xref: /llvm-project/llvm/test/Transforms/SLPVectorizer/insert-element-build-vector-inseltpoison.ll (revision 38fffa630ee80163dc65e759392ad29798905679) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: %if x86-registered-target %{ opt -S -passes=slp-vectorizer -slp-threshold=-10000 -mtriple=x86_64-unknown-unknown < %s | FileCheck %s --check-prefixes=CHECK,THRESHOLD %}
3; RUN: %if x86-registered-target %{ opt -S -passes=slp-vectorizer -slp-threshold=0 -mtriple=x86_64-unknown-unknown < %s | FileCheck %s --check-prefixes=CHECK,NOTHRESHOLD %}
4; RUN: %if x86-registered-target %{ opt -S -passes=slp-vectorizer -slp-threshold=-10000 -slp-min-tree-size=0 -mtriple=x86_64-unknown-unknown < %s | FileCheck %s --check-prefixes=CHECK,MINTREESIZE %}
5; RUN: %if aarch64-registered-target %{ opt -S -passes=slp-vectorizer -slp-threshold=-10000 -mtriple=aarch64-unknown-linux-gnu < %s | FileCheck %s --check-prefixes=CHECK,THRESHOLD %}
6; RUN: %if aarch64-registered-target %{ opt -S -passes=slp-vectorizer -slp-threshold=0 -mtriple=aarch64-unknown-linux-gnu < %s | FileCheck %s --check-prefixes=CHECK,NOTHRESHOLD %}
7; RUN: %if aarch64-registered-target %{ opt -S -passes=slp-vectorizer -slp-threshold=-10000 -slp-min-tree-size=0 -mtriple=aarch64-unknown-linux-gnu < %s | FileCheck %s --check-prefixes=CHECK,MINTREESIZE %}
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> poison, 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
42declare void @llvm.assume(i1) nounwind
43
44; This entire tree is ephemeral, don't vectorize any of it.
45define <4 x float> @simple_select_eph(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
46; THRESHOLD-LABEL: @simple_select_eph(
47; THRESHOLD-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
48; THRESHOLD-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
49; THRESHOLD-NEXT:    [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2
50; THRESHOLD-NEXT:    [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3
51; THRESHOLD-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
52; THRESHOLD-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
53; THRESHOLD-NEXT:    [[A2:%.*]] = extractelement <4 x float> [[A]], i32 2
54; THRESHOLD-NEXT:    [[A3:%.*]] = extractelement <4 x float> [[A]], i32 3
55; THRESHOLD-NEXT:    [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0
56; THRESHOLD-NEXT:    [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1
57; THRESHOLD-NEXT:    [[B2:%.*]] = extractelement <4 x float> [[B]], i32 2
58; THRESHOLD-NEXT:    [[B3:%.*]] = extractelement <4 x float> [[B]], i32 3
59; THRESHOLD-NEXT:    [[CMP0:%.*]] = icmp ne i32 [[C0]], 0
60; THRESHOLD-NEXT:    [[CMP1:%.*]] = icmp ne i32 [[C1]], 0
61; THRESHOLD-NEXT:    [[CMP2:%.*]] = icmp ne i32 [[C2]], 0
62; THRESHOLD-NEXT:    [[CMP3:%.*]] = icmp ne i32 [[C3]], 0
63; THRESHOLD-NEXT:    [[S0:%.*]] = select i1 [[CMP0]], float [[A0]], float [[B0]]
64; THRESHOLD-NEXT:    [[S1:%.*]] = select i1 [[CMP1]], float [[A1]], float [[B1]]
65; THRESHOLD-NEXT:    [[S2:%.*]] = select i1 [[CMP2]], float [[A2]], float [[B2]]
66; THRESHOLD-NEXT:    [[S3:%.*]] = select i1 [[CMP3]], float [[A3]], float [[B3]]
67; THRESHOLD-NEXT:    [[RA:%.*]] = insertelement <4 x float> poison, float [[S0]], i32 0
68; THRESHOLD-NEXT:    [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[S1]], i32 1
69; THRESHOLD-NEXT:    [[RC:%.*]] = insertelement <4 x float> [[RB]], float [[S2]], i32 2
70; THRESHOLD-NEXT:    [[RD:%.*]] = insertelement <4 x float> [[RC]], float [[S3]], i32 3
71; THRESHOLD-NEXT:    [[Q0:%.*]] = extractelement <4 x float> [[RD]], i32 0
72; THRESHOLD-NEXT:    [[Q1:%.*]] = extractelement <4 x float> [[RD]], i32 1
