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=ANY,BIGTHRESH
3; RUN: opt -S -slp-vectorizer -slp-threshold=0 < %s | FileCheck %s --check-prefixes=ANY,ZEROTHRESH
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; ANY-LABEL: @simple_select(
10; ANY-NEXT:    [[TMP1:%.*]] = icmp ne <4 x i32> [[C:%.*]], zeroinitializer
11; ANY-NEXT:    [[TMP2:%.*]] = select <4 x i1> [[TMP1]], <4 x float> [[A:%.*]], <4 x float> [[B:%.*]]
12; ANY-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[TMP2]], i32 0
13; ANY-NEXT:    [[RA:%.*]] = insertelement <4 x float> undef, float [[TMP3]], i32 0
14; ANY-NEXT:    [[TMP4:%.*]] = extractelement <4 x float> [[TMP2]], i32 1
15; ANY-NEXT:    [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[TMP4]], i32 1
16; ANY-NEXT:    [[TMP5:%.*]] = extractelement <4 x float> [[TMP2]], i32 2
17; ANY-NEXT:    [[RC:%.*]] = insertelement <4 x float> [[RB]], float [[TMP5]], i32 2
18; ANY-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[TMP2]], i32 3
19; ANY-NEXT:    [[RD:%.*]] = insertelement <4 x float> [[RC]], float [[TMP6]], i32 3
20; ANY-NEXT:    ret <4 x float> [[RD]]
21;
22  %c0 = extractelement <4 x i32> %c, i32 0
23  %c1 = extractelement <4 x i32> %c, i32 1
24  %c2 = extractelement <4 x i32> %c, i32 2
25  %c3 = extractelement <4 x i32> %c, i32 3
26  %a0 = extractelement <4 x float> %a, i32 0
27  %a1 = extractelement <4 x float> %a, i32 1
28  %a2 = extractelement <4 x float> %a, i32 2
29  %a3 = extractelement <4 x float> %a, i32 3
30  %b0 = extractelement <4 x float> %b, i32 0
31  %b1 = extractelement <4 x float> %b, i32 1
32  %b2 = extractelement <4 x float> %b, i32 2
33  %b3 = extractelement <4 x float> %b, i32 3
34  %cmp0 = icmp ne i32 %c0, 0
35  %cmp1 = icmp ne i32 %c1, 0
36  %cmp2 = icmp ne i32 %c2, 0
37  %cmp3 = icmp ne i32 %c3, 0
38  %s0 = select i1 %cmp0, float %a0, float %b0
39  %s1 = select i1 %cmp1, float %a1, float %b1
40  %s2 = select i1 %cmp2, float %a2, float %b2
41  %s3 = select i1 %cmp3, float %a3, float %b3
42  %ra = insertelement <4 x float> undef, float %s0, i32 0
43  %rb = insertelement <4 x float> %ra, float %s1, i32 1
44  %rc = insertelement <4 x float> %rb, float %s2, i32 2
45  %rd = insertelement <4 x float> %rc, float %s3, i32 3
46  ret <4 x float> %rd
47}
48
49declare void @llvm.assume(i1) nounwind
50
51; This entire tree is ephemeral, don't vectorize any of it.
52define <4 x float> @simple_select_eph(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
53; ANY-LABEL: @simple_select_eph(
54; ANY-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
55; ANY-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
56; ANY-NEXT:    [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2
57; ANY-NEXT:    [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3
58; ANY-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
59; ANY-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
60; ANY-NEXT:    [[A2:%.*]] = extractelement <4 x float> [[A]], i32 2
61; ANY-NEXT:    [[A3:%.*]] = extractelement <4 x float> [[A]], i32 3
62; ANY-NEXT:    [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0
63; ANY-NEXT:    [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1
64; ANY-NEXT:    [[B2:%.*]] = extractelement <4 x float> [[B]], i32 2
65; ANY-NEXT:    [[B3:%.*]] = extractelement <4 x float> [[B]], i32 3
66; ANY-NEXT:    [[CMP0:%.*]] = icmp ne i32 [[C0]], 0
67; ANY-NEXT:    [[CMP1:%.*]] = icmp ne i32 [[C1]], 0
68; ANY-NEXT:    [[CMP2:%.*]] = icmp ne i32 [[C2]], 0
69; ANY-NEXT:    [[CMP3:%.*]] = icmp ne i32 [[C3]], 0
70; ANY-NEXT:    [[S0:%.*]] = select i1 [[CMP0]], float [[A0]], float [[B0]]
71; ANY-NEXT:    [[S1:%.*]] = select i1 [[CMP1]], float [[A1]], float [[B1]]
72; ANY-NEXT:    [[S2:%.*]] = select i1 [[CMP2]], float [[A2]], float [[B2]]
