEAVE-NEXT:    [[TMP10:%.*]] = phi <4 x i16> [ poison, [[VECTOR_BODY]] ], [ [[TMP9]], [[PRED_UREM_IF]] ]
5571; INTERLEAVE-NEXT:    [[TMP11:%.*]] = extractelement <4 x i1> [[TMP4]], i64 1
5572; INTERLEAVE-NEXT:    br i1 [[TMP11]], label [[PRED_UREM_IF3:%.*]], label [[PRED_UREM_CONTINUE4:%.*]]
5573; INTERLEAVE:       pred.urem.if3:
5574; INTERLEAVE-NEXT:    [[TMP12:%.*]] = add i16 [[TMP1]], -19
5575; INTERLEAVE-NEXT:    [[TMP13:%.*]] = urem i16 [[B]], [[TMP12]]
5576; INTERLEAVE-NEXT:    [[TMP14:%.*]] = insertelement <4 x i16> [[TMP10]], i16 [[TMP13]], i64 1
5577; INTERLEAVE-NEXT:    br label [[PRED_UREM_CONTINUE4]]
5578; INTERLEAVE:       pred.urem.continue4:
5579; INTERLEAVE-NEXT:    [[TMP15:%.*]] = phi <4 x i16> [ [[TMP10]], [[PRED_UREM_CONTINUE]] ], [ [[TMP14]], [[PRED_UREM_IF3]] ]
5580; INTERLEAVE-NEXT:    [[TMP16:%.*]] = extractelement <4 x i1> [[TMP4]], i64 2
5581; INTERLEAVE-NEXT:    br i1 [[TMP16]], label [[PRED_UREM_IF5:%.*]], label [[PRED_UREM_CONTINUE6:%.*]]
5582; INTERLEAVE:       pred.urem.if5:
5583; INTERLEAVE-NEXT:    [[TMP17:%.*]] = add i16 [[TMP1]], -18
5584; INTERLEAVE-NEXT:    [[TMP18:%.*]] = urem i16 [[B]], [[TMP17]]
5585; INTERLEAVE-NEXT:    [[TMP19:%.*]] = insertelement <4 x i16> [[TMP15]], i16 [[TMP18]], i64 2
5586; INTERLEAVE-NEXT:    br label [[PRED_UREM_CONTINUE6]]
5587; INTERLEAVE:       pred.urem.continue6:
5588; INTERLEAVE-NEXT:    [[TMP20:%.*]] = phi <4 x i16> [ [[TMP15]], [[PRED_UREM_CONTINUE4]] ], [ [[TMP19]], [[PRED_UREM_IF5]] ]
5589; INTERLEAVE-NEXT:    [[TMP21:%.*]] = extractelement <4 x i1> [[TMP4]], i64 3
5590; INTERLEAVE-NEXT:    br i1 [[TMP21]], label [[PRED_UREM_IF7:%.*]], label [[PRED_UREM_CONTINUE8:%.*]]
5591; INTERLEAVE:       pred.urem.if7:
5592; INTERLEAVE-NEXT:    [[TMP22:%.*]] = add i16 [[TMP1]], -17
5593; INTERLEAVE-NEXT:    [[TMP23:%.*]] = urem i16 [[B]], [[TMP22]]
5594; INTERLEAVE-NEXT:    [[TMP24:%.*]] = insertelement <4 x i16> [[TMP20]], i16 [[TMP23]], i64 3
5595; INTERLEAVE-NEXT:    br label [[PRED_UREM_CONTINUE8]]
5596; INTERLEAVE:       pred.urem.continue8:
5597; INTERLEAVE-NEXT:    [[TMP25:%.*]] = phi <4 x i16> [ [[TMP20]], [[PRED_UREM_CONTINUE6]] ], [ [[TMP24]], [[PRED_UREM_IF7]] ]
5598; INTERLEAVE-NEXT:    [[TMP26:%.*]] = extractelement <4 x i1> [[TMP5]], i64 0
5599; INTERLEAVE-NEXT:    br i1 [[TMP26]], label [[PRED_UREM_IF9:%.*]], label [[PRED_UREM_CONTINUE10:%.*]]
5600; INTERLEAVE:       pred.urem.if9:
5601; INTERLEAVE-NEXT:    [[TMP27:%.*]] = add i16 [[TMP1]], -16
5602; INTERLEAVE-NEXT:    [[TMP28:%.*]] = urem i16 [[B]], [[TMP27]]
5603; INTERLEAVE-NEXT:    [[TMP29:%.*]] = insertelement <4 x i16> poison, i16 [[TMP28]], i64 0
5604; INTERLEAVE-NEXT:    br label [[PRED_UREM_CONTINUE10]]
5605; INTERLEAVE:       pred.urem.continue10:
5606; INTERLEAVE-NEXT:    [[TMP30:%.*]] = phi <4 x i16> [ poison, [[PRED_UREM_CONTINUE8]] ], [ [[TMP29]], [[PRED_UREM_IF9]] ]
5607; INTERLEAVE-NEXT:    [[TMP31:%.*]] = extractelement <4 x i1> [[TMP5]], i64 1
5608; INTERLEAVE-NEXT:    br i1 [[TMP31]], label [[PRED_UREM_IF11:%.*]], label [[PRED_UREM_CONTINUE12:%.*]]
5609; INTERLEAVE:       pred.urem.if11:
5610; INTERLEAVE-NEXT:    [[TMP32:%.*]] = add i16 [[TMP1]], -15
5611; INTERLEAVE-NEXT:    [[TMP33:%.*]] = urem i16 [[B]], [[TMP32]]
5612; INTERLEAVE-NEXT:    [[TMP34:%.*]] = insertelement <4 x i16> [[TMP30]], i16 [[TMP33]], i64 1
5613; INTERLEAVE-NEXT:    br label [[PRED_UREM_CONTINUE12]]
5614; INTERLEAVE:       pred.urem.continue12:
5615; INTERLEAVE-NEXT:    [[TMP35:%.*]] = phi <4 x i16> [ [[TMP30]], [[PRED_UREM_CONTINUE10]] ], [ [[TMP34]], [[PRED_UREM_IF11]] ]
5616; INTERLEAVE-NEXT:    [[TMP36:%.*]] = extractelement <4 x i1> [[TMP5]], i64 2
5617; INTERLEAVE-NEXT:    br i1 [[TMP36]], label [[PRED_UREM_IF13:%.*]], label [[PRED_UREM_CONTINUE14:%.*]]
5618; INTERLEAVE:       pred.urem.if13:
5619; INTERLEAVE-NEXT:    [[TMP37:%.*]] = add i16 [[TMP1]], -14
5620; INTERLEAVE-NEXT:    [[TMP38:%.*]] = urem i16 [[B]], [[TMP37]]
5621; INTERLEAVE-NEXT:    [[TMP39:%.*]] = insertelement <4 x i16> [[TMP35]], i16 [[TMP38]], i64 2
5622; INTERLEAVE-NEXT:    br label [[PRED_UREM_CONTINUE14]]
5623; INTERLEAVE:       pred.urem.continue14:
5624; INTERLEAVE-NEXT:    [[TMP40:%.*]] = phi <4 x i16> [ [[TMP35]], [[PRED_UREM_CONTINUE12]] ], [ [[TMP39]], [[PRED_UREM_IF13]] ]
5625; INTERLEAVE-NEXT:    [[TMP41:%.*]] = extractelement <4 x i1> [[TMP5]], i64 3
5626; INTERLEAVE-NEXT:    br i1 [[TMP41]], label [[PRED_UREM_IF15:%.*]], label [[PRED_UREM_CONTINUE16]]
5627; INTERLEAVE:       pred.urem.if15:
5628; INTERLEAVE-NEXT:    [[TMP42:%.*]] = add i16 [[TMP1]], -13
5629; INTERLEAVE-NEXT:    [[TMP43:%.*]] = urem i16 [[B]], [[TMP42]]
5630; INTERLEAVE-NEXT:    [[TMP44:%.*]] = insertelement <4 x i16> [[TMP40]], i16 [[TMP43]], i64 3
5631; INTERLEAVE-NEXT:    br label [[PRED_UREM_CONTINUE16]]
5632; INTERLEAVE:       pred.urem.continue16:
5633; INTERLEAVE-NEXT:    [[TMP45:%.*]] = phi <4 x i16> [ [[TMP40]], [[PRED_UREM_CONTINUE14]] ], [ [[TMP44]], [[PRED_UREM_IF15]] ]
5634; INTERLEAVE-NEXT:    [[PREDPHI:%.*]] = select <4 x i1> [[TMP2]], <4 x i16> zeroinitializer, <4 x i16> [[TMP25]]
5635; INTERLEAVE-NEXT:    [[PREDPHI17:%.*]] = select <4 x i1> [[TMP3]], <4 x i16> zeroinitializer, <4 x i16> [[TMP45]]
5636; INTERLEAVE-NEXT:    [[TMP46:%.*]] = sext <4 x i16> [[PREDPHI]] to <4 x i32>
5637; INTERLEAVE-NEXT:    [[TMP47:%.*]] = sext <4 x i16> [[PREDPHI17]] to <4 x i32>
5638; INTERLEAVE-NEXT:    [[TMP48]] = or <4 x i32> [[VEC_PHI]], [[TMP46]]
5639; INTERLEAVE-NEXT:    [[TMP49]] = or <4 x i32> [[VEC_PHI1]], [[TMP47]]
5640; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 8
5641; INTERLEAVE-NEXT:    [[VEC_IND_NEXT]] = add <4 x i16> [[VEC_IND]], <i16 8, i16 8, i16 8, i16 8>
5642; INTERLEAVE-NEXT:    [[TMP50:%.*]] = icmp eq i32 [[INDEX_NEXT]], 16
5643; INTERLEAVE-NEXT:    br i1 [[TMP50]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP48:![0-9]+]]
5644; INTERLEAVE:       middle.block:
5645; INTERLEAVE-NEXT:    [[BIN_RDX:%.*]] = or <4 x i32> [[TMP49]], [[TMP48]]
5646; INTERLEAVE-NEXT:    [[TMP51:%.*]] = call i32 @llvm.vector.reduce.or.v4i32(<4 x i32> [[BIN_RDX]])
5647; INTERLEAVE-NEXT:    br i1 false, label [[FOR_END:%.*]], label [[SCALAR_PH]]
5648; INTERLEAVE:       scalar.ph:
5649; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ -4, [[MIDDLE_BLOCK]] ], [ -20, [[ENTRY:%.*]] ]
5650; INTERLEAVE-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP51]], [[MIDDLE_BLOCK]] ], [ [[A]], [[ENTRY]] ]
5651; INTERLEAVE-NEXT:    br label [[FOR_BODY:%.*]]
5652; INTERLEAVE:       for.body:
5653; INTERLEAVE-NEXT:    [[I:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[FOR_INC:%.*]] ]
5654; INTERLEAVE-NEXT:    [[VAR0:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[VAR6:%.*]], [[FOR_INC]] ]
5655; INTERLEAVE-NEXT:    [[VAR1:%.*]] = trunc i32 [[I]] to i16
5656; INTERLEAVE-NEXT:    [[VAR2:%.*]] = icmp eq i16 [[VAR1]], 0
5657; INTERLEAVE-NEXT:    br i1 [[VAR2]], label [[FOR_INC]], label [[FOR_COND:%.*]]
5658; INTERLEAVE:       for.cond:
5659; INTERLEAVE-NEXT:    [[VAR3:%.*]] = urem i16 [[B]], [[VAR1]]
5660; INTERLEAVE-NEXT:    br label [[FOR_INC]]
5661; INTERLEAVE:       for.inc:
5662; INTERLEAVE-NEXT:    [[VAR4:%.*]] = phi i16 [ [[VAR3]], [[FOR_COND]] ], [ 0, [[FOR_BODY]] ]
5663; INTERLEAVE-NEXT:    [[VAR5:%.*]] = sext i16 [[VAR4]] to i32
5664; INTERLEAVE-NEXT:    [[VAR6]] = or i32 [[VAR0]], [[VAR5]]
5665; INTERLEAVE-NEXT:    [[I_NEXT]] = add nsw i32 [[I]], 1
5666; INTERLEAVE-NEXT:    [[COND:%.*]] = icmp eq i32 [[I_NEXT]], 0
5667; INTERLEAVE-NEXT:    br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP49:![0-9]+]]
5668; INTERLEAVE:       for.end:
5669; INTERLEAVE-NEXT:    [[VAR7:%.*]] = phi i32 [ [[VAR6]], [[FOR_INC]] ], [ [[TMP51]], [[MIDDLE_BLOCK]] ]
5670; INTERLEAVE-NEXT:    ret i32 [[VAR7]]
5671;
5672entry:
5673  br label %for.body
5674
5675for.body:
5676  %i = phi i32 [ -20, %entry ], [ %i.next, %for.inc ]
5677  %var0 = phi i32 [ %a, %entry ], [ %var6, %for.inc ]
5678  %var1 = trunc i32 %i to i16
5679  %var2 = icmp eq i16 %var1, 0
5680  br i1 %var2, label %for.inc, label %for.cond
5681
5682for.cond:
5683  %var3 = urem i16 %b, %var1
5684  br label %for.inc
5685
5686for.inc:
5687  %var4 = phi i16 [ %var3, %for.cond ], [ 0, %for.body ]
5688  %var5 = sext i16 %var4 to i32
5689  %var6 = or i32 %var0, %var5
5690  %i.next = add nsw i32 %i, 1
5691  %cond = icmp eq i32 %i.next, 0
5692  br i1 %cond, label %for.end, label %for.body
5693
5694for.end:
5695  %var7 = phi i32 [ %var6, %for.inc ]
5696  ret i32 %var7
5697}
5698
5699; Ensure that the shuffle vector for first order recurrence is inserted
5700; correctly after all the phis. These new phis correspond to new IVs
5701; that are generated by optimizing non-free truncs of IVs to IVs themselves.
5702; This also ensures the first-order recurrence splice recipe is placed
5703; correctly if it is fed by an induction.
