xref: /llvm-project/llvm/test/Transforms/PhaseOrdering/X86/excessive-unrolling.ll (revision d1e5e6735a845f1281f11389da1e5a55a0d2e87a) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -passes="default<O3>" -unroll-runtime -S %s | FileCheck %s
3
4target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
5target triple = "x86_64-apple-macosx"
6
7@b = global [58 x double] zeroinitializer, align 16
8@c = global [58 x double] zeroinitializer, align 16
9@a = global [58 x double] zeroinitializer, align 16
10
11; Test case for #42332, showing excessive unrolling of vector loop.
12define void @test_known_trip_count() {
13; CHECK-LABEL: @test_known_trip_count(
14; CHECK-NEXT:  entry:
15; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x double>, ptr @b, align 16
16; CHECK-NEXT:    [[WIDE_LOAD3:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 16), align 16
17; CHECK-NEXT:    [[WIDE_LOAD4:%.*]] = load <2 x double>, ptr @c, align 16
18; CHECK-NEXT:    [[WIDE_LOAD5:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 16), align 16
19; CHECK-NEXT:    [[TMP0:%.*]] = fadd <2 x double> [[WIDE_LOAD]], [[WIDE_LOAD4]]
20; CHECK-NEXT:    [[TMP1:%.*]] = fadd <2 x double> [[WIDE_LOAD3]], [[WIDE_LOAD5]]
21; CHECK-NEXT:    store <2 x double> [[TMP0]], ptr @a, align 16
22; CHECK-NEXT:    store <2 x double> [[TMP1]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 16), align 16
23; CHECK-NEXT:    [[WIDE_LOAD_1:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 32), align 16
24; CHECK-NEXT:    [[WIDE_LOAD3_1:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 48), align 16
25; CHECK-NEXT:    [[WIDE_LOAD4_1:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 32), align 16
26; CHECK-NEXT:    [[WIDE_LOAD5_1:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 48), align 16
27; CHECK-NEXT:    [[TMP2:%.*]] = fadd <2 x double> [[WIDE_LOAD_1]], [[WIDE_LOAD4_1]]
28; CHECK-NEXT:    [[TMP3:%.*]] = fadd <2 x double> [[WIDE_LOAD3_1]], [[WIDE_LOAD5_1]]
29; CHECK-NEXT:    store <2 x double> [[TMP2]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 32), align 16
30; CHECK-NEXT:    store <2 x double> [[TMP3]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 48), align 16
31; CHECK-NEXT:    [[WIDE_LOAD_2:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 64), align 16
32; CHECK-NEXT:    [[WIDE_LOAD3_2:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 80), align 16
33; CHECK-NEXT:    [[WIDE_LOAD4_2:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 64), align 16
34; CHECK-NEXT:    [[WIDE_LOAD5_2:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 80), align 16
35; CHECK-NEXT:    [[TMP4:%.*]] = fadd <2 x double> [[WIDE_LOAD_2]], [[WIDE_LOAD4_2]]
36; CHECK-NEXT:    [[TMP5:%.*]] = fadd <2 x double> [[WIDE_LOAD3_2]], [[WIDE_LOAD5_2]]
37; CHECK-NEXT:    store <2 x double> [[TMP4]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 64), align 16
38; CHECK-NEXT:    store <2 x double> [[TMP5]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 80), align 16
39; CHECK-NEXT:    [[WIDE_LOAD_3:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 96), align 16
40; CHECK-NEXT:    [[WIDE_LOAD3_3:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 112), align 16
41; CHECK-NEXT:    [[WIDE_LOAD4_3:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 96), align 16
42; CHECK-NEXT:    [[WIDE_LOAD5_3:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 112), align 16
43; CHECK-NEXT:    [[TMP6:%.*]] = fadd <2 x double> [[WIDE_LOAD_3]], [[WIDE_LOAD4_3]]
44; CHECK-NEXT:    [[TMP7:%.*]] = fadd <2 x double> [[WIDE_LOAD3_3]], [[WIDE_LOAD5_3]]
45; CHECK-NEXT:    store <2 x double> [[TMP6]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 96), align 16