73; ANY-NEXT:    [[S3:%.*]] = select i1 [[CMP3]], float [[A3]], float [[B3]]
74; ANY-NEXT:    [[RA:%.*]] = insertelement <4 x float> undef, float [[S0]], i32 0
75; ANY-NEXT:    [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[S1]], i32 1
76; ANY-NEXT:    [[RC:%.*]] = insertelement <4 x float> [[RB]], float [[S2]], i32 2
77; ANY-NEXT:    [[RD:%.*]] = insertelement <4 x float> [[RC]], float [[S3]], i32 3
78; ANY-NEXT:    [[Q0:%.*]] = extractelement <4 x float> [[RD]], i32 0
79; ANY-NEXT:    [[Q1:%.*]] = extractelement <4 x float> [[RD]], i32 1
80; ANY-NEXT:    [[Q2:%.*]] = extractelement <4 x float> [[RD]], i32 2
81; ANY-NEXT:    [[Q3:%.*]] = extractelement <4 x float> [[RD]], i32 3
82; ANY-NEXT:    [[Q4:%.*]] = fadd float [[Q0]], [[Q1]]
83; ANY-NEXT:    [[Q5:%.*]] = fadd float [[Q2]], [[Q3]]
84; ANY-NEXT:    [[Q6:%.*]] = fadd float [[Q4]], [[Q5]]
85; ANY-NEXT:    [[QI:%.*]] = fcmp olt float [[Q6]], [[Q5]]
86; ANY-NEXT:    call void @llvm.assume(i1 [[QI]])
87; ANY-NEXT:    ret <4 x float> undef
88;
89  %c0 = extractelement <4 x i32> %c, i32 0
90  %c1 = extractelement <4 x i32> %c, i32 1
91  %c2 = extractelement <4 x i32> %c, i32 2
92  %c3 = extractelement <4 x i32> %c, i32 3
93  %a0 = extractelement <4 x float> %a, i32 0
94  %a1 = extractelement <4 x float> %a, i32 1
95  %a2 = extractelement <4 x float> %a, i32 2
96  %a3 = extractelement <4 x float> %a, i32 3
97  %b0 = extractelement <4 x float> %b, i32 0
98  %b1 = extractelement <4 x float> %b, i32 1
99  %b2 = extractelement <4 x float> %b, i32 2
100  %b3 = extractelement <4 x float> %b, i32 3
101  %cmp0 = icmp ne i32 %c0, 0
102  %cmp1 = icmp ne i32 %c1, 0
103  %cmp2 = icmp ne i32 %c2, 0
104  %cmp3 = icmp ne i32 %c3, 0
105  %s0 = select i1 %cmp0, float %a0, float %b0
106  %s1 = select i1 %cmp1, float %a1, float %b1
107  %s2 = select i1 %cmp2, float %a2, float %b2
108  %s3 = select i1 %cmp3, float %a3, float %b3
109  %ra = insertelement <4 x float> undef, float %s0, i32 0
110  %rb = insertelement <4 x float> %ra, float %s1, i32 1
111  %rc = insertelement <4 x float> %rb, float %s2, i32 2
112  %rd = insertelement <4 x float> %rc, float %s3, i32 3
113  %q0 = extractelement <4 x float> %rd, i32 0
114  %q1 = extractelement <4 x float> %rd, i32 1
115  %q2 = extractelement <4 x float> %rd, i32 2
116  %q3 = extractelement <4 x float> %rd, i32 3
117  %q4 = fadd float %q0, %q1
118  %q5 = fadd float %q2, %q3
119  %q6 = fadd float %q4, %q5
120  %qi = fcmp olt float %q6, %q5
121  call void @llvm.assume(i1 %qi)
122  ret <4 x float> undef
123}
124
125; Insert in an order different from the vector indices to make sure it
126; doesn't matter
127define <4 x float> @simple_select_insert_out_of_order(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
128; ANY-LABEL: @simple_select_insert_out_of_order(
129; ANY-NEXT:    [[TMP1:%.*]] = icmp ne <4 x i32> [[C:%.*]], zeroinitializer
130; ANY-NEXT:    [[TMP2:%.*]] = select <4 x i1> [[TMP1]], <4 x float> [[A:%.*]], <4 x float> [[B:%.*]]
131; ANY-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[TMP2]], i32 0
132; ANY-NEXT:    [[RA:%.*]] = insertelement <4 x float> undef, float [[TMP3]], i32 2
133; ANY-NEXT:    [[TMP4:%.*]] = extractelement <4 x float> [[TMP2]], i32 1
134; ANY-NEXT:    [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[TMP4]], i32 1
135; ANY-NEXT:    [[TMP5:%.*]] = extractelement <4 x float> [[TMP2]], i32 2
136; ANY-NEXT:    [[RC:%.*]] = insertelement <4 x float> [[RB]], float [[TMP5]], i32 0
137; ANY-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[TMP2]], i32 3
138; ANY-NEXT:    [[RD:%.*]] = insertelement <4 x float> [[RC]], float [[TMP6]], i32 3
139; ANY-NEXT:    ret <4 x float> [[RD]]
140;
141  %c0 = extractelement <4 x i32> %c, i32 0
142  %c1 = extractelement <4 x i32> %c, i32 1