73; THRESHOLD-NEXT:    [[TMP1:%.*]] = shufflevector <4 x float> [[RD]], <4 x float> poison, <2 x i32> <i32 0, i32 1>
74; THRESHOLD-NEXT:    [[Q2:%.*]] = extractelement <4 x float> [[RD]], i32 2
75; THRESHOLD-NEXT:    [[Q3:%.*]] = extractelement <4 x float> [[RD]], i32 3
76; THRESHOLD-NEXT:    [[TMP2:%.*]] = shufflevector <4 x float> [[RD]], <4 x float> poison, <2 x i32> <i32 2, i32 3>
77; THRESHOLD-NEXT:    [[Q4:%.*]] = fadd float [[Q0]], [[Q1]]
78; THRESHOLD-NEXT:    [[Q5:%.*]] = fadd float [[Q2]], [[Q3]]
79; THRESHOLD-NEXT:    [[Q6:%.*]] = fadd float [[Q4]], [[Q5]]
80; THRESHOLD-NEXT:    [[QI:%.*]] = fcmp olt float [[Q6]], [[Q5]]
81; THRESHOLD-NEXT:    call void @llvm.assume(i1 [[QI]])
82; THRESHOLD-NEXT:    ret <4 x float> undef
83;
84; NOTHRESHOLD-LABEL: @simple_select_eph(
85; NOTHRESHOLD-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
86; NOTHRESHOLD-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
87; NOTHRESHOLD-NEXT:    [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2
88; NOTHRESHOLD-NEXT:    [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3
89; NOTHRESHOLD-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
90; NOTHRESHOLD-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
91; NOTHRESHOLD-NEXT:    [[A2:%.*]] = extractelement <4 x float> [[A]], i32 2
92; NOTHRESHOLD-NEXT:    [[A3:%.*]] = extractelement <4 x float> [[A]], i32 3
93; NOTHRESHOLD-NEXT:    [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0
94; NOTHRESHOLD-NEXT:    [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1
95; NOTHRESHOLD-NEXT:    [[B2:%.*]] = extractelement <4 x float> [[B]], i32 2
96; NOTHRESHOLD-NEXT:    [[B3:%.*]] = extractelement <4 x float> [[B]], i32 3
97; NOTHRESHOLD-NEXT:    [[CMP0:%.*]] = icmp ne i32 [[C0]], 0
98; NOTHRESHOLD-NEXT:    [[CMP1:%.*]] = icmp ne i32 [[C1]], 0
99; NOTHRESHOLD-NEXT:    [[CMP2:%.*]] = icmp ne i32 [[C2]], 0
100; NOTHRESHOLD-NEXT:    [[CMP3:%.*]] = icmp ne i32 [[C3]], 0
101; NOTHRESHOLD-NEXT:    [[S0:%.*]] = select i1 [[CMP0]], float [[A0]], float [[B0]]
102; NOTHRESHOLD-NEXT:    [[S1:%.*]] = select i1 [[CMP1]], float [[A1]], float [[B1]]
103; NOTHRESHOLD-NEXT:    [[S2:%.*]] = select i1 [[CMP2]], float [[A2]], float [[B2]]
104; NOTHRESHOLD-NEXT:    [[S3:%.*]] = select i1 [[CMP3]], float [[A3]], float [[B3]]
105; NOTHRESHOLD-NEXT:    [[RA:%.*]] = insertelement <4 x float> poison, float [[S0]], i32 0
106; NOTHRESHOLD-NEXT:    [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[S1]], i32 1
107; NOTHRESHOLD-NEXT:    [[RC:%.*]] = insertelement <4 x float> [[RB]], float [[S2]], i32 2
108; NOTHRESHOLD-NEXT:    [[RD:%.*]] = insertelement <4 x float> [[RC]], float [[S3]], i32 3
109; NOTHRESHOLD-NEXT:    [[Q0:%.*]] = extractelement <4 x float> [[RD]], i32 0
110; NOTHRESHOLD-NEXT:    [[Q1:%.*]] = extractelement <4 x float> [[RD]], i32 1
111; NOTHRESHOLD-NEXT:    [[Q2:%.*]] = extractelement <4 x float> [[RD]], i32 2
112; NOTHRESHOLD-NEXT:    [[Q3:%.*]] = extractelement <4 x float> [[RD]], i32 3
113; NOTHRESHOLD-NEXT:    [[Q4:%.*]] = fadd float [[Q0]], [[Q1]]
114; NOTHRESHOLD-NEXT:    [[Q5:%.*]] = fadd float [[Q2]], [[Q3]]
115; NOTHRESHOLD-NEXT:    [[Q6:%.*]] = fadd float [[Q4]], [[Q5]]
116; NOTHRESHOLD-NEXT:    [[QI:%.*]] = fcmp olt float [[Q6]], [[Q5]]
117; NOTHRESHOLD-NEXT:    call void @llvm.assume(i1 [[QI]])
118; NOTHRESHOLD-NEXT:    ret <4 x float> undef
119;
120; MINTREESIZE-LABEL: @simple_select_eph(
121; MINTREESIZE-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
122; MINTREESIZE-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
123; MINTREESIZE-NEXT:    [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2
124; MINTREESIZE-NEXT:    [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3
125; MINTREESIZE-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
