1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -S -slp-vectorizer -slp-threshold=-10000 < %s | FileCheck %s
3; RUN: opt -S -slp-vectorizer -slp-threshold=0 < %s | FileCheck %s
4
5target 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"
6target triple = "x86_64-apple-macosx10.8.0"
7
8define <4 x float> @simple_select(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
9; CHECK-LABEL: @simple_select(
10; CHECK-NEXT:    [[TMP1:%.*]] = icmp ne <4 x i32> [[C:%.*]], zeroinitializer
11; CHECK-NEXT:    [[TMP2:%.*]] = select <4 x i1> [[TMP1]], <4 x float> [[A:%.*]], <4 x float> [[B:%.*]]
12; CHECK-NEXT:    ret <4 x float> [[TMP2]]
13;
14  %c0 = extractelement <4 x i32> %c, i32 0
15  %c1 = extractelement <4 x i32> %c, i32 1
16  %c2 = extractelement <4 x i32> %c, i32 2
17  %c3 = extractelement <4 x i32> %c, i32 3
18  %a0 = extractelement <4 x float> %a, i32 0
19  %a1 = extractelement <4 x float> %a, i32 1
20  %a2 = extractelement <4 x float> %a, i32 2
21  %a3 = extractelement <4 x float> %a, i32 3
22  %b0 = extractelement <4 x float> %b, i32 0
23  %b1 = extractelement <4 x float> %b, i32 1
24  %b2 = extractelement <4 x float> %b, i32 2
25  %b3 = extractelement <4 x float> %b, i32 3
26  %cmp0 = icmp ne i32 %c0, 0
27  %cmp1 = icmp ne i32 %c1, 0
28  %cmp2 = icmp ne i32 %c2, 0
29  %cmp3 = icmp ne i32 %c3, 0
30  %s0 = select i1 %cmp0, float %a0, float %b0
31  %s1 = select i1 %cmp1, float %a1, float %b1
32  %s2 = select i1 %cmp2, float %a2, float %b2
33  %s3 = select i1 %cmp3, float %a3, float %b3
34  %ra = insertelement <4 x float> undef, float %s0, i32 0
35  %rb = insertelement <4 x float> %ra, float %s1, i32 1
36  %rc = insertelement <4 x float> %rb, float %s2, i32 2
37  %rd = insertelement <4 x float> %rc, float %s3, i32 3
38  ret <4 x float> %rd
39}
40
41declare void @llvm.assume(i1) nounwind
42
43; This entire tree is ephemeral, don't vectorize any of it.
44define <4 x float> @simple_select_eph(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
45; CHECK-LABEL: @simple_select_eph(
46; CHECK-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
47; CHECK-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
48; CHECK-NEXT:    [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2
49; CHECK-NEXT:    [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3
50; CHECK-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
51; CHECK-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
52; CHECK-NEXT:    [[A2:%.*]] = extractelement <4 x float> [[A]], i32 2
53; CHECK-NEXT:    [[A3:%.*]] = extractelement <4 x float> [[A]], i32 3
54; CHECK-NEXT:    [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0
55; CHECK-NEXT:    [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1
56; CHECK-NEXT:    [[B2:%.*]] = extractelement <4 x float> [[B]], i32 2
57; CHECK-NEXT:    [[B3:%.*]] = extractelement <4 x float> [[B]], i32 3
58; CHECK-NEXT:    [[CMP0:%.*]] = icmp ne i32 [[C0]], 0
59; CHECK-NEXT:    [[CMP1:%.*]] = icmp ne i32 [[C1]], 0
60; CHECK-NEXT:    [[CMP2:%.*]] = icmp ne i32 [[C2]], 0
61; CHECK-NEXT:    [[CMP3:%.*]] = icmp ne i32 [[C3]], 0
62; CHECK-NEXT:    [[S0:%.*]] = select i1 [[CMP0]], float [[A0]], float [[B0]]
63; CHECK-NEXT:    [[S1:%.*]] = select i1 [[CMP1]], float [[A1]], float [[B1]]