143  %c2 = extractelement <4 x i32> %c, i32 2
144  %c3 = extractelement <4 x i32> %c, i32 3
145  %a0 = extractelement <4 x float> %a, i32 0
146  %a1 = extractelement <4 x float> %a, i32 1
147  %a2 = extractelement <4 x float> %a, i32 2
148  %a3 = extractelement <4 x float> %a, i32 3
149  %b0 = extractelement <4 x float> %b, i32 0
150  %b1 = extractelement <4 x float> %b, i32 1
151  %b2 = extractelement <4 x float> %b, i32 2
152  %b3 = extractelement <4 x float> %b, i32 3
153  %cmp0 = icmp ne i32 %c0, 0
154  %cmp1 = icmp ne i32 %c1, 0
155  %cmp2 = icmp ne i32 %c2, 0
156  %cmp3 = icmp ne i32 %c3, 0
157  %s0 = select i1 %cmp0, float %a0, float %b0
158  %s1 = select i1 %cmp1, float %a1, float %b1
159  %s2 = select i1 %cmp2, float %a2, float %b2
160  %s3 = select i1 %cmp3, float %a3, float %b3
161  %ra = insertelement <4 x float> undef, float %s0, i32 2
162  %rb = insertelement <4 x float> %ra, float %s1, i32 1
163  %rc = insertelement <4 x float> %rb, float %s2, i32 0
164  %rd = insertelement <4 x float> %rc, float %s3, i32 3
165  ret <4 x float> %rd
166}
167
168declare void @v4f32_user(<4 x float>) #0
169declare void @f32_user(float) #0
170
171; Multiple users of the final constructed vector
172define <4 x float> @simple_select_users(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
173; ANY-LABEL: @simple_select_users(
174; ANY-NEXT:    [[TMP1:%.*]] = icmp ne <4 x i32> [[C:%.*]], zeroinitializer
175; ANY-NEXT:    [[TMP2:%.*]] = select <4 x i1> [[TMP1]], <4 x float> [[A:%.*]], <4 x float> [[B:%.*]]
176; ANY-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[TMP2]], i32 0
177; ANY-NEXT:    [[RA:%.*]] = insertelement <4 x float> undef, float [[TMP3]], i32 0
178; ANY-NEXT:    [[TMP4:%.*]] = extractelement <4 x float> [[TMP2]], i32 1
179; ANY-NEXT:    [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[TMP4]], i32 1
180; ANY-NEXT:    [[TMP5:%.*]] = extractelement <4 x float> [[TMP2]], i32 2
181; ANY-NEXT:    [[RC:%.*]] = insertelement <4 x float> [[RB]], float [[TMP5]], i32 2
182; ANY-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[TMP2]], i32 3
183; ANY-NEXT:    [[RD:%.*]] = insertelement <4 x float> [[RC]], float [[TMP6]], i32 3
184; ANY-NEXT:    call void @v4f32_user(<4 x float> [[RD]]) #0
185; ANY-NEXT:    ret <4 x float> [[RD]]
186;
187  %c0 = extractelement <4 x i32> %c, i32 0
188  %c1 = extractelement <4 x i32> %c, i32 1
189  %c2 = extractelement <4 x i32> %c, i32 2
190  %c3 = extractelement <4 x i32> %c, i32 3
191  %a0 = extractelement <4 x float> %a, i32 0
192  %a1 = extractelement <4 x float> %a, i32 1
193  %a2 = extractelement <4 x float> %a, i32 2
194  %a3 = extractelement <4 x float> %a, i32 3
195  %b0 = extractelement <4 x float> %b, i32 0
196  %b1 = extractelement <4 x float> %b, i32 1
197  %b2 = extractelement <4 x float> %b, i32 2
198  %b3 = extractelement <4 x float> %b, i32 3
199  %cmp0 = icmp ne i32 %c0, 0
200  %cmp1 = icmp ne i32 %c1, 0
201  %cmp2 = icmp ne i32 %c2, 0
202  %cmp3 = icmp ne i32 %c3, 0
203  %s0 = select i1 %cmp0, float %a0, float %b0
204  %s1 = select i1 %cmp1, float %a1, float %b1
205  %s2 = select i1 %cmp2, float %a2, float %b2
206  %s3 = select i1 %cmp3, float %a3, float %b3
207  %ra = insertelement <4 x float> undef, float %s0, i32 0
208  %rb = insertelement <4 x float> %ra, float %s1, i32 1
209  %rc = insertelement <4 x float> %rb, float %s2, i32 2
210  %rd = insertelement <4 x float> %rc, float %s3, i32 3
211  call void @v4f32_user(<4 x float> %rd) #0
212  ret <4 x float> %rd
213}
214
215; Unused insertelement
216define <4 x float> @simple_select_no_users(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
217; ANY-LABEL: @simple_select_no_users(
218; ANY-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
