1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -basic-aa -slp-vectorizer -S -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7-avx | FileCheck %s
3
4define void @julia_2xdouble([2 x double]* sret([2 x double]), [2 x double]*, [2 x double]*, [2 x double]*) {
5; CHECK-LABEL: @julia_2xdouble(
6; CHECK-NEXT:  top:
7; CHECK-NEXT:    [[PX0:%.*]] = getelementptr inbounds [2 x double], [2 x double]* [[TMP2:%.*]], i64 0, i64 0
8; CHECK-NEXT:    [[PY0:%.*]] = getelementptr inbounds [2 x double], [2 x double]* [[TMP3:%.*]], i64 0, i64 0
9; CHECK-NEXT:    [[PX1:%.*]] = getelementptr inbounds [2 x double], [2 x double]* [[TMP2]], i64 0, i64 1
10; CHECK-NEXT:    [[TMP4:%.*]] = bitcast double* [[PX0]] to <2 x double>*
11; CHECK-NEXT:    [[TMP5:%.*]] = load <2 x double>, <2 x double>* [[TMP4]], align 4
12; CHECK-NEXT:    [[PY1:%.*]] = getelementptr inbounds [2 x double], [2 x double]* [[TMP3]], i64 0, i64 1
13; CHECK-NEXT:    [[TMP6:%.*]] = bitcast double* [[PY0]] to <2 x double>*
14; CHECK-NEXT:    [[TMP7:%.*]] = load <2 x double>, <2 x double>* [[TMP6]], align 4
15; CHECK-NEXT:    [[TMP8:%.*]] = fmul <2 x double> [[TMP5]], [[TMP7]]
16; CHECK-NEXT:    [[PZ0:%.*]] = getelementptr inbounds [2 x double], [2 x double]* [[TMP1:%.*]], i64 0, i64 0
17; CHECK-NEXT:    [[PZ1:%.*]] = getelementptr inbounds [2 x double], [2 x double]* [[TMP1]], i64 0, i64 1
18; CHECK-NEXT:    [[TMP9:%.*]] = bitcast double* [[PZ0]] to <2 x double>*
19; CHECK-NEXT:    [[TMP10:%.*]] = load <2 x double>, <2 x double>* [[TMP9]], align 4
20; CHECK-NEXT:    [[TMP11:%.*]] = fadd <2 x double> [[TMP8]], [[TMP10]]
21; CHECK-NEXT:    [[TMP12:%.*]] = extractelement <2 x double> [[TMP11]], i32 0
22; CHECK-NEXT:    [[I0:%.*]] = insertvalue [2 x double] undef, double [[TMP12]], 0
23; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <2 x double> [[TMP11]], i32 1
24; CHECK-NEXT:    [[I1:%.*]] = insertvalue [2 x double] [[I0]], double [[TMP13]], 1
25; CHECK-NEXT:    store [2 x double] [[I1]], [2 x double]* [[TMP0:%.*]], align 4
26; CHECK-NEXT:    ret void
27;
28top:
29  %px0 = getelementptr inbounds [2 x double], [2 x double]* %2, i64 0, i64 0
30  %x0 = load double, double* %px0, align 4
31  %py0 = getelementptr inbounds [2 x double], [2 x double]* %3, i64 0, i64 0
32  %y0 = load double, double* %py0, align 4
33  %m0 = fmul double %x0, %y0
34  %px1 = getelementptr inbounds [2 x double], [2 x double]* %2, i64 0, i64 1
35  %x1 = load double, double* %px1, align 4
36  %py1 = getelementptr inbounds [2 x double], [2 x double]* %3, i64 0, i64 1
37  %y1 = load double, double* %py1, align 4
38  %m1 = fmul double %x1, %y1
39  %pz0 = getelementptr inbounds [2 x double], [2 x double]* %1, i64 0, i64 0
40  %z0 = load double, double* %pz0, align 4
41  %a0 = fadd double %m0, %z0
42  %i0 = insertvalue [2 x double] undef, double %a0, 0
43  %pz1 = getelementptr inbounds [2 x double], [2 x double]* %1, i64 0, i64 1
44  %z1 = load double, double* %pz1, align 4
45  %a1 = fadd double %m1, %z1
46  %i1 = insertvalue [2 x double] %i0, double %a1, 1
47  store [2 x double] %i1, [2 x double]* %0, align 4
48  ret void
49}
50
51define void @julia_4xfloat([4 x float]* sret([4 x float]), [4 x float]*, [4 x float]*, [4 x float]*) {
