xref: /llvm-project/llvm/test/Transforms/LoopVectorize/iv-select-cmp-nested-loop.ll (revision 7f3428d3ed71d87a2088b77b6cab9f3d86544234) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
2; RUN: opt -passes=loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -S < %s | FileCheck %s --check-prefix=CHECK-VF4IC1
3; RUN: opt -passes=loop-vectorize -force-vector-interleave=4 -force-vector-width=4 -S < %s | FileCheck %s --check-prefix=CHECK-VF4IC4
4; RUN: opt -passes=loop-vectorize -force-vector-interleave=4 -force-vector-width=1 -S < %s | FileCheck %s --check-prefix=CHECK-VF1IC4
5
6; This should be an AnyOf reduction instead of FindLastIV reduction.
7; The reason is the outer induction variable is invariant for inner loop.
8define i64 @select_iv_def_from_outer_loop(ptr %a, i64 %start, i64 %n) {
9; CHECK-VF4IC1-LABEL: define i64 @select_iv_def_from_outer_loop(
10; CHECK-VF4IC1-SAME: ptr [[A:%.*]], i64 [[START:%.*]], i64 [[N:%.*]]) {
11; CHECK-VF4IC1-NEXT:  [[ENTRY:.*]]:
12; CHECK-VF4IC1-NEXT:    br label %[[OUTER_LOOP:.*]]
13; CHECK-VF4IC1:       [[OUTER_LOOP]]:
14; CHECK-VF4IC1-NEXT:    [[RDX_OUTER:%.*]] = phi i64 [ [[START]], %[[ENTRY]] ], [ [[SELECT_LCSSA:%.*]], %[[OUTER_LOOP_EXIT:.*]] ]
15; CHECK-VF4IC1-NEXT:    [[IV_OUTER:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[IV_OUTER_NEXT:%.*]], %[[OUTER_LOOP_EXIT]] ]
16; CHECK-VF4IC1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds ptr, ptr [[A]], i64 [[IV_OUTER]]
17; CHECK-VF4IC1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[ARRAYIDX]], align 8
18; CHECK-VF4IC1-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4
19; CHECK-VF4IC1-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
20; CHECK-VF4IC1:       [[VECTOR_PH]]:
21; CHECK-VF4IC1-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], 4
22; CHECK-VF4IC1-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
23; CHECK-VF4IC1-NEXT:    br label %[[VECTOR_BODY:.*]]
24; CHECK-VF4IC1:       [[VECTOR_BODY]]:
25; CHECK-VF4IC1-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
26; CHECK-VF4IC1-NEXT:    [[VEC_PHI:%.*]] = phi <4 x i1> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP5:%.*]], %[[VECTOR_BODY]] ]
27; CHECK-VF4IC1-NEXT:    [[TMP1:%.*]] = add i64 [[INDEX]], 0
28; CHECK-VF4IC1-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[TMP0]], i64 [[TMP1]]
29; CHECK-VF4IC1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 0
30; CHECK-VF4IC1-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP3]], align 8
31; CHECK-VF4IC1-NEXT:    [[TMP4:%.*]] = icmp eq <4 x i64> [[WIDE_LOAD]], splat (i64 3)
32; CHECK-VF4IC1-NEXT:    [[TMP5]] = or <4 x i1> [[VEC_PHI]], [[TMP4]]
33; CHECK-VF4IC1-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
34; CHECK-VF4IC1-NEXT:    [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
