xref: /llvm-project/llvm/test/Transforms/LoopVectorize/X86/scatter_crash.ll (revision f0d5104c944b329c479802788571ed6df41e0e86) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -passes=loop-vectorize -S | FileCheck %s
3target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
4target triple = "x86_64-apple-macosx10.11.0"
5
6; This test checks vector GEP before scatter.
7; The code bellow crashed due to destroyed SSA while incorrect vectorization of
8; the GEP.
9
10@d = global [10 x [10 x i32]] zeroinitializer, align 16
11@c = external global i32, align 4
12@a = external global i32, align 4
13@b = external global i64, align 8
14
15; Function Attrs: norecurse nounwind ssp uwtable
16define void @_Z3fn1v() #0 {
17; CHECK-LABEL: @_Z3fn1v(
18; CHECK-NEXT:  entry:
19; CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr @c, align 4
20; CHECK-NEXT:    [[CMP34:%.*]] = icmp sgt i32 [[TMP0]], 8
21; CHECK-NEXT:    br i1 [[CMP34]], label [[FOR_BODY_LR_PH:%.*]], label [[FOR_COND_CLEANUP:%.*]]
22; CHECK:       for.body.lr.ph:
23; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr @a, align 4
24; CHECK-NEXT:    [[TOBOOL:%.*]] = icmp eq i32 [[TMP1]], 0
25; CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr @b, align 8
26; CHECK-NEXT:    [[MUL:%.*]] = mul i64 [[TMP2]], 4063299859190
27; CHECK-NEXT:    [[TOBOOL6:%.*]] = icmp eq i64 [[MUL]], 0
28; CHECK-NEXT:    [[TMP3:%.*]] = sext i32 [[TMP0]] to i64
29; CHECK-NEXT:    br i1 [[TOBOOL]], label [[ITER_CHECK27:%.*]], label [[ITER_CHECK:%.*]]
30; CHECK:       iter.check:
31; CHECK-NEXT:    [[TMP4:%.*]] = add nsw i64 [[TMP3]], -9
32; CHECK-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP4]], 1
33; CHECK-NEXT:    [[TMP6:%.*]] = add nuw i64 [[TMP5]], 1
34; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP6]], 8
35; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
36; CHECK:       vector.main.loop.iter.check:
37; CHECK-NEXT:    [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[TMP6]], 16
38; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
39; CHECK:       vector.ph:
40; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[TMP6]], 16
41; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[TMP6]], [[N_MOD_VF]]
42; CHECK-NEXT:    [[TMP7:%.*]] = mul i64 [[N_VEC]], 2
43; CHECK-NEXT:    [[IND_END:%.*]] = add i64 8, [[TMP7]]
44; CHECK-NEXT:    [[IND_END4:%.*]] = mul i64 [[N_VEC]], 2
45; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
46; CHECK:       vector.body:
47; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
48; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <16 x i64> [ <i64 8, i64 10, i64 12, i64 14, i64 16, i64 18, i64 20, i64 22, i64 24, i64 26, i64 28, i64 30, i64 32, i64 34, i64 36, i64 38>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
49; CHECK-NEXT:    [[VEC_IND3:%.*]] = phi <16 x i64> [ <i64 0, i64 2, i64 4, i64 6, i64 8, i64 10, i64 12, i64 14, i64 16, i64 18, i64 20, i64 22, i64 24, i64 26, i64 28, i64 30>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT3:%.*]], [[VECTOR_BODY]] ]
50; CHECK-NEXT:    [[TMP10:%.*]] = sub nsw <16 x i64> splat (i64 8), [[VEC_IND]]
51; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr @d, i64 0, <16 x i64> [[VEC_IND]]
52; CHECK-NEXT:    [[TMP12:%.*]] = add nsw <16 x i64> [[TMP10]], [[VEC_IND3]]
53; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [10 x i32], <16 x ptr> [[TMP11]], <16 x i64> [[TMP12]], i64 0