52; CHECK-LABEL: @julia_4xfloat(
53; CHECK-NEXT:  top:
54; CHECK-NEXT:    [[PX0:%.*]] = getelementptr inbounds [4 x float], [4 x float]* [[TMP2:%.*]], i64 0, i64 0
55; CHECK-NEXT:    [[PY0:%.*]] = getelementptr inbounds [4 x float], [4 x float]* [[TMP3:%.*]], i64 0, i64 0
56; CHECK-NEXT:    [[PX1:%.*]] = getelementptr inbounds [4 x float], [4 x float]* [[TMP2]], i64 0, i64 1
57; CHECK-NEXT:    [[PY1:%.*]] = getelementptr inbounds [4 x float], [4 x float]* [[TMP3]], i64 0, i64 1
58; CHECK-NEXT:    [[PX2:%.*]] = getelementptr inbounds [4 x float], [4 x float]* [[TMP2]], i64 0, i64 2
59; CHECK-NEXT:    [[PY2:%.*]] = getelementptr inbounds [4 x float], [4 x float]* [[TMP3]], i64 0, i64 2
60; CHECK-NEXT:    [[PX3:%.*]] = getelementptr inbounds [4 x float], [4 x float]* [[TMP2]], i64 0, i64 3
61; CHECK-NEXT:    [[TMP4:%.*]] = bitcast float* [[PX0]] to <4 x float>*
62; CHECK-NEXT:    [[TMP5:%.*]] = load <4 x float>, <4 x float>* [[TMP4]], align 4
63; CHECK-NEXT:    [[PY3:%.*]] = getelementptr inbounds [4 x float], [4 x float]* [[TMP3]], i64 0, i64 3
64; CHECK-NEXT:    [[TMP6:%.*]] = bitcast float* [[PY0]] to <4 x float>*
65; CHECK-NEXT:    [[TMP7:%.*]] = load <4 x float>, <4 x float>* [[TMP6]], align 4
66; CHECK-NEXT:    [[TMP8:%.*]] = fmul <4 x float> [[TMP5]], [[TMP7]]
67; CHECK-NEXT:    [[PZ0:%.*]] = getelementptr inbounds [4 x float], [4 x float]* [[TMP1:%.*]], i64 0, i64 0
68; CHECK-NEXT:    [[PZ1:%.*]] = getelementptr inbounds [4 x float], [4 x float]* [[TMP1]], i64 0, i64 1
69; CHECK-NEXT:    [[PZ2:%.*]] = getelementptr inbounds [4 x float], [4 x float]* [[TMP1]], i64 0, i64 2
70; CHECK-NEXT:    [[PZ3:%.*]] = getelementptr inbounds [4 x float], [4 x float]* [[TMP1]], i64 0, i64 3
71; CHECK-NEXT:    [[TMP9:%.*]] = bitcast float* [[PZ0]] to <4 x float>*
72; CHECK-NEXT:    [[TMP10:%.*]] = load <4 x float>, <4 x float>* [[TMP9]], align 4
73; CHECK-NEXT:    [[TMP11:%.*]] = fadd <4 x float> [[TMP8]], [[TMP10]]
74; CHECK-NEXT:    [[TMP12:%.*]] = extractelement <4 x float> [[TMP11]], i32 0
75; CHECK-NEXT:    [[I0:%.*]] = insertvalue [4 x float] undef, float [[TMP12]], 0
76; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <4 x float> [[TMP11]], i32 1
77; CHECK-NEXT:    [[I1:%.*]] = insertvalue [4 x float] [[I0]], float [[TMP13]], 1
78; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <4 x float> [[TMP11]], i32 2
79; CHECK-NEXT:    [[I2:%.*]] = insertvalue [4 x float] [[I1]], float [[TMP14]], 2
80; CHECK-NEXT:    [[TMP15:%.*]] = extractelement <4 x float> [[TMP11]], i32 3
81; CHECK-NEXT:    [[I3:%.*]] = insertvalue [4 x float] [[I2]], float [[TMP15]], 3
82; CHECK-NEXT:    store [4 x float] [[I3]], [4 x float]* [[TMP0:%.*]], align 4
83; CHECK-NEXT:    ret void
84;
85top:
86  %px0 = getelementptr inbounds [4 x float], [4 x float]* %2, i64 0, i64 0
87  %x0 = load float, float* %px0, align 4
88  %py0 = getelementptr inbounds [4 x float], [4 x float]* %3, i64 0, i64 0
89  %y0 = load float, float* %py0, align 4
90  %m0 = fmul float %x0, %y0
91  %px1 = getelementptr inbounds [4 x float], [4 x float]* %2, i64 0, i64 1
92  %x1 = load float, float* %px1, align 4
93  %py1 = getelementptr inbounds [4 x float], [4 x float]* %3, i64 0, i64 1
94  %y1 = load float, float* %py1, align 4