219; ANY-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
220; ANY-NEXT:    [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2
221; ANY-NEXT:    [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3
222; ANY-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
223; ANY-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
224; ANY-NEXT:    [[A2:%.*]] = extractelement <4 x float> [[A]], i32 2
225; ANY-NEXT:    [[A3:%.*]] = extractelement <4 x float> [[A]], i32 3
226; ANY-NEXT:    [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0
227; ANY-NEXT:    [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1
228; ANY-NEXT:    [[B2:%.*]] = extractelement <4 x float> [[B]], i32 2
229; ANY-NEXT:    [[B3:%.*]] = extractelement <4 x float> [[B]], i32 3
230; ANY-NEXT:    [[TMP1:%.*]] = insertelement <2 x i32> undef, i32 [[C0]], i32 0
231; ANY-NEXT:    [[TMP2:%.*]] = insertelement <2 x i32> [[TMP1]], i32 [[C1]], i32 1
232; ANY-NEXT:    [[TMP3:%.*]] = icmp ne <2 x i32> [[TMP2]], zeroinitializer
233; ANY-NEXT:    [[TMP4:%.*]] = insertelement <2 x i32> undef, i32 [[C2]], i32 0
234; ANY-NEXT:    [[TMP5:%.*]] = insertelement <2 x i32> [[TMP4]], i32 [[C3]], i32 1
235; ANY-NEXT:    [[TMP6:%.*]] = icmp ne <2 x i32> [[TMP5]], zeroinitializer
236; ANY-NEXT:    [[TMP7:%.*]] = insertelement <2 x float> undef, float [[A0]], i32 0
237; ANY-NEXT:    [[TMP8:%.*]] = insertelement <2 x float> [[TMP7]], float [[A1]], i32 1
238; ANY-NEXT:    [[TMP9:%.*]] = insertelement <2 x float> undef, float [[B0]], i32 0
239; ANY-NEXT:    [[TMP10:%.*]] = insertelement <2 x float> [[TMP9]], float [[B1]], i32 1
240; ANY-NEXT:    [[TMP11:%.*]] = select <2 x i1> [[TMP3]], <2 x float> [[TMP8]], <2 x float> [[TMP10]]
241; ANY-NEXT:    [[TMP12:%.*]] = insertelement <2 x float> undef, float [[A2]], i32 0
242; ANY-NEXT:    [[TMP13:%.*]] = insertelement <2 x float> [[TMP12]], float [[A3]], i32 1
243; ANY-NEXT:    [[TMP14:%.*]] = insertelement <2 x float> undef, float [[B2]], i32 0
244; ANY-NEXT:    [[TMP15:%.*]] = insertelement <2 x float> [[TMP14]], float [[B3]], i32 1
245; ANY-NEXT:    [[TMP16:%.*]] = select <2 x i1> [[TMP6]], <2 x float> [[TMP13]], <2 x float> [[TMP15]]
246; ANY-NEXT:    [[TMP17:%.*]] = extractelement <2 x float> [[TMP11]], i32 0
247; ANY-NEXT:    [[RA:%.*]] = insertelement <4 x float> undef, float [[TMP17]], i32 0
248; ANY-NEXT:    [[TMP18:%.*]] = extractelement <2 x float> [[TMP11]], i32 1
249; ANY-NEXT:    [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[TMP18]], i32 1
250; ANY-NEXT:    [[TMP19:%.*]] = extractelement <2 x float> [[TMP16]], i32 0
251; ANY-NEXT:    [[RC:%.*]] = insertelement <4 x float> undef, float [[TMP19]], i32 2
252; ANY-NEXT:    [[TMP20:%.*]] = extractelement <2 x float> [[TMP16]], i32 1
253; ANY-NEXT:    [[RD:%.*]] = insertelement <4 x float> [[RC]], float [[TMP20]], i32 3
254; ANY-NEXT:    ret <4 x float> [[RD]]
255;
256  %c0 = extractelement <4 x i32> %c, i32 0
257  %c1 = extractelement <4 x i32> %c, i32 1
258  %c2 = extractelement <4 x i32> %c, i32 2
259  %c3 = extractelement <4 x i32> %c, i32 3
260  %a0 = extractelement <4 x float> %a, i32 0
261  %a1 = extractelement <4 x float> %a, i32 1
262  %a2 = extractelement <4 x float> %a, i32 2
263  %a3 = extractelement <4 x float> %a, i32 3
264  %b0 = extractelement <4 x float> %b, i32 0
265  %b1 = extractelement <4 x float> %b, i32 1
266  %b2 = extractelement <4 x float> %b, i32 2
267  %b3 = extractelement <4 x float> %b, i32 3
268  %cmp0 = icmp ne i32 %c0, 0
269  %cmp1 = icmp ne i32 %c1, 0
270  %cmp2 = icmp ne i32 %c2, 0
271  %cmp3 = icmp ne i32 %c3, 0
272  %s0 = select i1 %cmp0, float %a0, float %b0
273  %s1 = select i1 %cmp1, float %a1, float %b1
274  %s2 = select i1 %cmp2, float %a2, float %b2
275  %s3 = select i1 %cmp3, float %a3, float %b3