54; CHECK-NEXT:    call void @llvm.masked.scatter.v16i32.v16p0(<16 x i32> splat (i32 8), <16 x ptr> [[TMP13]], i32 16, <16 x i1> splat (i1 true))
55; CHECK-NEXT:    [[TMP14:%.*]] = or disjoint <16 x i64> [[VEC_IND3]], splat (i64 1)
56; CHECK-NEXT:    [[TMP15:%.*]] = add nsw <16 x i64> [[TMP10]], [[TMP14]]
57; CHECK-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [10 x i32], <16 x ptr> [[TMP11]], <16 x i64> [[TMP15]], i64 0
58; CHECK-NEXT:    call void @llvm.masked.scatter.v16i32.v16p0(<16 x i32> splat (i32 8), <16 x ptr> [[TMP16]], i32 8, <16 x i1> splat (i1 true))
59; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
60; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <16 x i64> [[VEC_IND]], splat (i64 32)
61; CHECK-NEXT:    [[VEC_IND_NEXT3]] = add <16 x i64> [[VEC_IND3]], splat (i64 32)
62; CHECK-NEXT:    [[TMP63:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
63; CHECK-NEXT:    br i1 [[TMP63]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
64; CHECK:       middle.block:
65; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP6]], [[N_VEC]]
66; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_COND_CLEANUP_LOOPEXIT99:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
67; CHECK:       vec.epilog.iter.check:
68; CHECK-NEXT:    [[TMP64:%.*]] = mul i64 [[N_VEC]], 2
69; CHECK-NEXT:    [[IND_END9:%.*]] = add i64 8, [[TMP64]]
70; CHECK-NEXT:    [[IND_END12:%.*]] = mul i64 [[N_VEC]], 2
71; CHECK-NEXT:    [[N_VEC_REMAINING:%.*]] = sub i64 [[TMP6]], [[N_VEC]]
72; CHECK-NEXT:    [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 8
73; CHECK-NEXT:    br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
74; CHECK:       vec.epilog.ph:
75; CHECK-NEXT:    [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
76; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ 8, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
77; CHECK-NEXT:    [[BC_RESUME_VAL5:%.*]] = phi i64 [ [[IND_END4]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
78; CHECK-NEXT:    [[N_MOD_VF6:%.*]] = urem i64 [[TMP6]], 8
79; CHECK-NEXT:    [[N_VEC7:%.*]] = sub i64 [[TMP6]], [[N_MOD_VF6]]
80; CHECK-NEXT:    [[TMP17:%.*]] = mul i64 [[N_VEC7]], 2
81; CHECK-NEXT:    [[IND_END8:%.*]] = add i64 8, [[TMP17]]
82; CHECK-NEXT:    [[IND_END11:%.*]] = mul i64 [[N_VEC7]], 2
83; CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <8 x i64> poison, i64 [[BC_RESUME_VAL]], i64 0
84; CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <8 x i64> [[DOTSPLATINSERT]], <8 x i64> poison, <8 x i32> zeroinitializer
85; CHECK-NEXT:    [[INDUCTION:%.*]] = add <8 x i64> [[DOTSPLAT]], <i64 0, i64 2, i64 4, i64 6, i64 8, i64 10, i64 12, i64 14>
86; CHECK-NEXT:    [[DOTSPLATINSERT17:%.*]] = insertelement <8 x i64> poison, i64 [[BC_RESUME_VAL5]], i64 0
87; CHECK-NEXT:    [[DOTSPLAT18:%.*]] = shufflevector <8 x i64> [[DOTSPLATINSERT17]], <8 x i64> poison, <8 x i32> zeroinitializer
88; CHECK-NEXT:    [[INDUCTION19:%.*]] = add <8 x i64> [[DOTSPLAT18]], <i64 0, i64 2, i64 4, i64 6, i64 8, i64 10, i64 12, i64 14>
89; CHECK-NEXT:    br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
90; CHECK:       vec.epilog.vector.body:
91; CHECK-NEXT:    [[INDEX14:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT22:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
92; CHECK-NEXT:    [[VEC_IND15:%.*]] = phi <8 x i64> [ [[INDUCTION]], [[VEC_EPILOG_PH]] ], [ [[VEC_IND_NEXT16:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
93; CHECK-NEXT:    [[VEC_IND20:%.*]] = phi <8 x i64> [ [[INDUCTION19]], [[VEC_EPILOG_PH]] ], [ [[VEC_IND_NEXT21:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