95  %m1 = fmul float %x1, %y1
96  %px2 = getelementptr inbounds [4 x float], [4 x float]* %2, i64 0, i64 2
97  %x2 = load float, float* %px2, align 4
98  %py2 = getelementptr inbounds [4 x float], [4 x float]* %3, i64 0, i64 2
99  %y2 = load float, float* %py2, align 4
100  %m2 = fmul float %x2, %y2
101  %px3 = getelementptr inbounds [4 x float], [4 x float]* %2, i64 0, i64 3
102  %x3 = load float, float* %px3, align 4
103  %py3 = getelementptr inbounds [4 x float], [4 x float]* %3, i64 0, i64 3
104  %y3 = load float, float* %py3, align 4
105  %m3 = fmul float %x3, %y3
106  %pz0 = getelementptr inbounds [4 x float], [4 x float]* %1, i64 0, i64 0
107  %z0 = load float, float* %pz0, align 4
108  %a0 = fadd float %m0, %z0
109  %i0 = insertvalue [4 x float] undef, float %a0, 0
110  %pz1 = getelementptr inbounds [4 x float], [4 x float]* %1, i64 0, i64 1
111  %z1 = load float, float* %pz1, align 4
112  %a1 = fadd float %m1, %z1
113  %i1 = insertvalue [4 x float] %i0, float %a1, 1
114  %pz2 = getelementptr inbounds [4 x float], [4 x float]* %1, i64 0, i64 2
115  %z2 = load float, float* %pz2, align 4
116  %a2 = fadd float %m2, %z2
117  %i2 = insertvalue [4 x float] %i1, float %a2, 2
118  %pz3 = getelementptr inbounds [4 x float], [4 x float]* %1, i64 0, i64 3
119  %z3 = load float, float* %pz3, align 4
120  %a3 = fadd float %m3, %z3
121  %i3 = insertvalue [4 x float] %i2, float %a3, 3
122  store [4 x float] %i3, [4 x float]* %0, align 4
123  ret void
124}
125
126define void @julia_load_array_of_float([4 x float]* %a, [4 x float]* %b, [4 x float]* %c) {
127; CHECK-LABEL: @julia_load_array_of_float(
128; CHECK-NEXT:  top:
129; CHECK-NEXT:    [[TMP0:%.*]] = bitcast [4 x float]* [[A:%.*]] to <4 x float>*
130; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* [[TMP0]], align 4
131; CHECK-NEXT:    [[A_ARR:%.*]] = load [4 x float], [4 x float]* [[A]], align 4
132; CHECK-NEXT:    [[TMP2:%.*]] = bitcast [4 x float]* [[B:%.*]] to <4 x float>*
133; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x float>, <4 x float>* [[TMP2]], align 4
134; CHECK-NEXT:    [[B_ARR:%.*]] = load [4 x float], [4 x float]* [[B]], align 4
135; CHECK-NEXT:    [[TMP4:%.*]] = fsub <4 x float> [[TMP1]], [[TMP3]]
136; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x float> [[TMP4]], i32 0
137; CHECK-NEXT:    [[C_ARR0:%.*]] = insertvalue [4 x float] undef, float [[TMP5]], 0
138; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[TMP4]], i32 1
139; CHECK-NEXT:    [[C_ARR1:%.*]] = insertvalue [4 x float] [[C_ARR0]], float [[TMP6]], 1
140; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x float> [[TMP4]], i32 2
141; CHECK-NEXT:    [[C_ARR2:%.*]] = insertvalue [4 x float] [[C_ARR1]], float [[TMP7]], 2
142; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <4 x float> [[TMP4]], i32 3
143; CHECK-NEXT:    [[C_ARR3:%.*]] = insertvalue [4 x float] [[C_ARR2]], float [[TMP8]], 3
144; CHECK-NEXT:    store [4 x float] [[C_ARR3]], [4 x float]* [[C:%.*]], align 4
145; CHECK-NEXT:    ret void
146;
147top:
148  %a_arr = load [4 x float], [4 x float]* %a, align 4
149  %a0 = extractvalue [4 x float] %a_arr, 0
150  %a2 = extractvalue [4 x float] %a_arr, 2