276  %ra = insertelement <4 x float> undef, float %s0, i32 0
277  %rb = insertelement <4 x float> %ra, float %s1, i32 1
278  %rc = insertelement <4 x float> undef, float %s2, i32 2
279  %rd = insertelement <4 x float> %rc, float %s3, i32 3
280  ret <4 x float> %rd
281}
282
283; Make sure infinite loop doesn't happen which I ran into when trying
284; to do this backwards this backwards
285define <4 x i32> @reconstruct(<4 x i32> %c) #0 {
286; ANY-LABEL: @reconstruct(
287; ANY-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
288; ANY-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
289; ANY-NEXT:    [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2
290; ANY-NEXT:    [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3
291; ANY-NEXT:    [[RA:%.*]] = insertelement <4 x i32> undef, i32 [[C0]], i32 0
292; ANY-NEXT:    [[RB:%.*]] = insertelement <4 x i32> [[RA]], i32 [[C1]], i32 1
293; ANY-NEXT:    [[RC:%.*]] = insertelement <4 x i32> [[RB]], i32 [[C2]], i32 2
294; ANY-NEXT:    [[RD:%.*]] = insertelement <4 x i32> [[RC]], i32 [[C3]], i32 3
295; ANY-NEXT:    ret <4 x i32> [[RD]]
296;
297  %c0 = extractelement <4 x i32> %c, i32 0
298  %c1 = extractelement <4 x i32> %c, i32 1
299  %c2 = extractelement <4 x i32> %c, i32 2
300  %c3 = extractelement <4 x i32> %c, i32 3
301  %ra = insertelement <4 x i32> undef, i32 %c0, i32 0
302  %rb = insertelement <4 x i32> %ra, i32 %c1, i32 1
303  %rc = insertelement <4 x i32> %rb, i32 %c2, i32 2
304  %rd = insertelement <4 x i32> %rc, i32 %c3, i32 3
305  ret <4 x i32> %rd
306}
307
308define <2 x float> @simple_select_v2(<2 x float> %a, <2 x float> %b, <2 x i32> %c) #0 {
309; ANY-LABEL: @simple_select_v2(
310; ANY-NEXT:    [[TMP1:%.*]] = icmp ne <2 x i32> [[C:%.*]], zeroinitializer
311; ANY-NEXT:    [[TMP2:%.*]] = select <2 x i1> [[TMP1]], <2 x float> [[A:%.*]], <2 x float> [[B:%.*]]
312; ANY-NEXT:    [[TMP3:%.*]] = extractelement <2 x float> [[TMP2]], i32 0
313; ANY-NEXT:    [[RA:%.*]] = insertelement <2 x float> undef, float [[TMP3]], i32 0
314; ANY-NEXT:    [[TMP4:%.*]] = extractelement <2 x float> [[TMP2]], i32 1
315; ANY-NEXT:    [[RB:%.*]] = insertelement <2 x float> [[RA]], float [[TMP4]], i32 1
316; ANY-NEXT:    ret <2 x float> [[RB]]
317;
318  %c0 = extractelement <2 x i32> %c, i32 0
319  %c1 = extractelement <2 x i32> %c, i32 1
320  %a0 = extractelement <2 x float> %a, i32 0
321  %a1 = extractelement <2 x float> %a, i32 1
322  %b0 = extractelement <2 x float> %b, i32 0
323  %b1 = extractelement <2 x float> %b, i32 1
324  %cmp0 = icmp ne i32 %c0, 0
325  %cmp1 = icmp ne i32 %c1, 0
326  %s0 = select i1 %cmp0, float %a0, float %b0
327  %s1 = select i1 %cmp1, float %a1, float %b1
328  %ra = insertelement <2 x float> undef, float %s0, i32 0
329  %rb = insertelement <2 x float> %ra, float %s1, i32 1
330  ret <2 x float> %rb
331}
332
333; Make sure when we construct partial vectors, we don't keep
334; re-visiting the insertelement chains starting with undef
335; (low cost threshold needed to force this to happen)
336define <4 x float> @simple_select_partial_vector(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
337; ANY-LABEL: @simple_select_partial_vector(
338; ANY-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
339; ANY-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
340; ANY-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
341; ANY-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
342; ANY-NEXT:    [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0
343; ANY-NEXT:    [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1
344; ANY-NEXT:    [[TMP1:%.*]] = insertelement <2 x i32> undef, i32 [[C0]], i32 0
345; ANY-NEXT:    [[TMP2:%.*]] = insertelement <2 x i32> [[TMP1]], i32 [[C1]], i32 1
346; ANY-NEXT:    [[TMP3:%.*]] = icmp ne <2 x i32> [[TMP2]], zeroinitializer