46; CHECK-NEXT:    store <2 x double> [[TMP7]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 112), align 16
47; CHECK-NEXT:    [[WIDE_LOAD_4:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 128), align 16
48; CHECK-NEXT:    [[WIDE_LOAD3_4:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 144), align 16
49; CHECK-NEXT:    [[WIDE_LOAD4_4:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 128), align 16
50; CHECK-NEXT:    [[WIDE_LOAD5_4:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 144), align 16
51; CHECK-NEXT:    [[TMP8:%.*]] = fadd <2 x double> [[WIDE_LOAD_4]], [[WIDE_LOAD4_4]]
52; CHECK-NEXT:    [[TMP9:%.*]] = fadd <2 x double> [[WIDE_LOAD3_4]], [[WIDE_LOAD5_4]]
53; CHECK-NEXT:    store <2 x double> [[TMP8]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 128), align 16
54; CHECK-NEXT:    store <2 x double> [[TMP9]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 144), align 16
55; CHECK-NEXT:    [[WIDE_LOAD_5:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 160), align 16
56; CHECK-NEXT:    [[WIDE_LOAD3_5:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 176), align 16
57; CHECK-NEXT:    [[WIDE_LOAD4_5:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 160), align 16
58; CHECK-NEXT:    [[WIDE_LOAD5_5:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 176), align 16
59; CHECK-NEXT:    [[TMP10:%.*]] = fadd <2 x double> [[WIDE_LOAD_5]], [[WIDE_LOAD4_5]]
60; CHECK-NEXT:    [[TMP11:%.*]] = fadd <2 x double> [[WIDE_LOAD3_5]], [[WIDE_LOAD5_5]]
61; CHECK-NEXT:    store <2 x double> [[TMP10]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 160), align 16
62; CHECK-NEXT:    store <2 x double> [[TMP11]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 176), align 16
63; CHECK-NEXT:    [[WIDE_LOAD_6:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 192), align 16
64; CHECK-NEXT:    [[WIDE_LOAD3_6:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 208), align 16
65; CHECK-NEXT:    [[WIDE_LOAD4_6:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 192), align 16
66; CHECK-NEXT:    [[WIDE_LOAD5_6:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 208), align 16
67; CHECK-NEXT:    [[TMP12:%.*]] = fadd <2 x double> [[WIDE_LOAD_6]], [[WIDE_LOAD4_6]]
68; CHECK-NEXT:    [[TMP13:%.*]] = fadd <2 x double> [[WIDE_LOAD3_6]], [[WIDE_LOAD5_6]]
69; CHECK-NEXT:    store <2 x double> [[TMP12]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 192), align 16
70; CHECK-NEXT:    store <2 x double> [[TMP13]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 208), align 16
71; CHECK-NEXT:    [[WIDE_LOAD_7:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 224), align 16
72; CHECK-NEXT:    [[WIDE_LOAD3_7:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 240), align 16
73; CHECK-NEXT:    [[WIDE_LOAD4_7:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 224), align 16
74; CHECK-NEXT:    [[WIDE_LOAD5_7:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 240), align 16
75; CHECK-NEXT:    [[TMP14:%.*]] = fadd <2 x double> [[WIDE_LOAD_7]], [[WIDE_LOAD4_7]]
76; CHECK-NEXT:    [[TMP15:%.*]] = fadd <2 x double> [[WIDE_LOAD3_7]], [[WIDE_LOAD5_7]]
77; CHECK-NEXT:    store <2 x double> [[TMP14]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 224), align 16
78; CHECK-NEXT:    store <2 x double> [[TMP15]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 240), align 16
79; CHECK-NEXT:    [[WIDE_LOAD_8:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 256), align 16
80; CHECK-NEXT:    [[WIDE_LOAD3_8:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 272), align 16
81; CHECK-NEXT:    [[WIDE_LOAD4_8:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 256), align 16
82; CHECK-NEXT:    [[WIDE_LOAD5_8:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 272), align 16
