1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -S -slp-vectorizer -slp-threshold=-10000 < %s | FileCheck %s --check-prefixes=CHECK,THRESHOLD
3; RUN: opt -S -slp-vectorizer -slp-threshold=0 < %s | FileCheck %s --check-prefixes=CHECK,NOTHRESHOLD
4; RUN: opt -S -slp-vectorizer -slp-threshold=-10000 -slp-min-tree-size=0 < %s | FileCheck %s --check-prefixes=CHECK,MINTREESIZE
5
6target 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-n8:16:32:64-S128"
7target triple = "x86_64-apple-macosx10.8.0"
8
9define <4 x float> @simple_select(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
10; CHECK-LABEL: @simple_select(
11; CHECK-NEXT:    [[TMP1:%.*]] = icmp ne <4 x i32> [[C:%.*]], zeroinitializer
12; CHECK-NEXT:    [[TMP2:%.*]] = select <4 x i1> [[TMP1]], <4 x float> [[A:%.*]], <4 x float> [[B:%.*]]
13; CHECK-NEXT:    ret <4 x float> [[TMP2]]
14;
15  %c0 = extractelement <4 x i32> %c, i32 0
16  %c1 = extractelement <4 x i32> %c, i32 1
17  %c2 = extractelement <4 x i32> %c, i32 2
18  %c3 = extractelement <4 x i32> %c, i32 3
19  %a0 = extractelement <4 x float> %a, i32 0
20  %a1 = extractelement <4 x float> %a, i32 1
21  %a2 = extractelement <4 x float> %a, i32 2
22  %a3 = extractelement <4 x float> %a, i32 3
23  %b0 = extractelement <4 x float> %b, i32 0
24  %b1 = extractelement <4 x float> %b, i32 1
25  %b2 = extractelement <4 x float> %b, i32 2
26  %b3 = extractelement <4 x float> %b, i32 3
27  %cmp0 = icmp ne i32 %c0, 0
28  %cmp1 = icmp ne i32 %c1, 0
29  %cmp2 = icmp ne i32 %c2, 0
30  %cmp3 = icmp ne i32 %c3, 0
31  %s0 = select i1 %cmp0, float %a0, float %b0
32  %s1 = select i1 %cmp1, float %a1, float %b1
33  %s2 = select i1 %cmp2, float %a2, float %b2
34  %s3 = select i1 %cmp3, float %a3, float %b3
35  %ra = insertelement <4 x float> undef, float %s0, i32 0
36  %rb = insertelement <4 x float> %ra, float %s1, i32 1
37  %rc = insertelement <4 x float> %rb, float %s2, i32 2
38  %rd = insertelement <4 x float> %rc, float %s3, i32 3
39  ret <4 x float> %rd
40}
41
42define <8 x float> @simple_select2(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
43; CHECK-LABEL: @simple_select2(
44; CHECK-NEXT:    [[TMP1:%.*]] = icmp ne <4 x i32> [[C:%.*]], zeroinitializer
45; CHECK-NEXT:    [[TMP2:%.*]] = select <4 x i1> [[TMP1]], <4 x float> [[A:%.*]], <4 x float> [[B:%.*]]
46; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <4 x float> [[TMP2]], <4 x float> poison, <8 x i32> <i32 0, i32 undef, i32 1, i32 undef, i32 2, i32 undef, i32 undef, i32 3>
47; CHECK-NEXT:    [[RD1:%.*]] = shufflevector <8 x float> undef, <8 x float> [[TMP3]], <8 x i32> <i32 8, i32 1, i32 10, i32 3, i32 12, i32 5, i32 6, i32 15>
48; CHECK-NEXT:    ret <8 x float> [[RD1]]
49;
50  %c0 = extractelement <4 x i32> %c, i32 0
51  %c1 = extractelement <4 x i32> %c, i32 1
52  %c2 = extractelement <4 x i32> %c, i32 2
53  %c3 = extractelement <4 x i32> %c, i32 3
54  %a0 = extractelement <4 x float> %a, i32 0
55  %a1 = extractelement <4 x float> %a, i32 1
56  %a2 = extractelement <4 x float> %a, i32 2
57  %a3 = extractelement <4 x float> %a, i32 3
58  %b0 = extractelement <4 x float> %b, i32 0
59  %b1 = extractelement <4 x float> %b, i32 1
60  %b2 = extractelement <4 x float> %b, i32 2
61  %b3 = extractelement <4 x float> %b, i32 3
62  %cmp0 = icmp ne i32 %c0, 0
63  %cmp1 = icmp ne i32 %c1, 0
64  %cmp2 = icmp ne i32 %c2, 0
65  %cmp3 = icmp ne i32 %c3, 0
66  %s0 = select i1 %cmp0, float %a0, float %b0
67  %s1 = select i1 %cmp1, float %a1, float %b1
68  %s2 = select i1 %cmp2, float %a2, float %b2
69  %s3 = select i1 %cmp3, float %a3, float %b3
70  %ra = insertelement <8 x float> undef, float %s0, i32 0
71  %rb = insertelement <8 x float> %ra, float %s1, i32 2
72  %rc = insertelement <8 x float> %rb, float %s2, i32 4
73  %rd = insertelement <8 x float> %rc, float %s3, i32 7
74  ret <8 x float> %rd
75}
76
77declare void @llvm.assume(i1) nounwind
78
79; This entire tree is ephemeral, don't vectorize any of it.