64; CHECK-NEXT:    [[S2:%.*]] = select i1 [[CMP2]], float [[A2]], float [[B2]]
65; CHECK-NEXT:    [[S3:%.*]] = select i1 [[CMP3]], float [[A3]], float [[B3]]
66; CHECK-NEXT:    [[RA:%.*]] = insertelement <4 x float> undef, float [[S0]], i32 0
67; CHECK-NEXT:    [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[S1]], i32 1
68; CHECK-NEXT:    [[RC:%.*]] = insertelement <4 x float> [[RB]], float [[S2]], i32 2
69; CHECK-NEXT:    [[RD:%.*]] = insertelement <4 x float> [[RC]], float [[S3]], i32 3
70; CHECK-NEXT:    [[Q0:%.*]] = extractelement <4 x float> [[RD]], i32 0
71; CHECK-NEXT:    [[Q1:%.*]] = extractelement <4 x float> [[RD]], i32 1
72; CHECK-NEXT:    [[Q2:%.*]] = extractelement <4 x float> [[RD]], i32 2
73; CHECK-NEXT:    [[Q3:%.*]] = extractelement <4 x float> [[RD]], i32 3
74; CHECK-NEXT:    [[Q4:%.*]] = fadd float [[Q0]], [[Q1]]
75; CHECK-NEXT:    [[Q5:%.*]] = fadd float [[Q2]], [[Q3]]
76; CHECK-NEXT:    [[Q6:%.*]] = fadd float [[Q4]], [[Q5]]
77; CHECK-NEXT:    [[QI:%.*]] = fcmp olt float [[Q6]], [[Q5]]
78; CHECK-NEXT:    call void @llvm.assume(i1 [[QI]])
79; CHECK-NEXT:    ret <4 x float> undef
80;
81  %c0 = extractelement <4 x i32> %c, i32 0
82  %c1 = extractelement <4 x i32> %c, i32 1
83  %c2 = extractelement <4 x i32> %c, i32 2
84  %c3 = extractelement <4 x i32> %c, i32 3
85  %a0 = extractelement <4 x float> %a, i32 0
86  %a1 = extractelement <4 x float> %a, i32 1
87  %a2 = extractelement <4 x float> %a, i32 2
88  %a3 = extractelement <4 x float> %a, i32 3
89  %b0 = extractelement <4 x float> %b, i32 0
90  %b1 = extractelement <4 x float> %b, i32 1
91  %b2 = extractelement <4 x float> %b, i32 2
92  %b3 = extractelement <4 x float> %b, i32 3
93  %cmp0 = icmp ne i32 %c0, 0
94  %cmp1 = icmp ne i32 %c1, 0
95  %cmp2 = icmp ne i32 %c2, 0
96  %cmp3 = icmp ne i32 %c3, 0
97  %s0 = select i1 %cmp0, float %a0, float %b0
98  %s1 = select i1 %cmp1, float %a1, float %b1
99  %s2 = select i1 %cmp2, float %a2, float %b2
100  %s3 = select i1 %cmp3, float %a3, float %b3
101  %ra = insertelement <4 x float> undef, float %s0, i32 0
102  %rb = insertelement <4 x float> %ra, float %s1, i32 1
103  %rc = insertelement <4 x float> %rb, float %s2, i32 2
104  %rd = insertelement <4 x float> %rc, float %s3, i32 3
105  %q0 = extractelement <4 x float> %rd, i32 0
106  %q1 = extractelement <4 x float> %rd, i32 1
107  %q2 = extractelement <4 x float> %rd, i32 2
108  %q3 = extractelement <4 x float> %rd, i32 3
109  %q4 = fadd float %q0, %q1
110  %q5 = fadd float %q2, %q3
111  %q6 = fadd float %q4, %q5
112  %qi = fcmp olt float %q6, %q5
113  call void @llvm.assume(i1 %qi)
114  ret <4 x float> undef
115}
116
117; Insert in an order different from the vector indices to make sure it
118; doesn't matter
119define <4 x float> @simple_select_insert_out_of_order(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
120; CHECK-LABEL: @simple_select_insert_out_of_order(
121; CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[C:%.*]], <4 x i32> poison, <4 x i32> <i32 2, i32 1, i32 0, i32 3>
122; CHECK-NEXT:    [[SHUFFLE1:%.*]] = shufflevector <4 x float> [[A:%.*]], <4 x float> poison, <4 x i32> <i32 2, i32 1, i32 0, i32 3>