35; CHECK-VF4IC1-NEXT:    br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
36; CHECK-VF4IC1:       [[MIDDLE_BLOCK]]:
37; CHECK-VF4IC1-NEXT:    [[TMP7:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP5]])
38; CHECK-VF4IC1-NEXT:    [[TMP8:%.*]] = freeze i1 [[TMP7]]
39; CHECK-VF4IC1-NEXT:    [[RDX_SELECT:%.*]] = select i1 [[TMP8]], i64 [[IV_OUTER]], i64 [[RDX_OUTER]]
40; CHECK-VF4IC1-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
41; CHECK-VF4IC1-NEXT:    br i1 [[CMP_N]], label %[[OUTER_LOOP_EXIT]], label %[[SCALAR_PH]]
42; CHECK-VF4IC1:       [[SCALAR_PH]]:
43; CHECK-VF4IC1-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i64 [ [[RDX_SELECT]], %[[MIDDLE_BLOCK]] ], [ [[RDX_OUTER]], %[[OUTER_LOOP]] ]
44; CHECK-VF4IC1-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[OUTER_LOOP]] ]
45; CHECK-VF4IC1-NEXT:    br label %[[INNER_LOOP:.*]]
46; CHECK-VF4IC1:       [[INNER_LOOP]]:
47; CHECK-VF4IC1-NEXT:    [[RDX_INNER:%.*]] = phi i64 [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ], [ [[SELECT:%.*]], %[[INNER_LOOP]] ]
48; CHECK-VF4IC1-NEXT:    [[IV_INNER:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_INNER_NEXT:%.*]], %[[INNER_LOOP]] ]
49; CHECK-VF4IC1-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds i64, ptr [[TMP0]], i64 [[IV_INNER]]
50; CHECK-VF4IC1-NEXT:    [[TMP9:%.*]] = load i64, ptr [[ARRAYIDX2]], align 8
51; CHECK-VF4IC1-NEXT:    [[CMP:%.*]] = icmp eq i64 [[TMP9]], 3
52; CHECK-VF4IC1-NEXT:    [[SELECT]] = select i1 [[CMP]], i64 [[IV_OUTER]], i64 [[RDX_INNER]]
53; CHECK-VF4IC1-NEXT:    [[IV_INNER_NEXT]] = add nuw nsw i64 [[IV_INNER]], 1
54; CHECK-VF4IC1-NEXT:    [[EXITCOND_NOT_INNER:%.*]] = icmp eq i64 [[IV_INNER_NEXT]], [[N]]
55; CHECK-VF4IC1-NEXT:    br i1 [[EXITCOND_NOT_INNER]], label %[[OUTER_LOOP_EXIT]], label %[[INNER_LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
56; CHECK-VF4IC1:       [[OUTER_LOOP_EXIT]]:
57; CHECK-VF4IC1-NEXT:    [[SELECT_LCSSA]] = phi i64 [ [[SELECT]], %[[INNER_LOOP]] ], [ [[RDX_SELECT]], %[[MIDDLE_BLOCK]] ]
58; CHECK-VF4IC1-NEXT:    [[IV_OUTER_NEXT]] = add nuw nsw i64 [[IV_OUTER]], 1
59; CHECK-VF4IC1-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_OUTER_NEXT]], [[N]]
60; CHECK-VF4IC1-NEXT:    br i1 [[EXITCOND_NOT]], label %[[EXIT:.*]], label %[[OUTER_LOOP]]
61; CHECK-VF4IC1:       [[EXIT]]:
62; CHECK-VF4IC1-NEXT:    [[SELECT_LCSSA_LCSSA:%.*]] = phi i64 [ [[SELECT_LCSSA]], %[[OUTER_LOOP_EXIT]] ]
63; CHECK-VF4IC1-NEXT:    ret i64 [[SELECT_LCSSA_LCSSA]]
64;
65; CHECK-VF4IC4-LABEL: define i64 @select_iv_def_from_outer_loop(
66; CHECK-VF4IC4-SAME: ptr [[A:%.*]], i64 [[START:%.*]], i64 [[N:%.*]]) {
67; CHECK-VF4IC4-NEXT:  [[ENTRY:.*]]:
68; CHECK-VF4IC4-NEXT:    br label %[[OUTER_LOOP:.*]]
69; CHECK-VF4IC4:       [[OUTER_LOOP]]:
70; CHECK-VF4IC4-NEXT:    [[RDX_OUTER:%.*]] = phi i64 [ [[START]], %[[ENTRY]] ], [ [[SELECT_LCSSA:%.*]], %[[OUTER_LOOP_EXIT:.*]] ]
71; CHECK-VF4IC4-NEXT:    [[IV_OUTER:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[IV_OUTER_NEXT:%.*]], %[[OUTER_LOOP_EXIT]] ]
72; CHECK-VF4IC4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds ptr, ptr [[A]], i64 [[IV_OUTER]]
73; CHECK-VF4IC4-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[ARRAYIDX]], align 8
74; CHECK-VF4IC4-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 16
75; CHECK-VF4IC4-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
76; CHECK-VF4IC4:       [[VECTOR_PH]]:
77; CHECK-VF4IC4-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], 16
78; CHECK-VF4IC4-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
79; CHECK-VF4IC4-NEXT:    br label %[[VECTOR_BODY:.*]]
80; CHECK-VF4IC4:       [[VECTOR_BODY]]:
81; CHECK-VF4IC4-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
82; CHECK-VF4IC4-NEXT:    [[VEC_PHI:%.*]] = phi <4 x i1> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP11:%.*]], %[[VECTOR_BODY]] ]
83; CHECK-VF4IC4-NEXT:    [[VEC_PHI1:%.*]] = phi <4 x i1> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP12:%.*]], %[[VECTOR_BODY]] ]
84; CHECK-VF4IC4-NEXT:    [[VEC_PHI2:%.*]] = phi <4 x i1> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP13:%.*]], %[[VECTOR_BODY]] ]
85; CHECK-VF4IC4-NEXT:    [[VEC_PHI3:%.*]] = phi <4 x i1> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP14:%.*]], %[[VECTOR_BODY]] ]
86; CHECK-VF4IC4-NEXT:    [[TMP1:%.*]] = add i64 [[INDEX]], 0
87; CHECK-VF4IC4-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[TMP0]], i64 [[TMP1]]
88; CHECK-VF4IC4-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 0
89; CHECK-VF4IC4-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 4
90; CHECK-VF4IC4-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 8
91; CHECK-VF4IC4-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i64, ptr [[TMP2]], i32 12
92; CHECK-VF4IC4-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP3]], align 8
93; CHECK-VF4IC4-NEXT:    [[WIDE_LOAD4:%.*]] = load <4 x i64>, ptr [[TMP4]], align 8
94; CHECK-VF4IC4-NEXT:    [[WIDE_LOAD5:%.*]] = load <4 x i64>, ptr [[TMP5]], align 8
95; CHECK-VF4IC4-NEXT:    [[WIDE_LOAD6:%.*]] = load <4 x i64>, ptr [[TMP6]], align 8
96; CHECK-VF4IC4-NEXT:    [[TMP7:%.*]] = icmp eq <4 x i64> [[WIDE_LOAD]], splat (i64 3)
97; CHECK-VF4IC4-NEXT:    [[TMP8:%.*]] = icmp eq <4 x i64> [[WIDE_LOAD4]], splat (i64 3)
98; CHECK-VF4IC4-NEXT:    [[TMP9:%.*]] = icmp eq <4 x i64> [[WIDE_LOAD5]], splat (i64 3)
99; CHECK-VF4IC4-NEXT:    [[TMP10:%.*]] = icmp eq <4 x i64> [[WIDE_LOAD6]], splat (i64 3)
100; CHECK-VF4IC4-NEXT:    [[TMP11]] = or <4 x i1> [[VEC_PHI]], [[TMP7]]
101; CHECK-VF4IC4-NEXT:    [[TMP12]] = or <4 x i1> [[VEC_PHI1]], [[TMP8]]