80define <4 x float> @simple_select_eph(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
81; THRESHOLD-LABEL: @simple_select_eph(
82; THRESHOLD-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
83; THRESHOLD-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
84; THRESHOLD-NEXT:    [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2
85; THRESHOLD-NEXT:    [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3
86; THRESHOLD-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
87; THRESHOLD-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
88; THRESHOLD-NEXT:    [[A2:%.*]] = extractelement <4 x float> [[A]], i32 2
89; THRESHOLD-NEXT:    [[A3:%.*]] = extractelement <4 x float> [[A]], i32 3
90; THRESHOLD-NEXT:    [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0
91; THRESHOLD-NEXT:    [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1
92; THRESHOLD-NEXT:    [[B2:%.*]] = extractelement <4 x float> [[B]], i32 2
93; THRESHOLD-NEXT:    [[B3:%.*]] = extractelement <4 x float> [[B]], i32 3
94; THRESHOLD-NEXT:    [[CMP0:%.*]] = icmp ne i32 [[C0]], 0
95; THRESHOLD-NEXT:    [[CMP1:%.*]] = icmp ne i32 [[C1]], 0
96; THRESHOLD-NEXT:    [[CMP2:%.*]] = icmp ne i32 [[C2]], 0
97; THRESHOLD-NEXT:    [[CMP3:%.*]] = icmp ne i32 [[C3]], 0
98; THRESHOLD-NEXT:    [[S0:%.*]] = select i1 [[CMP0]], float [[A0]], float [[B0]]
99; THRESHOLD-NEXT:    [[S1:%.*]] = select i1 [[CMP1]], float [[A1]], float [[B1]]
100; THRESHOLD-NEXT:    [[S2:%.*]] = select i1 [[CMP2]], float [[A2]], float [[B2]]
101; THRESHOLD-NEXT:    [[S3:%.*]] = select i1 [[CMP3]], float [[A3]], float [[B3]]
102; THRESHOLD-NEXT:    [[RA:%.*]] = insertelement <4 x float> undef, float [[S0]], i32 0
103; THRESHOLD-NEXT:    [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[S1]], i32 1
104; THRESHOLD-NEXT:    [[RC:%.*]] = insertelement <4 x float> [[RB]], float [[S2]], i32 2
105; THRESHOLD-NEXT:    [[RD:%.*]] = insertelement <4 x float> [[RC]], float [[S3]], i32 3
106; THRESHOLD-NEXT:    [[Q0:%.*]] = extractelement <4 x float> [[RD]], i32 0
107; THRESHOLD-NEXT:    [[Q1:%.*]] = extractelement <4 x float> [[RD]], i32 1
108; THRESHOLD-NEXT:    [[Q2:%.*]] = extractelement <4 x float> [[RD]], i32 2
109; THRESHOLD-NEXT:    [[Q3:%.*]] = extractelement <4 x float> [[RD]], i32 3
110; THRESHOLD-NEXT:    [[Q4:%.*]] = fadd float [[Q0]], [[Q1]]
111; THRESHOLD-NEXT:    [[Q5:%.*]] = fadd float [[Q2]], [[Q3]]
112; THRESHOLD-NEXT:    [[Q6:%.*]] = fadd float [[Q4]], [[Q5]]
113; THRESHOLD-NEXT:    [[QI:%.*]] = fcmp olt float [[Q6]], [[Q5]]
114; THRESHOLD-NEXT:    call void @llvm.assume(i1 [[QI]])
115; THRESHOLD-NEXT:    ret <4 x float> undef
116;
117; NOTHRESHOLD-LABEL: @simple_select_eph(
118; NOTHRESHOLD-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
119; NOTHRESHOLD-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
120; NOTHRESHOLD-NEXT:    [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2
121; NOTHRESHOLD-NEXT:    [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3
122; NOTHRESHOLD-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
123; NOTHRESHOLD-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
124; NOTHRESHOLD-NEXT:    [[A2:%.*]] = extractelement <4 x float> [[A]], i32 2
125; NOTHRESHOLD-NEXT:    [[A3:%.*]] = extractelement <4 x float> [[A]], i32 3
126; NOTHRESHOLD-NEXT:    [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0
127; NOTHRESHOLD-NEXT:    [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1
128; NOTHRESHOLD-NEXT:    [[B2:%.*]] = extractelement <4 x float> [[B]], i32 2
129; NOTHRESHOLD-NEXT:    [[B3:%.*]] = extractelement <4 x float> [[B]], i32 3
130; NOTHRESHOLD-NEXT:    [[CMP0:%.*]] = icmp ne i32 [[C0]], 0
131; NOTHRESHOLD-NEXT:    [[CMP1:%.*]] = icmp ne i32 [[C1]], 0
132; NOTHRESHOLD-NEXT:    [[CMP2:%.*]] = icmp ne i32 [[C2]], 0
133; NOTHRESHOLD-NEXT:    [[CMP3:%.*]] = icmp ne i32 [[C3]], 0
134; NOTHRESHOLD-NEXT:    [[S0:%.*]] = select i1 [[CMP0]], float [[A0]], float [[B0]]
135; NOTHRESHOLD-NEXT:    [[S1:%.*]] = select i1 [[CMP1]], float [[A1]], float [[B1]]
136; NOTHRESHOLD-NEXT:    [[S2:%.*]] = select i1 [[CMP2]], float [[A2]], float [[B2]]
137; NOTHRESHOLD-NEXT:    [[S3:%.*]] = select i1 [[CMP3]], float [[A3]], float [[B3]]
138; NOTHRESHOLD-NEXT:    [[RA:%.*]] = insertelement <4 x float> undef, float [[S0]], i32 0
139; NOTHRESHOLD-NEXT:    [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[S1]], i32 1
140; NOTHRESHOLD-NEXT:    [[RC:%.*]] = insertelement <4 x float> [[RB]], float [[S2]], i32 2
141; NOTHRESHOLD-NEXT:    [[RD:%.*]] = insertelement <4 x float> [[RC]], float [[S3]], i32 3
142; NOTHRESHOLD-NEXT:    [[Q0:%.*]] = extractelement <4 x float> [[RD]], i32 0
143; NOTHRESHOLD-NEXT:    [[Q1:%.*]] = extractelement <4 x float> [[RD]], i32 1
144; NOTHRESHOLD-NEXT:    [[Q2:%.*]] = extractelement <4 x float> [[RD]], i32 2
145; NOTHRESHOLD-NEXT:    [[Q3:%.*]] = extractelement <4 x float> [[RD]], i32 3
146; NOTHRESHOLD-NEXT:    [[Q4:%.*]] = fadd float [[Q0]], [[Q1]]
147; NOTHRESHOLD-NEXT:    [[Q5:%.*]] = fadd float [[Q2]], [[Q3]]
148; NOTHRESHOLD-NEXT:    [[Q6:%.*]] = fadd float [[Q4]], [[Q5]]
149; NOTHRESHOLD-NEXT:    [[QI:%.*]] = fcmp olt float [[Q6]], [[Q5]]
150; NOTHRESHOLD-NEXT:    call void @llvm.assume(i1 [[QI]])
151; NOTHRESHOLD-NEXT:    ret <4 x float> undef
152;
153; MINTREESIZE-LABEL: @simple_select_eph(