5704define i64 @trunc_with_first_order_recurrence() {
5705; CHECK-LABEL: @trunc_with_first_order_recurrence(
5706; CHECK-NEXT:  entry:
5707; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
5708; CHECK:       vector.ph:
5709; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
5710; CHECK:       vector.body:
5711; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
5712; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP10:%.*]], [[VECTOR_BODY]] ]
5713; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 1, i32 2>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
5714; CHECK-NEXT:    [[VECTOR_RECUR:%.*]] = phi <2 x i32> [ <i32 poison, i32 42>, [[VECTOR_PH]] ], [ [[VEC_IND2:%.*]], [[VECTOR_BODY]] ]
5715; CHECK-NEXT:    [[VEC_IND2]] = phi <2 x i32> [ <i32 1, i32 2>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT3:%.*]], [[VECTOR_BODY]] ]
5716; CHECK-NEXT:    [[VEC_IND4:%.*]] = phi <2 x i32> [ <i32 1, i32 2>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT5:%.*]], [[VECTOR_BODY]] ]
5717; CHECK-NEXT:    [[TMP0:%.*]] = shufflevector <2 x i32> [[VECTOR_RECUR]], <2 x i32> [[VEC_IND2]], <2 x i32> <i32 1, i32 2>
5718; CHECK-NEXT:    [[TMP1:%.*]] = mul <2 x i32> [[VEC_IND]], [[VEC_IND2]]
5719; CHECK-NEXT:    [[TMP2:%.*]] = add <2 x i32> [[TMP1]], <i32 42, i32 42>
5720; CHECK-NEXT:    [[TMP3:%.*]] = add <2 x i32> [[TMP0]], [[VEC_IND2]]
5721; CHECK-NEXT:    [[TMP4:%.*]] = add <2 x i32> [[TMP3]], [[TMP2]]
5722; CHECK-NEXT:    [[TMP5:%.*]] = sext <2 x i32> [[TMP4]] to <2 x i64>
5723; CHECK-NEXT:    [[TMP6:%.*]] = add <2 x i64> [[VEC_PHI]], [[TMP5]]
5724; CHECK-NEXT:    [[TMP7:%.*]] = shl <2 x i32> [[VEC_IND4]], <i32 1, i32 1>
5725; CHECK-NEXT:    [[TMP8:%.*]] = add <2 x i32> [[TMP2]], [[TMP7]]
5726; CHECK-NEXT:    [[TMP9:%.*]] = sext <2 x i32> [[TMP8]] to <2 x i64>
5727; CHECK-NEXT:    [[TMP10]] = add <2 x i64> [[TMP6]], [[TMP9]]
5728; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
5729; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
5730; CHECK-NEXT:    [[VEC_IND_NEXT3]] = add <2 x i32> [[VEC_IND2]], <i32 2, i32 2>
5731; CHECK-NEXT:    [[VEC_IND_NEXT5]] = add <2 x i32> [[VEC_IND4]], <i32 2, i32 2>
5732; CHECK-NEXT:    [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], 112
5733; CHECK-NEXT:    br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP50:![0-9]+]]
5734; CHECK:       middle.block:
5735; CHECK-NEXT:    [[TMP12:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[TMP10]])
5736; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 113, 112
5737; CHECK-NEXT:    [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <2 x i32> [[VEC_IND2]], i32 1
5738; CHECK-NEXT:    [[VECTOR_RECUR_EXTRACT_FOR_PHI:%.*]] = extractelement <2 x i32> [[VEC_IND2]], i32 0
5739; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
5740; CHECK:       scalar.ph:
5741; CHECK-NEXT:    [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ 42, [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]
5742; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 113, [[MIDDLE_BLOCK]] ], [ 1, [[ENTRY]] ]
5743; CHECK-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ 113, [[MIDDLE_BLOCK]] ], [ 1, [[ENTRY]] ]
5744; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[TMP12]], [[MIDDLE_BLOCK]] ]
5745; CHECK-NEXT:    br label [[LOOP:%.*]]
5746; CHECK:       exit:
5747; CHECK-NEXT:    [[DOTLCSSA:%.*]] = phi i64 [ [[C23:%.*]], [[LOOP]] ], [ [[TMP12]], [[MIDDLE_BLOCK]] ]
5748; CHECK-NEXT:    ret i64 [[DOTLCSSA]]
5749; CHECK:       loop:
5750; CHECK-NEXT:    [[C5:%.*]] = phi i64 [ [[C23]], [[LOOP]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ]
5751; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
5752; CHECK-NEXT:    [[X:%.*]] = phi i32 [ [[C24:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]
5753; CHECK-NEXT:    [[SCALAR_RECUR:%.*]] = phi i32 [ [[C6:%.*]], [[LOOP]] ], [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ]
5754; CHECK-NEXT:    [[C6]] = trunc i64 [[INDVARS_IV]] to i32
5755; CHECK-NEXT:    [[C8:%.*]] = mul i32 [[X]], [[C6]]
5756; CHECK-NEXT:    [[C9:%.*]] = add i32 [[C8]], 42
5757; CHECK-NEXT:    [[C10:%.*]] = add i32 [[SCALAR_RECUR]], [[C6]]
5758; CHECK-NEXT:    [[C11:%.*]] = add i32 [[C10]], [[C9]]
5759; CHECK-NEXT:    [[C12:%.*]] = sext i32 [[C11]] to i64
5760; CHECK-NEXT:    [[C13:%.*]] = add i64 [[C5]], [[C12]]
5761; CHECK-NEXT:    [[INDVARS_IV_TR:%.*]] = trunc i64 [[INDVARS_IV]] to i32
5762; CHECK-NEXT:    [[C14:%.*]] = shl i32 [[INDVARS_IV_TR]], 1
5763; CHECK-NEXT:    [[C15:%.*]] = add i32 [[C9]], [[C14]]
5764; CHECK-NEXT:    [[C16:%.*]] = sext i32 [[C15]] to i64
5765; CHECK-NEXT:    [[C23]] = add i64 [[C13]], [[C16]]
5766; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
5767; CHECK-NEXT:    [[C24]] = add nuw nsw i32 [[X]], 1
5768; CHECK-NEXT:    [[EXITCOND_I:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 114
5769; CHECK-NEXT:    br i1 [[EXITCOND_I]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP51:![0-9]+]]
5770;
5771; IND-LABEL: @trunc_with_first_order_recurrence(
5772; IND-NEXT:  entry:
5773; IND-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
5774; IND:       vector.ph:
5775; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
5776; IND:       vector.body:
5777; IND-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
5778; IND-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP10:%.*]], [[VECTOR_BODY]] ]
5779; IND-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 1, i32 2>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
5780; IND-NEXT:    [[VECTOR_RECUR:%.*]] = phi <2 x i32> [ <i32 poison, i32 42>, [[VECTOR_PH]] ], [ [[VEC_IND2:%.*]], [[VECTOR_BODY]] ]
5781; IND-NEXT:    [[VEC_IND2]] = phi <2 x i32> [ <i32 1, i32 2>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT3:%.*]], [[VECTOR_BODY]] ]
5782; IND-NEXT:    [[VEC_IND4:%.*]] = phi <2 x i32> [ <i32 1, i32 2>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT5:%.*]], [[VECTOR_BODY]] ]
5783; IND-NEXT:    [[TMP0:%.*]] = shufflevector <2 x i32> [[VECTOR_RECUR]], <2 x i32> [[VEC_IND2]], <2 x i32> <i32 1, i32 2>
5784; IND-NEXT:    [[TMP1:%.*]] = mul <2 x i32> [[VEC_IND]], [[VEC_IND2]]
5785; IND-NEXT:    [[TMP2:%.*]] = add <2 x i32> [[TMP1]], <i32 42, i32 42>
5786; IND-NEXT:    [[TMP3:%.*]] = add <2 x i32> [[TMP0]], [[VEC_IND2]]
5787; IND-NEXT:    [[TMP4:%.*]] = add <2 x i32> [[TMP3]], [[TMP2]]
5788; IND-NEXT:    [[TMP5:%.*]] = sext <2 x i32> [[TMP4]] to <2 x i64>
5789; IND-NEXT:    [[TMP6:%.*]] = add <2 x i64> [[VEC_PHI]], [[TMP5]]
5790; IND-NEXT:    [[TMP7:%.*]] = shl <2 x i32> [[VEC_IND4]], <i32 1, i32 1>
5791; IND-NEXT:    [[TMP8:%.*]] = add <2 x i32> [[TMP2]], [[TMP7]]
5792; IND-NEXT:    [[TMP9:%.*]] = sext <2 x i32> [[TMP8]] to <2 x i64>
5793; IND-NEXT:    [[TMP10]] = add <2 x i64> [[TMP6]], [[TMP9]]
5794; IND-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
5795; IND-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
5796; IND-NEXT:    [[VEC_IND_NEXT3]] = add <2 x i32> [[VEC_IND2]], <i32 2, i32 2>
5797; IND-NEXT:    [[VEC_IND_NEXT5]] = add <2 x i32> [[VEC_IND4]], <i32 2, i32 2>
5798; IND-NEXT:    [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], 112
5799; IND-NEXT:    br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP50:![0-9]+]]
5800; IND:       middle.block:
5801; IND-NEXT:    [[TMP12:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[TMP10]])
5802; IND-NEXT:    [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <2 x i32> [[VEC_IND2]], i64 1
5803; IND-NEXT:    br i1 false, label [[EXIT:%.*]], label [[SCALAR_PH]]
5804; IND:       scalar.ph:
5805; IND-NEXT:    [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ 42, [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]
5806; IND-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 1, [[ENTRY]] ], [ 113, [[MIDDLE_BLOCK]] ]
5807; IND-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ 1, [[ENTRY]] ], [ 113, [[MIDDLE_BLOCK]] ]
5808; IND-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[TMP12]], [[MIDDLE_BLOCK]] ]
5809; IND-NEXT:    br label [[LOOP:%.*]]
5810; IND:       exit:
5811; IND-NEXT:    [[DOTLCSSA:%.*]] = phi i64 [ [[C23:%.*]], [[LOOP]] ], [ [[TMP12]], [[MIDDLE_BLOCK]] ]
5812; IND-NEXT:    ret i64 [[DOTLCSSA]]
5813; IND:       loop:
5814; IND-NEXT:    [[C5:%.*]] = phi i64 [ [[C23]], [[LOOP]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ]
5815; IND-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
5816; IND-NEXT:    [[X:%.*]] = phi i32 [ [[C24:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]
5817; IND-NEXT:    [[SCALAR_RECUR:%.*]] = phi i32 [ [[C6:%.*]], [[LOOP]] ], [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ]
5818; IND-NEXT:    [[C6]] = trunc i64 [[INDVARS_IV]] to i32
5819; IND-NEXT:    [[C8:%.*]] = mul i32 [[X]], [[C6]]
5820; IND-NEXT:    [[C9:%.*]] = add i32 [[C8]], 42
5821; IND-NEXT:    [[C10:%.*]] = add i32 [[SCALAR_RECUR]], [[C6]]
5822; IND-NEXT:    [[C11:%.*]] = add i32 [[C10]], [[C9]]
5823; IND-NEXT:    [[C12:%.*]] = sext i32 [[C11]] to i64
5824; IND-NEXT:    [[C13:%.*]] = add i64 [[C5]], [[C12]]
5825; IND-NEXT:    [[INDVARS_IV_TR:%.*]] = trunc i64 [[INDVARS_IV]] to i32
5826; IND-NEXT:    [[C14:%.*]] = shl i32 [[INDVARS_IV_TR]], 1
5827; IND-NEXT:    [[C15:%.*]] = add i32 [[C9]], [[C14]]
5828; IND-NEXT:    [[C16:%.*]] = sext i32 [[C15]] to i64
5829; IND-NEXT:    [[C23]] = add i64 [[C13]], [[C16]]
5830; IND-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
5831; IND-NEXT:    [[C24]] = add nuw nsw i32 [[X]], 1
5832; IND-NEXT:    [[EXITCOND_I:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 114
5833; IND-NEXT:    br i1 [[EXITCOND_I]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP51:![0-9]+]]
5834;
5835; UNROLL-LABEL: @trunc_with_first_order_recurrence(
5836; UNROLL-NEXT:  entry:
5837; UNROLL-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
5838; UNROLL:       vector.ph:
5839; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
5840; UNROLL:       vector.body:
5841; UNROLL-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
5842; UNROLL-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP20:%.*]], [[VECTOR_BODY]] ]
5843; UNROLL-NEXT:    [[VEC_PHI2:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP21:%.*]], [[VECTOR_BODY]] ]
5844; UNROLL-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 1, i32 2>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
5845; UNROLL-NEXT:    [[VECTOR_RECUR:%.*]] = phi <2 x i32> [ <i32 poison, i32 42>, [[VECTOR_PH]] ], [ [[STEP_ADD5:%.*]], [[VECTOR_BODY]] ]
5846; UNROLL-NEXT:    [[VEC_IND4:%.*]] = phi <2 x i32> [ <i32 1, i32 2>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT7:%.*]], [[VECTOR_BODY]] ]
5847; UNROLL-NEXT:    [[VEC_IND8:%.*]] = phi <2 x i32> [ <i32 1, i32 2>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT11:%.*]], [[VECTOR_BODY]] ]
5848; UNROLL-NEXT:    [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
5849; UNROLL-NEXT:    [[STEP_ADD5]] = add <2 x i32> [[VEC_IND4]], <i32 2, i32 2>
5850; UNROLL-NEXT:    [[TMP0:%.*]] = shufflevector <2 x i32> [[VECTOR_RECUR]], <2 x i32> [[VEC_IND4]], <2 x i32> <i32 1, i32 2>
5851; UNROLL-NEXT:    [[TMP1:%.*]] = shufflevector <2 x i32> [[VEC_IND4]], <2 x i32> [[STEP_ADD5]], <2 x i32> <i32 1, i32 2>
5852; UNROLL-NEXT:    [[TMP2:%.*]] = mul <2 x i32> [[VEC_IND]], [[VEC_IND4]]
5853; UNROLL-NEXT:    [[TMP3:%.*]] = mul <2 x i32> [[STEP_ADD]], [[STEP_ADD5]]
5854; UNROLL-NEXT:    [[TMP4:%.*]] = add <2 x i32> [[TMP2]], <i32 42, i32 42>
5855; UNROLL-NEXT:    [[TMP5:%.*]] = add <2 x i32> [[TMP3]], <i32 42, i32 42>
5856; UNROLL-NEXT:    [[TMP6:%.*]] = add <2 x i32> [[TMP0]], [[VEC_IND4]]
5857; UNROLL-NEXT:    [[TMP7:%.*]] = add <2 x i32> [[TMP1]], [[STEP_ADD5]]
5858; UNROLL-NEXT:    [[TMP8:%.*]] = add <2 x i32> [[TMP6]], [[TMP4]]
5859; UNROLL-NEXT:    [[TMP9:%.*]] = add <2 x i32> [[TMP7]], [[TMP5]]
5860; UNROLL-NEXT:    [[TMP10:%.*]] = sext <2 x i32> [[TMP8]] to <2 x i64>
5861; UNROLL-NEXT:    [[TMP11:%.*]] = sext <2 x i32> [[TMP9]] to <2 x i64>
5862; UNROLL-NEXT:    [[TMP12:%.*]] = add <2 x i64> [[VEC_PHI]], [[TMP10]]
5863; UNROLL-NEXT:    [[TMP13:%.*]] = add <2 x i64> [[VEC_PHI2]], [[TMP11]]
5864; UNROLL-NEXT:    [[TMP14:%.*]] = shl <2 x i32> [[VEC_IND8]], <i32 1, i32 1>
5865; UNROLL-NEXT:    [[STEP_ADD9:%.*]] = shl <2 x i32> [[VEC_IND8]], <i32 1, i32 1>
5866; UNROLL-NEXT:    [[TMP15:%.*]] = add <2 x i32> [[STEP_ADD9]], <i32 4, i32 4>
5867; UNROLL-NEXT:    [[TMP16:%.*]] = add <2 x i32> [[TMP4]], [[TMP14]]
5868; UNROLL-NEXT:    [[TMP17:%.*]] = add <2 x i32> [[TMP5]], [[TMP15]]
5869; UNROLL-NEXT:    [[TMP18:%.*]] = sext <2 x i32> [[TMP16]] to <2 x i64>