102; CHECK-VF4IC4-NEXT:    [[TMP13]] = or <4 x i1> [[VEC_PHI2]], [[TMP9]]
103; CHECK-VF4IC4-NEXT:    [[TMP14]] = or <4 x i1> [[VEC_PHI3]], [[TMP10]]
104; CHECK-VF4IC4-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
105; CHECK-VF4IC4-NEXT:    [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
106; CHECK-VF4IC4-NEXT:    br i1 [[TMP15]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
107; CHECK-VF4IC4:       [[MIDDLE_BLOCK]]:
108; CHECK-VF4IC4-NEXT:    [[BIN_RDX:%.*]] = or <4 x i1> [[TMP12]], [[TMP11]]
109; CHECK-VF4IC4-NEXT:    [[BIN_RDX7:%.*]] = or <4 x i1> [[TMP13]], [[BIN_RDX]]
110; CHECK-VF4IC4-NEXT:    [[BIN_RDX8:%.*]] = or <4 x i1> [[TMP14]], [[BIN_RDX7]]
111; CHECK-VF4IC4-NEXT:    [[TMP16:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[BIN_RDX8]])
112; CHECK-VF4IC4-NEXT:    [[TMP17:%.*]] = freeze i1 [[TMP16]]
113; CHECK-VF4IC4-NEXT:    [[RDX_SELECT:%.*]] = select i1 [[TMP17]], i64 [[IV_OUTER]], i64 [[RDX_OUTER]]
114; CHECK-VF4IC4-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
115; CHECK-VF4IC4-NEXT:    br i1 [[CMP_N]], label %[[OUTER_LOOP_EXIT]], label %[[SCALAR_PH]]
116; CHECK-VF4IC4:       [[SCALAR_PH]]:
117; CHECK-VF4IC4-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i64 [ [[RDX_SELECT]], %[[MIDDLE_BLOCK]] ], [ [[RDX_OUTER]], %[[OUTER_LOOP]] ]
118; CHECK-VF4IC4-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[OUTER_LOOP]] ]
119; CHECK-VF4IC4-NEXT:    br label %[[INNER_LOOP:.*]]
120; CHECK-VF4IC4:       [[INNER_LOOP]]:
121; CHECK-VF4IC4-NEXT:    [[RDX_INNER:%.*]] = phi i64 [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ], [ [[SELECT:%.*]], %[[INNER_LOOP]] ]
122; CHECK-VF4IC4-NEXT:    [[IV_INNER:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_INNER_NEXT:%.*]], %[[INNER_LOOP]] ]
123; CHECK-VF4IC4-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds i64, ptr [[TMP0]], i64 [[IV_INNER]]
124; CHECK-VF4IC4-NEXT:    [[TMP18:%.*]] = load i64, ptr [[ARRAYIDX2]], align 8
125; CHECK-VF4IC4-NEXT:    [[CMP:%.*]] = icmp eq i64 [[TMP18]], 3
126; CHECK-VF4IC4-NEXT:    [[SELECT]] = select i1 [[CMP]], i64 [[IV_OUTER]], i64 [[RDX_INNER]]
127; CHECK-VF4IC4-NEXT:    [[IV_INNER_NEXT]] = add nuw nsw i64 [[IV_INNER]], 1
128; CHECK-VF4IC4-NEXT:    [[EXITCOND_NOT_INNER:%.*]] = icmp eq i64 [[IV_INNER_NEXT]], [[N]]
129; CHECK-VF4IC4-NEXT:    br i1 [[EXITCOND_NOT_INNER]], label %[[OUTER_LOOP_EXIT]], label %[[INNER_LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
130; CHECK-VF4IC4:       [[OUTER_LOOP_EXIT]]:
131; CHECK-VF4IC4-NEXT:    [[SELECT_LCSSA]] = phi i64 [ [[SELECT]], %[[INNER_LOOP]] ], [ [[RDX_SELECT]], %[[MIDDLE_BLOCK]] ]
132; CHECK-VF4IC4-NEXT:    [[IV_OUTER_NEXT]] = add nuw nsw i64 [[IV_OUTER]], 1
133; CHECK-VF4IC4-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_OUTER_NEXT]], [[N]]