151  %a1 = extractvalue [4 x float] %a_arr, 1
152  %b_arr = load [4 x float], [4 x float]* %b, align 4
153  %b0 = extractvalue [4 x float] %b_arr, 0
154  %b2 = extractvalue [4 x float] %b_arr, 2
155  %b1 = extractvalue [4 x float] %b_arr, 1
156  %a3 = extractvalue [4 x float] %a_arr, 3
157  %c1 = fsub float %a1, %b1
158  %b3 = extractvalue [4 x float] %b_arr, 3
159  %c0 = fsub float %a0, %b0
160  %c2 = fsub float %a2, %b2
161  %c_arr0 = insertvalue [4 x float] undef, float %c0, 0
162  %c_arr1 = insertvalue [4 x float] %c_arr0, float %c1, 1
163  %c3 = fsub float %a3, %b3
164  %c_arr2 = insertvalue [4 x float] %c_arr1, float %c2, 2
165  %c_arr3 = insertvalue [4 x float] %c_arr2, float %c3, 3
166  store [4 x float] %c_arr3, [4 x float]* %c, align 4
167  ret void
168}
169
170define void @julia_load_array_of_i32([4 x i32]* %a, [4 x i32]* %b, [4 x i32]* %c) {
171; CHECK-LABEL: @julia_load_array_of_i32(
172; CHECK-NEXT:  top:
173; CHECK-NEXT:    [[TMP0:%.*]] = bitcast [4 x i32]* [[A:%.*]] to <4 x i32>*
174; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, <4 x i32>* [[TMP0]], align 4
175; CHECK-NEXT:    [[A_ARR:%.*]] = load [4 x i32], [4 x i32]* [[A]], align 4
176; CHECK-NEXT:    [[TMP2:%.*]] = bitcast [4 x i32]* [[B:%.*]] to <4 x i32>*
177; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, <4 x i32>* [[TMP2]], align 4
178; CHECK-NEXT:    [[B_ARR:%.*]] = load [4 x i32], [4 x i32]* [[B]], align 4
179; CHECK-NEXT:    [[TMP4:%.*]] = sub <4 x i32> [[TMP1]], [[TMP3]]
180; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x i32> [[TMP4]], i32 0
181; CHECK-NEXT:    [[C_ARR0:%.*]] = insertvalue [4 x i32] undef, i32 [[TMP5]], 0
182; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x i32> [[TMP4]], i32 1
183; CHECK-NEXT:    [[C_ARR1:%.*]] = insertvalue [4 x i32] [[C_ARR0]], i32 [[TMP6]], 1
184; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x i32> [[TMP4]], i32 2
185; CHECK-NEXT:    [[C_ARR2:%.*]] = insertvalue [4 x i32] [[C_ARR1]], i32 [[TMP7]], 2
186; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <4 x i32> [[TMP4]], i32 3
187; CHECK-NEXT:    [[C_ARR3:%.*]] = insertvalue [4 x i32] [[C_ARR2]], i32 [[TMP8]], 3
188; CHECK-NEXT:    store [4 x i32] [[C_ARR3]], [4 x i32]* [[C:%.*]], align 4
189; CHECK-NEXT:    ret void
190;
191top:
192  %a_arr = load [4 x i32], [4 x i32]* %a, align 4
193  %a0 = extractvalue [4 x i32] %a_arr, 0
194  %a2 = extractvalue [4 x i32] %a_arr, 2
195  %a1 = extractvalue [4 x i32] %a_arr, 1
196  %b_arr = load [4 x i32], [4 x i32]* %b, align 4
197  %b0 = extractvalue [4 x i32] %b_arr, 0
198  %b2 = extractvalue [4 x i32] %b_arr, 2
199  %b1 = extractvalue [4 x i32] %b_arr, 1
200  %a3 = extractvalue [4 x i32] %a_arr, 3
201  %c1 = sub i32 %a1, %b1
202  %b3 = extractvalue [4 x i32] %b_arr, 3
203  %c0 = sub i32 %a0, %b0
204  %c2 = sub i32 %a2, %b2
205  %c_arr0 = insertvalue [4 x i32] undef, i32 %c0, 0
206  %c_arr1 = insertvalue [4 x i32] %c_arr0, i32 %c1, 1
207  %c3 = sub i32 %a3, %b3
208  %c_arr2 = insertvalue [4 x i32] %c_arr1, i32 %c2, 2
209  %c_arr3 = insertvalue [4 x i32] %c_arr2, i32 %c3, 3
210  store [4 x i32] %c_arr3, [4 x i32]* %c, align 4
211  ret void
212}
213
214; Almost identical to previous test, but for type that should NOT be vectorized.