83; CHECK-NEXT:    [[TMP16:%.*]] = fadd <2 x double> [[WIDE_LOAD_8]], [[WIDE_LOAD4_8]]
84; CHECK-NEXT:    [[TMP17:%.*]] = fadd <2 x double> [[WIDE_LOAD3_8]], [[WIDE_LOAD5_8]]
85; CHECK-NEXT:    store <2 x double> [[TMP16]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 256), align 16
86; CHECK-NEXT:    store <2 x double> [[TMP17]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 272), align 16
87; CHECK-NEXT:    [[WIDE_LOAD_9:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 288), align 16
88; CHECK-NEXT:    [[WIDE_LOAD3_9:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 304), align 16
89; CHECK-NEXT:    [[WIDE_LOAD4_9:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 288), align 16
90; CHECK-NEXT:    [[WIDE_LOAD5_9:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 304), align 16
91; CHECK-NEXT:    [[TMP18:%.*]] = fadd <2 x double> [[WIDE_LOAD_9]], [[WIDE_LOAD4_9]]
92; CHECK-NEXT:    [[TMP19:%.*]] = fadd <2 x double> [[WIDE_LOAD3_9]], [[WIDE_LOAD5_9]]
93; CHECK-NEXT:    store <2 x double> [[TMP18]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 288), align 16
94; CHECK-NEXT:    store <2 x double> [[TMP19]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 304), align 16
95; CHECK-NEXT:    [[WIDE_LOAD_10:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 320), align 16
96; CHECK-NEXT:    [[WIDE_LOAD3_10:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 336), align 16
97; CHECK-NEXT:    [[WIDE_LOAD4_10:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 320), align 16
98; CHECK-NEXT:    [[WIDE_LOAD5_10:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 336), align 16
99; CHECK-NEXT:    [[TMP20:%.*]] = fadd <2 x double> [[WIDE_LOAD_10]], [[WIDE_LOAD4_10]]
100; CHECK-NEXT:    [[TMP21:%.*]] = fadd <2 x double> [[WIDE_LOAD3_10]], [[WIDE_LOAD5_10]]
101; CHECK-NEXT:    store <2 x double> [[TMP20]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 320), align 16
102; CHECK-NEXT:    store <2 x double> [[TMP21]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 336), align 16
103; CHECK-NEXT:    [[WIDE_LOAD_11:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 352), align 16
104; CHECK-NEXT:    [[WIDE_LOAD3_11:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 368), align 16
105; CHECK-NEXT:    [[WIDE_LOAD4_11:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 352), align 16
106; CHECK-NEXT:    [[WIDE_LOAD5_11:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 368), align 16
107; CHECK-NEXT:    [[TMP22:%.*]] = fadd <2 x double> [[WIDE_LOAD_11]], [[WIDE_LOAD4_11]]
108; CHECK-NEXT:    [[TMP23:%.*]] = fadd <2 x double> [[WIDE_LOAD3_11]], [[WIDE_LOAD5_11]]
109; CHECK-NEXT:    store <2 x double> [[TMP22]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 352), align 16
110; CHECK-NEXT:    store <2 x double> [[TMP23]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 368), align 16
111; CHECK-NEXT:    [[WIDE_LOAD_12:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 384), align 16
112; CHECK-NEXT:    [[WIDE_LOAD3_12:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 400), align 16
113; CHECK-NEXT:    [[WIDE_LOAD4_12:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 384), align 16
114; CHECK-NEXT:    [[WIDE_LOAD5_12:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 400), align 16
115; CHECK-NEXT:    [[TMP24:%.*]] = fadd <2 x double> [[WIDE_LOAD_12]], [[WIDE_LOAD4_12]]
116; CHECK-NEXT:    [[TMP25:%.*]] = fadd <2 x double> [[WIDE_LOAD3_12]], [[WIDE_LOAD5_12]]
117; CHECK-NEXT:    store <2 x double> [[TMP24]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 384), align 16
118; CHECK-NEXT:    store <2 x double> [[TMP25]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 400), align 16