134; CHECK-VF4IC4-NEXT:    br i1 [[EXITCOND_NOT]], label %[[EXIT:.*]], label %[[OUTER_LOOP]]
135; CHECK-VF4IC4:       [[EXIT]]:
136; CHECK-VF4IC4-NEXT:    [[SELECT_LCSSA_LCSSA:%.*]] = phi i64 [ [[SELECT_LCSSA]], %[[OUTER_LOOP_EXIT]] ]
137; CHECK-VF4IC4-NEXT:    ret i64 [[SELECT_LCSSA_LCSSA]]
138;
139; CHECK-VF1IC4-LABEL: define i64 @select_iv_def_from_outer_loop(
140; CHECK-VF1IC4-SAME: ptr [[A:%.*]], i64 [[START:%.*]], i64 [[N:%.*]]) {
141; CHECK-VF1IC4-NEXT:  [[ENTRY:.*]]:
142; CHECK-VF1IC4-NEXT:    br label %[[OUTER_LOOP:.*]]
143; CHECK-VF1IC4:       [[OUTER_LOOP]]:
144; CHECK-VF1IC4-NEXT:    [[RDX_OUTER:%.*]] = phi i64 [ [[START]], %[[ENTRY]] ], [ [[SELECT_LCSSA:%.*]], %[[OUTER_LOOP_EXIT:.*]] ]
145; CHECK-VF1IC4-NEXT:    [[IV_OUTER:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[IV_OUTER_NEXT:%.*]], %[[OUTER_LOOP_EXIT]] ]
146; CHECK-VF1IC4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds ptr, ptr [[A]], i64 [[IV_OUTER]]
147; CHECK-VF1IC4-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[ARRAYIDX]], align 8
148; CHECK-VF1IC4-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4
149; CHECK-VF1IC4-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
150; CHECK-VF1IC4:       [[VECTOR_PH]]:
151; CHECK-VF1IC4-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], 4
152; CHECK-VF1IC4-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
153; CHECK-VF1IC4-NEXT:    br label %[[VECTOR_BODY:.*]]
154; CHECK-VF1IC4:       [[VECTOR_BODY]]:
155; CHECK-VF1IC4-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
156; CHECK-VF1IC4-NEXT:    [[VEC_PHI:%.*]] = phi i1 [ false, %[[VECTOR_PH]] ], [ [[TMP17:%.*]], %[[VECTOR_BODY]] ]
157; CHECK-VF1IC4-NEXT:    [[VEC_PHI1:%.*]] = phi i1 [ false, %[[VECTOR_PH]] ], [ [[TMP18:%.*]], %[[VECTOR_BODY]] ]
158; CHECK-VF1IC4-NEXT:    [[VEC_PHI2:%.*]] = phi i1 [ false, %[[VECTOR_PH]] ], [ [[TMP19:%.*]], %[[VECTOR_BODY]] ]
159; CHECK-VF1IC4-NEXT:    [[VEC_PHI3:%.*]] = phi i1 [ false, %[[VECTOR_PH]] ], [ [[TMP20:%.*]], %[[VECTOR_BODY]] ]
160; CHECK-VF1IC4-NEXT:    [[TMP1:%.*]] = add i64 [[INDEX]], 0
161; CHECK-VF1IC4-NEXT:    [[TMP2:%.*]] = add i64 [[INDEX]], 1
162; CHECK-VF1IC4-NEXT:    [[TMP3:%.*]] = add i64 [[INDEX]], 2
163; CHECK-VF1IC4-NEXT:    [[TMP4:%.*]] = add i64 [[INDEX]], 3
164; CHECK-VF1IC4-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i64, ptr [[TMP0]], i64 [[TMP1]]
165; CHECK-VF1IC4-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i64, ptr [[TMP0]], i64 [[TMP2]]
166; CHECK-VF1IC4-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i64, ptr [[TMP0]], i64 [[TMP3]]
167; CHECK-VF1IC4-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i64, ptr [[TMP0]], i64 [[TMP4]]
168; CHECK-VF1IC4-NEXT:    [[TMP9:%.*]] = load i64, ptr [[TMP5]], align 8