215;
216define void @julia_load_array_of_i16([4 x i16]* %a, [4 x i16]* %b, [4 x i16]* %c) {
217; CHECK-LABEL: @julia_load_array_of_i16(
218; CHECK-NEXT:  top:
219; CHECK-NEXT:    [[A_ARR:%.*]] = load [4 x i16], [4 x i16]* [[A:%.*]], align 4
220; CHECK-NEXT:    [[A0:%.*]] = extractvalue [4 x i16] [[A_ARR]], 0
221; CHECK-NEXT:    [[A2:%.*]] = extractvalue [4 x i16] [[A_ARR]], 2
222; CHECK-NEXT:    [[A1:%.*]] = extractvalue [4 x i16] [[A_ARR]], 1
223; CHECK-NEXT:    [[B_ARR:%.*]] = load [4 x i16], [4 x i16]* [[B:%.*]], align 4
224; CHECK-NEXT:    [[B0:%.*]] = extractvalue [4 x i16] [[B_ARR]], 0
225; CHECK-NEXT:    [[B2:%.*]] = extractvalue [4 x i16] [[B_ARR]], 2
226; CHECK-NEXT:    [[B1:%.*]] = extractvalue [4 x i16] [[B_ARR]], 1
227; CHECK-NEXT:    [[A3:%.*]] = extractvalue [4 x i16] [[A_ARR]], 3
228; CHECK-NEXT:    [[C1:%.*]] = sub i16 [[A1]], [[B1]]
229; CHECK-NEXT:    [[B3:%.*]] = extractvalue [4 x i16] [[B_ARR]], 3
230; CHECK-NEXT:    [[C0:%.*]] = sub i16 [[A0]], [[B0]]
231; CHECK-NEXT:    [[C2:%.*]] = sub i16 [[A2]], [[B2]]
232; CHECK-NEXT:    [[C_ARR0:%.*]] = insertvalue [4 x i16] undef, i16 [[C0]], 0
233; CHECK-NEXT:    [[C_ARR1:%.*]] = insertvalue [4 x i16] [[C_ARR0]], i16 [[C1]], 1
234; CHECK-NEXT:    [[C3:%.*]] = sub i16 [[A3]], [[B3]]
235; CHECK-NEXT:    [[C_ARR2:%.*]] = insertvalue [4 x i16] [[C_ARR1]], i16 [[C2]], 2
236; CHECK-NEXT:    [[C_ARR3:%.*]] = insertvalue [4 x i16] [[C_ARR2]], i16 [[C3]], 3
237; CHECK-NEXT:    store [4 x i16] [[C_ARR3]], [4 x i16]* [[C:%.*]], align 4
238; CHECK-NEXT:    ret void
239;
240top:
241  %a_arr = load [4 x i16], [4 x i16]* %a, align 4
242  %a0 = extractvalue [4 x i16] %a_arr, 0
243  %a2 = extractvalue [4 x i16] %a_arr, 2
244  %a1 = extractvalue [4 x i16] %a_arr, 1
245  %b_arr = load [4 x i16], [4 x i16]* %b, align 4
246  %b0 = extractvalue [4 x i16] %b_arr, 0
247  %b2 = extractvalue [4 x i16] %b_arr, 2
248  %b1 = extractvalue [4 x i16] %b_arr, 1
249  %a3 = extractvalue [4 x i16] %a_arr, 3
250  %c1 = sub i16 %a1, %b1
251  %b3 = extractvalue [4 x i16] %b_arr, 3
252  %c0 = sub i16 %a0, %b0
253  %c2 = sub i16 %a2, %b2
254  %c_arr0 = insertvalue [4 x i16] undef, i16 %c0, 0
255  %c_arr1 = insertvalue [4 x i16] %c_arr0, i16 %c1, 1
256  %c3 = sub i16 %a3, %b3
257  %c_arr2 = insertvalue [4 x i16] %c_arr1, i16 %c2, 2
258  %c_arr3 = insertvalue [4 x i16] %c_arr2, i16 %c3, 3
259  store [4 x i16] %c_arr3, [4 x i16]* %c, align 4
260  ret void
261}
262
263%pseudovec = type { float, float, float, float }
264