126; MINTREESIZE-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
127; MINTREESIZE-NEXT:    [[A2:%.*]] = extractelement <4 x float> [[A]], i32 2
128; MINTREESIZE-NEXT:    [[A3:%.*]] = extractelement <4 x float> [[A]], i32 3
129; MINTREESIZE-NEXT:    [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0
130; MINTREESIZE-NEXT:    [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1
131; MINTREESIZE-NEXT:    [[B2:%.*]] = extractelement <4 x float> [[B]], i32 2
132; MINTREESIZE-NEXT:    [[B3:%.*]] = extractelement <4 x float> [[B]], i32 3
133; MINTREESIZE-NEXT:    [[CMP0:%.*]] = icmp ne i32 [[C0]], 0
134; MINTREESIZE-NEXT:    [[CMP1:%.*]] = icmp ne i32 [[C1]], 0
135; MINTREESIZE-NEXT:    [[CMP2:%.*]] = icmp ne i32 [[C2]], 0
136; MINTREESIZE-NEXT:    [[CMP3:%.*]] = icmp ne i32 [[C3]], 0
137; MINTREESIZE-NEXT:    [[TMP1:%.*]] = insertelement <4 x i1> poison, i1 [[CMP3]], i32 0
138; MINTREESIZE-NEXT:    [[TMP2:%.*]] = insertelement <4 x i1> [[TMP1]], i1 [[CMP2]], i32 1
139; MINTREESIZE-NEXT:    [[TMP3:%.*]] = insertelement <4 x i1> [[TMP2]], i1 [[CMP1]], i32 2
140; MINTREESIZE-NEXT:    [[TMP4:%.*]] = insertelement <4 x i1> [[TMP3]], i1 [[CMP0]], i32 3
141; MINTREESIZE-NEXT:    [[S0:%.*]] = select i1 [[CMP0]], float [[A0]], float [[B0]]
142; MINTREESIZE-NEXT:    [[S1:%.*]] = select i1 [[CMP1]], float [[A1]], float [[B1]]
143; MINTREESIZE-NEXT:    [[S2:%.*]] = select i1 [[CMP2]], float [[A2]], float [[B2]]
144; MINTREESIZE-NEXT:    [[S3:%.*]] = select i1 [[CMP3]], float [[A3]], float [[B3]]
145; MINTREESIZE-NEXT:    [[RA:%.*]] = insertelement <4 x float> poison, float [[S0]], i32 0
146; MINTREESIZE-NEXT:    [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[S1]], i32 1
147; MINTREESIZE-NEXT:    [[RC:%.*]] = insertelement <4 x float> [[RB]], float [[S2]], i32 2
148; MINTREESIZE-NEXT:    [[RD:%.*]] = insertelement <4 x float> [[RC]], float [[S3]], i32 3
149; MINTREESIZE-NEXT:    [[Q0:%.*]] = extractelement <4 x float> [[RD]], i32 0
150; MINTREESIZE-NEXT:    [[Q1:%.*]] = extractelement <4 x float> [[RD]], i32 1
151; MINTREESIZE-NEXT:    [[TMP5:%.*]] = shufflevector <4 x float> [[RD]], <4 x float> poison, <2 x i32> <i32 0, i32 1>
152; MINTREESIZE-NEXT:    [[Q2:%.*]] = extractelement <4 x float> [[RD]], i32 2
153; MINTREESIZE-NEXT:    [[Q3:%.*]] = extractelement <4 x float> [[RD]], i32 3
154; MINTREESIZE-NEXT:    [[TMP6:%.*]] = shufflevector <4 x float> [[RD]], <4 x float> poison, <2 x i32> <i32 2, i32 3>
155; MINTREESIZE-NEXT:    [[Q4:%.*]] = fadd float [[Q0]], [[Q1]]
156; MINTREESIZE-NEXT:    [[Q5:%.*]] = fadd float [[Q2]], [[Q3]]
157; MINTREESIZE-NEXT:    [[TMP7:%.*]] = insertelement <2 x float> poison, float [[Q4]], i32 0
158; MINTREESIZE-NEXT:    [[TMP8:%.*]] = insertelement <2 x float> [[TMP7]], float [[Q5]], i32 1
159; MINTREESIZE-NEXT:    [[Q6:%.*]] = fadd float [[Q4]], [[Q5]]
160; MINTREESIZE-NEXT:    [[QI:%.*]] = fcmp olt float [[Q6]], [[Q5]]
161; MINTREESIZE-NEXT:    call void @llvm.assume(i1 [[QI]])
162; MINTREESIZE-NEXT:    ret <4 x float> undef
163;
164  %c0 = extractelement <4 x i32> %c, i32 0
165  %c1 = extractelement <4 x i32> %c, i32 1
166  %c2 = extractelement <4 x i32> %c, i32 2
167  %c3 = extractelement <4 x i32> %c, i32 3
168  %a0 = extractelement <4 x float> %a, i32 0
169  %a1 = extractelement <4 x float> %a, i32 1
170  %a2 = extractelement <4 x float> %a, i32 2
171  %a3 = extractelement <4 x float> %a, i32 3
172  %b0 = extractelement <4 x float> %b, i32 0
173  %b1 = extractelement <4 x float> %b, i32 1
174  %b2 = extractelement <4 x float> %b, i32 2
175  %b3 = extractelement <4 x float> %b, i32 3
176  %cmp0 = icmp ne i32 %c0, 0
177  %cmp1 = icmp ne i32 %c1, 0
178  %cmp2 = icmp ne i32 %c2, 0