169; CHECK-VF1IC4-NEXT:    [[TMP10:%.*]] = load i64, ptr [[TMP6]], align 8
170; CHECK-VF1IC4-NEXT:    [[TMP11:%.*]] = load i64, ptr [[TMP7]], align 8
171; CHECK-VF1IC4-NEXT:    [[TMP12:%.*]] = load i64, ptr [[TMP8]], align 8
172; CHECK-VF1IC4-NEXT:    [[TMP13:%.*]] = icmp eq i64 [[TMP9]], 3
173; CHECK-VF1IC4-NEXT:    [[TMP14:%.*]] = icmp eq i64 [[TMP10]], 3
174; CHECK-VF1IC4-NEXT:    [[TMP15:%.*]] = icmp eq i64 [[TMP11]], 3
175; CHECK-VF1IC4-NEXT:    [[TMP16:%.*]] = icmp eq i64 [[TMP12]], 3
176; CHECK-VF1IC4-NEXT:    [[TMP17]] = or i1 [[VEC_PHI]], [[TMP13]]
177; CHECK-VF1IC4-NEXT:    [[TMP18]] = or i1 [[VEC_PHI1]], [[TMP14]]
178; CHECK-VF1IC4-NEXT:    [[TMP19]] = or i1 [[VEC_PHI2]], [[TMP15]]
179; CHECK-VF1IC4-NEXT:    [[TMP20]] = or i1 [[VEC_PHI3]], [[TMP16]]
180; CHECK-VF1IC4-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
181; CHECK-VF1IC4-NEXT:    [[TMP21:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
182; CHECK-VF1IC4-NEXT:    br i1 [[TMP21]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
183; CHECK-VF1IC4:       [[MIDDLE_BLOCK]]:
184; CHECK-VF1IC4-NEXT:    [[BIN_RDX:%.*]] = or i1 [[TMP18]], [[TMP17]]
185; CHECK-VF1IC4-NEXT:    [[BIN_RDX4:%.*]] = or i1 [[TMP19]], [[BIN_RDX]]
186; CHECK-VF1IC4-NEXT:    [[BIN_RDX5:%.*]] = or i1 [[TMP20]], [[BIN_RDX4]]
187; CHECK-VF1IC4-NEXT:    [[TMP22:%.*]] = freeze i1 [[BIN_RDX5]]
188; CHECK-VF1IC4-NEXT:    [[RDX_SELECT:%.*]] = select i1 [[TMP22]], i64 [[IV_OUTER]], i64 [[RDX_OUTER]]
189; CHECK-VF1IC4-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
190; CHECK-VF1IC4-NEXT:    br i1 [[CMP_N]], label %[[OUTER_LOOP_EXIT]], label %[[SCALAR_PH]]
191; CHECK-VF1IC4:       [[SCALAR_PH]]:
192; CHECK-VF1IC4-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i64 [ [[RDX_SELECT]], %[[MIDDLE_BLOCK]] ], [ [[RDX_OUTER]], %[[OUTER_LOOP]] ]
193; CHECK-VF1IC4-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[OUTER_LOOP]] ]
194; CHECK-VF1IC4-NEXT:    br label %[[INNER_LOOP:.*]]
195; CHECK-VF1IC4:       [[INNER_LOOP]]:
196; CHECK-VF1IC4-NEXT:    [[RDX_INNER:%.*]] = phi i64 [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ], [ [[SELECT:%.*]], %[[INNER_LOOP]] ]
197; CHECK-VF1IC4-NEXT:    [[IV_INNER:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_INNER_NEXT:%.*]], %[[INNER_LOOP]] ]
198; CHECK-VF1IC4-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds i64, ptr [[TMP0]], i64 [[IV_INNER]]
199; CHECK-VF1IC4-NEXT:    [[TMP23:%.*]] = load i64, ptr [[ARRAYIDX2]], align 8
200; CHECK-VF1IC4-NEXT:    [[CMP:%.*]] = icmp eq i64 [[TMP23]], 3
201; CHECK-VF1IC4-NEXT:    [[SELECT]] = select i1 [[CMP]], i64 [[IV_OUTER]], i64 [[RDX_INNER]]
202; CHECK-VF1IC4-NEXT:    [[IV_INNER_NEXT]] = add nuw nsw i64 [[IV_INNER]], 1
203; CHECK-VF1IC4-NEXT:    [[EXITCOND_NOT_INNER:%.*]] = icmp eq i64 [[IV_INNER_NEXT]], [[N]]