265define void @julia_load_struct_of_float(%pseudovec* %a, %pseudovec* %b, %pseudovec* %c) {
266; CHECK-LABEL: @julia_load_struct_of_float(
267; CHECK-NEXT:  top:
268; CHECK-NEXT:    [[TMP0:%.*]] = bitcast %pseudovec* [[A:%.*]] to <4 x float>*
269; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* [[TMP0]], align 4
270; CHECK-NEXT:    [[A_STRUCT:%.*]] = load [[PSEUDOVEC:%.*]], %pseudovec* [[A]], align 4
271; CHECK-NEXT:    [[TMP2:%.*]] = bitcast %pseudovec* [[B:%.*]] to <4 x float>*
272; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x float>, <4 x float>* [[TMP2]], align 4
273; CHECK-NEXT:    [[B_STRUCT:%.*]] = load [[PSEUDOVEC]], %pseudovec* [[B]], align 4
274; CHECK-NEXT:    [[TMP4:%.*]] = fsub <4 x float> [[TMP1]], [[TMP3]]
275; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x float> [[TMP4]], i32 0
276; CHECK-NEXT:    [[C_STRUCT0:%.*]] = insertvalue [[PSEUDOVEC]] undef, float [[TMP5]], 0
277; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[TMP4]], i32 1
278; CHECK-NEXT:    [[C_STRUCT1:%.*]] = insertvalue [[PSEUDOVEC]] %c_struct0, float [[TMP6]], 1
279; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x float> [[TMP4]], i32 2
280; CHECK-NEXT:    [[C_STRUCT2:%.*]] = insertvalue [[PSEUDOVEC]] %c_struct1, float [[TMP7]], 2
281; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <4 x float> [[TMP4]], i32 3
282; CHECK-NEXT:    [[C_STRUCT3:%.*]] = insertvalue [[PSEUDOVEC]] %c_struct2, float [[TMP8]], 3
283; CHECK-NEXT:    store [[PSEUDOVEC]] %c_struct3, %pseudovec* [[C:%.*]], align 4
284; CHECK-NEXT:    ret void
285;
286top:
287  %a_struct = load %pseudovec, %pseudovec* %a, align 4
288  %a0 = extractvalue %pseudovec %a_struct, 0
289  %a1 = extractvalue %pseudovec %a_struct, 1
290  %b_struct = load %pseudovec, %pseudovec* %b, align 4
291  %a2 = extractvalue %pseudovec %a_struct, 2
292  %b0 = extractvalue %pseudovec %b_struct, 0
293  %a3 = extractvalue %pseudovec %a_struct, 3
294  %c0 = fsub float %a0, %b0
295  %b1 = extractvalue %pseudovec %b_struct, 1
296  %b2 = extractvalue %pseudovec %b_struct, 2
297  %c1 = fsub float %a1, %b1
298  %c_struct0 = insertvalue %pseudovec undef, float %c0, 0
299  %b3 = extractvalue %pseudovec %b_struct, 3
300  %c3 = fsub float %a3, %b3
301  %c_struct1 = insertvalue %pseudovec %c_struct0, float %c1, 1
302  %c2 = fsub float %a2, %b2
303  %c_struct2 = insertvalue %pseudovec %c_struct1, float %c2, 2
304  %c_struct3 = insertvalue %pseudovec %c_struct2, float %c3, 3
305  store %pseudovec %c_struct3, %pseudovec* %c, align 4
306  ret void
307}
308