94; CHECK-NEXT:    [[TMP18:%.*]] = sub nsw <8 x i64> splat (i64 8), [[VEC_IND15]]
95; CHECK-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr @d, i64 0, <8 x i64> [[VEC_IND15]]
96; CHECK-NEXT:    [[TMP20:%.*]] = add nsw <8 x i64> [[TMP18]], [[VEC_IND20]]
97; CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [10 x i32], <8 x ptr> [[TMP19]], <8 x i64> [[TMP20]], i64 0
98; CHECK-NEXT:    call void @llvm.masked.scatter.v8i32.v8p0(<8 x i32> splat (i32 8), <8 x ptr> [[TMP21]], i32 16, <8 x i1> splat (i1 true))
99; CHECK-NEXT:    [[TMP22:%.*]] = or disjoint <8 x i64> [[VEC_IND20]], splat (i64 1)
100; CHECK-NEXT:    [[TMP23:%.*]] = add nsw <8 x i64> [[TMP18]], [[TMP22]]
101; CHECK-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [10 x i32], <8 x ptr> [[TMP19]], <8 x i64> [[TMP23]], i64 0
102; CHECK-NEXT:    call void @llvm.masked.scatter.v8i32.v8p0(<8 x i32> splat (i32 8), <8 x ptr> [[TMP24]], i32 8, <8 x i1> splat (i1 true))
103; CHECK-NEXT:    [[INDEX_NEXT22]] = add nuw i64 [[INDEX14]], 8
104; CHECK-NEXT:    [[VEC_IND_NEXT16]] = add <8 x i64> [[VEC_IND15]], splat (i64 16)
105; CHECK-NEXT:    [[VEC_IND_NEXT21]] = add <8 x i64> [[VEC_IND20]], splat (i64 16)
106; CHECK-NEXT:    [[TMP25:%.*]] = icmp eq i64 [[INDEX_NEXT22]], [[N_VEC7]]
107; CHECK-NEXT:    br i1 [[TMP25]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
108; CHECK:       vec.epilog.middle.block:
109; CHECK-NEXT:    [[CMP_N23:%.*]] = icmp eq i64 [[TMP6]], [[N_VEC7]]
110; CHECK-NEXT:    br i1 [[CMP_N23]], label [[FOR_COND_CLEANUP_LOOPEXIT99]], label [[VEC_EPILOG_SCALAR_PH]]
111; CHECK:       vec.epilog.scalar.ph:
112; CHECK-NEXT:    [[BC_RESUME_VAL10:%.*]] = phi i64 [ [[IND_END8]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 8, [[ITER_CHECK]] ], [ [[IND_END9]], [[VEC_EPILOG_ITER_CHECK]] ]
113; CHECK-NEXT:    [[BC_RESUME_VAL13:%.*]] = phi i64 [ [[IND_END11]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 0, [[ITER_CHECK]] ], [ [[IND_END12]], [[VEC_EPILOG_ITER_CHECK]] ]
114; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
115; CHECK:       iter.check23:
116; CHECK-NEXT:    [[TMP26:%.*]] = add nsw i64 [[TMP3]], -9
117; CHECK-NEXT:    [[TMP27:%.*]] = lshr i64 [[TMP26]], 1
118; CHECK-NEXT:    [[TMP28:%.*]] = add nuw i64 [[TMP27]], 1
119; CHECK-NEXT:    [[MIN_ITERS_CHECK25:%.*]] = icmp ult i64 [[TMP28]], 8
120; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK25]], label [[VEC_EPILOG_SCALAR_PH41:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK25:%.*]]
121; CHECK:       vector.main.loop.iter.check25:
122; CHECK-NEXT:    [[MIN_ITERS_CHECK28:%.*]] = icmp ult i64 [[TMP28]], 16
123; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK28]], label [[VEC_EPILOG_PH42:%.*]], label [[VECTOR_PH30:%.*]]
124; CHECK:       vector.ph26:
125; CHECK-NEXT:    [[N_MOD_VF31:%.*]] = urem i64 [[TMP28]], 16
126; CHECK-NEXT:    [[N_VEC32:%.*]] = sub i64 [[TMP28]], [[N_MOD_VF31]]
127; CHECK-NEXT:    [[TMP29:%.*]] = mul i64 [[N_VEC32]], 2
128; CHECK-NEXT:    [[IND_END41:%.*]] = add i64 8, [[TMP29]]
129; CHECK-NEXT:    [[IND_END43:%.*]] = mul i64 [[N_VEC32]], 2
130; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i1> poison, i1 [[TOBOOL6]], i64 0
131; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i1> [[BROADCAST_SPLATINSERT]], <16 x i1> poison, <16 x i32> zeroinitializer
132; CHECK-NEXT:    [[TMP34:%.*]] = xor <16 x i1> [[BROADCAST_SPLAT]], splat (i1 true)