154; MINTREESIZE-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
155; MINTREESIZE-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
156; MINTREESIZE-NEXT:    [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2
157; MINTREESIZE-NEXT:    [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3
158; MINTREESIZE-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
159; MINTREESIZE-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
160; MINTREESIZE-NEXT:    [[A2:%.*]] = extractelement <4 x float> [[A]], i32 2
161; MINTREESIZE-NEXT:    [[A3:%.*]] = extractelement <4 x float> [[A]], i32 3
162; MINTREESIZE-NEXT:    [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0
163; MINTREESIZE-NEXT:    [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1
164; MINTREESIZE-NEXT:    [[B2:%.*]] = extractelement <4 x float> [[B]], i32 2
165; MINTREESIZE-NEXT:    [[B3:%.*]] = extractelement <4 x float> [[B]], i32 3
166; MINTREESIZE-NEXT:    [[CMP0:%.*]] = icmp ne i32 [[C0]], 0
167; MINTREESIZE-NEXT:    [[CMP1:%.*]] = icmp ne i32 [[C1]], 0
168; MINTREESIZE-NEXT:    [[CMP2:%.*]] = icmp ne i32 [[C2]], 0
169; MINTREESIZE-NEXT:    [[CMP3:%.*]] = icmp ne i32 [[C3]], 0
170; MINTREESIZE-NEXT:    [[TMP1:%.*]] = insertelement <4 x i1> poison, i1 [[CMP3]], i32 0
171; MINTREESIZE-NEXT:    [[TMP2:%.*]] = insertelement <4 x i1> [[TMP1]], i1 [[CMP2]], i32 1
172; MINTREESIZE-NEXT:    [[TMP3:%.*]] = insertelement <4 x i1> [[TMP2]], i1 [[CMP1]], i32 2
173; MINTREESIZE-NEXT:    [[TMP4:%.*]] = insertelement <4 x i1> [[TMP3]], i1 [[CMP0]], i32 3
174; MINTREESIZE-NEXT:    [[S0:%.*]] = select i1 [[CMP0]], float [[A0]], float [[B0]]
175; MINTREESIZE-NEXT:    [[S1:%.*]] = select i1 [[CMP1]], float [[A1]], float [[B1]]
176; MINTREESIZE-NEXT:    [[S2:%.*]] = select i1 [[CMP2]], float [[A2]], float [[B2]]
177; MINTREESIZE-NEXT:    [[S3:%.*]] = select i1 [[CMP3]], float [[A3]], float [[B3]]
178; MINTREESIZE-NEXT:    [[RA:%.*]] = insertelement <4 x float> undef, float [[S0]], i32 0
179; MINTREESIZE-NEXT:    [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[S1]], i32 1
180; MINTREESIZE-NEXT:    [[RC:%.*]] = insertelement <4 x float> [[RB]], float [[S2]], i32 2
181; MINTREESIZE-NEXT:    [[RD:%.*]] = insertelement <4 x float> [[RC]], float [[S3]], i32 3
182; MINTREESIZE-NEXT:    [[Q0:%.*]] = extractelement <4 x float> [[RD]], i32 0
183; MINTREESIZE-NEXT:    [[Q1:%.*]] = extractelement <4 x float> [[RD]], i32 1
184; MINTREESIZE-NEXT:    [[TMP5:%.*]] = insertelement <2 x float> poison, float [[Q0]], i32 0
185; MINTREESIZE-NEXT:    [[TMP6:%.*]] = insertelement <2 x float> [[TMP5]], float [[Q1]], i32 1
186; MINTREESIZE-NEXT:    [[Q2:%.*]] = extractelement <4 x float> [[RD]], i32 2
187; MINTREESIZE-NEXT:    [[Q3:%.*]] = extractelement <4 x float> [[RD]], i32 3
188; MINTREESIZE-NEXT:    [[TMP7:%.*]] = insertelement <2 x float> poison, float [[Q2]], i32 0
189; MINTREESIZE-NEXT:    [[TMP8:%.*]] = insertelement <2 x float> [[TMP7]], float [[Q3]], i32 1
190; MINTREESIZE-NEXT:    [[Q4:%.*]] = fadd float [[Q0]], [[Q1]]
191; MINTREESIZE-NEXT:    [[Q5:%.*]] = fadd float [[Q2]], [[Q3]]
192; MINTREESIZE-NEXT:    [[TMP9:%.*]] = insertelement <2 x float> poison, float [[Q4]], i32 0
193; MINTREESIZE-NEXT:    [[TMP10:%.*]] = insertelement <2 x float> [[TMP9]], float [[Q5]], i32 1
194; MINTREESIZE-NEXT:    [[Q6:%.*]] = fadd float [[Q4]], [[Q5]]
195; MINTREESIZE-NEXT:    [[TMP11:%.*]] = insertelement <2 x float> poison, float [[Q6]], i32 0
196; MINTREESIZE-NEXT:    [[TMP12:%.*]] = insertelement <2 x float> [[TMP11]], float [[Q5]], i32 1
197; MINTREESIZE-NEXT:    [[QI:%.*]] = fcmp olt float [[Q6]], [[Q5]]
198; MINTREESIZE-NEXT:    call void @llvm.assume(i1 [[QI]])
199; MINTREESIZE-NEXT:    ret <4 x float> undef
200;
201  %c0 = extractelement <4 x i32> %c, i32 0
202  %c1 = extractelement <4 x i32> %c, i32 1
203  %c2 = extractelement <4 x i32> %c, i32 2
204  %c3 = extractelement <4 x i32> %c, i32 3
205  %a0 = extractelement <4 x float> %a, i32 0
206  %a1 = extractelement <4 x float> %a, i32 1
207  %a2 = extractelement <4 x float> %a, i32 2
208  %a3 = extractelement <4 x float> %a, i32 3
209  %b0 = extractelement <4 x float> %b, i32 0
210  %b1 = extractelement <4 x float> %b, i32 1
211  %b2 = extractelement <4 x float> %b, i32 2
212  %b3 = extractelement <4 x float> %b, i32 3
213  %cmp0 = icmp ne i32 %c0, 0
214  %cmp1 = icmp ne i32 %c1, 0
215  %cmp2 = icmp ne i32 %c2, 0
216  %cmp3 = icmp ne i32 %c3, 0
217  %s0 = select i1 %cmp0, float %a0, float %b0
218  %s1 = select i1 %cmp1, float %a1, float %b1
219  %s2 = select i1 %cmp2, float %a2, float %b2
220  %s3 = select i1 %cmp3, float %a3, float %b3
221  %ra = insertelement <4 x float> undef, float %s0, i32 0
222  %rb = insertelement <4 x float> %ra, float %s1, i32 1
223  %rc = insertelement <4 x float> %rb, float %s2, i32 2
224  %rd = insertelement <4 x float> %rc, float %s3, i32 3
225  %q0 = extractelement <4 x float> %rd, i32 0
226  %q1 = extractelement <4 x float> %rd, i32 1
227  %q2 = extractelement <4 x float> %rd, i32 2
228  %q3 = extractelement <4 x float> %rd, i32 3
229  %q4 = fadd float %q0, %q1
230  %q5 = fadd float %q2, %q3
231  %q6 = fadd float %q4, %q5
232  %qi = fcmp olt float %q6, %q5
233  call void @llvm.assume(i1 %qi)
234  ret <4 x float> undef
235}
236