5870; UNROLL-NEXT:    [[TMP19:%.*]] = sext <2 x i32> [[TMP17]] to <2 x i64>
5871; UNROLL-NEXT:    [[TMP20]] = add <2 x i64> [[TMP12]], [[TMP18]]
5872; UNROLL-NEXT:    [[TMP21]] = add <2 x i64> [[TMP13]], [[TMP19]]
5873; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
5874; UNROLL-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 4, i32 4>
5875; UNROLL-NEXT:    [[VEC_IND_NEXT7]] = add <2 x i32> [[VEC_IND4]], <i32 4, i32 4>
5876; UNROLL-NEXT:    [[VEC_IND_NEXT11]] = add <2 x i32> [[VEC_IND8]], <i32 4, i32 4>
5877; UNROLL-NEXT:    [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], 112
5878; UNROLL-NEXT:    br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP50:![0-9]+]]
5879; UNROLL:       middle.block:
5880; UNROLL-NEXT:    [[BIN_RDX:%.*]] = add <2 x i64> [[TMP21]], [[TMP20]]
5881; UNROLL-NEXT:    [[TMP23:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[BIN_RDX]])
5882; UNROLL-NEXT:    [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <2 x i32> [[STEP_ADD5]], i64 1
5883; UNROLL-NEXT:    br i1 false, label [[EXIT:%.*]], label [[SCALAR_PH]]
5884; UNROLL:       scalar.ph:
5885; UNROLL-NEXT:    [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ 42, [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]
5886; UNROLL-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 1, [[ENTRY]] ], [ 113, [[MIDDLE_BLOCK]] ]
5887; UNROLL-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ 1, [[ENTRY]] ], [ 113, [[MIDDLE_BLOCK]] ]
5888; UNROLL-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[TMP23]], [[MIDDLE_BLOCK]] ]
5889; UNROLL-NEXT:    br label [[LOOP:%.*]]
5890; UNROLL:       exit:
5891; UNROLL-NEXT:    [[DOTLCSSA:%.*]] = phi i64 [ [[C23:%.*]], [[LOOP]] ], [ [[TMP23]], [[MIDDLE_BLOCK]] ]
5892; UNROLL-NEXT:    ret i64 [[DOTLCSSA]]
5893; UNROLL:       loop:
5894; UNROLL-NEXT:    [[C5:%.*]] = phi i64 [ [[C23]], [[LOOP]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ]
5895; UNROLL-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
5896; UNROLL-NEXT:    [[X:%.*]] = phi i32 [ [[C24:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]
5897; UNROLL-NEXT:    [[SCALAR_RECUR:%.*]] = phi i32 [ [[C6:%.*]], [[LOOP]] ], [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ]
5898; UNROLL-NEXT:    [[C6]] = trunc i64 [[INDVARS_IV]] to i32
5899; UNROLL-NEXT:    [[C8:%.*]] = mul i32 [[X]], [[C6]]
5900; UNROLL-NEXT:    [[C9:%.*]] = add i32 [[C8]], 42
5901; UNROLL-NEXT:    [[C10:%.*]] = add i32 [[SCALAR_RECUR]], [[C6]]
5902; UNROLL-NEXT:    [[C11:%.*]] = add i32 [[C10]], [[C9]]
5903; UNROLL-NEXT:    [[C12:%.*]] = sext i32 [[C11]] to i64
5904; UNROLL-NEXT:    [[C13:%.*]] = add i64 [[C5]], [[C12]]
5905; UNROLL-NEXT:    [[INDVARS_IV_TR:%.*]] = trunc i64 [[INDVARS_IV]] to i32
5906; UNROLL-NEXT:    [[C14:%.*]] = shl i32 [[INDVARS_IV_TR]], 1
5907; UNROLL-NEXT:    [[C15:%.*]] = add i32 [[C9]], [[C14]]
5908; UNROLL-NEXT:    [[C16:%.*]] = sext i32 [[C15]] to i64
5909; UNROLL-NEXT:    [[C23]] = add i64 [[C13]], [[C16]]
5910; UNROLL-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
5911; UNROLL-NEXT:    [[C24]] = add nuw nsw i32 [[X]], 1
5912; UNROLL-NEXT:    [[EXITCOND_I:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 114
5913; UNROLL-NEXT:    br i1 [[EXITCOND_I]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP51:![0-9]+]]
5914;
5915; UNROLL-NO-IC-LABEL: @trunc_with_first_order_recurrence(
5916; UNROLL-NO-IC-NEXT:  entry:
5917; UNROLL-NO-IC-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
5918; UNROLL-NO-IC:       vector.ph:
5919; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
5920; UNROLL-NO-IC:       vector.body:
5921; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
5922; UNROLL-NO-IC-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP20:%.*]], [[VECTOR_BODY]] ]
5923; UNROLL-NO-IC-NEXT:    [[VEC_PHI2:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP21:%.*]], [[VECTOR_BODY]] ]
5924; UNROLL-NO-IC-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 1, i32 2>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
5925; UNROLL-NO-IC-NEXT:    [[VECTOR_RECUR:%.*]] = phi <2 x i32> [ <i32 poison, i32 42>, [[VECTOR_PH]] ], [ [[STEP_ADD5:%.*]], [[VECTOR_BODY]] ]
5926; UNROLL-NO-IC-NEXT:    [[VEC_IND4:%.*]] = phi <2 x i32> [ <i32 1, i32 2>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT7:%.*]], [[VECTOR_BODY]] ]
5927; UNROLL-NO-IC-NEXT:    [[VEC_IND8:%.*]] = phi <2 x i32> [ <i32 1, i32 2>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT11:%.*]], [[VECTOR_BODY]] ]
5928; UNROLL-NO-IC-NEXT:    [[STEP_ADD:%.*]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
5929; UNROLL-NO-IC-NEXT:    [[STEP_ADD5]] = add <2 x i32> [[VEC_IND4]], <i32 2, i32 2>
5930; UNROLL-NO-IC-NEXT:    [[STEP_ADD9:%.*]] = add <2 x i32> [[VEC_IND8]], <i32 2, i32 2>
5931; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = shufflevector <2 x i32> [[VECTOR_RECUR]], <2 x i32> [[VEC_IND4]], <2 x i32> <i32 1, i32 2>
5932; UNROLL-NO-IC-NEXT:    [[TMP1:%.*]] = shufflevector <2 x i32> [[VEC_IND4]], <2 x i32> [[STEP_ADD5]], <2 x i32> <i32 1, i32 2>
5933; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = mul <2 x i32> [[VEC_IND]], [[VEC_IND4]]
5934; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = mul <2 x i32> [[STEP_ADD]], [[STEP_ADD5]]
5935; UNROLL-NO-IC-NEXT:    [[TMP4:%.*]] = add <2 x i32> [[TMP2]], <i32 42, i32 42>
5936; UNROLL-NO-IC-NEXT:    [[TMP5:%.*]] = add <2 x i32> [[TMP3]], <i32 42, i32 42>
5937; UNROLL-NO-IC-NEXT:    [[TMP6:%.*]] = add <2 x i32> [[TMP0]], [[VEC_IND4]]
5938; UNROLL-NO-IC-NEXT:    [[TMP7:%.*]] = add <2 x i32> [[TMP1]], [[STEP_ADD5]]
5939; UNROLL-NO-IC-NEXT:    [[TMP8:%.*]] = add <2 x i32> [[TMP6]], [[TMP4]]
5940; UNROLL-NO-IC-NEXT:    [[TMP9:%.*]] = add <2 x i32> [[TMP7]], [[TMP5]]
5941; UNROLL-NO-IC-NEXT:    [[TMP10:%.*]] = sext <2 x i32> [[TMP8]] to <2 x i64>
5942; UNROLL-NO-IC-NEXT:    [[TMP11:%.*]] = sext <2 x i32> [[TMP9]] to <2 x i64>
5943; UNROLL-NO-IC-NEXT:    [[TMP12:%.*]] = add <2 x i64> [[VEC_PHI]], [[TMP10]]
5944; UNROLL-NO-IC-NEXT:    [[TMP13:%.*]] = add <2 x i64> [[VEC_PHI2]], [[TMP11]]
5945; UNROLL-NO-IC-NEXT:    [[TMP14:%.*]] = shl <2 x i32> [[VEC_IND8]], <i32 1, i32 1>
5946; UNROLL-NO-IC-NEXT:    [[TMP15:%.*]] = shl <2 x i32> [[STEP_ADD9]], <i32 1, i32 1>
5947; UNROLL-NO-IC-NEXT:    [[TMP16:%.*]] = add <2 x i32> [[TMP4]], [[TMP14]]
5948; UNROLL-NO-IC-NEXT:    [[TMP17:%.*]] = add <2 x i32> [[TMP5]], [[TMP15]]
5949; UNROLL-NO-IC-NEXT:    [[TMP18:%.*]] = sext <2 x i32> [[TMP16]] to <2 x i64>
5950; UNROLL-NO-IC-NEXT:    [[TMP19:%.*]] = sext <2 x i32> [[TMP17]] to <2 x i64>
5951; UNROLL-NO-IC-NEXT:    [[TMP20]] = add <2 x i64> [[TMP12]], [[TMP18]]
5952; UNROLL-NO-IC-NEXT:    [[TMP21]] = add <2 x i64> [[TMP13]], [[TMP19]]
5953; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
5954; UNROLL-NO-IC-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[STEP_ADD]], <i32 2, i32 2>
5955; UNROLL-NO-IC-NEXT:    [[VEC_IND_NEXT7]] = add <2 x i32> [[STEP_ADD5]], <i32 2, i32 2>
5956; UNROLL-NO-IC-NEXT:    [[VEC_IND_NEXT11]] = add <2 x i32> [[STEP_ADD9]], <i32 2, i32 2>
5957; UNROLL-NO-IC-NEXT:    [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], 112
5958; UNROLL-NO-IC-NEXT:    br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP50:![0-9]+]]
5959; UNROLL-NO-IC:       middle.block:
5960; UNROLL-NO-IC-NEXT:    [[BIN_RDX:%.*]] = add <2 x i64> [[TMP21]], [[TMP20]]
5961; UNROLL-NO-IC-NEXT:    [[TMP23:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[BIN_RDX]])
5962; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i64 113, 112
5963; UNROLL-NO-IC-NEXT:    [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <2 x i32> [[STEP_ADD5]], i32 1
5964; UNROLL-NO-IC-NEXT:    [[VECTOR_RECUR_EXTRACT_FOR_PHI:%.*]] = extractelement <2 x i32> [[STEP_ADD5]], i32 0
5965; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
5966; UNROLL-NO-IC:       scalar.ph:
5967; UNROLL-NO-IC-NEXT:    [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ 42, [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]
5968; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 113, [[MIDDLE_BLOCK]] ], [ 1, [[ENTRY]] ]
5969; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ 113, [[MIDDLE_BLOCK]] ], [ 1, [[ENTRY]] ]
5970; UNROLL-NO-IC-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[TMP23]], [[MIDDLE_BLOCK]] ]
5971; UNROLL-NO-IC-NEXT:    br label [[LOOP:%.*]]
5972; UNROLL-NO-IC:       exit:
5973; UNROLL-NO-IC-NEXT:    [[DOTLCSSA:%.*]] = phi i64 [ [[C23:%.*]], [[LOOP]] ], [ [[TMP23]], [[MIDDLE_BLOCK]] ]
5974; UNROLL-NO-IC-NEXT:    ret i64 [[DOTLCSSA]]
5975; UNROLL-NO-IC:       loop:
5976; UNROLL-NO-IC-NEXT:    [[C5:%.*]] = phi i64 [ [[C23]], [[LOOP]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ]
5977; UNROLL-NO-IC-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
5978; UNROLL-NO-IC-NEXT:    [[X:%.*]] = phi i32 [ [[C24:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]
5979; UNROLL-NO-IC-NEXT:    [[SCALAR_RECUR:%.*]] = phi i32 [ [[C6:%.*]], [[LOOP]] ], [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ]
5980; UNROLL-NO-IC-NEXT:    [[C6]] = trunc i64 [[INDVARS_IV]] to i32
5981; UNROLL-NO-IC-NEXT:    [[C8:%.*]] = mul i32 [[X]], [[C6]]
5982; UNROLL-NO-IC-NEXT:    [[C9:%.*]] = add i32 [[C8]], 42
5983; UNROLL-NO-IC-NEXT:    [[C10:%.*]] = add i32 [[SCALAR_RECUR]], [[C6]]
5984; UNROLL-NO-IC-NEXT:    [[C11:%.*]] = add i32 [[C10]], [[C9]]
5985; UNROLL-NO-IC-NEXT:    [[C12:%.*]] = sext i32 [[C11]] to i64
5986; UNROLL-NO-IC-NEXT:    [[C13:%.*]] = add i64 [[C5]], [[C12]]
5987; UNROLL-NO-IC-NEXT:    [[INDVARS_IV_TR:%.*]] = trunc i64 [[INDVARS_IV]] to i32
5988; UNROLL-NO-IC-NEXT:    [[C14:%.*]] = shl i32 [[INDVARS_IV_TR]], 1
5989; UNROLL-NO-IC-NEXT:    [[C15:%.*]] = add i32 [[C9]], [[C14]]
5990; UNROLL-NO-IC-NEXT:    [[C16:%.*]] = sext i32 [[C15]] to i64
5991; UNROLL-NO-IC-NEXT:    [[C23]] = add i64 [[C13]], [[C16]]
5992; UNROLL-NO-IC-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
5993; UNROLL-NO-IC-NEXT:    [[C24]] = add nuw nsw i32 [[X]], 1
5994; UNROLL-NO-IC-NEXT:    [[EXITCOND_I:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 114
5995; UNROLL-NO-IC-NEXT:    br i1 [[EXITCOND_I]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP51:![0-9]+]]
5996;
5997; INTERLEAVE-LABEL: @trunc_with_first_order_recurrence(
5998; INTERLEAVE-NEXT:  entry:
5999; INTERLEAVE-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
6000; INTERLEAVE:       vector.ph:
6001; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
6002; INTERLEAVE:       vector.body:
6003; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
6004; INTERLEAVE-NEXT:    [[VEC_PHI:%.*]] = phi <4 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP20:%.*]], [[VECTOR_BODY]] ]
6005; INTERLEAVE-NEXT:    [[VEC_PHI2:%.*]] = phi <4 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP21:%.*]], [[VECTOR_BODY]] ]
6006; INTERLEAVE-NEXT:    [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 1, i32 2, i32 3, i32 4>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
6007; INTERLEAVE-NEXT:    [[VECTOR_RECUR:%.*]] = phi <4 x i32> [ <i32 poison, i32 poison, i32 poison, i32 42>, [[VECTOR_PH]] ], [ [[STEP_ADD5:%.*]], [[VECTOR_BODY]] ]
6008; INTERLEAVE-NEXT:    [[VEC_IND4:%.*]] = phi <4 x i32> [ <i32 1, i32 2, i32 3, i32 4>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT7:%.*]], [[VECTOR_BODY]] ]
6009; INTERLEAVE-NEXT:    [[VEC_IND8:%.*]] = phi <4 x i32> [ <i32 1, i32 2, i32 3, i32 4>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT11:%.*]], [[VECTOR_BODY]] ]
6010; INTERLEAVE-NEXT:    [[STEP_ADD:%.*]] = add <4 x i32> [[VEC_IND]], <i32 4, i32 4, i32 4, i32 4>
6011; INTERLEAVE-NEXT:    [[STEP_ADD5]] = add <4 x i32> [[VEC_IND4]], <i32 4, i32 4, i32 4, i32 4>
6012; INTERLEAVE-NEXT:    [[TMP0:%.*]] = shufflevector <4 x i32> [[VECTOR_RECUR]], <4 x i32> [[VEC_IND4]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
6013; INTERLEAVE-NEXT:    [[TMP1:%.*]] = shufflevector <4 x i32> [[VEC_IND4]], <4 x i32> [[STEP_ADD5]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
6014; INTERLEAVE-NEXT:    [[TMP2:%.*]] = mul <4 x i32> [[VEC_IND]], [[VEC_IND4]]
6015; INTERLEAVE-NEXT:    [[TMP3:%.*]] = mul <4 x i32> [[STEP_ADD]], [[STEP_ADD5]]
6016; INTERLEAVE-NEXT:    [[TMP4:%.*]] = add <4 x i32> [[TMP2]], <i32 42, i32 42, i32 42, i32 42>
6017; INTERLEAVE-NEXT:    [[TMP5:%.*]] = add <4 x i32> [[TMP3]], <i32 42, i32 42, i32 42, i32 42>
6018; INTERLEAVE-NEXT:    [[TMP6:%.*]] = add <4 x i32> [[TMP0]], [[VEC_IND4]]
6019; INTERLEAVE-NEXT:    [[TMP7:%.*]] = add <4 x i32> [[TMP1]], [[STEP_ADD5]]
6020; INTERLEAVE-NEXT:    [[TMP8:%.*]] = add <4 x i32> [[TMP6]], [[TMP4]]