133; CHECK-NEXT:    br label [[VECTOR_BODY29:%.*]]
134; CHECK:       vector.body29:
135; CHECK-NEXT:    [[INDEX34:%.*]] = phi i64 [ 0, [[VECTOR_PH30]] ], [ [[INDEX_NEXT39:%.*]], [[VECTOR_BODY29]] ]
136; CHECK-NEXT:    [[VEC_IND35:%.*]] = phi <16 x i64> [ <i64 8, i64 10, i64 12, i64 14, i64 16, i64 18, i64 20, i64 22, i64 24, i64 26, i64 28, i64 30, i64 32, i64 34, i64 36, i64 38>, [[VECTOR_PH30]] ], [ [[VEC_IND_NEXT36:%.*]], [[VECTOR_BODY29]] ]
137; CHECK-NEXT:    [[VEC_IND37:%.*]] = phi <16 x i64> [ <i64 0, i64 2, i64 4, i64 6, i64 8, i64 10, i64 12, i64 14, i64 16, i64 18, i64 20, i64 22, i64 24, i64 26, i64 28, i64 30>, [[VECTOR_PH30]] ], [ [[VEC_IND_NEXT38:%.*]], [[VECTOR_BODY29]] ]
138; CHECK-NEXT:    [[TMP30:%.*]] = sub nsw <16 x i64> splat (i64 8), [[VEC_IND35]]
139; CHECK-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr @d, i64 0, <16 x i64> [[VEC_IND35]]
140; CHECK-NEXT:    [[TMP32:%.*]] = add nsw <16 x i64> [[TMP30]], [[VEC_IND37]]
141; CHECK-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [10 x i32], <16 x ptr> [[TMP31]], <16 x i64> [[TMP32]], i64 0
142; CHECK-NEXT:    call void @llvm.masked.scatter.v16i32.v16p0(<16 x i32> splat (i32 8), <16 x ptr> [[TMP33]], i32 16, <16 x i1> [[TMP34]])
143; CHECK-NEXT:    [[TMP35:%.*]] = or disjoint <16 x i64> [[VEC_IND37]], splat (i64 1)
144; CHECK-NEXT:    [[TMP36:%.*]] = add nsw <16 x i64> [[TMP30]], [[TMP35]]
145; CHECK-NEXT:    [[TMP37:%.*]] = getelementptr inbounds [10 x i32], <16 x ptr> [[TMP31]], <16 x i64> [[TMP36]], i64 0
146; CHECK-NEXT:    call void @llvm.masked.scatter.v16i32.v16p0(<16 x i32> splat (i32 8), <16 x ptr> [[TMP37]], i32 8, <16 x i1> [[TMP34]])
147; CHECK-NEXT:    call void @llvm.masked.scatter.v16i32.v16p0(<16 x i32> splat (i32 7), <16 x ptr> [[TMP33]], i32 16, <16 x i1> [[BROADCAST_SPLAT]])
148; CHECK-NEXT:    [[TMP38:%.*]] = or disjoint <16 x i64> [[VEC_IND37]], splat (i64 1)
149; CHECK-NEXT:    [[TMP39:%.*]] = add nsw <16 x i64> [[TMP30]], [[TMP38]]
150; CHECK-NEXT:    [[TMP40:%.*]] = getelementptr inbounds [10 x i32], <16 x ptr> [[TMP31]], <16 x i64> [[TMP39]], i64 0
151; CHECK-NEXT:    call void @llvm.masked.scatter.v16i32.v16p0(<16 x i32> splat (i32 7), <16 x ptr> [[TMP40]], i32 8, <16 x i1> [[BROADCAST_SPLAT]])
152; CHECK-NEXT:    [[INDEX_NEXT39]] = add nuw i64 [[INDEX34]], 16
153; CHECK-NEXT:    [[VEC_IND_NEXT36]] = add <16 x i64> [[VEC_IND35]], splat (i64 32)
154; CHECK-NEXT:    [[VEC_IND_NEXT38]] = add <16 x i64> [[VEC_IND37]], splat (i64 32)
155; CHECK-NEXT:    [[TMP41:%.*]] = icmp eq i64 [[INDEX_NEXT39]], [[N_VEC32]]
156; CHECK-NEXT:    br i1 [[TMP41]], label [[MIDDLE_BLOCK20:%.*]], label [[VECTOR_BODY29]], !llvm.loop [[LOOP4:![0-9]+]]
157; CHECK:       middle.block20:
158; CHECK-NEXT:    [[CMP_N40:%.*]] = icmp eq i64 [[TMP28]], [[N_VEC32]]
159; CHECK-NEXT:    br i1 [[CMP_N40]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK43:%.*]]
160; CHECK:       vec.epilog.iter.check43:
161; CHECK-NEXT:    [[TMP42:%.*]] = mul i64 [[N_VEC32]], 2
162; CHECK-NEXT:    [[IND_END55:%.*]] = add i64 8, [[TMP42]]
163; CHECK-NEXT:    [[IND_END58:%.*]] = mul i64 [[N_VEC32]], 2
164; CHECK-NEXT:    [[N_VEC_REMAINING49:%.*]] = sub i64 [[TMP28]], [[N_VEC32]]
165; CHECK-NEXT:    [[MIN_EPILOG_ITERS_CHECK50:%.*]] = icmp ult i64 [[N_VEC_REMAINING49]], 8
166; CHECK-NEXT:    br i1 [[MIN_EPILOG_ITERS_CHECK50]], label [[VEC_EPILOG_SCALAR_PH41]], label [[VEC_EPILOG_PH42]]
167; CHECK:       vec.epilog.ph42:
168; CHECK-NEXT:    [[VEC_EPILOG_RESUME_VAL51:%.*]] = phi i64 [ [[N_VEC32]], [[VEC_EPILOG_ITER_CHECK43]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK25]] ]
169; CHECK-NEXT:    [[BC_RESUME_VAL42:%.*]] = phi i64 [ [[IND_END41]], [[VEC_EPILOG_ITER_CHECK43]] ], [ 8, [[VECTOR_MAIN_LOOP_ITER_CHECK25]] ]
170; CHECK-NEXT:    [[BC_RESUME_VAL44:%.*]] = phi i64 [ [[IND_END43]], [[VEC_EPILOG_ITER_CHECK43]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK25]] ]
171; CHECK-NEXT:    [[N_MOD_VF52:%.*]] = urem i64 [[TMP28]], 8
172; CHECK-NEXT:    [[N_VEC53:%.*]] = sub i64 [[TMP28]], [[N_MOD_VF52]]
173; CHECK-NEXT:    [[TMP43:%.*]] = mul i64 [[N_VEC53]], 2
174; CHECK-NEXT:    [[IND_END54:%.*]] = add i64 8, [[TMP43]]
175; CHECK-NEXT:    [[IND_END57:%.*]] = mul i64 [[N_VEC53]], 2
176; CHECK-NEXT:    [[BROADCAST_SPLATINSERT50:%.*]] = insertelement <8 x i1> poison, i1 [[TOBOOL6]], i64 0
177; CHECK-NEXT:    [[BROADCAST_SPLAT73:%.*]] = shufflevector <8 x i1> [[BROADCAST_SPLATINSERT50]], <8 x i1> poison, <8 x i32> zeroinitializer
178; CHECK-NEXT:    [[TMP48:%.*]] = xor <8 x i1> [[BROADCAST_SPLAT73]], splat (i1 true)
179; CHECK-NEXT:    [[DOTSPLATINSERT62:%.*]] = insertelement <8 x i64> poison, i64 [[BC_RESUME_VAL42]], i64 0
180; CHECK-NEXT:    [[DOTSPLAT63:%.*]] = shufflevector <8 x i64> [[DOTSPLATINSERT62]], <8 x i64> poison, <8 x i32> zeroinitializer
181; CHECK-NEXT:    [[INDUCTION64:%.*]] = add <8 x i64> [[DOTSPLAT63]], <i64 0, i64 2, i64 4, i64 6, i64 8, i64 10, i64 12, i64 14>
182; CHECK-NEXT:    [[DOTSPLATINSERT67:%.*]] = insertelement <8 x i64> poison, i64 [[BC_RESUME_VAL44]], i64 0
183; CHECK-NEXT:    [[DOTSPLAT68:%.*]] = shufflevector <8 x i64> [[DOTSPLATINSERT67]], <8 x i64> poison, <8 x i32> zeroinitializer
184; CHECK-NEXT:    [[INDUCTION69:%.*]] = add <8 x i64> [[DOTSPLAT68]], <i64 0, i64 2, i64 4, i64 6, i64 8, i64 10, i64 12, i64 14>
185; CHECK-NEXT:    br label [[VEC_EPILOG_VECTOR_BODY50:%.*]]
186; CHECK:       vec.epilog.vector.body52:
187; CHECK-NEXT:    [[INDEX61:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL51]], [[VEC_EPILOG_PH42]] ], [ [[INDEX_NEXT74:%.*]], [[VEC_EPILOG_VECTOR_BODY50]] ]
188; CHECK-NEXT:    [[VEC_IND65:%.*]] = phi <8 x i64> [ [[INDUCTION64]], [[VEC_EPILOG_PH42]] ], [ [[VEC_IND_NEXT66:%.*]], [[VEC_EPILOG_VECTOR_BODY50]] ]
189; CHECK-NEXT:    [[VEC_IND70:%.*]] = phi <8 x i64> [ [[INDUCTION69]], [[VEC_EPILOG_PH42]] ], [ [[VEC_IND_NEXT71:%.*]], [[VEC_EPILOG_VECTOR_BODY50]] ]
190; CHECK-NEXT:    [[TMP44:%.*]] = sub nsw <8 x i64> splat (i64 8), [[VEC_IND65]]
191; CHECK-NEXT:    [[TMP45:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr @d, i64 0, <8 x i64> [[VEC_IND65]]
192; CHECK-NEXT:    [[TMP46:%.*]] = add nsw <8 x i64> [[TMP44]], [[VEC_IND70]]
193; CHECK-NEXT:    [[TMP47:%.*]] = getelementptr inbounds [10 x i32], <8 x ptr> [[TMP45]], <8 x i64> [[TMP46]], i64 0
194; CHECK-NEXT:    call void @llvm.masked.scatter.v8i32.v8p0(<8 x i32> splat (i32 8), <8 x ptr> [[TMP47]], i32 16, <8 x i1> [[TMP48]])
195; CHECK-NEXT:    [[TMP49:%.*]] = or disjoint <8 x i64> [[VEC_IND70]], splat (i64 1)
196; CHECK-NEXT:    [[TMP50:%.*]] = add nsw <8 x i64> [[TMP44]], [[TMP49]]
197; CHECK-NEXT:    [[TMP51:%.*]] = getelementptr inbounds [10 x i32], <8 x ptr> [[TMP45]], <8 x i64> [[TMP50]], i64 0
198; CHECK-NEXT:    call void @llvm.masked.scatter.v8i32.v8p0(<8 x i32> splat (i32 8), <8 x ptr> [[TMP51]], i32 8, <8 x i1> [[TMP48]])
199; CHECK-NEXT:    call void @llvm.masked.scatter.v8i32.v8p0(<8 x i32> splat (i32 7), <8 x ptr> [[TMP47]], i32 16, <8 x i1> [[BROADCAST_SPLAT73]])