237; Insert in an order different from the vector indices to make sure it
238; doesn't matter
239define <4 x float> @simple_select_insert_out_of_order(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
240; CHECK-LABEL: @simple_select_insert_out_of_order(
241; CHECK-NEXT:    [[TMP1:%.*]] = icmp ne <4 x i32> [[C:%.*]], zeroinitializer
242; CHECK-NEXT:    [[TMP2:%.*]] = select <4 x i1> [[TMP1]], <4 x float> [[A:%.*]], <4 x float> [[B:%.*]]
243; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <4 x float> [[TMP2]], <4 x float> poison, <4 x i32> <i32 2, i32 1, i32 0, i32 3>
244; CHECK-NEXT:    ret <4 x float> [[TMP3]]
245;
246  %c0 = extractelement <4 x i32> %c, i32 0
247  %c1 = extractelement <4 x i32> %c, i32 1
248  %c2 = extractelement <4 x i32> %c, i32 2
249  %c3 = extractelement <4 x i32> %c, i32 3
250  %a0 = extractelement <4 x float> %a, i32 0
251  %a1 = extractelement <4 x float> %a, i32 1
252  %a2 = extractelement <4 x float> %a, i32 2
253  %a3 = extractelement <4 x float> %a, i32 3
254  %b0 = extractelement <4 x float> %b, i32 0
255  %b1 = extractelement <4 x float> %b, i32 1
256  %b2 = extractelement <4 x float> %b, i32 2
257  %b3 = extractelement <4 x float> %b, i32 3
258  %cmp0 = icmp ne i32 %c0, 0
259  %cmp1 = icmp ne i32 %c1, 0
260  %cmp2 = icmp ne i32 %c2, 0
261  %cmp3 = icmp ne i32 %c3, 0
262  %s0 = select i1 %cmp0, float %a0, float %b0
263  %s1 = select i1 %cmp1, float %a1, float %b1
264  %s2 = select i1 %cmp2, float %a2, float %b2
265  %s3 = select i1 %cmp3, float %a3, float %b3
266  %ra = insertelement <4 x float> undef, float %s0, i32 2
267  %rb = insertelement <4 x float> %ra, float %s1, i32 1
268  %rc = insertelement <4 x float> %rb, float %s2, i32 0
269  %rd = insertelement <4 x float> %rc, float %s3, i32 3
270  ret <4 x float> %rd
271}
272
273declare void @v4f32_user(<4 x float>) #0
274declare void @f32_user(float) #0
275
276; Multiple users of the final constructed vector
277define <4 x float> @simple_select_users(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
278; CHECK-LABEL: @simple_select_users(
279; CHECK-NEXT:    [[TMP1:%.*]] = icmp ne <4 x i32> [[C:%.*]], zeroinitializer
280; CHECK-NEXT:    [[TMP2:%.*]] = select <4 x i1> [[TMP1]], <4 x float> [[A:%.*]], <4 x float> [[B:%.*]]
281; CHECK-NEXT:    call void @v4f32_user(<4 x float> [[TMP2]]) #[[ATTR0:[0-9]+]]
282; CHECK-NEXT:    ret <4 x float> [[TMP2]]
283;
284  %c0 = extractelement <4 x i32> %c, i32 0
285  %c1 = extractelement <4 x i32> %c, i32 1
286  %c2 = extractelement <4 x i32> %c, i32 2
287  %c3 = extractelement <4 x i32> %c, i32 3
288  %a0 = extractelement <4 x float> %a, i32 0
289  %a1 = extractelement <4 x float> %a, i32 1
290  %a2 = extractelement <4 x float> %a, i32 2
291  %a3 = extractelement <4 x float> %a, i32 3
292  %b0 = extractelement <4 x float> %b, i32 0
293  %b1 = extractelement <4 x float> %b, i32 1
294  %b2 = extractelement <4 x float> %b, i32 2
295  %b3 = extractelement <4 x float> %b, i32 3
296  %cmp0 = icmp ne i32 %c0, 0
297  %cmp1 = icmp ne i32 %c1, 0
298  %cmp2 = icmp ne i32 %c2, 0
299  %cmp3 = icmp ne i32 %c3, 0
300  %s0 = select i1 %cmp0, float %a0, float %b0
301  %s1 = select i1 %cmp1, float %a1, float %b1
302  %s2 = select i1 %cmp2, float %a2, float %b2
303  %s3 = select i1 %cmp3, float %a3, float %b3
304  %ra = insertelement <4 x float> undef, float %s0, i32 0
305  %rb = insertelement <4 x float> %ra, float %s1, i32 1
306  %rc = insertelement <4 x float> %rb, float %s2, i32 2
307  %rd = insertelement <4 x float> %rc, float %s3, i32 3
308  call void @v4f32_user(<4 x float> %rd) #0
309  ret <4 x float> %rd
310}
311
312; Unused insertelement
313define <4 x float> @simple_select_no_users(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
314; CHECK-LABEL: @simple_select_no_users(
315; CHECK-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
316; CHECK-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
317; CHECK-NEXT:    [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2
318; CHECK-NEXT:    [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3
319; CHECK-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
320; CHECK-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
321; CHECK-NEXT:    [[A2:%.*]] = extractelement <4 x float> [[A]], i32 2
322; CHECK-NEXT:    [[A3:%.*]] = extractelement <4 x float> [[A]], i32 3
323; CHECK-NEXT:    [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0
324; CHECK-NEXT:    [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1
325; CHECK-NEXT:    [[B2:%.*]] = extractelement <4 x float> [[B]], i32 2
326; CHECK-NEXT:    [[B3:%.*]] = extractelement <4 x float> [[B]], i32 3
327; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x i32> poison, i32 [[C0]], i32 0
328; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x i32> [[TMP1]], i32 [[C1]], i32 1
329; CHECK-NEXT:    [[TMP3:%.*]] = icmp ne <2 x i32> [[TMP2]], zeroinitializer
330; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <2 x i32> poison, i32 [[C2]], i32 0
331; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <2 x i32> [[TMP4]], i32 [[C3]], i32 1
332; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <2 x i32> [[TMP5]], zeroinitializer
333; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x float> poison, float [[A0]], i32 0
334; CHECK-NEXT:    [[TMP8:%.*]] = insertelement <2 x float> [[TMP7]], float [[A1]], i32 1