123; CHECK-NEXT:    [[SHUFFLE2:%.*]] = shufflevector <4 x float> [[B:%.*]], <4 x float> poison, <4 x i32> <i32 2, i32 1, i32 0, i32 3>
124; CHECK-NEXT:    [[TMP1:%.*]] = icmp ne <4 x i32> [[SHUFFLE]], zeroinitializer
125; CHECK-NEXT:    [[TMP2:%.*]] = select <4 x i1> [[TMP1]], <4 x float> [[SHUFFLE1]], <4 x float> [[SHUFFLE2]]
126; CHECK-NEXT:    ret <4 x float> [[TMP2]]
127;
128  %c0 = extractelement <4 x i32> %c, i32 0
129  %c1 = extractelement <4 x i32> %c, i32 1
130  %c2 = extractelement <4 x i32> %c, i32 2
131  %c3 = extractelement <4 x i32> %c, i32 3
132  %a0 = extractelement <4 x float> %a, i32 0
133  %a1 = extractelement <4 x float> %a, i32 1
134  %a2 = extractelement <4 x float> %a, i32 2
135  %a3 = extractelement <4 x float> %a, i32 3
136  %b0 = extractelement <4 x float> %b, i32 0
137  %b1 = extractelement <4 x float> %b, i32 1
138  %b2 = extractelement <4 x float> %b, i32 2
139  %b3 = extractelement <4 x float> %b, i32 3
140  %cmp0 = icmp ne i32 %c0, 0
141  %cmp1 = icmp ne i32 %c1, 0
142  %cmp2 = icmp ne i32 %c2, 0
143  %cmp3 = icmp ne i32 %c3, 0
144  %s0 = select i1 %cmp0, float %a0, float %b0
145  %s1 = select i1 %cmp1, float %a1, float %b1
146  %s2 = select i1 %cmp2, float %a2, float %b2
147  %s3 = select i1 %cmp3, float %a3, float %b3
148  %ra = insertelement <4 x float> undef, float %s0, i32 2
149  %rb = insertelement <4 x float> %ra, float %s1, i32 1
150  %rc = insertelement <4 x float> %rb, float %s2, i32 0
151  %rd = insertelement <4 x float> %rc, float %s3, i32 3
152  ret <4 x float> %rd
153}
154
155declare void @v4f32_user(<4 x float>) #0
156declare void @f32_user(float) #0
157
158; Multiple users of the final constructed vector
159define <4 x float> @simple_select_users(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
160; CHECK-LABEL: @simple_select_users(
161; CHECK-NEXT:    [[TMP1:%.*]] = icmp ne <4 x i32> [[C:%.*]], zeroinitializer
162; CHECK-NEXT:    [[TMP2:%.*]] = select <4 x i1> [[TMP1]], <4 x float> [[A:%.*]], <4 x float> [[B:%.*]]
163; CHECK-NEXT:    call void @v4f32_user(<4 x float> [[TMP2]]) #[[ATTR0:[0-9]+]]
164; CHECK-NEXT:    ret <4 x float> [[TMP2]]
165;
166  %c0 = extractelement <4 x i32> %c, i32 0
167  %c1 = extractelement <4 x i32> %c, i32 1
168  %c2 = extractelement <4 x i32> %c, i32 2
169  %c3 = extractelement <4 x i32> %c, i32 3
170  %a0 = extractelement <4 x float> %a, i32 0
171  %a1 = extractelement <4 x float> %a, i32 1
172  %a2 = extractelement <4 x float> %a, i32 2
173  %a3 = extractelement <4 x float> %a, i32 3
174  %b0 = extractelement <4 x float> %b, i32 0
175  %b1 = extractelement <4 x float> %b, i32 1
176  %b2 = extractelement <4 x float> %b, i32 2
177  %b3 = extractelement <4 x float> %b, i32 3
178  %cmp0 = icmp ne i32 %c0, 0
179  %cmp1 = icmp ne i32 %c1, 0
180  %cmp2 = icmp ne i32 %c2, 0
181  %cmp3 = icmp ne i32 %c3, 0
182  %s0 = select i1 %cmp0, float %a0, float %b0
183  %s1 = select i1 %cmp1, float %a1, float %b1
184  %s2 = select i1 %cmp2, float %a2, float %b2
185  %s3 = select i1 %cmp3, float %a3, float %b3
186  %ra = insertelement <4 x float> undef, float %s0, i32 0
187  %rb = insertelement <4 x float> %ra, float %s1, i32 1
188  %rc = insertelement <4 x float> %rb, float %s2, i32 2
189  %rd = insertelement <4 x float> %rc, float %s3, i32 3