200; CHECK-NEXT:    [[TMP52:%.*]] = or disjoint <8 x i64> [[VEC_IND70]], splat (i64 1)
201; CHECK-NEXT:    [[TMP53:%.*]] = add nsw <8 x i64> [[TMP44]], [[TMP52]]
202; CHECK-NEXT:    [[TMP54:%.*]] = getelementptr inbounds [10 x i32], <8 x ptr> [[TMP45]], <8 x i64> [[TMP53]], i64 0
203; CHECK-NEXT:    call void @llvm.masked.scatter.v8i32.v8p0(<8 x i32> splat (i32 7), <8 x ptr> [[TMP54]], i32 8, <8 x i1> [[BROADCAST_SPLAT73]])
204; CHECK-NEXT:    [[INDEX_NEXT74]] = add nuw i64 [[INDEX61]], 8
205; CHECK-NEXT:    [[VEC_IND_NEXT66]] = add <8 x i64> [[VEC_IND65]], splat (i64 16)
206; CHECK-NEXT:    [[VEC_IND_NEXT71]] = add <8 x i64> [[VEC_IND70]], splat (i64 16)
207; CHECK-NEXT:    [[TMP55:%.*]] = icmp eq i64 [[INDEX_NEXT74]], [[N_VEC53]]
208; CHECK-NEXT:    br i1 [[TMP55]], label [[VEC_EPILOG_MIDDLE_BLOCK40:%.*]], label [[VEC_EPILOG_VECTOR_BODY50]], !llvm.loop [[LOOP5:![0-9]+]]
209; CHECK:       vec.epilog.middle.block40:
210; CHECK-NEXT:    [[CMP_N65:%.*]] = icmp eq i64 [[TMP28]], [[N_VEC53]]
211; CHECK-NEXT:    br i1 [[CMP_N65]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[VEC_EPILOG_SCALAR_PH41]]
212; CHECK:       vec.epilog.scalar.ph41:
213; CHECK-NEXT:    [[BC_RESUME_VAL56:%.*]] = phi i64 [ [[IND_END54]], [[VEC_EPILOG_MIDDLE_BLOCK40]] ], [ 8, [[ITER_CHECK27]] ], [ [[IND_END55]], [[VEC_EPILOG_ITER_CHECK43]] ]
214; CHECK-NEXT:    [[BC_RESUME_VAL67:%.*]] = phi i64 [ [[IND_END57]], [[VEC_EPILOG_MIDDLE_BLOCK40]] ], [ 0, [[ITER_CHECK27]] ], [ [[IND_END58]], [[VEC_EPILOG_ITER_CHECK43]] ]
215; CHECK-NEXT:    br label [[FOR_BODY_US:%.*]]
216; CHECK:       for.body.us:
217; CHECK-NEXT:    [[INDVARS_IV78:%.*]] = phi i64 [ [[INDVARS_IV_NEXT79:%.*]], [[FOR_COND_CLEANUP4_US_LCSSA_US_US:%.*]] ], [ [[BC_RESUME_VAL56]], [[VEC_EPILOG_SCALAR_PH41]] ]
218; CHECK-NEXT:    [[INDVARS_IV70:%.*]] = phi i64 [ [[INDVARS_IV_NEXT71:%.*]], [[FOR_COND_CLEANUP4_US_LCSSA_US_US]] ], [ [[BC_RESUME_VAL67]], [[VEC_EPILOG_SCALAR_PH41]] ]
219; CHECK-NEXT:    [[TMP56:%.*]] = sub nsw i64 8, [[INDVARS_IV78]]
220; CHECK-NEXT:    [[ADD_PTR_US:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr @d, i64 0, i64 [[INDVARS_IV78]]
221; CHECK-NEXT:    [[TMP57:%.*]] = add nsw i64 [[TMP56]], [[INDVARS_IV70]]
222; CHECK-NEXT:    [[ARRAYDECAY_US_US_US:%.*]] = getelementptr inbounds [10 x i32], ptr [[ADD_PTR_US]], i64 [[TMP57]], i64 0
223; CHECK-NEXT:    br i1 [[TOBOOL6]], label [[FOR_BODY5_US_US_US_PREHEADER:%.*]], label [[FOR_BODY5_US_US48_PREHEADER:%.*]]
224; CHECK:       for.body5.us.us48.preheader:
225; CHECK-NEXT:    store i32 8, ptr [[ARRAYDECAY_US_US_US]], align 16
226; CHECK-NEXT:    [[INDVARS_IV_NEXT66:%.*]] = or disjoint i64 [[INDVARS_IV70]], 1
227; CHECK-NEXT:    [[TMP58:%.*]] = add nsw i64 [[TMP56]], [[INDVARS_IV_NEXT66]]
228; CHECK-NEXT:    [[ARRAYDECAY_US_US55_1:%.*]] = getelementptr inbounds [10 x i32], ptr [[ADD_PTR_US]], i64 [[TMP58]], i64 0
229; CHECK-NEXT:    store i32 8, ptr [[ARRAYDECAY_US_US55_1]], align 8
230; CHECK-NEXT:    br label [[FOR_COND_CLEANUP4_US_LCSSA_US_US]]
231; CHECK:       for.body5.us.us.us.preheader:
232; CHECK-NEXT:    store i32 7, ptr [[ARRAYDECAY_US_US_US]], align 16
233; CHECK-NEXT:    [[INDVARS_IV_NEXT73:%.*]] = or disjoint i64 [[INDVARS_IV70]], 1
234; CHECK-NEXT:    [[TMP59:%.*]] = add nsw i64 [[TMP56]], [[INDVARS_IV_NEXT73]]
235; CHECK-NEXT:    [[ARRAYDECAY_US_US_US_1:%.*]] = getelementptr inbounds [10 x i32], ptr [[ADD_PTR_US]], i64 [[TMP59]], i64 0