6021; INTERLEAVE-NEXT:    [[TMP9:%.*]] = add <4 x i32> [[TMP7]], [[TMP5]]
6022; INTERLEAVE-NEXT:    [[TMP10:%.*]] = sext <4 x i32> [[TMP8]] to <4 x i64>
6023; INTERLEAVE-NEXT:    [[TMP11:%.*]] = sext <4 x i32> [[TMP9]] to <4 x i64>
6024; INTERLEAVE-NEXT:    [[TMP12:%.*]] = add <4 x i64> [[VEC_PHI]], [[TMP10]]
6025; INTERLEAVE-NEXT:    [[TMP13:%.*]] = add <4 x i64> [[VEC_PHI2]], [[TMP11]]
6026; INTERLEAVE-NEXT:    [[TMP14:%.*]] = shl <4 x i32> [[VEC_IND8]], <i32 1, i32 1, i32 1, i32 1>
6027; INTERLEAVE-NEXT:    [[STEP_ADD9:%.*]] = shl <4 x i32> [[VEC_IND8]], <i32 1, i32 1, i32 1, i32 1>
6028; INTERLEAVE-NEXT:    [[TMP15:%.*]] = add <4 x i32> [[STEP_ADD9]], <i32 8, i32 8, i32 8, i32 8>
6029; INTERLEAVE-NEXT:    [[TMP16:%.*]] = add <4 x i32> [[TMP4]], [[TMP14]]
6030; INTERLEAVE-NEXT:    [[TMP17:%.*]] = add <4 x i32> [[TMP5]], [[TMP15]]
6031; INTERLEAVE-NEXT:    [[TMP18:%.*]] = sext <4 x i32> [[TMP16]] to <4 x i64>
6032; INTERLEAVE-NEXT:    [[TMP19:%.*]] = sext <4 x i32> [[TMP17]] to <4 x i64>
6033; INTERLEAVE-NEXT:    [[TMP20]] = add <4 x i64> [[TMP12]], [[TMP18]]
6034; INTERLEAVE-NEXT:    [[TMP21]] = add <4 x i64> [[TMP13]], [[TMP19]]
6035; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
6036; INTERLEAVE-NEXT:    [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 8, i32 8, i32 8, i32 8>
6037; INTERLEAVE-NEXT:    [[VEC_IND_NEXT7]] = add <4 x i32> [[VEC_IND4]], <i32 8, i32 8, i32 8, i32 8>
6038; INTERLEAVE-NEXT:    [[VEC_IND_NEXT11]] = add <4 x i32> [[VEC_IND8]], <i32 8, i32 8, i32 8, i32 8>
6039; INTERLEAVE-NEXT:    [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], 112
6040; INTERLEAVE-NEXT:    br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP50:![0-9]+]]
6041; INTERLEAVE:       middle.block:
6042; INTERLEAVE-NEXT:    [[BIN_RDX:%.*]] = add <4 x i64> [[TMP21]], [[TMP20]]
6043; INTERLEAVE-NEXT:    [[TMP23:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[BIN_RDX]])
6044; INTERLEAVE-NEXT:    [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <4 x i32> [[STEP_ADD5]], i64 3
6045; INTERLEAVE-NEXT:    br i1 false, label [[EXIT:%.*]], label [[SCALAR_PH]]
6046; INTERLEAVE:       scalar.ph:
6047; INTERLEAVE-NEXT:    [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ 42, [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]
6048; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 1, [[ENTRY]] ], [ 113, [[MIDDLE_BLOCK]] ]
6049; INTERLEAVE-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ 1, [[ENTRY]] ], [ 113, [[MIDDLE_BLOCK]] ]
6050; INTERLEAVE-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[TMP23]], [[MIDDLE_BLOCK]] ]
6051; INTERLEAVE-NEXT:    br label [[LOOP:%.*]]
6052; INTERLEAVE:       exit:
6053; INTERLEAVE-NEXT:    [[DOTLCSSA:%.*]] = phi i64 [ [[C23:%.*]], [[LOOP]] ], [ [[TMP23]], [[MIDDLE_BLOCK]] ]
6054; INTERLEAVE-NEXT:    ret i64 [[DOTLCSSA]]
6055; INTERLEAVE:       loop:
6056; INTERLEAVE-NEXT:    [[C5:%.*]] = phi i64 [ [[C23]], [[LOOP]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ]
6057; INTERLEAVE-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
6058; INTERLEAVE-NEXT:    [[X:%.*]] = phi i32 [ [[C24:%.*]], [[LOOP]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]
6059; INTERLEAVE-NEXT:    [[SCALAR_RECUR:%.*]] = phi i32 [ [[C6:%.*]], [[LOOP]] ], [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ]
6060; INTERLEAVE-NEXT:    [[C6]] = trunc i64 [[INDVARS_IV]] to i32
6061; INTERLEAVE-NEXT:    [[C8:%.*]] = mul i32 [[X]], [[C6]]
6062; INTERLEAVE-NEXT:    [[C9:%.*]] = add i32 [[C8]], 42
6063; INTERLEAVE-NEXT:    [[C10:%.*]] = add i32 [[SCALAR_RECUR]], [[C6]]
6064; INTERLEAVE-NEXT:    [[C11:%.*]] = add i32 [[C10]], [[C9]]
6065; INTERLEAVE-NEXT:    [[C12:%.*]] = sext i32 [[C11]] to i64
6066; INTERLEAVE-NEXT:    [[C13:%.*]] = add i64 [[C5]], [[C12]]
6067; INTERLEAVE-NEXT:    [[INDVARS_IV_TR:%.*]] = trunc i64 [[INDVARS_IV]] to i32
6068; INTERLEAVE-NEXT:    [[C14:%.*]] = shl i32 [[INDVARS_IV_TR]], 1
6069; INTERLEAVE-NEXT:    [[C15:%.*]] = add i32 [[C9]], [[C14]]
6070; INTERLEAVE-NEXT:    [[C16:%.*]] = sext i32 [[C15]] to i64
6071; INTERLEAVE-NEXT:    [[C23]] = add i64 [[C13]], [[C16]]
6072; INTERLEAVE-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
6073; INTERLEAVE-NEXT:    [[C24]] = add nuw nsw i32 [[X]], 1
6074; INTERLEAVE-NEXT:    [[EXITCOND_I:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 114
6075; INTERLEAVE-NEXT:    br i1 [[EXITCOND_I]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP51:![0-9]+]]
6076;
6077entry:
6078  br label %loop
6079
6080exit:                                        ; preds = %loop
6081  %.lcssa = phi i64 [ %c23, %loop ]
6082  ret i64 %.lcssa
6083
6084loop:                                         ; preds = %loop, %entry
6085  %c5 = phi i64 [ %c23, %loop ], [ 0, %entry ]
6086  %indvars.iv = phi i64 [ %indvars.iv.next, %loop ], [ 1, %entry ]
6087  %x = phi i32 [ %c24, %loop ], [ 1, %entry ]
6088  %y = phi i32 [ %c6, %loop ], [ 42, %entry ]
6089  %c6 = trunc i64 %indvars.iv to i32
6090  %c8 = mul i32 %x, %c6
6091  %c9 = add i32 %c8, 42
6092  %c10 = add i32 %y, %c6
6093  %c11 = add i32 %c10, %c9
6094  %c12 = sext i32 %c11 to i64
6095  %c13 = add i64 %c5, %c12
6096  %indvars.iv.tr = trunc i64 %indvars.iv to i32
6097  %c14 = shl i32 %indvars.iv.tr, 1
6098  %c15 = add i32 %c9, %c14
6099  %c16 = sext i32 %c15 to i64
6100  %c23 = add i64 %c13, %c16
6101  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
6102  %c24 = add nuw nsw i32 %x, 1
6103  %exitcond.i = icmp eq i64 %indvars.iv.next, 114
6104  br i1 %exitcond.i, label %exit, label %loop
6105
6106}
6107
6108; Test case for PR52460.
6109define void @pr52460_first_order_recurrence_truncated_iv(i32* noalias %src, i32* %dst) {
6110;
6111; CHECK-LABEL: @pr52460_first_order_recurrence_truncated_iv(
6112; CHECK-NEXT:  entry:
6113; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
6114; CHECK:       vector.ph:
6115; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
6116; CHECK:       vector.body:
6117; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
6118; CHECK-NEXT:    [[VECTOR_RECUR:%.*]] = phi <2 x i32> [ <i32 poison, i32 0>, [[VECTOR_PH]] ], [ [[VEC_IND:%.*]], [[VECTOR_BODY]] ]
6119; CHECK-NEXT:    [[VEC_IND]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
6120; CHECK-NEXT:    [[TMP0:%.*]] = trunc i64 [[INDEX]] to i32
6121; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[TMP0]], 0
6122; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <2 x i32> [[VECTOR_RECUR]], <2 x i32> [[VEC_IND]], <2 x i32> <i32 1, i32 2>
6123; CHECK-NEXT:    [[TMP3:%.*]] = load i32, i32* [[SRC:%.*]], align 4
6124; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[TMP3]], i32 0
6125; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
6126; CHECK-NEXT:    [[TMP4:%.*]] = mul nsw <2 x i32> [[BROADCAST_SPLAT]], [[TMP2]]
6127; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr i32, i32* [[DST:%.*]], i32 [[TMP1]]
6128; CHECK-NEXT:    [[TMP6:%.*]] = add <2 x i32> [[VEC_IND]], [[TMP4]]
6129; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr i32, i32* [[TMP5]], i32 0
6130; CHECK-NEXT:    [[TMP8:%.*]] = bitcast i32* [[TMP7]] to <2 x i32>*
6131; CHECK-NEXT:    store <2 x i32> [[TMP6]], <2 x i32>* [[TMP8]], align 4
6132; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
6133; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
6134; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], 100
6135; CHECK-NEXT:    br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP52:![0-9]+]]
6136; CHECK:       middle.block:
6137; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 100, 100
6138; CHECK-NEXT:    [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <2 x i32> [[VEC_IND]], i32 1
6139; CHECK-NEXT:    [[VECTOR_RECUR_EXTRACT_FOR_PHI:%.*]] = extractelement <2 x i32> [[VEC_IND]], i32 0
6140; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
6141; CHECK:       scalar.ph:
6142; CHECK-NEXT:    [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]
6143; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 100, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
6144; CHECK-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ 100, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
6145; CHECK-NEXT:    br label [[LOOP:%.*]]
6146; CHECK:       loop:
6147; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
6148; CHECK-NEXT:    [[TRUNC_IV:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[TRUNC_IV_NEXT:%.*]], [[LOOP]] ]
6149; CHECK-NEXT:    [[SCALAR_RECUR:%.*]] = phi i32 [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[IV_TRUNC:%.*]], [[LOOP]] ]
6150; CHECK-NEXT:    [[LV:%.*]] = load i32, i32* [[SRC]], align 4
6151; CHECK-NEXT:    [[MUL:%.*]] = mul nsw i32 [[LV]], [[SCALAR_RECUR]]
6152; CHECK-NEXT:    [[TRUNC_IV_NEXT]] = add i32 [[TRUNC_IV]], 1
6153; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
6154; CHECK-NEXT:    [[IV_TRUNC]] = trunc i64 [[IV]] to i32
6155; CHECK-NEXT:    [[DST_GEP:%.*]] = getelementptr i32, i32* [[DST]], i32 [[IV_TRUNC]]
6156; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[IV_TRUNC]], [[MUL]]
6157; CHECK-NEXT:    store i32 [[ADD]], i32* [[DST_GEP]], align 4
6158; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[TRUNC_IV_NEXT]], 100
6159; CHECK-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP53:![0-9]+]]
6160; CHECK:       exit:
6161; CHECK-NEXT:    ret void
6162;
6163; IND-LABEL: @pr52460_first_order_recurrence_truncated_iv(
6164; IND-NEXT:  entry:
6165; IND-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
6166; IND:       vector.ph:
6167; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
6168; IND:       vector.body:
6169; IND-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
6170; IND-NEXT:    [[VECTOR_RECUR:%.*]] = phi <2 x i32> [ <i32 poison, i32 0>, [[VECTOR_PH]] ], [ [[VEC_IND:%.*]], [[VECTOR_BODY]] ]
6171; IND-NEXT:    [[VEC_IND]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
6172; IND-NEXT:    [[TMP0:%.*]] = shufflevector <2 x i32> [[VECTOR_RECUR]], <2 x i32> [[VEC_IND]], <2 x i32> <i32 1, i32 2>
6173; IND-NEXT:    [[TMP1:%.*]] = load i32, i32* [[SRC:%.*]], align 4
6174; IND-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[TMP1]], i64 0
6175; IND-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
6176; IND-NEXT:    [[TMP2:%.*]] = mul nsw <2 x i32> [[BROADCAST_SPLAT]], [[TMP0]]
6177; IND-NEXT:    [[SEXT:%.*]] = shl i64 [[INDEX]], 32
6178; IND-NEXT:    [[TMP3:%.*]] = ashr exact i64 [[SEXT]], 32
6179; IND-NEXT:    [[TMP4:%.*]] = getelementptr i32, i32* [[DST:%.*]], i64 [[TMP3]]
6180; IND-NEXT:    [[TMP5:%.*]] = add <2 x i32> [[VEC_IND]], [[TMP2]]
6181; IND-NEXT:    [[TMP6:%.*]] = bitcast i32* [[TMP4]] to <2 x i32>*
6182; IND-NEXT:    store <2 x i32> [[TMP5]], <2 x i32>* [[TMP6]], align 4
6183; IND-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
6184; IND-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
6185; IND-NEXT:    [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], 100
6186; IND-NEXT:    br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP52:![0-9]+]]
6187; IND:       middle.block:
6188; IND-NEXT:    br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
6189; IND:       scalar.ph:
6190; IND-NEXT:    br label [[LOOP:%.*]]
6191; IND:       loop:
6192; IND-NEXT:    br i1 poison, label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP53:![0-9]+]]
6193; IND:       exit:
6194; IND-NEXT:    ret void
6195;
6196; UNROLL-LABEL: @pr52460_first_order_recurrence_truncated_iv(
6197; UNROLL-NEXT:  entry:
6198; UNROLL-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
6199; UNROLL:       vector.ph:
6200; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
6201; UNROLL:       vector.body:
6202; UNROLL-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
6203; UNROLL-NEXT:    [[VECTOR_RECUR:%.*]] = phi <2 x i32> [ <i32 poison, i32 0>, [[VECTOR_PH]] ], [ [[STEP_ADD:%.*]], [[VECTOR_BODY]] ]
6204; UNROLL-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
6205; UNROLL-NEXT:    [[STEP_ADD]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
6206; UNROLL-NEXT:    [[TMP0:%.*]] = shufflevector <2 x i32> [[VECTOR_RECUR]], <2 x i32> [[VEC_IND]], <2 x i32> <i32 1, i32 2>
6207; UNROLL-NEXT:    [[TMP1:%.*]] = shufflevector <2 x i32> [[VEC_IND]], <2 x i32> [[STEP_ADD]], <2 x i32> <i32 1, i32 2>
6208; UNROLL-NEXT:    [[TMP2:%.*]] = load i32, i32* [[SRC:%.*]], align 4
6209; UNROLL-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[TMP2]], i64 0
6210; UNROLL-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