335; CHECK-NEXT:    [[TMP9:%.*]] = insertelement <2 x float> poison, float [[B0]], i32 0
336; CHECK-NEXT:    [[TMP10:%.*]] = insertelement <2 x float> [[TMP9]], float [[B1]], i32 1
337; CHECK-NEXT:    [[TMP11:%.*]] = select <2 x i1> [[TMP3]], <2 x float> [[TMP8]], <2 x float> [[TMP10]]
338; CHECK-NEXT:    [[TMP12:%.*]] = insertelement <2 x float> poison, float [[A2]], i32 0
339; CHECK-NEXT:    [[TMP13:%.*]] = insertelement <2 x float> [[TMP12]], float [[A3]], i32 1
340; CHECK-NEXT:    [[TMP14:%.*]] = insertelement <2 x float> poison, float [[B2]], i32 0
341; CHECK-NEXT:    [[TMP15:%.*]] = insertelement <2 x float> [[TMP14]], float [[B3]], i32 1
342; CHECK-NEXT:    [[TMP16:%.*]] = select <2 x i1> [[TMP6]], <2 x float> [[TMP13]], <2 x float> [[TMP15]]
343; CHECK-NEXT:    [[TMP17:%.*]] = shufflevector <2 x float> [[TMP11]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
344; CHECK-NEXT:    [[TMP18:%.*]] = shufflevector <2 x float> [[TMP16]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
345; CHECK-NEXT:    [[RD1:%.*]] = shufflevector <4 x float> undef, <4 x float> [[TMP18]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
346; CHECK-NEXT:    ret <4 x float> [[RD1]]
347;
348  %c0 = extractelement <4 x i32> %c, i32 0
349  %c1 = extractelement <4 x i32> %c, i32 1
350  %c2 = extractelement <4 x i32> %c, i32 2
351  %c3 = extractelement <4 x i32> %c, i32 3
352  %a0 = extractelement <4 x float> %a, i32 0
353  %a1 = extractelement <4 x float> %a, i32 1
354  %a2 = extractelement <4 x float> %a, i32 2
355  %a3 = extractelement <4 x float> %a, i32 3
356  %b0 = extractelement <4 x float> %b, i32 0
357  %b1 = extractelement <4 x float> %b, i32 1
358  %b2 = extractelement <4 x float> %b, i32 2
359  %b3 = extractelement <4 x float> %b, i32 3
360  %cmp0 = icmp ne i32 %c0, 0
361  %cmp1 = icmp ne i32 %c1, 0
362  %cmp2 = icmp ne i32 %c2, 0
363  %cmp3 = icmp ne i32 %c3, 0
364  %s0 = select i1 %cmp0, float %a0, float %b0
365  %s1 = select i1 %cmp1, float %a1, float %b1
366  %s2 = select i1 %cmp2, float %a2, float %b2
367  %s3 = select i1 %cmp3, float %a3, float %b3
368  %ra = insertelement <4 x float> undef, float %s0, i32 0
369  %rb = insertelement <4 x float> %ra, float %s1, i32 1
370  %rc = insertelement <4 x float> undef, float %s2, i32 2
371  %rd = insertelement <4 x float> %rc, float %s3, i32 3
372  ret <4 x float> %rd
373}
374
375; Make sure infinite loop doesn't happen which I ran into when trying
376; to do this backwards this backwards
377define <4 x i32> @reconstruct(<4 x i32> %c) #0 {
378; CHECK-LABEL: @reconstruct(
379; CHECK-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
380; CHECK-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
381; CHECK-NEXT:    [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2
382; CHECK-NEXT:    [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3
383; CHECK-NEXT:    [[RA:%.*]] = insertelement <4 x i32> undef, i32 [[C0]], i32 0
384; CHECK-NEXT:    [[RB:%.*]] = insertelement <4 x i32> [[RA]], i32 [[C1]], i32 1
385; CHECK-NEXT:    [[RC:%.*]] = insertelement <4 x i32> [[RB]], i32 [[C2]], i32 2
386; CHECK-NEXT:    [[RD:%.*]] = insertelement <4 x i32> [[RC]], i32 [[C3]], i32 3
387; CHECK-NEXT:    ret <4 x i32> [[RD]]
388;
389  %c0 = extractelement <4 x i32> %c, i32 0
390  %c1 = extractelement <4 x i32> %c, i32 1
391  %c2 = extractelement <4 x i32> %c, i32 2
392  %c3 = extractelement <4 x i32> %c, i32 3
393  %ra = insertelement <4 x i32> undef, i32 %c0, i32 0
394  %rb = insertelement <4 x i32> %ra, i32 %c1, i32 1
395  %rc = insertelement <4 x i32> %rb, i32 %c2, i32 2
396  %rd = insertelement <4 x i32> %rc, i32 %c3, i32 3
397  ret <4 x i32> %rd
398}
399
400define <2 x float> @simple_select_v2(<2 x float> %a, <2 x float> %b, <2 x i32> %c) #0 {
401; CHECK-LABEL: @simple_select_v2(
402; CHECK-NEXT:    [[TMP1:%.*]] = icmp ne <2 x i32> [[C:%.*]], zeroinitializer
403; CHECK-NEXT:    [[TMP2:%.*]] = select <2 x i1> [[TMP1]], <2 x float> [[A:%.*]], <2 x float> [[B:%.*]]
404; CHECK-NEXT:    ret <2 x float> [[TMP2]]
405;
406  %c0 = extractelement <2 x i32> %c, i32 0
407  %c1 = extractelement <2 x i32> %c, i32 1
408  %a0 = extractelement <2 x float> %a, i32 0
409  %a1 = extractelement <2 x float> %a, i32 1
410  %b0 = extractelement <2 x float> %b, i32 0
411  %b1 = extractelement <2 x float> %b, i32 1
412  %cmp0 = icmp ne i32 %c0, 0
413  %cmp1 = icmp ne i32 %c1, 0
414  %s0 = select i1 %cmp0, float %a0, float %b0
415  %s1 = select i1 %cmp1, float %a1, float %b1
416  %ra = insertelement <2 x float> undef, float %s0, i32 0
417  %rb = insertelement <2 x float> %ra, float %s1, i32 1
418  ret <2 x float> %rb
419}
420
421; Make sure when we construct partial vectors, we don't keep
422; re-visiting the insertelement chains starting with undef
423; (low cost threshold needed to force this to happen)
424define <4 x float> @simple_select_partial_vector(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
425; CHECK-LABEL: @simple_select_partial_vector(
426; CHECK-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
427; CHECK-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
428; CHECK-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
429; CHECK-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
430; CHECK-NEXT:    [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0