190  call void @v4f32_user(<4 x float> %rd) #0
191  ret <4 x float> %rd
192}
193
194; Unused insertelement
195define <4 x float> @simple_select_no_users(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
196; CHECK-LABEL: @simple_select_no_users(
197; CHECK-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
198; CHECK-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
199; CHECK-NEXT:    [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2
200; CHECK-NEXT:    [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3
201; CHECK-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
202; CHECK-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
203; CHECK-NEXT:    [[A2:%.*]] = extractelement <4 x float> [[A]], i32 2
204; CHECK-NEXT:    [[A3:%.*]] = extractelement <4 x float> [[A]], i32 3
205; CHECK-NEXT:    [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0
206; CHECK-NEXT:    [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1
207; CHECK-NEXT:    [[B2:%.*]] = extractelement <4 x float> [[B]], i32 2
208; CHECK-NEXT:    [[B3:%.*]] = extractelement <4 x float> [[B]], i32 3
209; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x i32> poison, i32 [[C0]], i32 0
210; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x i32> [[TMP1]], i32 [[C1]], i32 1
211; CHECK-NEXT:    [[TMP3:%.*]] = icmp ne <2 x i32> [[TMP2]], zeroinitializer
212; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <2 x i32> poison, i32 [[C2]], i32 0
213; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <2 x i32> [[TMP4]], i32 [[C3]], i32 1
214; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <2 x i32> [[TMP5]], zeroinitializer
215; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x float> poison, float [[A0]], i32 0
216; CHECK-NEXT:    [[TMP8:%.*]] = insertelement <2 x float> [[TMP7]], float [[A1]], i32 1
217; CHECK-NEXT:    [[TMP9:%.*]] = insertelement <2 x float> poison, float [[B0]], i32 0
218; CHECK-NEXT:    [[TMP10:%.*]] = insertelement <2 x float> [[TMP9]], float [[B1]], i32 1
219; CHECK-NEXT:    [[TMP11:%.*]] = select <2 x i1> [[TMP3]], <2 x float> [[TMP8]], <2 x float> [[TMP10]]
220; CHECK-NEXT:    [[TMP12:%.*]] = insertelement <2 x float> poison, float [[A2]], i32 0
221; CHECK-NEXT:    [[TMP13:%.*]] = insertelement <2 x float> [[TMP12]], float [[A3]], i32 1
222; CHECK-NEXT:    [[TMP14:%.*]] = insertelement <2 x float> poison, float [[B2]], i32 0
223; CHECK-NEXT:    [[TMP15:%.*]] = insertelement <2 x float> [[TMP14]], float [[B3]], i32 1
224; CHECK-NEXT:    [[TMP16:%.*]] = select <2 x i1> [[TMP6]], <2 x float> [[TMP13]], <2 x float> [[TMP15]]
225; CHECK-NEXT:    [[TMP17:%.*]] = shufflevector <2 x float> [[TMP11]], <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
226; CHECK-NEXT:    [[RB2:%.*]] = shufflevector <4 x float> undef, <4 x float> [[TMP17]], <4 x i32> <i32 4, i32 5, i32 2, i32 3>
227; CHECK-NEXT:    [[TMP18:%.*]] = shufflevector <2 x float> [[TMP16]], <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
228; CHECK-NEXT:    [[RD1:%.*]] = shufflevector <4 x float> undef, <4 x float> [[TMP18]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
229; CHECK-NEXT:    ret <4 x float> [[RD1]]
230;