6211; UNROLL-NEXT:    [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <2 x i32> poison, i32 [[TMP2]], i64 0
6212; UNROLL-NEXT:    [[BROADCAST_SPLAT4:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT3]], <2 x i32> poison, <2 x i32> zeroinitializer
6213; UNROLL-NEXT:    [[TMP3:%.*]] = mul nsw <2 x i32> [[BROADCAST_SPLAT]], [[TMP0]]
6214; UNROLL-NEXT:    [[TMP4:%.*]] = mul nsw <2 x i32> [[BROADCAST_SPLAT4]], [[TMP1]]
6215; UNROLL-NEXT:    [[SEXT:%.*]] = shl i64 [[INDEX]], 32
6216; UNROLL-NEXT:    [[TMP5:%.*]] = ashr exact i64 [[SEXT]], 32
6217; UNROLL-NEXT:    [[TMP6:%.*]] = getelementptr i32, i32* [[DST:%.*]], i64 [[TMP5]]
6218; UNROLL-NEXT:    [[TMP7:%.*]] = add <2 x i32> [[VEC_IND]], [[TMP3]]
6219; UNROLL-NEXT:    [[TMP8:%.*]] = add <2 x i32> [[STEP_ADD]], [[TMP4]]
6220; UNROLL-NEXT:    [[TMP9:%.*]] = bitcast i32* [[TMP6]] to <2 x i32>*
6221; UNROLL-NEXT:    store <2 x i32> [[TMP7]], <2 x i32>* [[TMP9]], align 4
6222; UNROLL-NEXT:    [[TMP10:%.*]] = getelementptr i32, i32* [[TMP6]], i64 2
6223; UNROLL-NEXT:    [[TMP11:%.*]] = bitcast i32* [[TMP10]] to <2 x i32>*
6224; UNROLL-NEXT:    store <2 x i32> [[TMP8]], <2 x i32>* [[TMP11]], align 4
6225; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
6226; UNROLL-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 4, i32 4>
6227; UNROLL-NEXT:    [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 100
6228; UNROLL-NEXT:    br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP52:![0-9]+]]
6229; UNROLL:       middle.block:
6230; UNROLL-NEXT:    br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
6231; UNROLL:       scalar.ph:
6232; UNROLL-NEXT:    br label [[LOOP:%.*]]
6233; UNROLL:       loop:
6234; UNROLL-NEXT:    br i1 poison, label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP53:![0-9]+]]
6235; UNROLL:       exit:
6236; UNROLL-NEXT:    ret void
6237;
6238; UNROLL-NO-IC-LABEL: @pr52460_first_order_recurrence_truncated_iv(
6239; UNROLL-NO-IC-NEXT:  entry:
6240; UNROLL-NO-IC-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
6241; UNROLL-NO-IC:       vector.ph:
6242; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
6243; UNROLL-NO-IC:       vector.body:
6244; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
6245; UNROLL-NO-IC-NEXT:    [[VECTOR_RECUR:%.*]] = phi <2 x i32> [ <i32 poison, i32 0>, [[VECTOR_PH]] ], [ [[STEP_ADD:%.*]], [[VECTOR_BODY]] ]
6246; UNROLL-NO-IC-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
6247; UNROLL-NO-IC-NEXT:    [[STEP_ADD]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
6248; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = trunc i64 [[INDEX]] to i32
6249; UNROLL-NO-IC-NEXT:    [[TMP1:%.*]] = add i32 [[TMP0]], 0
6250; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = add i32 [[TMP0]], 2
6251; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = shufflevector <2 x i32> [[VECTOR_RECUR]], <2 x i32> [[VEC_IND]], <2 x i32> <i32 1, i32 2>
6252; UNROLL-NO-IC-NEXT:    [[TMP4:%.*]] = shufflevector <2 x i32> [[VEC_IND]], <2 x i32> [[STEP_ADD]], <2 x i32> <i32 1, i32 2>
6253; UNROLL-NO-IC-NEXT:    [[TMP5:%.*]] = load i32, i32* [[SRC:%.*]], align 4
6254; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[TMP5]], i32 0
6255; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
6256; UNROLL-NO-IC-NEXT:    [[TMP6:%.*]] = load i32, i32* [[SRC]], align 4
6257; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <2 x i32> poison, i32 [[TMP6]], i32 0
6258; UNROLL-NO-IC-NEXT:    [[BROADCAST_SPLAT4:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT3]], <2 x i32> poison, <2 x i32> zeroinitializer
6259; UNROLL-NO-IC-NEXT:    [[TMP7:%.*]] = mul nsw <2 x i32> [[BROADCAST_SPLAT]], [[TMP3]]
6260; UNROLL-NO-IC-NEXT:    [[TMP8:%.*]] = mul nsw <2 x i32> [[BROADCAST_SPLAT4]], [[TMP4]]
6261; UNROLL-NO-IC-NEXT:    [[TMP9:%.*]] = getelementptr i32, i32* [[DST:%.*]], i32 [[TMP1]]
6262; UNROLL-NO-IC-NEXT:    [[TMP10:%.*]] = getelementptr i32, i32* [[DST]], i32 [[TMP2]]
6263; UNROLL-NO-IC-NEXT:    [[TMP11:%.*]] = add <2 x i32> [[VEC_IND]], [[TMP7]]
6264; UNROLL-NO-IC-NEXT:    [[TMP12:%.*]] = add <2 x i32> [[STEP_ADD]], [[TMP8]]
6265; UNROLL-NO-IC-NEXT:    [[TMP13:%.*]] = getelementptr i32, i32* [[TMP9]], i32 0
6266; UNROLL-NO-IC-NEXT:    [[TMP14:%.*]] = bitcast i32* [[TMP13]] to <2 x i32>*
6267; UNROLL-NO-IC-NEXT:    store <2 x i32> [[TMP11]], <2 x i32>* [[TMP14]], align 4
6268; UNROLL-NO-IC-NEXT:    [[TMP15:%.*]] = getelementptr i32, i32* [[TMP9]], i32 2
6269; UNROLL-NO-IC-NEXT:    [[TMP16:%.*]] = bitcast i32* [[TMP15]] to <2 x i32>*
6270; UNROLL-NO-IC-NEXT:    store <2 x i32> [[TMP12]], <2 x i32>* [[TMP16]], align 4
6271; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
6272; UNROLL-NO-IC-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[STEP_ADD]], <i32 2, i32 2>
6273; UNROLL-NO-IC-NEXT:    [[TMP17:%.*]] = icmp eq i64 [[INDEX_NEXT]], 100
6274; UNROLL-NO-IC-NEXT:    br i1 [[TMP17]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP52:![0-9]+]]
6275; UNROLL-NO-IC:       middle.block:
6276; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i64 100, 100
6277; UNROLL-NO-IC-NEXT:    [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <2 x i32> [[STEP_ADD]], i32 1
6278; UNROLL-NO-IC-NEXT:    [[VECTOR_RECUR_EXTRACT_FOR_PHI:%.*]] = extractelement <2 x i32> [[STEP_ADD]], i32 0
6279; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
6280; UNROLL-NO-IC:       scalar.ph:
6281; UNROLL-NO-IC-NEXT:    [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]
6282; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 100, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
6283; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ 100, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
6284; UNROLL-NO-IC-NEXT:    br label [[LOOP:%.*]]
6285; UNROLL-NO-IC:       loop:
6286; UNROLL-NO-IC-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
6287; UNROLL-NO-IC-NEXT:    [[TRUNC_IV:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[TRUNC_IV_NEXT:%.*]], [[LOOP]] ]
6288; UNROLL-NO-IC-NEXT:    [[SCALAR_RECUR:%.*]] = phi i32 [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[IV_TRUNC:%.*]], [[LOOP]] ]
6289; UNROLL-NO-IC-NEXT:    [[LV:%.*]] = load i32, i32* [[SRC]], align 4
6290; UNROLL-NO-IC-NEXT:    [[MUL:%.*]] = mul nsw i32 [[LV]], [[SCALAR_RECUR]]
6291; UNROLL-NO-IC-NEXT:    [[TRUNC_IV_NEXT]] = add i32 [[TRUNC_IV]], 1
6292; UNROLL-NO-IC-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
6293; UNROLL-NO-IC-NEXT:    [[IV_TRUNC]] = trunc i64 [[IV]] to i32
6294; UNROLL-NO-IC-NEXT:    [[DST_GEP:%.*]] = getelementptr i32, i32* [[DST]], i32 [[IV_TRUNC]]
6295; UNROLL-NO-IC-NEXT:    [[ADD:%.*]] = add i32 [[IV_TRUNC]], [[MUL]]
6296; UNROLL-NO-IC-NEXT:    store i32 [[ADD]], i32* [[DST_GEP]], align 4
6297; UNROLL-NO-IC-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[TRUNC_IV_NEXT]], 100
6298; UNROLL-NO-IC-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP53:![0-9]+]]
6299; UNROLL-NO-IC:       exit:
6300; UNROLL-NO-IC-NEXT:    ret void
6301;
6302; INTERLEAVE-LABEL: @pr52460_first_order_recurrence_truncated_iv(
6303; INTERLEAVE-NEXT:  entry:
6304; INTERLEAVE-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
6305; INTERLEAVE:       vector.ph:
6306; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
6307; INTERLEAVE:       vector.body:
6308; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
6309; INTERLEAVE-NEXT:    [[VECTOR_RECUR:%.*]] = phi <4 x i32> [ <i32 poison, i32 poison, i32 poison, i32 0>, [[VECTOR_PH]] ], [ [[STEP_ADD:%.*]], [[VECTOR_BODY]] ]
6310; INTERLEAVE-NEXT:    [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
6311; INTERLEAVE-NEXT:    [[STEP_ADD]] = add <4 x i32> [[VEC_IND]], <i32 4, i32 4, i32 4, i32 4>
6312; INTERLEAVE-NEXT:    [[TMP0:%.*]] = shufflevector <4 x i32> [[VECTOR_RECUR]], <4 x i32> [[VEC_IND]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
6313; INTERLEAVE-NEXT:    [[TMP1:%.*]] = shufflevector <4 x i32> [[VEC_IND]], <4 x i32> [[STEP_ADD]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
6314; INTERLEAVE-NEXT:    [[TMP2:%.*]] = load i32, i32* [[SRC:%.*]], align 4
6315; INTERLEAVE-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[TMP2]], i64 0
6316; INTERLEAVE-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer
6317; INTERLEAVE-NEXT:    [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <4 x i32> poison, i32 [[TMP2]], i64 0
6318; INTERLEAVE-NEXT:    [[BROADCAST_SPLAT4:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT3]], <4 x i32> poison, <4 x i32> zeroinitializer
6319; INTERLEAVE-NEXT:    [[TMP3:%.*]] = mul nsw <4 x i32> [[BROADCAST_SPLAT]], [[TMP0]]
6320; INTERLEAVE-NEXT:    [[TMP4:%.*]] = mul nsw <4 x i32> [[BROADCAST_SPLAT4]], [[TMP1]]
6321; INTERLEAVE-NEXT:    [[SEXT:%.*]] = shl i64 [[INDEX]], 32
6322; INTERLEAVE-NEXT:    [[TMP5:%.*]] = ashr exact i64 [[SEXT]], 32
6323; INTERLEAVE-NEXT:    [[TMP6:%.*]] = getelementptr i32, i32* [[DST:%.*]], i64 [[TMP5]]
6324; INTERLEAVE-NEXT:    [[TMP7:%.*]] = add <4 x i32> [[VEC_IND]], [[TMP3]]
6325; INTERLEAVE-NEXT:    [[TMP8:%.*]] = add <4 x i32> [[STEP_ADD]], [[TMP4]]
6326; INTERLEAVE-NEXT:    [[TMP9:%.*]] = bitcast i32* [[TMP6]] to <4 x i32>*
6327; INTERLEAVE-NEXT:    store <4 x i32> [[TMP7]], <4 x i32>* [[TMP9]], align 4
6328; INTERLEAVE-NEXT:    [[TMP10:%.*]] = getelementptr i32, i32* [[TMP6]], i64 4
6329; INTERLEAVE-NEXT:    [[TMP11:%.*]] = bitcast i32* [[TMP10]] to <4 x i32>*
6330; INTERLEAVE-NEXT:    store <4 x i32> [[TMP8]], <4 x i32>* [[TMP11]], align 4
6331; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
6332; INTERLEAVE-NEXT:    [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 8, i32 8, i32 8, i32 8>
6333; INTERLEAVE-NEXT:    [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 96
6334; INTERLEAVE-NEXT:    br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP52:![0-9]+]]
6335; INTERLEAVE:       middle.block:
6336; INTERLEAVE-NEXT:    [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <4 x i32> [[STEP_ADD]], i64 3
6337; INTERLEAVE-NEXT:    br i1 false, label [[EXIT:%.*]], label [[SCALAR_PH]]
6338; INTERLEAVE:       scalar.ph:
6339; INTERLEAVE-NEXT:    [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]
6340; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ 96, [[MIDDLE_BLOCK]] ]
6341; INTERLEAVE-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ 96, [[MIDDLE_BLOCK]] ]
6342; INTERLEAVE-NEXT:    br label [[LOOP:%.*]]
6343; INTERLEAVE:       loop:
6344; INTERLEAVE-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
6345; INTERLEAVE-NEXT:    [[TRUNC_IV:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[TRUNC_IV_NEXT:%.*]], [[LOOP]] ]
6346; INTERLEAVE-NEXT:    [[SCALAR_RECUR:%.*]] = phi i32 [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[IV_TRUNC:%.*]], [[LOOP]] ]
6347; INTERLEAVE-NEXT:    [[LV:%.*]] = load i32, i32* [[SRC]], align 4
6348; INTERLEAVE-NEXT:    [[MUL:%.*]] = mul nsw i32 [[LV]], [[SCALAR_RECUR]]
6349; INTERLEAVE-NEXT:    [[TRUNC_IV_NEXT]] = add i32 [[TRUNC_IV]], 1
6350; INTERLEAVE-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
6351; INTERLEAVE-NEXT:    [[IV_TRUNC]] = trunc i64 [[IV]] to i32
6352; INTERLEAVE-NEXT:    [[SEXT5:%.*]] = shl i64 [[IV]], 32
6353; INTERLEAVE-NEXT:    [[TMP13:%.*]] = ashr exact i64 [[SEXT5]], 32
6354; INTERLEAVE-NEXT:    [[DST_GEP:%.*]] = getelementptr i32, i32* [[DST]], i64 [[TMP13]]
6355; INTERLEAVE-NEXT:    [[ADD:%.*]] = add i32 [[MUL]], [[IV_TRUNC]]
6356; INTERLEAVE-NEXT:    store i32 [[ADD]], i32* [[DST_GEP]], align 4
6357; INTERLEAVE-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[TRUNC_IV_NEXT]], 100
6358; INTERLEAVE-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP53:![0-9]+]]
6359; INTERLEAVE:       exit:
6360; INTERLEAVE-NEXT:    ret void
6361;
6362entry:
6363  br label %loop
6364
6365loop:
6366  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
6367  %trunc.iv = phi i32 [ 0, %entry ], [ %trunc.iv.next, %loop ]
6368  %recur = phi i32 [ 0, %entry ], [ %iv.trunc, %loop ]
6369  %lv = load i32, i32* %src, align 4
6370  %mul = mul nsw i32 %lv, %recur
6371  %trunc.iv.next  = add i32 %trunc.iv, 1
6372  %iv.next = add nuw nsw i64 %iv, 1
6373  %iv.trunc = trunc i64 %iv to i32
6374  %dst.gep = getelementptr i32, i32* %dst, i32 %iv.trunc
6375  %add = add i32 %iv.trunc, %mul
6376  store i32 %add, i32* %dst.gep
6377  %exitcond = icmp eq i32 %trunc.iv.next, 100
6378  br i1 %exitcond, label %exit, label %loop
6379
6380exit:
6381  ret void
6382}
6383
6384; Test case where %iv.2.ext and %iv.2.conv become redundant due to the SCEV
6385; predicates generated for the vector loop. They should be removed in the
6386; vector loop.