204; CHECK-VF1IC4-NEXT:    br i1 [[EXITCOND_NOT_INNER]], label %[[OUTER_LOOP_EXIT]], label %[[INNER_LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
205; CHECK-VF1IC4:       [[OUTER_LOOP_EXIT]]:
206; CHECK-VF1IC4-NEXT:    [[SELECT_LCSSA]] = phi i64 [ [[SELECT]], %[[INNER_LOOP]] ], [ [[RDX_SELECT]], %[[MIDDLE_BLOCK]] ]
207; CHECK-VF1IC4-NEXT:    [[IV_OUTER_NEXT]] = add nuw nsw i64 [[IV_OUTER]], 1
208; CHECK-VF1IC4-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_OUTER_NEXT]], [[N]]
209; CHECK-VF1IC4-NEXT:    br i1 [[EXITCOND_NOT]], label %[[EXIT:.*]], label %[[OUTER_LOOP]]
210; CHECK-VF1IC4:       [[EXIT]]:
211; CHECK-VF1IC4-NEXT:    [[SELECT_LCSSA_LCSSA:%.*]] = phi i64 [ [[SELECT_LCSSA]], %[[OUTER_LOOP_EXIT]] ]
212; CHECK-VF1IC4-NEXT:    ret i64 [[SELECT_LCSSA_LCSSA]]
213;
214entry:
215  br label %outer.loop
216
217outer.loop:
218  %rdx.outer = phi i64 [ %start, %entry ], [ %select, %outer.loop.exit ]
219  %iv.outer = phi i64 [ 0, %entry ], [ %iv.outer.next, %outer.loop.exit ]
220  %arrayidx = getelementptr inbounds ptr, ptr %a, i64 %iv.outer
221  %0 = load ptr, ptr %arrayidx, align 8
222  br label %inner.loop
223
224inner.loop:
225  %rdx.inner = phi i64 [ %rdx.outer, %outer.loop ], [ %select, %inner.loop ]
226  %iv.inner = phi i64 [ 0, %outer.loop ], [ %iv.inner.next, %inner.loop ]
227  %arrayidx2 = getelementptr inbounds i64, ptr %0, i64 %iv.inner
228  %1 = load i64, ptr %arrayidx2, align 8
229  %cmp = icmp eq i64 %1, 3
230  %select = select i1 %cmp, i64 %iv.outer, i64 %rdx.inner
231  %iv.inner.next = add nuw nsw i64 %iv.inner, 1
232  %exitcond.not.inner = icmp eq i64 %iv.inner.next, %n
233  br i1 %exitcond.not.inner, label %outer.loop.exit, label %inner.loop
234
235outer.loop.exit:
236  %iv.outer.next = add nuw nsw i64 %iv.outer, 1
237  %exitcond.not = icmp eq i64 %iv.outer.next, %n
238  br i1 %exitcond.not, label %exit, label %outer.loop
239
240exit:
241  ret i64 %select
242}
243;.
244; CHECK-VF4IC1: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
245; CHECK-VF4IC1: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
246; CHECK-VF4IC1: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
247; CHECK-VF4IC1: [[LOOP3]] = distinct !{[[LOOP3]], [[META2]], [[META1]]}
248;.
249; CHECK-VF4IC4: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
250; CHECK-VF4IC4: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
251; CHECK-VF4IC4: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
252; CHECK-VF4IC4: [[LOOP3]] = distinct !{[[LOOP3]], [[META2]], [[META1]]}
253;.
254; CHECK-VF1IC4: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
255; CHECK-VF1IC4: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
256; CHECK-VF1IC4: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
257; CHECK-VF1IC4: [[LOOP3]] = distinct !{[[LOOP3]], [[META1]]}
258;.
259