347; ANY-NEXT:    [[TMP4:%.*]] = insertelement <2 x float> undef, float [[A0]], i32 0
348; ANY-NEXT:    [[TMP5:%.*]] = insertelement <2 x float> [[TMP4]], float [[A1]], i32 1
349; ANY-NEXT:    [[TMP6:%.*]] = insertelement <2 x float> undef, float [[B0]], i32 0
350; ANY-NEXT:    [[TMP7:%.*]] = insertelement <2 x float> [[TMP6]], float [[B1]], i32 1
351; ANY-NEXT:    [[TMP8:%.*]] = select <2 x i1> [[TMP3]], <2 x float> [[TMP5]], <2 x float> [[TMP7]]
352; ANY-NEXT:    [[TMP9:%.*]] = extractelement <2 x float> [[TMP8]], i32 0
353; ANY-NEXT:    [[RA:%.*]] = insertelement <4 x float> undef, float [[TMP9]], i32 0
354; ANY-NEXT:    [[TMP10:%.*]] = extractelement <2 x float> [[TMP8]], i32 1
355; ANY-NEXT:    [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[TMP10]], i32 1
356; ANY-NEXT:    ret <4 x float> [[RB]]
357;
358  %c0 = extractelement <4 x i32> %c, i32 0
359  %c1 = extractelement <4 x i32> %c, i32 1
360  %a0 = extractelement <4 x float> %a, i32 0
361  %a1 = extractelement <4 x float> %a, i32 1
362  %b0 = extractelement <4 x float> %b, i32 0
363  %b1 = extractelement <4 x float> %b, i32 1
364  %1 = insertelement <2 x i32> undef, i32 %c0, i32 0
365  %2 = insertelement <2 x i32> %1, i32 %c1, i32 1
366  %3 = icmp ne <2 x i32> %2, zeroinitializer
367  %4 = insertelement <2 x float> undef, float %a0, i32 0
368  %5 = insertelement <2 x float> %4, float %a1, i32 1
369  %6 = insertelement <2 x float> undef, float %b0, i32 0
370  %7 = insertelement <2 x float> %6, float %b1, i32 1
371  %8 = select <2 x i1> %3, <2 x float> %5, <2 x float> %7
372  %9 = extractelement <2 x float> %8, i32 0
373  %ra = insertelement <4 x float> undef, float %9, i32 0
374  %10 = extractelement <2 x float> %8, i32 1
375  %rb = insertelement <4 x float> %ra, float %10, i32 1
376  ret <4 x float> %rb
377}
378
379; Make sure that vectorization happens even if insertelements operations
380; must be rescheduled. The case here is from compiling Julia.
381define <4 x float> @reschedule_extract(<4 x float> %a, <4 x float> %b) {
382; ANY-LABEL: @reschedule_extract(
383; ANY-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A:%.*]], [[B:%.*]]
384; ANY-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> [[TMP1]], i32 0
385; ANY-NEXT:    [[V0:%.*]] = insertelement <4 x float> undef, float [[TMP2]], i32 0
386; ANY-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[TMP1]], i32 1
387; ANY-NEXT:    [[V1:%.*]] = insertelement <4 x float> [[V0]], float [[TMP3]], i32 1
388; ANY-NEXT:    [[TMP4:%.*]] = extractelement <4 x float> [[TMP1]], i32 2
389; ANY-NEXT:    [[V2:%.*]] = insertelement <4 x float> [[V1]], float [[TMP4]], i32 2
390; ANY-NEXT:    [[TMP5:%.*]] = extractelement <4 x float> [[TMP1]], i32 3
391; ANY-NEXT:    [[V3:%.*]] = insertelement <4 x float> [[V2]], float [[TMP5]], i32 3
392; ANY-NEXT:    ret <4 x float> [[V3]]
393;
394  %a0 = extractelement <4 x float> %a, i32 0
395  %b0 = extractelement <4 x float> %b, i32 0
396  %c0 = fadd float %a0, %b0
397  %v0 = insertelement <4 x float> undef, float %c0, i32 0
398  %a1 = extractelement <4 x float> %a, i32 1
399  %b1 = extractelement <4 x float> %b, i32 1
400  %c1 = fadd float %a1, %b1
401  %v1 = insertelement <4 x float> %v0, float %c1, i32 1
402  %a2 = extractelement <4 x float> %a, i32 2
403  %b2 = extractelement <4 x float> %b, i32 2
404  %c2 = fadd float %a2, %b2
405  %v2 = insertelement <4 x float> %v1, float %c2, i32 2
406  %a3 = extractelement <4 x float> %a, i32 3
407  %b3 = extractelement <4 x float> %b, i32 3
408  %c3 = fadd float %a3, %b3
409  %v3 = insertelement <4 x float> %v2, float %c3, i32 3
410  ret <4 x float> %v3
411}
412
413; Check that cost model for vectorization takes credit for
414; instructions that are erased.