6387define void @test_optimized_cast_induction_feeding_first_order_recurrence(i64 %n, i32 %step, i32* %ptr) {
6388;
6389; CHECK-LABEL: @test_optimized_cast_induction_feeding_first_order_recurrence(
6390; CHECK-NEXT:  entry:
6391; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 2
6392; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
6393; CHECK:       vector.scevcheck:
6394; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[N]], -1
6395; CHECK-NEXT:    [[TMP1:%.*]] = trunc i32 [[STEP:%.*]] to i8
6396; CHECK-NEXT:    [[TMP2:%.*]] = sub i8 0, [[TMP1]]
6397; CHECK-NEXT:    [[TMP3:%.*]] = icmp slt i8 [[TMP1]], 0
6398; CHECK-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], i8 [[TMP2]], i8 [[TMP1]]
6399; CHECK-NEXT:    [[TMP5:%.*]] = trunc i64 [[TMP0]] to i8
6400; CHECK-NEXT:    [[MUL:%.*]] = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 [[TMP4]], i8 [[TMP5]])
6401; CHECK-NEXT:    [[MUL_RESULT:%.*]] = extractvalue { i8, i1 } [[MUL]], 0
6402; CHECK-NEXT:    [[MUL_OVERFLOW:%.*]] = extractvalue { i8, i1 } [[MUL]], 1
6403; CHECK-NEXT:    [[TMP6:%.*]] = sub i8 0, [[MUL_RESULT]]
6404; CHECK-NEXT:    [[TMP7:%.*]] = icmp slt i8 [[MUL_RESULT]], 0
6405; CHECK-NEXT:    [[TMP8:%.*]] = icmp sgt i8 [[TMP6]], 0
6406; CHECK-NEXT:    [[TMP9:%.*]] = select i1 [[TMP3]], i1 [[TMP8]], i1 [[TMP7]]
6407; CHECK-NEXT:    [[TMP10:%.*]] = or i1 [[TMP9]], [[MUL_OVERFLOW]]
6408; CHECK-NEXT:    [[TMP11:%.*]] = icmp ugt i64 [[TMP0]], 255
6409; CHECK-NEXT:    [[TMP12:%.*]] = icmp ne i8 [[TMP1]], 0
6410; CHECK-NEXT:    [[TMP13:%.*]] = and i1 [[TMP11]], [[TMP12]]
6411; CHECK-NEXT:    [[TMP14:%.*]] = or i1 [[TMP10]], [[TMP13]]
6412; CHECK-NEXT:    [[TMP15:%.*]] = sext i8 [[TMP1]] to i32
6413; CHECK-NEXT:    [[IDENT_CHECK:%.*]] = icmp ne i32 [[STEP]], [[TMP15]]
6414; CHECK-NEXT:    [[TMP16:%.*]] = or i1 [[TMP14]], [[IDENT_CHECK]]
6415; CHECK-NEXT:    br i1 [[TMP16]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
6416; CHECK:       vector.ph:
6417; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], 2
6418; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
6419; CHECK-NEXT:    [[CAST_VTC:%.*]] = trunc i64 [[N_VEC]] to i32
6420; CHECK-NEXT:    [[IND_END:%.*]] = mul i32 [[CAST_VTC]], [[STEP]]
6421; CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[STEP]], i32 0
6422; CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
6423; CHECK-NEXT:    [[TMP17:%.*]] = mul <2 x i32> <i32 0, i32 1>, [[DOTSPLAT]]
6424; CHECK-NEXT:    [[INDUCTION:%.*]] = add <2 x i32> zeroinitializer, [[TMP17]]
6425; CHECK-NEXT:    [[TMP18:%.*]] = mul i32 [[STEP]], 2
6426; CHECK-NEXT:    [[DOTSPLATINSERT2:%.*]] = insertelement <2 x i32> poison, i32 [[TMP18]], i32 0
6427; CHECK-NEXT:    [[DOTSPLAT3:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT2]], <2 x i32> poison, <2 x i32> zeroinitializer
6428; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
6429; CHECK:       vector.body:
6430; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
6431; CHECK-NEXT:    [[VECTOR_RECUR:%.*]] = phi <2 x i32> [ <i32 poison, i32 0>, [[VECTOR_PH]] ], [ [[VEC_IND:%.*]], [[VECTOR_BODY]] ]
6432; CHECK-NEXT:    [[VEC_IND]] = phi <2 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
6433; CHECK-NEXT:    [[TMP19:%.*]] = add i64 [[INDEX]], 0
6434; CHECK-NEXT:    [[TMP20:%.*]] = shufflevector <2 x i32> [[VECTOR_RECUR]], <2 x i32> [[VEC_IND]], <2 x i32> <i32 1, i32 2>
6435; CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds i32, i32* [[PTR:%.*]], i64 [[TMP19]]
6436; CHECK-NEXT:    [[TMP22:%.*]] = getelementptr inbounds i32, i32* [[TMP21]], i32 0
6437; CHECK-NEXT:    [[TMP23:%.*]] = bitcast i32* [[TMP22]] to <2 x i32>*
6438; CHECK-NEXT:    store <2 x i32> [[TMP20]], <2 x i32>* [[TMP23]], align 4
6439; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
6440; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], [[DOTSPLAT3]]
6441; CHECK-NEXT:    [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
6442; CHECK-NEXT:    br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP54:![0-9]+]]
6443; CHECK:       middle.block:
6444; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
6445; CHECK-NEXT:    [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <2 x i32> [[VEC_IND]], i32 1
6446; CHECK-NEXT:    [[VECTOR_RECUR_EXTRACT_FOR_PHI:%.*]] = extractelement <2 x i32> [[VEC_IND]], i32 0
6447; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
6448; CHECK:       scalar.ph:
6449; CHECK-NEXT:    [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]
6450; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
6451; CHECK-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
6452; CHECK-NEXT:    br label [[LOOP:%.*]]
6453; CHECK:       loop:
6454; CHECK-NEXT:    [[SCALAR_RECUR:%.*]] = phi i32 [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[IV_2_CONV:%.*]], [[LOOP]] ]
6455; CHECK-NEXT:    [[IV_1:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_1_NEXT:%.*]], [[LOOP]] ]
6456; CHECK-NEXT:    [[IV_2:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[IV_2_NEXT:%.*]], [[LOOP]] ]
6457; CHECK-NEXT:    [[IV_2_EXT:%.*]] = shl i32 [[IV_2]], 24
6458; CHECK-NEXT:    [[IV_2_CONV]] = ashr exact i32 [[IV_2_EXT]], 24
6459; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds i32, i32* [[PTR]], i64 [[IV_1]]
6460; CHECK-NEXT:    store i32 [[SCALAR_RECUR]], i32* [[GEP]], align 4
6461; CHECK-NEXT:    [[IV_2_NEXT]] = add nsw i32 [[IV_2_CONV]], [[STEP]]
6462; CHECK-NEXT:    [[IV_1_NEXT]] = add nuw nsw i64 [[IV_1]], 1
6463; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[IV_1_NEXT]], [[N]]
6464; CHECK-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP55:![0-9]+]]
6465; CHECK:       exit:
6466; CHECK-NEXT:    ret void
6467;
6468; IND-LABEL: @test_optimized_cast_induction_feeding_first_order_recurrence(
6469; IND-NEXT:  entry:
6470; IND-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 2
6471; IND-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
6472; IND:       vector.scevcheck:
6473; IND-NEXT:    [[TMP0:%.*]] = add i64 [[N]], -1
6474; IND-NEXT:    [[TMP1:%.*]] = trunc i32 [[STEP:%.*]] to i8
6475; IND-NEXT:    [[TMP2:%.*]] = icmp slt i8 [[TMP1]], 0
6476; IND-NEXT:    [[TMP3:%.*]] = call i8 @llvm.abs.i8(i8 [[TMP1]], i1 false)
6477; IND-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP0]] to i8
6478; IND-NEXT:    [[MUL:%.*]] = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 [[TMP3]], i8 [[TMP4]])
6479; IND-NEXT:    [[MUL_RESULT:%.*]] = extractvalue { i8, i1 } [[MUL]], 0
6480; IND-NEXT:    [[MUL_OVERFLOW:%.*]] = extractvalue { i8, i1 } [[MUL]], 1
6481; IND-NEXT:    [[TMP5:%.*]] = icmp slt i8 [[MUL_RESULT]], 0
6482; IND-NEXT:    [[TMP6:%.*]] = icmp ugt i8 [[MUL_RESULT]], -128
6483; IND-NEXT:    [[TMP7:%.*]] = select i1 [[TMP2]], i1 [[TMP6]], i1 [[TMP5]]
6484; IND-NEXT:    [[TMP8:%.*]] = or i1 [[TMP7]], [[MUL_OVERFLOW]]
6485; IND-NEXT:    [[TMP9:%.*]] = icmp ugt i64 [[TMP0]], 255
6486; IND-NEXT:    [[TMP10:%.*]] = icmp ne i8 [[TMP1]], 0
6487; IND-NEXT:    [[TMP11:%.*]] = and i1 [[TMP9]], [[TMP10]]
6488; IND-NEXT:    [[TMP12:%.*]] = or i1 [[TMP8]], [[TMP11]]
6489; IND-NEXT:    [[TMP13:%.*]] = add i32 [[STEP]], -128
6490; IND-NEXT:    [[TMP14:%.*]] = icmp ult i32 [[TMP13]], -256
6491; IND-NEXT:    [[TMP15:%.*]] = or i1 [[TMP12]], [[TMP14]]
6492; IND-NEXT:    br i1 [[TMP15]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
6493; IND:       vector.ph:
6494; IND-NEXT:    [[N_VEC:%.*]] = and i64 [[N]], -2
6495; IND-NEXT:    [[CAST_VTC:%.*]] = trunc i64 [[N_VEC]] to i32
6496; IND-NEXT:    [[IND_END:%.*]] = mul i32 [[CAST_VTC]], [[STEP]]
6497; IND-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[STEP]], i64 0
6498; IND-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
6499; IND-NEXT:    [[TMP16:%.*]] = mul nuw <2 x i32> [[DOTSPLAT]], <i32 0, i32 1>
6500; IND-NEXT:    [[TMP17:%.*]] = shl i32 [[STEP]], 1
6501; IND-NEXT:    [[DOTSPLATINSERT2:%.*]] = insertelement <2 x i32> poison, i32 [[TMP17]], i64 0
6502; IND-NEXT:    [[DOTSPLAT3:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT2]], <2 x i32> poison, <2 x i32> zeroinitializer
6503; IND-NEXT:    br label [[VECTOR_BODY:%.*]]
6504; IND:       vector.body:
6505; IND-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
6506; IND-NEXT:    [[VECTOR_RECUR:%.*]] = phi <2 x i32> [ <i32 poison, i32 0>, [[VECTOR_PH]] ], [ [[VEC_IND:%.*]], [[VECTOR_BODY]] ]
6507; IND-NEXT:    [[VEC_IND]] = phi <2 x i32> [ [[TMP16]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
6508; IND-NEXT:    [[TMP18:%.*]] = shufflevector <2 x i32> [[VECTOR_RECUR]], <2 x i32> [[VEC_IND]], <2 x i32> <i32 1, i32 2>
6509; IND-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i32, i32* [[PTR:%.*]], i64 [[INDEX]]
6510; IND-NEXT:    [[TMP20:%.*]] = bitcast i32* [[TMP19]] to <2 x i32>*
6511; IND-NEXT:    store <2 x i32> [[TMP18]], <2 x i32>* [[TMP20]], align 4
6512; IND-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
6513; IND-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], [[DOTSPLAT3]]
6514; IND-NEXT:    [[TMP21:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
6515; IND-NEXT:    br i1 [[TMP21]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP54:![0-9]+]]
6516; IND:       middle.block:
6517; IND-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[N]]
6518; IND-NEXT:    [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <2 x i32> [[VEC_IND]], i64 1
6519; IND-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
6520; IND:       scalar.ph:
6521; IND-NEXT:    [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]
6522; IND-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[ENTRY]] ], [ [[N_VEC]], [[MIDDLE_BLOCK]] ]
6523; IND-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[ENTRY]] ], [ [[IND_END]], [[MIDDLE_BLOCK]] ]
6524; IND-NEXT:    br label [[LOOP:%.*]]
6525; IND:       loop:
6526; IND-NEXT:    [[SCALAR_RECUR:%.*]] = phi i32 [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[IV_2_CONV:%.*]], [[LOOP]] ]
6527; IND-NEXT:    [[IV_1:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_1_NEXT:%.*]], [[LOOP]] ]