179  %cmp3 = icmp ne i32 %c3, 0
180  %s0 = select i1 %cmp0, float %a0, float %b0
181  %s1 = select i1 %cmp1, float %a1, float %b1
182  %s2 = select i1 %cmp2, float %a2, float %b2
183  %s3 = select i1 %cmp3, float %a3, float %b3
184  %ra = insertelement <4 x float> poison, float %s0, i32 0
185  %rb = insertelement <4 x float> %ra, float %s1, i32 1
186  %rc = insertelement <4 x float> %rb, float %s2, i32 2
187  %rd = insertelement <4 x float> %rc, float %s3, i32 3
188  %q0 = extractelement <4 x float> %rd, i32 0
189  %q1 = extractelement <4 x float> %rd, i32 1
190  %q2 = extractelement <4 x float> %rd, i32 2
191  %q3 = extractelement <4 x float> %rd, i32 3
192  %q4 = fadd float %q0, %q1
193  %q5 = fadd float %q2, %q3
194  %q6 = fadd float %q4, %q5
195  %qi = fcmp olt float %q6, %q5
196  call void @llvm.assume(i1 %qi)
197  ret <4 x float> undef
198}
199
200; Insert in an order different from the vector indices to make sure it
201; doesn't matter
202define <4 x float> @simple_select_insert_out_of_order(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
203; CHECK-LABEL: @simple_select_insert_out_of_order(
204; CHECK-NEXT:    [[TMP1:%.*]] = icmp ne <4 x i32> [[C:%.*]], zeroinitializer
205; CHECK-NEXT:    [[TMP2:%.*]] = select <4 x i1> [[TMP1]], <4 x float> [[A:%.*]], <4 x float> [[B:%.*]]
206; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <4 x float> [[TMP2]], <4 x float> poison, <4 x i32> <i32 2, i32 1, i32 0, i32 3>
207; CHECK-NEXT:    ret <4 x float> [[TMP3]]
208;
209  %c0 = extractelement <4 x i32> %c, i32 0
210  %c1 = extractelement <4 x i32> %c, i32 1
211  %c2 = extractelement <4 x i32> %c, i32 2
212  %c3 = extractelement <4 x i32> %c, i32 3
213  %a0 = extractelement <4 x float> %a, i32 0
214  %a1 = extractelement <4 x float> %a, i32 1
215  %a2 = extractelement <4 x float> %a, i32 2
216  %a3 = extractelement <4 x float> %a, i32 3
217  %b0 = extractelement <4 x float> %b, i32 0
218  %b1 = extractelement <4 x float> %b, i32 1
219  %b2 = extractelement <4 x float> %b, i32 2
220  %b3 = extractelement <4 x float> %b, i32 3
221  %cmp0 = icmp ne i32 %c0, 0
222  %cmp1 = icmp ne i32 %c1, 0
223  %cmp2 = icmp ne i32 %c2, 0
224  %cmp3 = icmp ne i32 %c3, 0
225  %s0 = select i1 %cmp0, float %a0, float %b0
226  %s1 = select i1 %cmp1, float %a1, float %b1
227  %s2 = select i1 %cmp2, float %a2, float %b2
228  %s3 = select i1 %cmp3, float %a3, float %b3
229  %ra = insertelement <4 x float> poison, float %s0, i32 2
230  %rb = insertelement <4 x float> %ra, float %s1, i32 1
231  %rc = insertelement <4 x float> %rb, float %s2, i32 0
232  %rd = insertelement <4 x float> %rc, float %s3, i32 3
233  ret <4 x float> %rd
234}
235
236declare void @v4f32_user(<4 x float>) #0
237declare void @f32_user(float) #0
238
239; Multiple users of the final constructed vector
240define <4 x float> @simple_select_users(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
241; CHECK-LABEL: @simple_select_users(
242; CHECK-NEXT:    [[TMP1:%.*]] = icmp ne <4 x i32> [[C:%.*]], zeroinitializer
243; CHECK-NEXT:    [[TMP2:%.*]] = select <4 x i1> [[TMP1]], <4 x float> [[A:%.*]], <4 x float> [[B:%.*]]
244; CHECK-NEXT:    call void @v4f32_user(<4 x float> [[TMP2]]) #[[ATTR0:[0-9]+]]
245; CHECK-NEXT:    ret <4 x float> [[TMP2]]
246;
247  %c0 = extractelement <4 x i32> %c, i32 0
248  %c1 = extractelement <4 x i32> %c, i32 1
249  %c2 = extractelement <4 x i32> %c, i32 2
250  %c3 = extractelement <4 x i32> %c, i32 3
251  %a0 = extractelement <4 x float> %a, i32 0
252  %a1 = extractelement <4 x float> %a, i32 1
253  %a2 = extractelement <4 x float> %a, i32 2
254  %a3 = extractelement <4 x float> %a, i32 3
255  %b0 = extractelement <4 x float> %b, i32 0
256  %b1 = extractelement <4 x float> %b, i32 1