431; CHECK-NEXT:    [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1
432; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x i32> undef, i32 [[C0]], i32 0
433; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x i32> [[TMP1]], i32 [[C1]], i32 1
434; CHECK-NEXT:    [[TMP3:%.*]] = icmp ne <2 x i32> [[TMP2]], zeroinitializer
435; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <2 x float> undef, float [[A0]], i32 0
436; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <2 x float> [[TMP4]], float [[A1]], i32 1
437; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <2 x float> undef, float [[B0]], i32 0
438; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x float> [[TMP6]], float [[B1]], i32 1
439; CHECK-NEXT:    [[TMP8:%.*]] = select <2 x i1> [[TMP3]], <2 x float> [[TMP5]], <2 x float> [[TMP7]]
440; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <2 x float> [[TMP8]], i32 0
441; CHECK-NEXT:    [[RA:%.*]] = insertelement <4 x float> undef, float [[TMP9]], i32 0
442; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <2 x float> [[TMP8]], i32 1
443; CHECK-NEXT:    [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[TMP10]], i32 1
444; CHECK-NEXT:    ret <4 x float> [[RB]]
445;
446  %c0 = extractelement <4 x i32> %c, i32 0
447  %c1 = extractelement <4 x i32> %c, i32 1
448  %a0 = extractelement <4 x float> %a, i32 0
449  %a1 = extractelement <4 x float> %a, i32 1
450  %b0 = extractelement <4 x float> %b, i32 0
451  %b1 = extractelement <4 x float> %b, i32 1
452  %1 = insertelement <2 x i32> undef, i32 %c0, i32 0
453  %2 = insertelement <2 x i32> %1, i32 %c1, i32 1
454  %3 = icmp ne <2 x i32> %2, zeroinitializer
455  %4 = insertelement <2 x float> undef, float %a0, i32 0
456  %5 = insertelement <2 x float> %4, float %a1, i32 1
457  %6 = insertelement <2 x float> undef, float %b0, i32 0
458  %7 = insertelement <2 x float> %6, float %b1, i32 1
459  %8 = select <2 x i1> %3, <2 x float> %5, <2 x float> %7
460  %9 = extractelement <2 x float> %8, i32 0
461  %ra = insertelement <4 x float> undef, float %9, i32 0
462  %10 = extractelement <2 x float> %8, i32 1
463  %rb = insertelement <4 x float> %ra, float %10, i32 1
464  ret <4 x float> %rb
465}
466
467; Make sure that vectorization happens even if insertelements operations
468; must be rescheduled. The case here is from compiling Julia.
469define <4 x float> @reschedule_extract(<4 x float> %a, <4 x float> %b) {
470; THRESHOLD-LABEL: @reschedule_extract(
471; THRESHOLD-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A:%.*]], [[B:%.*]]
472; THRESHOLD-NEXT:    ret <4 x float> [[TMP1]]
473;
474; NOTHRESHOLD-LABEL: @reschedule_extract(
475; NOTHRESHOLD-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A:%.*]], [[B:%.*]]
476; NOTHRESHOLD-NEXT:    ret <4 x float> [[TMP1]]
477;
478; MINTREESIZE-LABEL: @reschedule_extract(
479; MINTREESIZE-NEXT:    [[TMP1:%.*]] = extractelement <4 x float> [[B:%.*]], i32 3
480; MINTREESIZE-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> [[B]], i32 2
481; MINTREESIZE-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[B]], i32 1
482; MINTREESIZE-NEXT:    [[TMP4:%.*]] = extractelement <4 x float> [[B]], i32 0
483; MINTREESIZE-NEXT:    [[TMP5:%.*]] = extractelement <4 x float> [[A:%.*]], i32 3
484; MINTREESIZE-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[A]], i32 2
485; MINTREESIZE-NEXT:    [[TMP7:%.*]] = extractelement <4 x float> [[A]], i32 1
486; MINTREESIZE-NEXT:    [[TMP8:%.*]] = extractelement <4 x float> [[A]], i32 0
487; MINTREESIZE-NEXT:    [[TMP9:%.*]] = insertelement <2 x float> poison, float [[TMP8]], i32 0
488; MINTREESIZE-NEXT:    [[TMP10:%.*]] = insertelement <2 x float> [[TMP9]], float [[TMP4]], i32 1
489; MINTREESIZE-NEXT:    [[TMP11:%.*]] = fadd <4 x float> [[A]], [[B]]
490; MINTREESIZE-NEXT:    [[TMP12:%.*]] = insertelement <2 x float> poison, float [[TMP7]], i32 0
491; MINTREESIZE-NEXT:    [[TMP13:%.*]] = insertelement <2 x float> [[TMP12]], float [[TMP3]], i32 1
492; MINTREESIZE-NEXT:    [[TMP14:%.*]] = insertelement <2 x float> poison, float [[TMP6]], i32 0
493; MINTREESIZE-NEXT:    [[TMP15:%.*]] = insertelement <2 x float> [[TMP14]], float [[TMP2]], i32 1
494; MINTREESIZE-NEXT:    [[TMP16:%.*]] = insertelement <2 x float> poison, float [[TMP5]], i32 0
495; MINTREESIZE-NEXT:    [[TMP17:%.*]] = insertelement <2 x float> [[TMP16]], float [[TMP1]], i32 1
496; MINTREESIZE-NEXT:    ret <4 x float> [[TMP11]]
497;
498  %a0 = extractelement <4 x float> %a, i32 0
499  %b0 = extractelement <4 x float> %b, i32 0
500  %c0 = fadd float %a0, %b0
501  %v0 = insertelement <4 x float> undef, float %c0, i32 0
502  %a1 = extractelement <4 x float> %a, i32 1
503  %b1 = extractelement <4 x float> %b, i32 1
504  %c1 = fadd float %a1, %b1
505  %v1 = insertelement <4 x float> %v0, float %c1, i32 1
506  %a2 = extractelement <4 x float> %a, i32 2
507  %b2 = extractelement <4 x float> %b, i32 2
508  %c2 = fadd float %a2, %b2
509  %v2 = insertelement <4 x float> %v1, float %c2, i32 2
510  %a3 = extractelement <4 x float> %a, i32 3
511  %b3 = extractelement <4 x float> %b, i32 3
512  %c3 = fadd float %a3, %b3
513  %v3 = insertelement <4 x float> %v2, float %c3, i32 3
514  ret <4 x float> %v3
515}
516
517; Check that cost model for vectorization takes credit for
518; instructions that are erased.