6528; IND-NEXT:    [[IV_2:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[IV_2_NEXT:%.*]], [[LOOP]] ]
6529; IND-NEXT:    [[IV_2_EXT:%.*]] = shl i32 [[IV_2]], 24
6530; IND-NEXT:    [[IV_2_CONV]] = ashr exact i32 [[IV_2_EXT]], 24
6531; IND-NEXT:    [[GEP:%.*]] = getelementptr inbounds i32, i32* [[PTR]], i64 [[IV_1]]
6532; IND-NEXT:    store i32 [[SCALAR_RECUR]], i32* [[GEP]], align 4
6533; IND-NEXT:    [[IV_2_NEXT]] = add nsw i32 [[IV_2_CONV]], [[STEP]]
6534; IND-NEXT:    [[IV_1_NEXT]] = add nuw nsw i64 [[IV_1]], 1
6535; IND-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[IV_1_NEXT]], [[N]]
6536; IND-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP55:![0-9]+]]
6537; IND:       exit:
6538; IND-NEXT:    ret void
6539;
6540; UNROLL-LABEL: @test_optimized_cast_induction_feeding_first_order_recurrence(
6541; UNROLL-NEXT:  entry:
6542; UNROLL-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4
6543; UNROLL-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
6544; UNROLL:       vector.scevcheck:
6545; UNROLL-NEXT:    [[TMP0:%.*]] = add i64 [[N]], -1
6546; UNROLL-NEXT:    [[TMP1:%.*]] = trunc i32 [[STEP:%.*]] to i8
6547; UNROLL-NEXT:    [[TMP2:%.*]] = icmp slt i8 [[TMP1]], 0
6548; UNROLL-NEXT:    [[TMP3:%.*]] = call i8 @llvm.abs.i8(i8 [[TMP1]], i1 false)
6549; UNROLL-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP0]] to i8
6550; UNROLL-NEXT:    [[MUL:%.*]] = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 [[TMP3]], i8 [[TMP4]])
6551; UNROLL-NEXT:    [[MUL_RESULT:%.*]] = extractvalue { i8, i1 } [[MUL]], 0
6552; UNROLL-NEXT:    [[MUL_OVERFLOW:%.*]] = extractvalue { i8, i1 } [[MUL]], 1
6553; UNROLL-NEXT:    [[TMP5:%.*]] = icmp slt i8 [[MUL_RESULT]], 0
6554; UNROLL-NEXT:    [[TMP6:%.*]] = icmp ugt i8 [[MUL_RESULT]], -128
6555; UNROLL-NEXT:    [[TMP7:%.*]] = select i1 [[TMP2]], i1 [[TMP6]], i1 [[TMP5]]
6556; UNROLL-NEXT:    [[TMP8:%.*]] = or i1 [[TMP7]], [[MUL_OVERFLOW]]
6557; UNROLL-NEXT:    [[TMP9:%.*]] = icmp ugt i64 [[TMP0]], 255
6558; UNROLL-NEXT:    [[TMP10:%.*]] = icmp ne i8 [[TMP1]], 0
6559; UNROLL-NEXT:    [[TMP11:%.*]] = and i1 [[TMP9]], [[TMP10]]
6560; UNROLL-NEXT:    [[TMP12:%.*]] = or i1 [[TMP8]], [[TMP11]]
6561; UNROLL-NEXT:    [[TMP13:%.*]] = add i32 [[STEP]], -128
6562; UNROLL-NEXT:    [[TMP14:%.*]] = icmp ult i32 [[TMP13]], -256
6563; UNROLL-NEXT:    [[TMP15:%.*]] = or i1 [[TMP12]], [[TMP14]]
6564; UNROLL-NEXT:    br i1 [[TMP15]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
6565; UNROLL:       vector.ph:
6566; UNROLL-NEXT:    [[N_VEC:%.*]] = and i64 [[N]], -4
6567; UNROLL-NEXT:    [[CAST_VTC:%.*]] = trunc i64 [[N_VEC]] to i32
6568; UNROLL-NEXT:    [[IND_END:%.*]] = mul i32 [[CAST_VTC]], [[STEP]]
6569; UNROLL-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[STEP]], i64 0
6570; UNROLL-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
6571; UNROLL-NEXT:    [[TMP16:%.*]] = mul nuw <2 x i32> [[DOTSPLAT]], <i32 0, i32 1>
6572; UNROLL-NEXT:    [[TMP17:%.*]] = shl i32 [[STEP]], 1
6573; UNROLL-NEXT:    [[DOTSPLATINSERT2:%.*]] = insertelement <2 x i32> poison, i32 [[TMP17]], i64 0
6574; UNROLL-NEXT:    [[DOTSPLAT3:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT2]], <2 x i32> poison, <2 x i32> zeroinitializer
6575; UNROLL-NEXT:    br label [[VECTOR_BODY:%.*]]
6576; UNROLL:       vector.body:
6577; UNROLL-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
6578; UNROLL-NEXT:    [[VECTOR_RECUR:%.*]] = phi <2 x i32> [ <i32 poison, i32 0>, [[VECTOR_PH]] ], [ [[STEP_ADD:%.*]], [[VECTOR_BODY]] ]
6579; UNROLL-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ [[TMP16]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
6580; UNROLL-NEXT:    [[STEP_ADD]] = add <2 x i32> [[VEC_IND]], [[DOTSPLAT3]]
6581; UNROLL-NEXT:    [[TMP18:%.*]] = shufflevector <2 x i32> [[VECTOR_RECUR]], <2 x i32> [[VEC_IND]], <2 x i32> <i32 1, i32 2>
6582; UNROLL-NEXT:    [[TMP19:%.*]] = shufflevector <2 x i32> [[VEC_IND]], <2 x i32> [[STEP_ADD]], <2 x i32> <i32 1, i32 2>
6583; UNROLL-NEXT:    [[TMP20:%.*]] = getelementptr inbounds i32, i32* [[PTR:%.*]], i64 [[INDEX]]
6584; UNROLL-NEXT:    [[TMP21:%.*]] = bitcast i32* [[TMP20]] to <2 x i32>*
6585; UNROLL-NEXT:    store <2 x i32> [[TMP18]], <2 x i32>* [[TMP21]], align 4
6586; UNROLL-NEXT:    [[TMP22:%.*]] = getelementptr inbounds i32, i32* [[TMP20]], i64 2
6587; UNROLL-NEXT:    [[TMP23:%.*]] = bitcast i32* [[TMP22]] to <2 x i32>*
6588; UNROLL-NEXT:    store <2 x i32> [[TMP19]], <2 x i32>* [[TMP23]], align 4
6589; UNROLL-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
6590; UNROLL-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[STEP_ADD]], [[DOTSPLAT3]]
6591; UNROLL-NEXT:    [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
6592; UNROLL-NEXT:    br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP54:![0-9]+]]
6593; UNROLL:       middle.block:
6594; UNROLL-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[N]]
6595; UNROLL-NEXT:    [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <2 x i32> [[STEP_ADD]], i64 1
6596; UNROLL-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
6597; UNROLL:       scalar.ph:
6598; UNROLL-NEXT:    [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]
6599; UNROLL-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[ENTRY]] ], [ [[N_VEC]], [[MIDDLE_BLOCK]] ]
6600; UNROLL-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[ENTRY]] ], [ [[IND_END]], [[MIDDLE_BLOCK]] ]
6601; UNROLL-NEXT:    br label [[LOOP:%.*]]
6602; UNROLL:       loop:
6603; UNROLL-NEXT:    [[SCALAR_RECUR:%.*]] = phi i32 [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[IV_2_CONV:%.*]], [[LOOP]] ]
6604; UNROLL-NEXT:    [[IV_1:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_1_NEXT:%.*]], [[LOOP]] ]
6605; UNROLL-NEXT:    [[IV_2:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[IV_2_NEXT:%.*]], [[LOOP]] ]
6606; UNROLL-NEXT:    [[IV_2_EXT:%.*]] = shl i32 [[IV_2]], 24
6607; UNROLL-NEXT:    [[IV_2_CONV]] = ashr exact i32 [[IV_2_EXT]], 24
6608; UNROLL-NEXT:    [[GEP:%.*]] = getelementptr inbounds i32, i32* [[PTR]], i64 [[IV_1]]
6609; UNROLL-NEXT:    store i32 [[SCALAR_RECUR]], i32* [[GEP]], align 4
6610; UNROLL-NEXT:    [[IV_2_NEXT]] = add nsw i32 [[IV_2_CONV]], [[STEP]]
6611; UNROLL-NEXT:    [[IV_1_NEXT]] = add nuw nsw i64 [[IV_1]], 1
6612; UNROLL-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[IV_1_NEXT]], [[N]]
6613; UNROLL-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP55:![0-9]+]]
6614; UNROLL:       exit:
6615; UNROLL-NEXT:    ret void
6616;
6617; UNROLL-NO-IC-LABEL: @test_optimized_cast_induction_feeding_first_order_recurrence(
6618; UNROLL-NO-IC-NEXT:  entry:
6619; UNROLL-NO-IC-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4
6620; UNROLL-NO-IC-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
6621; UNROLL-NO-IC:       vector.scevcheck:
6622; UNROLL-NO-IC-NEXT:    [[TMP0:%.*]] = add i64 [[N]], -1
6623; UNROLL-NO-IC-NEXT:    [[TMP1:%.*]] = trunc i32 [[STEP:%.*]] to i8
6624; UNROLL-NO-IC-NEXT:    [[TMP2:%.*]] = sub i8 0, [[TMP1]]
6625; UNROLL-NO-IC-NEXT:    [[TMP3:%.*]] = icmp slt i8 [[TMP1]], 0
6626; UNROLL-NO-IC-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], i8 [[TMP2]], i8 [[TMP1]]
6627; UNROLL-NO-IC-NEXT:    [[TMP5:%.*]] = trunc i64 [[TMP0]] to i8
6628; UNROLL-NO-IC-NEXT:    [[MUL:%.*]] = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 [[TMP4]], i8 [[TMP5]])
6629; UNROLL-NO-IC-NEXT:    [[MUL_RESULT:%.*]] = extractvalue { i8, i1 } [[MUL]], 0
6630; UNROLL-NO-IC-NEXT:    [[MUL_OVERFLOW:%.*]] = extractvalue { i8, i1 } [[MUL]], 1
6631; UNROLL-NO-IC-NEXT:    [[TMP6:%.*]] = sub i8 0, [[MUL_RESULT]]
6632; UNROLL-NO-IC-NEXT:    [[TMP7:%.*]] = icmp slt i8 [[MUL_RESULT]], 0
6633; UNROLL-NO-IC-NEXT:    [[TMP8:%.*]] = icmp sgt i8 [[TMP6]], 0
6634; UNROLL-NO-IC-NEXT:    [[TMP9:%.*]] = select i1 [[TMP3]], i1 [[TMP8]], i1 [[TMP7]]
6635; UNROLL-NO-IC-NEXT:    [[TMP10:%.*]] = or i1 [[TMP9]], [[MUL_OVERFLOW]]
6636; UNROLL-NO-IC-NEXT:    [[TMP11:%.*]] = icmp ugt i64 [[TMP0]], 255
6637; UNROLL-NO-IC-NEXT:    [[TMP12:%.*]] = icmp ne i8 [[TMP1]], 0
6638; UNROLL-NO-IC-NEXT:    [[TMP13:%.*]] = and i1 [[TMP11]], [[TMP12]]
6639; UNROLL-NO-IC-NEXT:    [[TMP14:%.*]] = or i1 [[TMP10]], [[TMP13]]
6640; UNROLL-NO-IC-NEXT:    [[TMP15:%.*]] = sext i8 [[TMP1]] to i32
6641; UNROLL-NO-IC-NEXT:    [[IDENT_CHECK:%.*]] = icmp ne i32 [[STEP]], [[TMP15]]
6642; UNROLL-NO-IC-NEXT:    [[TMP16:%.*]] = or i1 [[TMP14]], [[IDENT_CHECK]]
6643; UNROLL-NO-IC-NEXT:    br i1 [[TMP16]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
6644; UNROLL-NO-IC:       vector.ph:
6645; UNROLL-NO-IC-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], 4
6646; UNROLL-NO-IC-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
6647; UNROLL-NO-IC-NEXT:    [[CAST_VTC:%.*]] = trunc i64 [[N_VEC]] to i32
6648; UNROLL-NO-IC-NEXT:    [[IND_END:%.*]] = mul i32 [[CAST_VTC]], [[STEP]]
6649; UNROLL-NO-IC-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[STEP]], i32 0
6650; UNROLL-NO-IC-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
6651; UNROLL-NO-IC-NEXT:    [[TMP17:%.*]] = mul <2 x i32> <i32 0, i32 1>, [[DOTSPLAT]]
6652; UNROLL-NO-IC-NEXT:    [[INDUCTION:%.*]] = add <2 x i32> zeroinitializer, [[TMP17]]
6653; UNROLL-NO-IC-NEXT:    [[TMP18:%.*]] = mul i32 [[STEP]], 2
6654; UNROLL-NO-IC-NEXT:    [[DOTSPLATINSERT2:%.*]] = insertelement <2 x i32> poison, i32 [[TMP18]], i32 0
6655; UNROLL-NO-IC-NEXT:    [[DOTSPLAT3:%.*]] = shufflevector <2 x i32> [[DOTSPLATINSERT2]], <2 x i32> poison, <2 x i32> zeroinitializer
6656; UNROLL-NO-IC-NEXT:    br label [[VECTOR_BODY:%.*]]
6657; UNROLL-NO-IC:       vector.body:
6658; UNROLL-NO-IC-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
6659; UNROLL-NO-IC-NEXT:    [[VECTOR_RECUR:%.*]] = phi <2 x i32> [ <i32 poison, i32 0>, [[VECTOR_PH]] ], [ [[STEP_ADD:%.*]], [[VECTOR_BODY]] ]
6660; UNROLL-NO-IC-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
6661; UNROLL-NO-IC-NEXT:    [[STEP_ADD]] = add <2 x i32> [[VEC_IND]], [[DOTSPLAT3]]
6662; UNROLL-NO-IC-NEXT:    [[TMP19:%.*]] = add i64 [[INDEX]], 0
6663; UNROLL-NO-IC-NEXT:    [[TMP20:%.*]] = add i64 [[INDEX]], 2