415define <4 x float> @take_credit(<4 x float> %a, <4 x float> %b) {
416; ANY-LABEL: @take_credit(
417; ANY-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A:%.*]], [[B:%.*]]
418; ANY-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> [[TMP1]], i32 0
419; ANY-NEXT:    [[V0:%.*]] = insertelement <4 x float> undef, float [[TMP2]], i32 0
420; ANY-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[TMP1]], i32 1
421; ANY-NEXT:    [[V1:%.*]] = insertelement <4 x float> [[V0]], float [[TMP3]], i32 1
422; ANY-NEXT:    [[TMP4:%.*]] = extractelement <4 x float> [[TMP1]], i32 2
423; ANY-NEXT:    [[V2:%.*]] = insertelement <4 x float> [[V1]], float [[TMP4]], i32 2
424; ANY-NEXT:    [[TMP5:%.*]] = extractelement <4 x float> [[TMP1]], i32 3
425; ANY-NEXT:    [[V3:%.*]] = insertelement <4 x float> [[V2]], float [[TMP5]], i32 3
426; ANY-NEXT:    ret <4 x float> [[V3]]
427;
428  %a0 = extractelement <4 x float> %a, i32 0
429  %b0 = extractelement <4 x float> %b, i32 0
430  %c0 = fadd float %a0, %b0
431  %a1 = extractelement <4 x float> %a, i32 1
432  %b1 = extractelement <4 x float> %b, i32 1
433  %c1 = fadd float %a1, %b1
434  %a2 = extractelement <4 x float> %a, i32 2
435  %b2 = extractelement <4 x float> %b, i32 2
436  %c2 = fadd float %a2, %b2
437  %a3 = extractelement <4 x float> %a, i32 3
438  %b3 = extractelement <4 x float> %b, i32 3
439  %c3 = fadd float %a3, %b3
440  %v0 = insertelement <4 x float> undef, float %c0, i32 0
441  %v1 = insertelement <4 x float> %v0, float %c1, i32 1
442  %v2 = insertelement <4 x float> %v1, float %c2, i32 2
443  %v3 = insertelement <4 x float> %v2, float %c3, i32 3
444  ret <4 x float> %v3
445}
446
447; Make sure we handle multiple trees that feed one build vector correctly.
448define <4 x double> @multi_tree(double %w, double %x, double %y, double %z) {
449; ANY-LABEL: @multi_tree(
450; ANY-NEXT:    [[TMP1:%.*]] = insertelement <4 x double> undef, double [[W:%.*]], i32 0
451; ANY-NEXT:    [[TMP2:%.*]] = insertelement <4 x double> [[TMP1]], double [[X:%.*]], i32 1
452; ANY-NEXT:    [[TMP3:%.*]] = insertelement <4 x double> [[TMP2]], double [[Y:%.*]], i32 2
453; ANY-NEXT:    [[TMP4:%.*]] = insertelement <4 x double> [[TMP3]], double [[Z:%.*]], i32 3
454; ANY-NEXT:    [[TMP5:%.*]] = fadd <4 x double> [[TMP4]], <double 0.000000e+00, double 1.000000e+00, double 2.000000e+00, double 3.000000e+00>
455; ANY-NEXT:    [[TMP6:%.*]] = fmul <4 x double> [[TMP5]], <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00>
456; ANY-NEXT:    [[TMP7:%.*]] = extractelement <4 x double> [[TMP6]], i32 0
457; ANY-NEXT:    [[I1:%.*]] = insertelement <4 x double> undef, double [[TMP7]], i32 3
458; ANY-NEXT:    [[TMP8:%.*]] = extractelement <4 x double> [[TMP6]], i32 1
459; ANY-NEXT:    [[I2:%.*]] = insertelement <4 x double> [[I1]], double [[TMP8]], i32 2
460; ANY-NEXT:    [[TMP9:%.*]] = extractelement <4 x double> [[TMP6]], i32 2
461; ANY-NEXT:    [[I3:%.*]] = insertelement <4 x double> [[I2]], double [[TMP9]], i32 1
462; ANY-NEXT:    [[TMP10:%.*]] = extractelement <4 x double> [[TMP6]], i32 3
463; ANY-NEXT:    [[I4:%.*]] = insertelement <4 x double> [[I3]], double [[TMP10]], i32 0
464; ANY-NEXT:    ret <4 x double> [[I4]]
465;
466  %t0 = fadd double %w , 0.000000e+00
467  %t1 = fadd double %x , 1.000000e+00
468  %t2 = fadd double %y , 2.000000e+00
469  %t3 = fadd double %z , 3.000000e+00
470  %t4 = fmul double %t0, 1.000000e+00
471  %i1 = insertelement <4 x double> undef, double %t4, i32 3
472  %t5 = fmul double %t1, 1.000000e+00
473  %i2 = insertelement <4 x double> %i1, double %t5, i32 2
474  %t6 = fmul double %t2, 1.000000e+00
475  %i3 = insertelement <4 x double> %i2, double %t6, i32 1
476  %t7 = fmul double %t3, 1.000000e+00