257  %b2 = extractelement <4 x float> %b, i32 2
258  %b3 = extractelement <4 x float> %b, i32 3
259  %cmp0 = icmp ne i32 %c0, 0
260  %cmp1 = icmp ne i32 %c1, 0
261  %cmp2 = icmp ne i32 %c2, 0
262  %cmp3 = icmp ne i32 %c3, 0
263  %s0 = select i1 %cmp0, float %a0, float %b0
264  %s1 = select i1 %cmp1, float %a1, float %b1
265  %s2 = select i1 %cmp2, float %a2, float %b2
266  %s3 = select i1 %cmp3, float %a3, float %b3
267  %ra = insertelement <4 x float> poison, float %s0, i32 0
268  %rb = insertelement <4 x float> %ra, float %s1, i32 1
269  %rc = insertelement <4 x float> %rb, float %s2, i32 2
270  %rd = insertelement <4 x float> %rc, float %s3, i32 3
271  call void @v4f32_user(<4 x float> %rd) #0
272  ret <4 x float> %rd
273}
274
275; Unused insertelement
276define <4 x float> @simple_select_no_users(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
277; CHECK-LABEL: @simple_select_no_users(
278; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <4 x i32> [[C:%.*]], <4 x i32> poison, <2 x i32> <i32 0, i32 1>
279; CHECK-NEXT:    [[TMP2:%.*]] = icmp ne <2 x i32> [[TMP1]], zeroinitializer
280; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <4 x float> [[A:%.*]], <4 x float> poison, <2 x i32> <i32 0, i32 1>
281; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <4 x float> [[B:%.*]], <4 x float> poison, <2 x i32> <i32 0, i32 1>
282; CHECK-NEXT:    [[TMP5:%.*]] = select <2 x i1> [[TMP2]], <2 x float> [[TMP3]], <2 x float> [[TMP4]]
283; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <4 x i32> [[C]], <4 x i32> poison, <2 x i32> <i32 2, i32 3>
284; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne <2 x i32> [[TMP6]], zeroinitializer
285; CHECK-NEXT:    [[TMP8:%.*]] = shufflevector <4 x float> [[A]], <4 x float> poison, <2 x i32> <i32 2, i32 3>
286; CHECK-NEXT:    [[TMP9:%.*]] = shufflevector <4 x float> [[B]], <4 x float> poison, <2 x i32> <i32 2, i32 3>
287; CHECK-NEXT:    [[TMP10:%.*]] = select <2 x i1> [[TMP7]], <2 x float> [[TMP8]], <2 x float> [[TMP9]]
288; CHECK-NEXT:    [[TMP11:%.*]] = shufflevector <2 x float> [[TMP5]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
289; CHECK-NEXT:    [[TMP12:%.*]] = shufflevector <2 x float> [[TMP10]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
290; CHECK-NEXT:    [[RD1:%.*]] = shufflevector <4 x float> [[TMP12]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 0, i32 1>
291; CHECK-NEXT:    ret <4 x float> [[RD1]]
292;
293  %c0 = extractelement <4 x i32> %c, i32 0
294  %c1 = extractelement <4 x i32> %c, i32 1
295  %c2 = extractelement <4 x i32> %c, i32 2
296  %c3 = extractelement <4 x i32> %c, i32 3
297  %a0 = extractelement <4 x float> %a, i32 0
298  %a1 = extractelement <4 x float> %a, i32 1
299  %a2 = extractelement <4 x float> %a, i32 2
300  %a3 = extractelement <4 x float> %a, i32 3
301  %b0 = extractelement <4 x float> %b, i32 0
302  %b1 = extractelement <4 x float> %b, i32 1
303  %b2 = extractelement <4 x float> %b, i32 2
304  %b3 = extractelement <4 x float> %b, i32 3
305  %cmp0 = icmp ne i32 %c0, 0
306  %cmp1 = icmp ne i32 %c1, 0
307  %cmp2 = icmp ne i32 %c2, 0
308  %cmp3 = icmp ne i32 %c3, 0
309  %s0 = select i1 %cmp0, float %a0, float %b0
310  %s1 = select i1 %cmp1, float %a1, float %b1
311  %s2 = select i1 %cmp2, float %a2, float %b2
312  %s3 = select i1 %cmp3, float %a3, float %b3
313  %ra = insertelement <4 x float> poison, float %s0, i32 0
314  %rb = insertelement <4 x float> %ra, float %s1, i32 1
315  %rc = insertelement <4 x float> poison, float %s2, i32 2
316  %rd = insertelement <4 x float> %rc, float %s3, i32 3
317  ret <4 x float> %rd
318}
319
320; Make sure infinite loop doesn't happen which I ran into when trying