6664; UNROLL-NO-IC-NEXT:    [[TMP21:%.*]] = shufflevector <2 x i32> [[VECTOR_RECUR]], <2 x i32> [[VEC_IND]], <2 x i32> <i32 1, i32 2>
6665; UNROLL-NO-IC-NEXT:    [[TMP22:%.*]] = shufflevector <2 x i32> [[VEC_IND]], <2 x i32> [[STEP_ADD]], <2 x i32> <i32 1, i32 2>
6666; UNROLL-NO-IC-NEXT:    [[TMP23:%.*]] = getelementptr inbounds i32, i32* [[PTR:%.*]], i64 [[TMP19]]
6667; UNROLL-NO-IC-NEXT:    [[TMP24:%.*]] = getelementptr inbounds i32, i32* [[PTR]], i64 [[TMP20]]
6668; UNROLL-NO-IC-NEXT:    [[TMP25:%.*]] = getelementptr inbounds i32, i32* [[TMP23]], i32 0
6669; UNROLL-NO-IC-NEXT:    [[TMP26:%.*]] = bitcast i32* [[TMP25]] to <2 x i32>*
6670; UNROLL-NO-IC-NEXT:    store <2 x i32> [[TMP21]], <2 x i32>* [[TMP26]], align 4
6671; UNROLL-NO-IC-NEXT:    [[TMP27:%.*]] = getelementptr inbounds i32, i32* [[TMP23]], i32 2
6672; UNROLL-NO-IC-NEXT:    [[TMP28:%.*]] = bitcast i32* [[TMP27]] to <2 x i32>*
6673; UNROLL-NO-IC-NEXT:    store <2 x i32> [[TMP22]], <2 x i32>* [[TMP28]], align 4
6674; UNROLL-NO-IC-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
6675; UNROLL-NO-IC-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[STEP_ADD]], [[DOTSPLAT3]]
6676; UNROLL-NO-IC-NEXT:    [[TMP29:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
6677; UNROLL-NO-IC-NEXT:    br i1 [[TMP29]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP54:![0-9]+]]
6678; UNROLL-NO-IC:       middle.block:
6679; UNROLL-NO-IC-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
6680; UNROLL-NO-IC-NEXT:    [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <2 x i32> [[STEP_ADD]], i32 1
6681; UNROLL-NO-IC-NEXT:    [[VECTOR_RECUR_EXTRACT_FOR_PHI:%.*]] = extractelement <2 x i32> [[STEP_ADD]], i32 0
6682; UNROLL-NO-IC-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
6683; UNROLL-NO-IC:       scalar.ph:
6684; UNROLL-NO-IC-NEXT:    [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]
6685; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
6686; UNROLL-NO-IC-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ], [ 0, [[VECTOR_SCEVCHECK]] ]
6687; UNROLL-NO-IC-NEXT:    br label [[LOOP:%.*]]
6688; UNROLL-NO-IC:       loop:
6689; UNROLL-NO-IC-NEXT:    [[SCALAR_RECUR:%.*]] = phi i32 [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[IV_2_CONV:%.*]], [[LOOP]] ]
6690; UNROLL-NO-IC-NEXT:    [[IV_1:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_1_NEXT:%.*]], [[LOOP]] ]
6691; UNROLL-NO-IC-NEXT:    [[IV_2:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[IV_2_NEXT:%.*]], [[LOOP]] ]
6692; UNROLL-NO-IC-NEXT:    [[IV_2_EXT:%.*]] = shl i32 [[IV_2]], 24
6693; UNROLL-NO-IC-NEXT:    [[IV_2_CONV]] = ashr exact i32 [[IV_2_EXT]], 24
6694; UNROLL-NO-IC-NEXT:    [[GEP:%.*]] = getelementptr inbounds i32, i32* [[PTR]], i64 [[IV_1]]
6695; UNROLL-NO-IC-NEXT:    store i32 [[SCALAR_RECUR]], i32* [[GEP]], align 4
6696; UNROLL-NO-IC-NEXT:    [[IV_2_NEXT]] = add nsw i32 [[IV_2_CONV]], [[STEP]]
6697; UNROLL-NO-IC-NEXT:    [[IV_1_NEXT]] = add nuw nsw i64 [[IV_1]], 1
6698; UNROLL-NO-IC-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[IV_1_NEXT]], [[N]]
6699; UNROLL-NO-IC-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP55:![0-9]+]]
6700; UNROLL-NO-IC:       exit:
6701; UNROLL-NO-IC-NEXT:    ret void
6702;
6703; INTERLEAVE-LABEL: @test_optimized_cast_induction_feeding_first_order_recurrence(
6704; INTERLEAVE-NEXT:  entry:
6705; INTERLEAVE-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 8
6706; INTERLEAVE-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]]
6707; INTERLEAVE:       vector.scevcheck:
6708; INTERLEAVE-NEXT:    [[TMP0:%.*]] = add i64 [[N]], -1
6709; INTERLEAVE-NEXT:    [[TMP1:%.*]] = trunc i32 [[STEP:%.*]] to i8
6710; INTERLEAVE-NEXT:    [[TMP2:%.*]] = icmp slt i8 [[TMP1]], 0
6711; INTERLEAVE-NEXT:    [[TMP3:%.*]] = call i8 @llvm.abs.i8(i8 [[TMP1]], i1 false)
6712; INTERLEAVE-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP0]] to i8
6713; INTERLEAVE-NEXT:    [[MUL:%.*]] = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 [[TMP3]], i8 [[TMP4]])
6714; INTERLEAVE-NEXT:    [[MUL_RESULT:%.*]] = extractvalue { i8, i1 } [[MUL]], 0
6715; INTERLEAVE-NEXT:    [[MUL_OVERFLOW:%.*]] = extractvalue { i8, i1 } [[MUL]], 1
6716; INTERLEAVE-NEXT:    [[TMP5:%.*]] = icmp slt i8 [[MUL_RESULT]], 0
6717; INTERLEAVE-NEXT:    [[TMP6:%.*]] = icmp ugt i8 [[MUL_RESULT]], -128
6718; INTERLEAVE-NEXT:    [[TMP7:%.*]] = select i1 [[TMP2]], i1 [[TMP6]], i1 [[TMP5]]
6719; INTERLEAVE-NEXT:    [[TMP8:%.*]] = or i1 [[TMP7]], [[MUL_OVERFLOW]]
6720; INTERLEAVE-NEXT:    [[TMP9:%.*]] = icmp ugt i64 [[TMP0]], 255
6721; INTERLEAVE-NEXT:    [[TMP10:%.*]] = icmp ne i8 [[TMP1]], 0
6722; INTERLEAVE-NEXT:    [[TMP11:%.*]] = and i1 [[TMP9]], [[TMP10]]
6723; INTERLEAVE-NEXT:    [[TMP12:%.*]] = or i1 [[TMP8]], [[TMP11]]
6724; INTERLEAVE-NEXT:    [[TMP13:%.*]] = add i32 [[STEP]], -128
6725; INTERLEAVE-NEXT:    [[TMP14:%.*]] = icmp ult i32 [[TMP13]], -256
6726; INTERLEAVE-NEXT:    [[TMP15:%.*]] = or i1 [[TMP12]], [[TMP14]]
6727; INTERLEAVE-NEXT:    br i1 [[TMP15]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
6728; INTERLEAVE:       vector.ph:
6729; INTERLEAVE-NEXT:    [[N_VEC:%.*]] = and i64 [[N]], -8
6730; INTERLEAVE-NEXT:    [[CAST_VTC:%.*]] = trunc i64 [[N_VEC]] to i32
6731; INTERLEAVE-NEXT:    [[IND_END:%.*]] = mul i32 [[CAST_VTC]], [[STEP]]
6732; INTERLEAVE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[STEP]], i64 0
6733; INTERLEAVE-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <4 x i32> [[DOTSPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer
6734; INTERLEAVE-NEXT:    [[TMP16:%.*]] = mul <4 x i32> [[DOTSPLAT]], <i32 0, i32 1, i32 2, i32 3>
6735; INTERLEAVE-NEXT:    [[TMP17:%.*]] = shl i32 [[STEP]], 2
6736; INTERLEAVE-NEXT:    [[DOTSPLATINSERT2:%.*]] = insertelement <4 x i32> poison, i32 [[TMP17]], i64 0
6737; INTERLEAVE-NEXT:    [[DOTSPLAT3:%.*]] = shufflevector <4 x i32> [[DOTSPLATINSERT2]], <4 x i32> poison, <4 x i32> zeroinitializer
6738; INTERLEAVE-NEXT:    br label [[VECTOR_BODY:%.*]]
6739; INTERLEAVE:       vector.body:
6740; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
6741; INTERLEAVE-NEXT:    [[VECTOR_RECUR:%.*]] = phi <4 x i32> [ <i32 poison, i32 poison, i32 poison, i32 0>, [[VECTOR_PH]] ], [ [[STEP_ADD:%.*]], [[VECTOR_BODY]] ]
6742; INTERLEAVE-NEXT:    [[VEC_IND:%.*]] = phi <4 x i32> [ [[TMP16]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
6743; INTERLEAVE-NEXT:    [[STEP_ADD]] = add <4 x i32> [[VEC_IND]], [[DOTSPLAT3]]
6744; INTERLEAVE-NEXT:    [[TMP18:%.*]] = shufflevector <4 x i32> [[VECTOR_RECUR]], <4 x i32> [[VEC_IND]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
6745; INTERLEAVE-NEXT:    [[TMP19:%.*]] = shufflevector <4 x i32> [[VEC_IND]], <4 x i32> [[STEP_ADD]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
6746; INTERLEAVE-NEXT:    [[TMP20:%.*]] = getelementptr inbounds i32, i32* [[PTR:%.*]], i64 [[INDEX]]
6747; INTERLEAVE-NEXT:    [[TMP21:%.*]] = bitcast i32* [[TMP20]] to <4 x i32>*
6748; INTERLEAVE-NEXT:    store <4 x i32> [[TMP18]], <4 x i32>* [[TMP21]], align 4
6749; INTERLEAVE-NEXT:    [[TMP22:%.*]] = getelementptr inbounds i32, i32* [[TMP20]], i64 4
6750; INTERLEAVE-NEXT:    [[TMP23:%.*]] = bitcast i32* [[TMP22]] to <4 x i32>*
6751; INTERLEAVE-NEXT:    store <4 x i32> [[TMP19]], <4 x i32>* [[TMP23]], align 4
6752; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
6753; INTERLEAVE-NEXT:    [[VEC_IND_NEXT]] = add <4 x i32> [[STEP_ADD]], [[DOTSPLAT3]]
6754; INTERLEAVE-NEXT:    [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
6755; INTERLEAVE-NEXT:    br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP54:![0-9]+]]
6756; INTERLEAVE:       middle.block:
6757; INTERLEAVE-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[N]]
6758; INTERLEAVE-NEXT:    [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <4 x i32> [[STEP_ADD]], i64 3
6759; INTERLEAVE-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
6760; INTERLEAVE:       scalar.ph:
6761; INTERLEAVE-NEXT:    [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]
6762; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[ENTRY]] ], [ [[N_VEC]], [[MIDDLE_BLOCK]] ]
6763; INTERLEAVE-NEXT:    [[BC_RESUME_VAL1:%.*]] = phi i32 [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[ENTRY]] ], [ [[IND_END]], [[MIDDLE_BLOCK]] ]
6764; INTERLEAVE-NEXT:    br label [[LOOP:%.*]]
6765; INTERLEAVE:       loop:
6766; INTERLEAVE-NEXT:    [[SCALAR_RECUR:%.*]] = phi i32 [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[IV_2_CONV:%.*]], [[LOOP]] ]
6767; INTERLEAVE-NEXT:    [[IV_1:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_1_NEXT:%.*]], [[LOOP]] ]
6768; INTERLEAVE-NEXT:    [[IV_2:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[IV_2_NEXT:%.*]], [[LOOP]] ]
6769; INTERLEAVE-NEXT:    [[IV_2_EXT:%.*]] = shl i32 [[IV_2]], 24
6770; INTERLEAVE-NEXT:    [[IV_2_CONV]] = ashr exact i32 [[IV_2_EXT]], 24
6771; INTERLEAVE-NEXT:    [[GEP:%.*]] = getelementptr inbounds i32, i32* [[PTR]], i64 [[IV_1]]
6772; INTERLEAVE-NEXT:    store i32 [[SCALAR_RECUR]], i32* [[GEP]], align 4
6773; INTERLEAVE-NEXT:    [[IV_2_NEXT]] = add nsw i32 [[IV_2_CONV]], [[STEP]]
6774; INTERLEAVE-NEXT:    [[IV_1_NEXT]] = add nuw nsw i64 [[IV_1]], 1
6775; INTERLEAVE-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[IV_1_NEXT]], [[N]]
6776; INTERLEAVE-NEXT:    br i1 [[EXITCOND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP55:![0-9]+]]
6777; INTERLEAVE:       exit:
6778; INTERLEAVE-NEXT:    ret void
6779;
6780entry:
6781  br label %loop
6782
6783loop:
6784  %for = phi i32 [ 0, %entry ], [ %iv.2.conv, %loop ]
6785  %iv.1 = phi i64 [ 0, %entry ], [ %iv.1.next, %loop ]
6786  %iv.2 = phi i32 [ 0, %entry ], [ %iv.2.next, %loop ]
6787  %iv.2.ext = shl i32 %iv.2, 24
6788  %iv.2.conv = ashr exact i32 %iv.2.ext, 24
6789  %gep = getelementptr inbounds i32, i32* %ptr, i64 %iv.1
6790  store i32 %for, i32* %gep, align 4
6791  %iv.2.next = add nsw i32 %iv.2.conv, %step
6792  %iv.1.next = add nuw nsw i64 %iv.1, 1
6793  %exitcond = icmp eq i64 %iv.1.next, %n
6794  br i1 %exitcond, label %exit, label %loop
6795
6796exit:
6797  ret void
6798}
6799