231  %c0 = extractelement <4 x i32> %c, i32 0
232  %c1 = extractelement <4 x i32> %c, i32 1
233  %c2 = extractelement <4 x i32> %c, i32 2
234  %c3 = extractelement <4 x i32> %c, i32 3
235  %a0 = extractelement <4 x float> %a, i32 0
236  %a1 = extractelement <4 x float> %a, i32 1
237  %a2 = extractelement <4 x float> %a, i32 2
238  %a3 = extractelement <4 x float> %a, i32 3
239  %b0 = extractelement <4 x float> %b, i32 0
240  %b1 = extractelement <4 x float> %b, i32 1
241  %b2 = extractelement <4 x float> %b, i32 2
242  %b3 = extractelement <4 x float> %b, i32 3
243  %cmp0 = icmp ne i32 %c0, 0
244  %cmp1 = icmp ne i32 %c1, 0
245  %cmp2 = icmp ne i32 %c2, 0
246  %cmp3 = icmp ne i32 %c3, 0
247  %s0 = select i1 %cmp0, float %a0, float %b0
248  %s1 = select i1 %cmp1, float %a1, float %b1
249  %s2 = select i1 %cmp2, float %a2, float %b2
250  %s3 = select i1 %cmp3, float %a3, float %b3
251  %ra = insertelement <4 x float> undef, float %s0, i32 0
252  %rb = insertelement <4 x float> %ra, float %s1, i32 1
253  %rc = insertelement <4 x float> undef, float %s2, i32 2
254  %rd = insertelement <4 x float> %rc, float %s3, i32 3
255  ret <4 x float> %rd
256}
257
258; Make sure infinite loop doesn't happen which I ran into when trying
259; to do this backwards this backwards
260define <4 x i32> @reconstruct(<4 x i32> %c) #0 {
261; CHECK-LABEL: @reconstruct(
262; CHECK-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
263; CHECK-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
264; CHECK-NEXT:    [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2
265; CHECK-NEXT:    [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3
266; CHECK-NEXT:    [[RA:%.*]] = insertelement <4 x i32> undef, i32 [[C0]], i32 0
267; CHECK-NEXT:    [[RB:%.*]] = insertelement <4 x i32> [[RA]], i32 [[C1]], i32 1
268; CHECK-NEXT:    [[RC:%.*]] = insertelement <4 x i32> [[RB]], i32 [[C2]], i32 2
269; CHECK-NEXT:    [[RD:%.*]] = insertelement <4 x i32> [[RC]], i32 [[C3]], i32 3
270; CHECK-NEXT:    ret <4 x i32> [[RD]]
271;
272  %c0 = extractelement <4 x i32> %c, i32 0
273  %c1 = extractelement <4 x i32> %c, i32 1
274  %c2 = extractelement <4 x i32> %c, i32 2
275  %c3 = extractelement <4 x i32> %c, i32 3
276  %ra = insertelement <4 x i32> undef, i32 %c0, i32 0
277  %rb = insertelement <4 x i32> %ra, i32 %c1, i32 1
278  %rc = insertelement <4 x i32> %rb, i32 %c2, i32 2
279  %rd = insertelement <4 x i32> %rc, i32 %c3, i32 3
280  ret <4 x i32> %rd
281}
282
283define <2 x float> @simple_select_v2(<2 x float> %a, <2 x float> %b, <2 x i32> %c) #0 {
284; CHECK-LABEL: @simple_select_v2(
285; CHECK-NEXT:    [[TMP1:%.*]] = icmp ne <2 x i32> [[C:%.*]], zeroinitializer
286; CHECK-NEXT:    [[TMP2:%.*]] = select <2 x i1> [[TMP1]], <2 x float> [[A:%.*]], <2 x float> [[B:%.*]]
287; CHECK-NEXT:    ret <2 x float> [[TMP2]]
288;
289  %c0 = extractelement <2 x i32> %c, i32 0
290  %c1 = extractelement <2 x i32> %c, i32 1
291  %a0 = extractelement <2 x float> %a, i32 0
292  %a1 = extractelement <2 x float> %a, i32 1
293  %b0 = extractelement <2 x float> %b, i32 0
294  %b1 = extractelement <2 x float> %b, i32 1
295  %cmp0 = icmp ne i32 %c0, 0
296  %cmp1 = icmp ne i32 %c1, 0
297  %s0 = select i1 %cmp0, float %a0, float %b0