321; to do this backwards this backwards
322define <4 x i32> @reconstruct(<4 x i32> %c) #0 {
323; CHECK-LABEL: @reconstruct(
324; CHECK-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
325; CHECK-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
326; CHECK-NEXT:    [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2
327; CHECK-NEXT:    [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3
328; CHECK-NEXT:    [[RA:%.*]] = insertelement <4 x i32> poison, i32 [[C0]], i32 0
329; CHECK-NEXT:    [[RB:%.*]] = insertelement <4 x i32> [[RA]], i32 [[C1]], i32 1
330; CHECK-NEXT:    [[RC:%.*]] = insertelement <4 x i32> [[RB]], i32 [[C2]], i32 2
331; CHECK-NEXT:    [[RD:%.*]] = insertelement <4 x i32> [[RC]], i32 [[C3]], i32 3
332; CHECK-NEXT:    ret <4 x i32> [[RD]]
333;
334  %c0 = extractelement <4 x i32> %c, i32 0
335  %c1 = extractelement <4 x i32> %c, i32 1
336  %c2 = extractelement <4 x i32> %c, i32 2
337  %c3 = extractelement <4 x i32> %c, i32 3
338  %ra = insertelement <4 x i32> poison, i32 %c0, i32 0
339  %rb = insertelement <4 x i32> %ra, i32 %c1, i32 1
340  %rc = insertelement <4 x i32> %rb, i32 %c2, i32 2
341  %rd = insertelement <4 x i32> %rc, i32 %c3, i32 3
342  ret <4 x i32> %rd
343}
344
345define <2 x float> @simple_select_v2(<2 x float> %a, <2 x float> %b, <2 x i32> %c) #0 {
346; CHECK-LABEL: @simple_select_v2(
347; CHECK-NEXT:    [[TMP1:%.*]] = icmp ne <2 x i32> [[C:%.*]], zeroinitializer
348; CHECK-NEXT:    [[TMP2:%.*]] = select <2 x i1> [[TMP1]], <2 x float> [[A:%.*]], <2 x float> [[B:%.*]]
349; CHECK-NEXT:    ret <2 x float> [[TMP2]]
350;
351  %c0 = extractelement <2 x i32> %c, i32 0
352  %c1 = extractelement <2 x i32> %c, i32 1
353  %a0 = extractelement <2 x float> %a, i32 0
354  %a1 = extractelement <2 x float> %a, i32 1
355  %b0 = extractelement <2 x float> %b, i32 0
356  %b1 = extractelement <2 x float> %b, i32 1
357  %cmp0 = icmp ne i32 %c0, 0
358  %cmp1 = icmp ne i32 %c1, 0
359  %s0 = select i1 %cmp0, float %a0, float %b0
360  %s1 = select i1 %cmp1, float %a1, float %b1
361  %ra = insertelement <2 x float> poison, float %s0, i32 0
362  %rb = insertelement <2 x float> %ra, float %s1, i32 1
363  ret <2 x float> %rb
364}
365
366; Make sure when we construct partial vectors, we don't keep
367; re-visiting the insertelement chains starting with undef
368; (low cost threshold needed to force this to happen)
369define <4 x float> @simple_select_partial_vector(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
370; CHECK-LABEL: @simple_select_partial_vector(
371; CHECK-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
372; CHECK-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
373; CHECK-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
374; CHECK-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
375; CHECK-NEXT:    [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0
376; CHECK-NEXT:    [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1
377; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x i32> poison, i32 [[C0]], i32 0
378; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x i32> [[TMP1]], i32 [[C1]], i32 1
379; CHECK-NEXT:    [[TMP3:%.*]] = icmp ne <2 x i32> [[TMP2]], zeroinitializer
380; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <2 x float> poison, float [[A0]], i32 0
381; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <2 x float> [[TMP4]], float [[A1]], i32 1
382; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <2 x float> poison, float [[B0]], i32 0
383; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x float> [[TMP6]], float [[B1]], i32 1
384; CHECK-NEXT:    [[TMP8:%.*]] = select <2 x i1> [[TMP3]], <2 x float> [[TMP5]], <2 x float> [[TMP7]]
385; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <2 x float> [[TMP8]], i32 0
386; CHECK-NEXT:    [[RA:%.*]] = insertelement <4 x float> poison, float [[TMP9]], i32 0
387; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <2 x float> [[TMP8]], i32 1
388; CHECK-NEXT:    [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[TMP10]], i32 1
389; CHECK-NEXT:    ret <4 x float> [[RB]]
390;
391  %c0 = extractelement <4 x i32> %c, i32 0
392  %c1 = extractelement <4 x i32> %c, i32 1
393  %a0 = extractelement <4 x float> %a, i32 0
394  %a1 = extractelement <4 x float> %a, i32 1
395  %b0 = extractelement <4 x float> %b, i32 0
396  %b1 = extractelement <4 x float> %b, i32 1
397  %1 = insertelement <2 x i32> poison, i32 %c0, i32 0
398  %2 = insertelement <2 x i32> %1, i32 %c1, i32 1
399  %3 = icmp ne <2 x i32> %2, zeroinitializer
400  %4 = insertelement <2 x float> poison, float %a0, i32 0
401  %5 = insertelement <2 x float> %4, float %a1, i32 1
402  %6 = insertelement <2 x float> poison, float %b0, i32 0
403  %7 = insertelement <2 x float> %6, float %b1, i32 1
404  %8 = select <2 x i1> %3, <2 x float> %5, <2 x float> %7
405  %9 = extractelement <2 x float> %8, i32 0
406  %ra = insertelement <4 x float> poison, float %9, i32 0
407  %10 = extractelement <2 x float> %8, i32 1
408  %rb = insertelement <4 x float> %ra, float %10, i32 1
409  ret <4 x float> %rb
410}
411
412; Make sure that vectorization happens even if insertelements operations
413; must be rescheduled. The case here is from compiling Julia.
414define <4 x float> @reschedule_extract(<4 x float> %a, <4 x float> %b) {
415; CHECK-LABEL: @reschedule_extract(
416; CHECK-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A:%.*]], [[B:%.*]]
417; CHECK-NEXT:    ret <4 x float> [[TMP1]]
418;
419  %a0 = extractelement <4 x float> %a, i32 0
420  %b0 = extractelement <4 x float> %b, i32 0
421  %c0 = fadd float %a0, %b0
422  %v0 = insertelement <4 x float> poison, float %c0, i32 0
423  %a1 = extractelement <4 x float> %a, i32 1
424  %b1 = extractelement <4 x float> %b, i32 1
425  %c1 = fadd float %a1, %b1
426  %v1 = insertelement <4 x float> %v0, float %c1, i32 1
427  %a2 = extractelement <4 x float> %a, i32 2
428  %b2 = extractelement <4 x float> %b, i32 2
429  %c2 = fadd float %a2, %b2
430  %v2 = insertelement <4 x float> %v1, float %c2, i32 2
431  %a3 = extractelement <4 x float> %a, i32 3
432  %b3 = extractelement <4 x float> %b, i32 3
433  %c3 = fadd float %a3, %b3
434  %v3 = insertelement <4 x float> %v2, float %c3, i32 3
435  ret <4 x float> %v3
436}
437
438; Check that cost model for vectorization takes credit for
439; instructions that are erased.
440define <4 x float> @take_credit(<4 x float> %a, <4 x float> %b) {
441; CHECK-LABEL: @take_credit(
442; CHECK-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A:%.*]], [[B:%.*]]
443; CHECK-NEXT:    ret <4 x float> [[TMP1]]
444;
445  %a0 = extractelement <4 x float> %a, i32 0
446  %b0 = extractelement <4 x float> %b, i32 0
447  %c0 = fadd float %a0, %b0
448  %a1 = extractelement <4 x float> %a, i32 1
449  %b1 = extractelement <4 x float> %b, i32 1
450  %c1 = fadd float %a1, %b1
451  %a2 = extractelement <4 x float> %a, i32 2
452  %b2 = extractelement <4 x float> %b, i32 2
453  %c2 = fadd float %a2, %b2
454  %a3 = extractelement <4 x float> %a, i32 3
455  %b3 = extractelement <4 x float> %b, i32 3
456  %c3 = fadd float %a3, %b3
457  %v0 = insertelement <4 x float> poison, float %c0, i32 0
458  %v1 = insertelement <4 x float> %v0, float %c1, i32 1
459  %v2 = insertelement <4 x float> %v1, float %c2, i32 2
460  %v3 = insertelement <4 x float> %v2, float %c3, i32 3
461  ret <4 x float> %v3
462}
463
464; Make sure we handle multiple trees that feed one build vector correctly.