236; CHECK-NEXT:    store i32 7, ptr [[ARRAYDECAY_US_US_US_1]], align 8
237; CHECK-NEXT:    br label [[FOR_COND_CLEANUP4_US_LCSSA_US_US]]
238; CHECK:       for.cond.cleanup4.us-lcssa.us.us:
239; CHECK-NEXT:    [[INDVARS_IV_NEXT79]] = add nuw nsw i64 [[INDVARS_IV78]], 2
240; CHECK-NEXT:    [[CMP_US:%.*]] = icmp slt i64 [[INDVARS_IV_NEXT79]], [[TMP3]]
241; CHECK-NEXT:    [[INDVARS_IV_NEXT71]] = add nuw nsw i64 [[INDVARS_IV70]], 2
242; CHECK-NEXT:    br i1 [[CMP_US]], label [[FOR_BODY_US]], label [[FOR_COND_CLEANUP_LOOPEXIT]], !llvm.loop [[LOOP6:![0-9]+]]
243; CHECK:       for.cond.cleanup.loopexit:
244; CHECK-NEXT:    br label [[FOR_COND_CLEANUP]]
245; CHECK:       for.cond.cleanup.loopexit99:
246; CHECK-NEXT:    br label [[FOR_COND_CLEANUP]]
247; CHECK:       for.cond.cleanup:
248; CHECK-NEXT:    ret void
249; CHECK:       for.body:
250; CHECK-NEXT:    [[INDVARS_IV95:%.*]] = phi i64 [ [[INDVARS_IV_NEXT96:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL10]], [[VEC_EPILOG_SCALAR_PH]] ]
251; CHECK-NEXT:    [[INDVARS_IV87:%.*]] = phi i64 [ [[INDVARS_IV_NEXT88:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL13]], [[VEC_EPILOG_SCALAR_PH]] ]
252; CHECK-NEXT:    [[TMP60:%.*]] = sub nsw i64 8, [[INDVARS_IV95]]
253; CHECK-NEXT:    [[ADD_PTR:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr @d, i64 0, i64 [[INDVARS_IV95]]
254; CHECK-NEXT:    [[TMP61:%.*]] = add nsw i64 [[TMP60]], [[INDVARS_IV87]]
255; CHECK-NEXT:    [[ARRAYDECAY_US31:%.*]] = getelementptr inbounds [10 x i32], ptr [[ADD_PTR]], i64 [[TMP61]], i64 0
256; CHECK-NEXT:    store i32 8, ptr [[ARRAYDECAY_US31]], align 16
257; CHECK-NEXT:    [[INDVARS_IV_NEXT90:%.*]] = or disjoint i64 [[INDVARS_IV87]], 1
258; CHECK-NEXT:    [[TMP62:%.*]] = add nsw i64 [[TMP60]], [[INDVARS_IV_NEXT90]]
259; CHECK-NEXT:    [[ARRAYDECAY_US31_1:%.*]] = getelementptr inbounds [10 x i32], ptr [[ADD_PTR]], i64 [[TMP62]], i64 0
260; CHECK-NEXT:    store i32 8, ptr [[ARRAYDECAY_US31_1]], align 8
261; CHECK-NEXT:    [[INDVARS_IV_NEXT96]] = add nuw nsw i64 [[INDVARS_IV95]], 2
262; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i64 [[INDVARS_IV_NEXT96]], [[TMP3]]
263; CHECK-NEXT:    [[INDVARS_IV_NEXT88]] = add nuw nsw i64 [[INDVARS_IV87]], 2
264; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_COND_CLEANUP_LOOPEXIT99]], !llvm.loop [[LOOP7:![0-9]+]]
265;
266entry:
267  %0 = load i32, ptr @c, align 4
268  %cmp34 = icmp sgt i32 %0, 8
269  br i1 %cmp34, label %for.body.lr.ph, label %for.cond.cleanup
270
271for.body.lr.ph:                                   ; preds = %entry
272  %1 = load i32, ptr @a, align 4
273  %tobool = icmp eq i32 %1, 0
274  %2 = load i64, ptr @b, align 8
275  %mul = mul i64 %2, 4063299859190
276  %tobool6 = icmp eq i64 %mul, 0
277  %3 = sext i32 %0 to i64
278  br i1 %tobool, label %for.body.us.preheader, label %for.body.preheader
279
280for.body.preheader:                               ; preds = %for.body.lr.ph
281  br label %for.body
282
283for.body.us.preheader:                            ; preds = %for.body.lr.ph
284  br label %for.body.us
285
286for.body.us:                                      ; preds = %for.body.us.preheader, %for.cond.cleanup4.us-lcssa.us.us
287  %indvars.iv78 = phi i64 [ %indvars.iv.next79, %for.cond.cleanup4.us-lcssa.us.us ], [ 8, %for.body.us.preheader ]
288  %indvars.iv70 = phi i64 [ %indvars.iv.next71, %for.cond.cleanup4.us-lcssa.us.us ], [ 0, %for.body.us.preheader ]
289  %4 = sub nsw i64 8, %indvars.iv78