119; CHECK-NEXT:    [[WIDE_LOAD_13:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 416), align 16
120; CHECK-NEXT:    [[WIDE_LOAD3_13:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 432), align 16
121; CHECK-NEXT:    [[WIDE_LOAD4_13:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 416), align 16
122; CHECK-NEXT:    [[WIDE_LOAD5_13:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 432), align 16
123; CHECK-NEXT:    [[TMP26:%.*]] = fadd <2 x double> [[WIDE_LOAD_13]], [[WIDE_LOAD4_13]]
124; CHECK-NEXT:    [[TMP27:%.*]] = fadd <2 x double> [[WIDE_LOAD3_13]], [[WIDE_LOAD5_13]]
125; CHECK-NEXT:    store <2 x double> [[TMP26]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 416), align 16
126; CHECK-NEXT:    store <2 x double> [[TMP27]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 432), align 16
127; CHECK-NEXT:    [[WIDE_LOAD_14:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 448), align 16
128; CHECK-NEXT:    [[WIDE_LOAD3_14:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @b, i64 464), align 16
129; CHECK-NEXT:    [[WIDE_LOAD4_14:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 448), align 16
130; CHECK-NEXT:    [[WIDE_LOAD5_14:%.*]] = load <2 x double>, ptr getelementptr inbounds nuw (i8, ptr @c, i64 464), align 16
131; CHECK-NEXT:    [[TMP28:%.*]] = fadd <2 x double> [[WIDE_LOAD_14]], [[WIDE_LOAD4_14]]
132; CHECK-NEXT:    [[TMP29:%.*]] = fadd <2 x double> [[WIDE_LOAD3_14]], [[WIDE_LOAD5_14]]
133; CHECK-NEXT:    store <2 x double> [[TMP28]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 448), align 16
134; CHECK-NEXT:    store <2 x double> [[TMP29]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 464), align 16
135; CHECK-NEXT:    [[TMP30:%.*]] = load double, ptr getelementptr inbounds nuw (i8, ptr @b, i64 480), align 16
136; CHECK-NEXT:    [[TMP31:%.*]] = load double, ptr getelementptr inbounds nuw (i8, ptr @c, i64 480), align 16
137; CHECK-NEXT:    [[ADD:%.*]] = fadd double [[TMP30]], [[TMP31]]
138; CHECK-NEXT:    store double [[ADD]], ptr getelementptr inbounds nuw (i8, ptr @a, i64 480), align 16
139; CHECK-NEXT:    ret void
140;
141entry:
142  br label %for.cond
143
144for.cond:
145  %i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
146  %cmp = icmp slt i32 %i.0, 61
147  br i1 %cmp, label %for.body, label %exit
148
149for.body:
150  %idxprom = sext i32 %i.0 to i64
151  %arrayidx = getelementptr inbounds [58 x double], ptr @b, i64 0, i64 %idxprom
152  %0 = load double, ptr %arrayidx, align 8
153  %idxprom1 = sext i32 %i.0 to i64
154  %arrayidx2 = getelementptr inbounds [58 x double], ptr @c, i64 0, i64 %idxprom1
155  %1 = load double, ptr %arrayidx2, align 8
156  %add = fadd double %0, %1
157  %idxprom3 = sext i32 %i.0 to i64
158  %arrayidx4 = getelementptr inbounds [58 x double], ptr @a, i64 0, i64 %idxprom3
159  store double %add, ptr %arrayidx4, align 8
160  %inc = add nsw i32 %i.0, 1
161  br label %for.cond
162
163exit:
164  ret void
165}
166
167
168define void @test_runtime_trip_count(i32 %N) {
169; CHECK-LABEL: @test_runtime_trip_count(
170; CHECK-NEXT:  entry:
171; CHECK-NEXT:    [[CMP1:%.*]] = icmp sgt i32 [[N:%.*]], 0
172; CHECK-NEXT:    br i1 [[CMP1]], label [[FOR_BODY_PREHEADER:%.*]], label [[EXIT:%.*]]
173; CHECK:       for.body.preheader:
174; CHECK-NEXT:    [[WIDE_TRIP_COUNT:%.*]] = zext nneg i32 [[N]] to i64
175; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[N]], 4
176; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[FOR_BODY_PREHEADER7:%.*]], label [[VECTOR_PH:%.*]]
177; CHECK:       vector.ph:
178; CHECK-NEXT:    [[N_VEC:%.*]] = and i64 [[WIDE_TRIP_COUNT]], 2147483644
179; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
180; CHECK:       vector.body:
181; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
182; CHECK-NEXT:    [[TMP0:%.*]] = getelementptr inbounds nuw [58 x double], ptr @b, i64 0, i64 [[INDEX]]
183; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP0]], i64 16
184; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x double>, ptr [[TMP0]], align 16
185; CHECK-NEXT:    [[WIDE_LOAD4:%.*]] = load <2 x double>, ptr [[TMP1]], align 16
186; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds nuw [58 x double], ptr @c, i64 0, i64 [[INDEX]]
187; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP2]], i64 16
188; CHECK-NEXT:    [[WIDE_LOAD5:%.*]] = load <2 x double>, ptr [[TMP2]], align 16
189; CHECK-NEXT:    [[WIDE_LOAD6:%.*]] = load <2 x double>, ptr [[TMP3]], align 16
190; CHECK-NEXT:    [[TMP4:%.*]] = fadd <2 x double> [[WIDE_LOAD]], [[WIDE_LOAD5]]
191; CHECK-NEXT:    [[TMP5:%.*]] = fadd <2 x double> [[WIDE_LOAD4]], [[WIDE_LOAD6]]
192; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds nuw [58 x double], ptr @a, i64 0, i64 [[INDEX]]
193; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP6]], i64 16
194; CHECK-NEXT:    store <2 x double> [[TMP4]], ptr [[TMP6]], align 16
195; CHECK-NEXT:    store <2 x double> [[TMP5]], ptr [[TMP7]], align 16
196; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
197; CHECK-NEXT:    [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
198; CHECK-NEXT:    br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
199; CHECK:       middle.block:
200; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[WIDE_TRIP_COUNT]]
201; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT]], label [[FOR_BODY_PREHEADER7]]
202; CHECK:       for.body.preheader7:
203; CHECK-NEXT:    [[INDVARS_IV_PH:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[N_VEC]], [[MIDDLE_BLOCK]] ]
204; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
205; CHECK:       for.body:
206; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[INDVARS_IV_PH]], [[FOR_BODY_PREHEADER7]] ]
207; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds nuw [58 x double], ptr @b, i64 0, i64 [[INDVARS_IV]]
208; CHECK-NEXT:    [[TMP9:%.*]] = load double, ptr [[ARRAYIDX]], align 8
209; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds nuw [58 x double], ptr @c, i64 0, i64 [[INDVARS_IV]]
210; CHECK-NEXT:    [[TMP10:%.*]] = load double, ptr [[ARRAYIDX2]], align 8
211; CHECK-NEXT:    [[ADD:%.*]] = fadd double [[TMP9]], [[TMP10]]
212; CHECK-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds nuw [58 x double], ptr @a, i64 0, i64 [[INDVARS_IV]]
213; CHECK-NEXT:    store double [[ADD]], ptr [[ARRAYIDX4]], align 8
214; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
215; CHECK-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]
216; CHECK-NEXT:    br i1 [[EXITCOND_NOT]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
217; CHECK:       exit:
218; CHECK-NEXT:    ret void
219;
220entry:
221  br label %for.cond
222
223for.cond:
224  %i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
225  %cmp = icmp slt i32 %i.0, %N
226  br i1 %cmp, label %for.body, label %exit
227
228for.body:
229  %idxprom = sext i32 %i.0 to i64
230  %arrayidx = getelementptr inbounds [58 x double], ptr @b, i64 0, i64 %idxprom
231  %0 = load double, ptr %arrayidx, align 8
232  %idxprom1 = sext i32 %i.0 to i64
233  %arrayidx2 = getelementptr inbounds [58 x double], ptr @c, i64 0, i64 %idxprom1
234  %1 = load double, ptr %arrayidx2, align 8
235  %add = fadd double %0, %1
236  %idxprom3 = sext i32 %i.0 to i64
237  %arrayidx4 = getelementptr inbounds [58 x double], ptr @a, i64 0, i64 %idxprom3
238  store double %add, ptr %arrayidx4, align 8
239  %inc = add nsw i32 %i.0, 1
240  br label %for.cond
241
242exit:
243  ret void
244}
245