298  %s1 = select i1 %cmp1, float %a1, float %b1
299  %ra = insertelement <2 x float> undef, float %s0, i32 0
300  %rb = insertelement <2 x float> %ra, float %s1, i32 1
301  ret <2 x float> %rb
302}
303
304; Make sure when we construct partial vectors, we don't keep
305; re-visiting the insertelement chains starting with undef
306; (low cost threshold needed to force this to happen)
307define <4 x float> @simple_select_partial_vector(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
308; CHECK-LABEL: @simple_select_partial_vector(
309; CHECK-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
310; CHECK-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
311; CHECK-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
312; CHECK-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
313; CHECK-NEXT:    [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0
314; CHECK-NEXT:    [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1
315; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x i32> undef, i32 [[C0]], i32 0
316; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x i32> [[TMP1]], i32 [[C1]], i32 1
317; CHECK-NEXT:    [[TMP3:%.*]] = icmp ne <2 x i32> [[TMP2]], zeroinitializer
318; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <2 x float> undef, float [[A0]], i32 0
319; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <2 x float> [[TMP4]], float [[A1]], i32 1
320; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <2 x float> undef, float [[B0]], i32 0
321; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x float> [[TMP6]], float [[B1]], i32 1
322; CHECK-NEXT:    [[TMP8:%.*]] = select <2 x i1> [[TMP3]], <2 x float> [[TMP5]], <2 x float> [[TMP7]]
323; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <2 x float> [[TMP8]], i32 0
324; CHECK-NEXT:    [[RA:%.*]] = insertelement <4 x float> undef, float [[TMP9]], i32 0
325; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <2 x float> [[TMP8]], i32 1
326; CHECK-NEXT:    [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[TMP10]], i32 1
327; CHECK-NEXT:    ret <4 x float> [[RB]]
328;
329  %c0 = extractelement <4 x i32> %c, i32 0
330  %c1 = extractelement <4 x i32> %c, i32 1
331  %a0 = extractelement <4 x float> %a, i32 0
332  %a1 = extractelement <4 x float> %a, i32 1
333  %b0 = extractelement <4 x float> %b, i32 0
334  %b1 = extractelement <4 x float> %b, i32 1
335  %1 = insertelement <2 x i32> undef, i32 %c0, i32 0
336  %2 = insertelement <2 x i32> %1, i32 %c1, i32 1
337  %3 = icmp ne <2 x i32> %2, zeroinitializer
338  %4 = insertelement <2 x float> undef, float %a0, i32 0
339  %5 = insertelement <2 x float> %4, float %a1, i32 1
340  %6 = insertelement <2 x float> undef, float %b0, i32 0
341  %7 = insertelement <2 x float> %6, float %b1, i32 1
342  %8 = select <2 x i1> %3, <2 x float> %5, <2 x float> %7
343  %9 = extractelement <2 x float> %8, i32 0
344  %ra = insertelement <4 x float> undef, float %9, i32 0
345  %10 = extractelement <2 x float> %8, i32 1
346  %rb = insertelement <4 x float> %ra, float %10, i32 1
347  ret <4 x float> %rb
348}
349
350; Make sure that vectorization happens even if insertelements operations
351; must be rescheduled. The case here is from compiling Julia.