519define <4 x float> @take_credit(<4 x float> %a, <4 x float> %b) {
520; THRESHOLD-LABEL: @take_credit(
521; THRESHOLD-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A:%.*]], [[B:%.*]]
522; THRESHOLD-NEXT:    ret <4 x float> [[TMP1]]
523;
524; NOTHRESHOLD-LABEL: @take_credit(
525; NOTHRESHOLD-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A:%.*]], [[B:%.*]]
526; NOTHRESHOLD-NEXT:    ret <4 x float> [[TMP1]]
527;
528; MINTREESIZE-LABEL: @take_credit(
529; MINTREESIZE-NEXT:    [[TMP1:%.*]] = extractelement <4 x float> [[B:%.*]], i32 3
530; MINTREESIZE-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> [[B]], i32 2
531; MINTREESIZE-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[B]], i32 1
532; MINTREESIZE-NEXT:    [[TMP4:%.*]] = extractelement <4 x float> [[B]], i32 0
533; MINTREESIZE-NEXT:    [[TMP5:%.*]] = extractelement <4 x float> [[A:%.*]], i32 3
534; MINTREESIZE-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[A]], i32 2
535; MINTREESIZE-NEXT:    [[TMP7:%.*]] = extractelement <4 x float> [[A]], i32 1
536; MINTREESIZE-NEXT:    [[TMP8:%.*]] = extractelement <4 x float> [[A]], i32 0
537; MINTREESIZE-NEXT:    [[TMP9:%.*]] = insertelement <2 x float> poison, float [[TMP8]], i32 0
538; MINTREESIZE-NEXT:    [[TMP10:%.*]] = insertelement <2 x float> [[TMP9]], float [[TMP4]], i32 1
539; MINTREESIZE-NEXT:    [[TMP11:%.*]] = insertelement <2 x float> poison, float [[TMP7]], i32 0
540; MINTREESIZE-NEXT:    [[TMP12:%.*]] = insertelement <2 x float> [[TMP11]], float [[TMP3]], i32 1
541; MINTREESIZE-NEXT:    [[TMP13:%.*]] = insertelement <2 x float> poison, float [[TMP6]], i32 0
542; MINTREESIZE-NEXT:    [[TMP14:%.*]] = insertelement <2 x float> [[TMP13]], float [[TMP2]], i32 1
543; MINTREESIZE-NEXT:    [[TMP15:%.*]] = insertelement <2 x float> poison, float [[TMP5]], i32 0
544; MINTREESIZE-NEXT:    [[TMP16:%.*]] = insertelement <2 x float> [[TMP15]], float [[TMP1]], i32 1
545; MINTREESIZE-NEXT:    [[TMP17:%.*]] = fadd <4 x float> [[A]], [[B]]
546; MINTREESIZE-NEXT:    ret <4 x float> [[TMP17]]
547;
548  %a0 = extractelement <4 x float> %a, i32 0
549  %b0 = extractelement <4 x float> %b, i32 0
550  %c0 = fadd float %a0, %b0
551  %a1 = extractelement <4 x float> %a, i32 1
552  %b1 = extractelement <4 x float> %b, i32 1
553  %c1 = fadd float %a1, %b1
554  %a2 = extractelement <4 x float> %a, i32 2
555  %b2 = extractelement <4 x float> %b, i32 2
556  %c2 = fadd float %a2, %b2
557  %a3 = extractelement <4 x float> %a, i32 3
558  %b3 = extractelement <4 x float> %b, i32 3
559  %c3 = fadd float %a3, %b3
560  %v0 = insertelement <4 x float> undef, float %c0, i32 0
561  %v1 = insertelement <4 x float> %v0, float %c1, i32 1
562  %v2 = insertelement <4 x float> %v1, float %c2, i32 2
563  %v3 = insertelement <4 x float> %v2, float %c3, i32 3
564  ret <4 x float> %v3
565}
566
567; Make sure we handle multiple trees that feed one build vector correctly.
568define <4 x double> @multi_tree(double %w, double %x, double %y, double %z) {
569; CHECK-LABEL: @multi_tree(
570; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x double> poison, double [[Z:%.*]], i32 0
571; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x double> [[TMP1]], double [[Y:%.*]], i32 1
572; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x double> [[TMP2]], double [[X:%.*]], i32 2
573; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x double> [[TMP3]], double [[W:%.*]], i32 3
574; CHECK-NEXT:    [[TMP5:%.*]] = fadd <4 x double> [[TMP4]], <double 3.000000e+00, double 2.000000e+00, double 1.000000e+00, double 0.000000e+00>
575; CHECK-NEXT:    [[TMP6:%.*]] = fmul <4 x double> [[TMP5]], <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00>
576; CHECK-NEXT:    ret <4 x double> [[TMP6]]
577;
578  %t0 = fadd double %w , 0.000000e+00
579  %t1 = fadd double %x , 1.000000e+00
580  %t2 = fadd double %y , 2.000000e+00
581  %t3 = fadd double %z , 3.000000e+00
582  %t4 = fmul double %t0, 1.000000e+00
583  %i1 = insertelement <4 x double> undef, double %t4, i32 3
584  %t5 = fmul double %t1, 1.000000e+00
585  %i2 = insertelement <4 x double> %i1, double %t5, i32 2
586  %t6 = fmul double %t2, 1.000000e+00
587  %i3 = insertelement <4 x double> %i2, double %t6, i32 1
588  %t7 = fmul double %t3, 1.000000e+00
589  %i4 = insertelement <4 x double> %i3, double %t7, i32 0
590  ret <4 x double> %i4
591}
592
593define <8 x float> @_vadd256(<8 x float> %a, <8 x float> %b) local_unnamed_addr #0 {
594; THRESHOLD-LABEL: @_vadd256(
595; THRESHOLD-NEXT:    [[TMP1:%.*]] = fadd <8 x float> [[A:%.*]], [[B:%.*]]
596; THRESHOLD-NEXT:    ret <8 x float> [[TMP1]]
597;
598; NOTHRESHOLD-LABEL: @_vadd256(
599; NOTHRESHOLD-NEXT:    [[TMP1:%.*]] = fadd <8 x float> [[A:%.*]], [[B:%.*]]
600; NOTHRESHOLD-NEXT:    ret <8 x float> [[TMP1]]
601;
602; MINTREESIZE-LABEL: @_vadd256(
603; MINTREESIZE-NEXT:    [[TMP1:%.*]] = extractelement <8 x float> [[B:%.*]], i32 7