465define <4 x double> @multi_tree(double %w, double %x, double %y, double %z) {
466; CHECK-LABEL: @multi_tree(
467; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x double> poison, double [[Z:%.*]], i32 0
468; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x double> [[TMP1]], double [[Y:%.*]], i32 1
469; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x double> [[TMP2]], double [[X:%.*]], i32 2
470; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x double> [[TMP3]], double [[W:%.*]], i32 3
471; CHECK-NEXT:    [[TMP5:%.*]] = fadd <4 x double> [[TMP4]], <double 3.000000e+00, double 2.000000e+00, double 1.000000e+00, double 0.000000e+00>
472; CHECK-NEXT:    [[TMP6:%.*]] = fmul <4 x double> [[TMP5]], splat (double 1.000000e+00)
473; CHECK-NEXT:    ret <4 x double> [[TMP6]]
474;
475  %t0 = fadd double %w , 0.000000e+00
476  %t1 = fadd double %x , 1.000000e+00
477  %t2 = fadd double %y , 2.000000e+00
478  %t3 = fadd double %z , 3.000000e+00
479  %t4 = fmul double %t0, 1.000000e+00
480  %i1 = insertelement <4 x double> poison, double %t4, i32 3
481  %t5 = fmul double %t1, 1.000000e+00
482  %i2 = insertelement <4 x double> %i1, double %t5, i32 2
483  %t6 = fmul double %t2, 1.000000e+00
484  %i3 = insertelement <4 x double> %i2, double %t6, i32 1
485  %t7 = fmul double %t3, 1.000000e+00
486  %i4 = insertelement <4 x double> %i3, double %t7, i32 0
487  ret <4 x double> %i4
488}
489
490define <8 x float> @_vadd256(<8 x float> %a, <8 x float> %b) local_unnamed_addr #0 {
491; CHECK-LABEL: @_vadd256(
492; CHECK-NEXT:    [[TMP1:%.*]] = fadd <8 x float> [[A:%.*]], [[B:%.*]]
493; CHECK-NEXT:    ret <8 x float> [[TMP1]]
494;
495  %vecext = extractelement <8 x float> %a, i32 0
496  %vecext1 = extractelement <8 x float> %b, i32 0
497  %add = fadd float %vecext, %vecext1
498  %vecext2 = extractelement <8 x float> %a, i32 1
499  %vecext3 = extractelement <8 x float> %b, i32 1
500  %add4 = fadd float %vecext2, %vecext3
501  %vecext5 = extractelement <8 x float> %a, i32 2
502  %vecext6 = extractelement <8 x float> %b, i32 2
503  %add7 = fadd float %vecext5, %vecext6
504  %vecext8 = extractelement <8 x float> %a, i32 3
505  %vecext9 = extractelement <8 x float> %b, i32 3
506  %add10 = fadd float %vecext8, %vecext9
507  %vecext11 = extractelement <8 x float> %a, i32 4
508  %vecext12 = extractelement <8 x float> %b, i32 4
509  %add13 = fadd float %vecext11, %vecext12
510  %vecext14 = extractelement <8 x float> %a, i32 5
511  %vecext15 = extractelement <8 x float> %b, i32 5
512  %add16 = fadd float %vecext14, %vecext15
513  %vecext17 = extractelement <8 x float> %a, i32 6
514  %vecext18 = extractelement <8 x float> %b, i32 6
515  %add19 = fadd float %vecext17, %vecext18
516  %vecext20 = extractelement <8 x float> %a, i32 7
517  %vecext21 = extractelement <8 x float> %b, i32 7
518  %add22 = fadd float %vecext20, %vecext21
519  %vecinit.i = insertelement <8 x float> poison, float %add, i32 0
520  %vecinit1.i = insertelement <8 x float> %vecinit.i, float %add4, i32 1
521  %vecinit2.i = insertelement <8 x float> %vecinit1.i, float %add7, i32 2
522  %vecinit3.i = insertelement <8 x float> %vecinit2.i, float %add10, i32 3
523  %vecinit4.i = insertelement <8 x float> %vecinit3.i, float %add13, i32 4
524  %vecinit5.i = insertelement <8 x float> %vecinit4.i, float %add16, i32 5
525  %vecinit6.i = insertelement <8 x float> %vecinit5.i, float %add19, i32 6
526  %vecinit7.i = insertelement <8 x float> %vecinit6.i, float %add22, i32 7
527  ret <8 x float> %vecinit7.i
528}
529
530attributes #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" }
531