352define <4 x float> @reschedule_extract(<4 x float> %a, <4 x float> %b) {
353; CHECK-LABEL: @reschedule_extract(
354; CHECK-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A:%.*]], [[B:%.*]]
355; CHECK-NEXT:    ret <4 x float> [[TMP1]]
356;
357  %a0 = extractelement <4 x float> %a, i32 0
358  %b0 = extractelement <4 x float> %b, i32 0
359  %c0 = fadd float %a0, %b0
360  %v0 = insertelement <4 x float> undef, float %c0, i32 0
361  %a1 = extractelement <4 x float> %a, i32 1
362  %b1 = extractelement <4 x float> %b, i32 1
363  %c1 = fadd float %a1, %b1
364  %v1 = insertelement <4 x float> %v0, float %c1, i32 1
365  %a2 = extractelement <4 x float> %a, i32 2
366  %b2 = extractelement <4 x float> %b, i32 2
367  %c2 = fadd float %a2, %b2
368  %v2 = insertelement <4 x float> %v1, float %c2, i32 2
369  %a3 = extractelement <4 x float> %a, i32 3
370  %b3 = extractelement <4 x float> %b, i32 3
371  %c3 = fadd float %a3, %b3
372  %v3 = insertelement <4 x float> %v2, float %c3, i32 3
373  ret <4 x float> %v3
374}
375
376; Check that cost model for vectorization takes credit for
377; instructions that are erased.
378define <4 x float> @take_credit(<4 x float> %a, <4 x float> %b) {
379; CHECK-LABEL: @take_credit(
380; CHECK-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A:%.*]], [[B:%.*]]
381; CHECK-NEXT:    ret <4 x float> [[TMP1]]
382;
383  %a0 = extractelement <4 x float> %a, i32 0
384  %b0 = extractelement <4 x float> %b, i32 0
385  %c0 = fadd float %a0, %b0
386  %a1 = extractelement <4 x float> %a, i32 1
387  %b1 = extractelement <4 x float> %b, i32 1
388  %c1 = fadd float %a1, %b1
389  %a2 = extractelement <4 x float> %a, i32 2
390  %b2 = extractelement <4 x float> %b, i32 2
391  %c2 = fadd float %a2, %b2
392  %a3 = extractelement <4 x float> %a, i32 3
393  %b3 = extractelement <4 x float> %b, i32 3
394  %c3 = fadd float %a3, %b3
395  %v0 = insertelement <4 x float> undef, float %c0, i32 0
396  %v1 = insertelement <4 x float> %v0, float %c1, i32 1
397  %v2 = insertelement <4 x float> %v1, float %c2, i32 2
398  %v3 = insertelement <4 x float> %v2, float %c3, i32 3
399  ret <4 x float> %v3
400}
401
402; Make sure we handle multiple trees that feed one build vector correctly.