604; MINTREESIZE-NEXT:    [[TMP2:%.*]] = extractelement <8 x float> [[B]], i32 6
605; MINTREESIZE-NEXT:    [[TMP3:%.*]] = extractelement <8 x float> [[B]], i32 5
606; MINTREESIZE-NEXT:    [[TMP4:%.*]] = extractelement <8 x float> [[B]], i32 4
607; MINTREESIZE-NEXT:    [[TMP5:%.*]] = extractelement <8 x float> [[B]], i32 3
608; MINTREESIZE-NEXT:    [[TMP6:%.*]] = extractelement <8 x float> [[B]], i32 2
609; MINTREESIZE-NEXT:    [[TMP7:%.*]] = extractelement <8 x float> [[B]], i32 1
610; MINTREESIZE-NEXT:    [[TMP8:%.*]] = extractelement <8 x float> [[B]], i32 0
611; MINTREESIZE-NEXT:    [[TMP9:%.*]] = extractelement <8 x float> [[A:%.*]], i32 7
612; MINTREESIZE-NEXT:    [[TMP10:%.*]] = extractelement <8 x float> [[A]], i32 6
613; MINTREESIZE-NEXT:    [[TMP11:%.*]] = extractelement <8 x float> [[A]], i32 5
614; MINTREESIZE-NEXT:    [[TMP12:%.*]] = extractelement <8 x float> [[A]], i32 4
615; MINTREESIZE-NEXT:    [[TMP13:%.*]] = extractelement <8 x float> [[A]], i32 3
616; MINTREESIZE-NEXT:    [[TMP14:%.*]] = extractelement <8 x float> [[A]], i32 2
617; MINTREESIZE-NEXT:    [[TMP15:%.*]] = extractelement <8 x float> [[A]], i32 1
618; MINTREESIZE-NEXT:    [[TMP16:%.*]] = extractelement <8 x float> [[A]], i32 0
619; MINTREESIZE-NEXT:    [[TMP17:%.*]] = insertelement <2 x float> poison, float [[TMP16]], i32 0
620; MINTREESIZE-NEXT:    [[TMP18:%.*]] = insertelement <2 x float> [[TMP17]], float [[TMP8]], i32 1
621; MINTREESIZE-NEXT:    [[TMP19:%.*]] = insertelement <2 x float> poison, float [[TMP15]], i32 0
622; MINTREESIZE-NEXT:    [[TMP20:%.*]] = insertelement <2 x float> [[TMP19]], float [[TMP7]], i32 1
623; MINTREESIZE-NEXT:    [[TMP21:%.*]] = insertelement <2 x float> poison, float [[TMP14]], i32 0
624; MINTREESIZE-NEXT:    [[TMP22:%.*]] = insertelement <2 x float> [[TMP21]], float [[TMP6]], i32 1
625; MINTREESIZE-NEXT:    [[TMP23:%.*]] = insertelement <2 x float> poison, float [[TMP13]], i32 0
626; MINTREESIZE-NEXT:    [[TMP24:%.*]] = insertelement <2 x float> [[TMP23]], float [[TMP5]], i32 1
627; MINTREESIZE-NEXT:    [[TMP25:%.*]] = insertelement <2 x float> poison, float [[TMP12]], i32 0
628; MINTREESIZE-NEXT:    [[TMP26:%.*]] = insertelement <2 x float> [[TMP25]], float [[TMP4]], i32 1
629; MINTREESIZE-NEXT:    [[TMP27:%.*]] = insertelement <2 x float> poison, float [[TMP11]], i32 0
630; MINTREESIZE-NEXT:    [[TMP28:%.*]] = insertelement <2 x float> [[TMP27]], float [[TMP3]], i32 1
631; MINTREESIZE-NEXT:    [[TMP29:%.*]] = insertelement <2 x float> poison, float [[TMP10]], i32 0
632; MINTREESIZE-NEXT:    [[TMP30:%.*]] = insertelement <2 x float> [[TMP29]], float [[TMP2]], i32 1
633; MINTREESIZE-NEXT:    [[TMP31:%.*]] = insertelement <2 x float> poison, float [[TMP9]], i32 0
634; MINTREESIZE-NEXT:    [[TMP32:%.*]] = insertelement <2 x float> [[TMP31]], float [[TMP1]], i32 1
635; MINTREESIZE-NEXT:    [[TMP33:%.*]] = fadd <8 x float> [[A]], [[B]]
636; MINTREESIZE-NEXT:    ret <8 x float> [[TMP33]]
637;
638  %vecext = extractelement <8 x float> %a, i32 0
639  %vecext1 = extractelement <8 x float> %b, i32 0
640  %add = fadd float %vecext, %vecext1
641  %vecext2 = extractelement <8 x float> %a, i32 1
642  %vecext3 = extractelement <8 x float> %b, i32 1
643  %add4 = fadd float %vecext2, %vecext3
644  %vecext5 = extractelement <8 x float> %a, i32 2
645  %vecext6 = extractelement <8 x float> %b, i32 2
646  %add7 = fadd float %vecext5, %vecext6
647  %vecext8 = extractelement <8 x float> %a, i32 3
648  %vecext9 = extractelement <8 x float> %b, i32 3
649  %add10 = fadd float %vecext8, %vecext9
650  %vecext11 = extractelement <8 x float> %a, i32 4
651  %vecext12 = extractelement <8 x float> %b, i32 4
652  %add13 = fadd float %vecext11, %vecext12
653  %vecext14 = extractelement <8 x float> %a, i32 5
654  %vecext15 = extractelement <8 x float> %b, i32 5
655  %add16 = fadd float %vecext14, %vecext15
656  %vecext17 = extractelement <8 x float> %a, i32 6
657  %vecext18 = extractelement <8 x float> %b, i32 6
658  %add19 = fadd float %vecext17, %vecext18
659  %vecext20 = extractelement <8 x float> %a, i32 7
660  %vecext21 = extractelement <8 x float> %b, i32 7
661  %add22 = fadd float %vecext20, %vecext21
662  %vecinit.i = insertelement <8 x float> undef, float %add, i32 0
663  %vecinit1.i = insertelement <8 x float> %vecinit.i, float %add4, i32 1
664  %vecinit2.i = insertelement <8 x float> %vecinit1.i, float %add7, i32 2
665  %vecinit3.i = insertelement <8 x float> %vecinit2.i, float %add10, i32 3
666  %vecinit4.i = insertelement <8 x float> %vecinit3.i, float %add13, i32 4
667  %vecinit5.i = insertelement <8 x float> %vecinit4.i, float %add16, i32 5
668  %vecinit6.i = insertelement <8 x float> %vecinit5.i, float %add19, i32 6
669  %vecinit7.i = insertelement <8 x float> %vecinit6.i, float %add22, i32 7
670  ret <8 x float> %vecinit7.i
671}
672
673attributes #0 = { nounwind ssp uwtable "less-precise-fpmad"="false" "frame-pointer"="all" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
674