477  %i4 = insertelement <4 x double> %i3, double %t7, i32 0
478  ret <4 x double> %i4
479}
480
481define <8 x float> @_vadd256(<8 x float> %a, <8 x float> %b) local_unnamed_addr #0 {
482; ANY-LABEL: @_vadd256(
483; ANY-NEXT:    [[TMP1:%.*]] = fadd <8 x float> [[A:%.*]], [[B:%.*]]
484; ANY-NEXT:    [[TMP2:%.*]] = extractelement <8 x float> [[TMP1]], i32 0
485; ANY-NEXT:    [[VECINIT_I:%.*]] = insertelement <8 x float> undef, float [[TMP2]], i32 0
486; ANY-NEXT:    [[TMP3:%.*]] = extractelement <8 x float> [[TMP1]], i32 1
487; ANY-NEXT:    [[VECINIT1_I:%.*]] = insertelement <8 x float> [[VECINIT_I]], float [[TMP3]], i32 1
488; ANY-NEXT:    [[TMP4:%.*]] = extractelement <8 x float> [[TMP1]], i32 2
489; ANY-NEXT:    [[VECINIT2_I:%.*]] = insertelement <8 x float> [[VECINIT1_I]], float [[TMP4]], i32 2
490; ANY-NEXT:    [[TMP5:%.*]] = extractelement <8 x float> [[TMP1]], i32 3
491; ANY-NEXT:    [[VECINIT3_I:%.*]] = insertelement <8 x float> [[VECINIT2_I]], float [[TMP5]], i32 3
492; ANY-NEXT:    [[TMP6:%.*]] = extractelement <8 x float> [[TMP1]], i32 4
493; ANY-NEXT:    [[VECINIT4_I:%.*]] = insertelement <8 x float> [[VECINIT3_I]], float [[TMP6]], i32 4
494; ANY-NEXT:    [[TMP7:%.*]] = extractelement <8 x float> [[TMP1]], i32 5
495; ANY-NEXT:    [[VECINIT5_I:%.*]] = insertelement <8 x float> [[VECINIT4_I]], float [[TMP7]], i32 5
496; ANY-NEXT:    [[TMP8:%.*]] = extractelement <8 x float> [[TMP1]], i32 6
497; ANY-NEXT:    [[VECINIT6_I:%.*]] = insertelement <8 x float> [[VECINIT5_I]], float [[TMP8]], i32 6
498; ANY-NEXT:    [[TMP9:%.*]] = extractelement <8 x float> [[TMP1]], i32 7
499; ANY-NEXT:    [[VECINIT7_I:%.*]] = insertelement <8 x float> [[VECINIT6_I]], float [[TMP9]], i32 7
500; ANY-NEXT:    ret <8 x float> [[VECINIT7_I]]
501;
502  %vecext = extractelement <8 x float> %a, i32 0
503  %vecext1 = extractelement <8 x float> %b, i32 0
504  %add = fadd float %vecext, %vecext1
505  %vecext2 = extractelement <8 x float> %a, i32 1
506  %vecext3 = extractelement <8 x float> %b, i32 1
507  %add4 = fadd float %vecext2, %vecext3
508  %vecext5 = extractelement <8 x float> %a, i32 2
509  %vecext6 = extractelement <8 x float> %b, i32 2
510  %add7 = fadd float %vecext5, %vecext6
511  %vecext8 = extractelement <8 x float> %a, i32 3
512  %vecext9 = extractelement <8 x float> %b, i32 3
513  %add10 = fadd float %vecext8, %vecext9
514  %vecext11 = extractelement <8 x float> %a, i32 4
515  %vecext12 = extractelement <8 x float> %b, i32 4
516  %add13 = fadd float %vecext11, %vecext12
517  %vecext14 = extractelement <8 x float> %a, i32 5
518  %vecext15 = extractelement <8 x float> %b, i32 5
519  %add16 = fadd float %vecext14, %vecext15
520  %vecext17 = extractelement <8 x float> %a, i32 6
521  %vecext18 = extractelement <8 x float> %b, i32 6
522  %add19 = fadd float %vecext17, %vecext18
523  %vecext20 = extractelement <8 x float> %a, i32 7
524  %vecext21 = extractelement <8 x float> %b, i32 7
525  %add22 = fadd float %vecext20, %vecext21
526  %vecinit.i = insertelement <8 x float> undef, float %add, i32 0
527  %vecinit1.i = insertelement <8 x float> %vecinit.i, float %add4, i32 1
528  %vecinit2.i = insertelement <8 x float> %vecinit1.i, float %add7, i32 2
529  %vecinit3.i = insertelement <8 x float> %vecinit2.i, float %add10, i32 3
530  %vecinit4.i = insertelement <8 x float> %vecinit3.i, float %add13, i32 4
531  %vecinit5.i = insertelement <8 x float> %vecinit4.i, float %add16, i32 5
532  %vecinit6.i = insertelement <8 x float> %vecinit5.i, float %add19, i32 6
533  %vecinit7.i = insertelement <8 x float> %vecinit6.i, float %add22, i32 7
534  ret <8 x float> %vecinit7.i
535}
536
537attributes #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" }
538