290  %add.ptr.us = getelementptr inbounds [10 x [10 x i32]], ptr @d, i64 0, i64 %indvars.iv78
291  %5 = add nsw i64 %4, %indvars.iv70
292  %arraydecay.us.us.us = getelementptr inbounds [10 x i32], ptr %add.ptr.us, i64 %5, i64 0
293  br i1 %tobool6, label %for.body5.us.us.us.preheader, label %for.body5.us.us48.preheader
294
295for.body5.us.us48.preheader:                      ; preds = %for.body.us
296  store i32 8, ptr %arraydecay.us.us.us, align 16
297  %indvars.iv.next66 = or disjoint i64 %indvars.iv70, 1
298  %6 = add nsw i64 %4, %indvars.iv.next66
299  %arraydecay.us.us55.1 = getelementptr inbounds [10 x i32], ptr %add.ptr.us, i64 %6, i64 0
300  store i32 8, ptr %arraydecay.us.us55.1, align 8
301  br label %for.cond.cleanup4.us-lcssa.us.us
302
303for.body5.us.us.us.preheader:                     ; preds = %for.body.us
304  store i32 7, ptr %arraydecay.us.us.us, align 16
305  %indvars.iv.next73 = or disjoint i64 %indvars.iv70, 1
306  %7 = add nsw i64 %4, %indvars.iv.next73
307  %arraydecay.us.us.us.1 = getelementptr inbounds [10 x i32], ptr %add.ptr.us, i64 %7, i64 0
308  store i32 7, ptr %arraydecay.us.us.us.1, align 8
309  br label %for.cond.cleanup4.us-lcssa.us.us
310
311for.cond.cleanup4.us-lcssa.us.us:                 ; preds = %for.body5.us.us48.preheader, %for.body5.us.us.us.preheader
312  %indvars.iv.next79 = add nuw nsw i64 %indvars.iv78, 2
313  %cmp.us = icmp slt i64 %indvars.iv.next79, %3
314  %indvars.iv.next71 = add nuw nsw i64 %indvars.iv70, 2
315  br i1 %cmp.us, label %for.body.us, label %for.cond.cleanup.loopexit
316
317for.cond.cleanup.loopexit:                        ; preds = %for.cond.cleanup4.us-lcssa.us.us
318  br label %for.cond.cleanup
319
320for.cond.cleanup.loopexit99:                      ; preds = %for.body
321  br label %for.cond.cleanup
322
323for.cond.cleanup:                                 ; preds = %for.cond.cleanup.loopexit99, %for.cond.cleanup.loopexit, %entry
324  ret void
325
326for.body:                                         ; preds = %for.body.preheader, %for.body
327  %indvars.iv95 = phi i64 [ %indvars.iv.next96, %for.body ], [ 8, %for.body.preheader ]
328  %indvars.iv87 = phi i64 [ %indvars.iv.next88, %for.body ], [ 0, %for.body.preheader ]
329  %8 = sub nsw i64 8, %indvars.iv95
330  %add.ptr = getelementptr inbounds [10 x [10 x i32]], ptr @d, i64 0, i64 %indvars.iv95
331  %9 = add nsw i64 %8, %indvars.iv87
332  %arraydecay.us31 = getelementptr inbounds [10 x i32], ptr %add.ptr, i64 %9, i64 0
333  store i32 8, ptr %arraydecay.us31, align 16
334  %indvars.iv.next90 = or disjoint i64 %indvars.iv87, 1
335  %10 = add nsw i64 %8, %indvars.iv.next90
336  %arraydecay.us31.1 = getelementptr inbounds [10 x i32], ptr %add.ptr, i64 %10, i64 0
337  store i32 8, ptr %arraydecay.us31.1, align 8
338  %indvars.iv.next96 = add nuw nsw i64 %indvars.iv95, 2
339  %cmp = icmp slt i64 %indvars.iv.next96, %3
340  %indvars.iv.next88 = add nuw nsw i64 %indvars.iv87, 2
341  br i1 %cmp, label %for.body, label %for.cond.cleanup.loopexit99
342}
343
344attributes #0 = { norecurse nounwind ssp uwtable "disable-tail-calls"="false" "less-precise-fpmad"="false" "frame-pointer"="all" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="broadwell" "target-features"="+adx,+aes,+avx,+avx2,+avx512cd,+avx512f,+bmi,+bmi2,+cx16,+f16c,+fma,+fsgsbase,+fxsr,+evex512,+lzcnt,+mmx,+movbe,+pclmul,+popcnt,+rdrnd,+rdseed,+rtm,+sse,+sse2,+sse3,+sse4.1,+sse4.2,+ssse3,+x87,+xsave,+xsaveopt,-vzeroupper" "unsafe-fp-math"="false" "use-soft-float"="false" }
345