403define <4 x double> @multi_tree(double %w, double %x, double %y, double %z) {
404; CHECK-LABEL: @multi_tree(
405; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x double> poison, double [[Z:%.*]], i32 0
406; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x double> [[TMP1]], double [[Y:%.*]], i32 1
407; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x double> [[TMP2]], double [[X:%.*]], i32 2
408; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x double> [[TMP3]], double [[W:%.*]], i32 3
409; CHECK-NEXT:    [[TMP5:%.*]] = fadd <4 x double> [[TMP4]], <double 3.000000e+00, double 2.000000e+00, double 1.000000e+00, double 0.000000e+00>
410; CHECK-NEXT:    [[TMP6:%.*]] = fmul <4 x double> [[TMP5]], <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00>
411; CHECK-NEXT:    ret <4 x double> [[TMP6]]
412;
413  %t0 = fadd double %w , 0.000000e+00
414  %t1 = fadd double %x , 1.000000e+00
415  %t2 = fadd double %y , 2.000000e+00
416  %t3 = fadd double %z , 3.000000e+00
417  %t4 = fmul double %t0, 1.000000e+00
418  %i1 = insertelement <4 x double> undef, double %t4, i32 3
419  %t5 = fmul double %t1, 1.000000e+00
420  %i2 = insertelement <4 x double> %i1, double %t5, i32 2
421  %t6 = fmul double %t2, 1.000000e+00
422  %i3 = insertelement <4 x double> %i2, double %t6, i32 1
423  %t7 = fmul double %t3, 1.000000e+00
424  %i4 = insertelement <4 x double> %i3, double %t7, i32 0
425  ret <4 x double> %i4
426}
427
428define <8 x float> @_vadd256(<8 x float> %a, <8 x float> %b) local_unnamed_addr #0 {
429; CHECK-LABEL: @_vadd256(
430; CHECK-NEXT:    [[TMP1:%.*]] = fadd <8 x float> [[A:%.*]], [[B:%.*]]
431; CHECK-NEXT:    ret <8 x float> [[TMP1]]
432;
433  %vecext = extractelement <8 x float> %a, i32 0
434  %vecext1 = extractelement <8 x float> %b, i32 0
435  %add = fadd float %vecext, %vecext1
436  %vecext2 = extractelement <8 x float> %a, i32 1
437  %vecext3 = extractelement <8 x float> %b, i32 1
438  %add4 = fadd float %vecext2, %vecext3
439  %vecext5 = extractelement <8 x float> %a, i32 2
440  %vecext6 = extractelement <8 x float> %b, i32 2
441  %add7 = fadd float %vecext5, %vecext6
442  %vecext8 = extractelement <8 x float> %a, i32 3
443  %vecext9 = extractelement <8 x float> %b, i32 3
444  %add10 = fadd float %vecext8, %vecext9
445  %vecext11 = extractelement <8 x float> %a, i32 4
446  %vecext12 = extractelement <8 x float> %b, i32 4
447  %add13 = fadd float %vecext11, %vecext12
448  %vecext14 = extractelement <8 x float> %a, i32 5
449  %vecext15 = extractelement <8 x float> %b, i32 5
450  %add16 = fadd float %vecext14, %vecext15
451  %vecext17 = extractelement <8 x float> %a, i32 6
452  %vecext18 = extractelement <8 x float> %b, i32 6
453  %add19 = fadd float %vecext17, %vecext18
454  %vecext20 = extractelement <8 x float> %a, i32 7
455  %vecext21 = extractelement <8 x float> %b, i32 7
456  %add22 = fadd float %vecext20, %vecext21
457  %vecinit.i = insertelement <8 x float> undef, float %add, i32 0
458  %vecinit1.i = insertelement <8 x float> %vecinit.i, float %add4, i32 1
459  %vecinit2.i = insertelement <8 x float> %vecinit1.i, float %add7, i32 2
460  %vecinit3.i = insertelement <8 x float> %vecinit2.i, float %add10, i32 3
461  %vecinit4.i = insertelement <8 x float> %vecinit3.i, float %add13, i32 4
462  %vecinit5.i = insertelement <8 x float> %vecinit4.i, float %add16, i32 5
463  %vecinit6.i = insertelement <8 x float> %vecinit5.i, float %add19, i32 6
464  %vecinit7.i = insertelement <8 x float> %vecinit6.i, float %add22, i32 7
465  ret <8 x float> %vecinit7.i
466}
467
468attributes #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" }
469