1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt -slp-vectorizer -slp-vectorize-hor -S -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver2 -debug < %s 2>&1 | FileCheck %s --check-prefixes=CHECK,AVX
3 ; RUN: opt -slp-vectorizer -slp-vectorize-hor -S -mtriple=x86_64-unknown-linux-gnu -mcpu=core2 -debug < %s 2>&1 | FileCheck %s --check-prefixes=CHECK,SSE
6 ; int test_add(unsigned int *p) {
8 ; for (int i = 0; i < 8; i++)
13 ; Vector cost is 5, Scalar cost is 7
14 ; AVX: Adding cost -2 for reduction that starts with %7 = load i32, i32* %arrayidx.7, align 4 (It is a splitting reduction)
15 ; Vector cost is 6, Scalar cost is 7
16 ; SSE: Adding cost -1 for reduction that starts with %7 = load i32, i32* %arrayidx.7, align 4 (It is a splitting reduction)
17 define i32 @test_add(i32* nocapture readonly %p) {
18 ; CHECK-LABEL: @test_add(
20 ; CHECK-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, i32* [[P:%.*]], i64 1
21 ; CHECK-NEXT: [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 2
22 ; CHECK-NEXT: [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 3
23 ; CHECK-NEXT: [[ARRAYIDX_4:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 4
24 ; CHECK-NEXT: [[ARRAYIDX_5:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 5
25 ; CHECK-NEXT: [[ARRAYIDX_6:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 6
26 ; CHECK-NEXT: [[ARRAYIDX_7:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 7
27 ; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[P]] to <8 x i32>*
28 ; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i32>, <8 x i32>* [[TMP0]], align 4
29 ; CHECK-NEXT: [[MUL_18:%.*]] = add i32 undef, undef
30 ; CHECK-NEXT: [[MUL_29:%.*]] = add i32 undef, [[MUL_18]]
31 ; CHECK-NEXT: [[MUL_310:%.*]] = add i32 undef, [[MUL_29]]
32 ; CHECK-NEXT: [[MUL_411:%.*]] = add i32 undef, [[MUL_310]]
33 ; CHECK-NEXT: [[MUL_512:%.*]] = add i32 undef, [[MUL_411]]
34 ; CHECK-NEXT: [[MUL_613:%.*]] = add i32 undef, [[MUL_512]]
35 ; CHECK-NEXT: [[RDX_SHUF:%.*]] = shufflevector <8 x i32> [[TMP1]], <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
36 ; CHECK-NEXT: [[BIN_RDX:%.*]] = add <8 x i32> [[TMP1]], [[RDX_SHUF]]
37 ; CHECK-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <8 x i32> [[BIN_RDX]], <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
38 ; CHECK-NEXT: [[BIN_RDX2:%.*]] = add <8 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
39 ; CHECK-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <8 x i32> [[BIN_RDX2]], <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
40 ; CHECK-NEXT: [[BIN_RDX4:%.*]] = add <8 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
41 ; CHECK-NEXT: [[TMP2:%.*]] = extractelement <8 x i32> [[BIN_RDX4]], i32 0
42 ; CHECK-NEXT: [[MUL_714:%.*]] = add i32 undef, [[MUL_613]]
43 ; CHECK-NEXT: ret i32 [[TMP2]]
46 %0 = load i32, i32* %p, align 4
47 %arrayidx.1 = getelementptr inbounds i32, i32* %p, i64 1
48 %1 = load i32, i32* %arrayidx.1, align 4
49 %mul.18 = add i32 %1, %0
50 %arrayidx.2 = getelementptr inbounds i32, i32* %p, i64 2
51 %2 = load i32, i32* %arrayidx.2, align 4
52 %mul.29 = add i32 %2, %mul.18
53 %arrayidx.3 = getelementptr inbounds i32, i32* %p, i64 3
54 %3 = load i32, i32* %arrayidx.3, align 4
55 %mul.310 = add i32 %3, %mul.29
56 %arrayidx.4 = getelementptr inbounds i32, i32* %p, i64 4
57 %4 = load i32, i32* %arrayidx.4, align 4
58 %mul.411 = add i32 %4, %mul.310
59 %arrayidx.5 = getelementptr inbounds i32, i32* %p, i64 5
60 %5 = load i32, i32* %arrayidx.5, align 4
61 %mul.512 = add i32 %5, %mul.411
62 %arrayidx.6 = getelementptr inbounds i32, i32* %p, i64 6
63 %6 = load i32, i32* %arrayidx.6, align 4
64 %mul.613 = add i32 %6, %mul.512
65 %arrayidx.7 = getelementptr inbounds i32, i32* %p, i64 7
66 %7 = load i32, i32* %arrayidx.7, align 4
67 %mul.714 = add i32 %7, %mul.613
71 ; int test_mul(unsigned int *p) {
73 ; for (int i = 0; i < 8; i++)
78 define i32 @test_mul(i32* nocapture readonly %p) {
79 ; CHECK-LABEL: @test_mul(
81 ; CHECK-NEXT: [[TMP0:%.*]] = load i32, i32* [[P:%.*]], align 4
82 ; CHECK-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 1
83 ; CHECK-NEXT: [[TMP1:%.*]] = load i32, i32* [[ARRAYIDX_1]], align 4
84 ; CHECK-NEXT: [[MUL_18:%.*]] = mul i32 [[TMP1]], [[TMP0]]
85 ; CHECK-NEXT: [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 2
86 ; CHECK-NEXT: [[TMP2:%.*]] = load i32, i32* [[ARRAYIDX_2]], align 4
87 ; CHECK-NEXT: [[MUL_29:%.*]] = mul i32 [[TMP2]], [[MUL_18]]
88 ; CHECK-NEXT: [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 3
89 ; CHECK-NEXT: [[TMP3:%.*]] = load i32, i32* [[ARRAYIDX_3]], align 4
90 ; CHECK-NEXT: [[MUL_310:%.*]] = mul i32 [[TMP3]], [[MUL_29]]
91 ; CHECK-NEXT: [[ARRAYIDX_4:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 4
92 ; CHECK-NEXT: [[TMP4:%.*]] = load i32, i32* [[ARRAYIDX_4]], align 4
93 ; CHECK-NEXT: [[MUL_411:%.*]] = mul i32 [[TMP4]], [[MUL_310]]
94 ; CHECK-NEXT: [[ARRAYIDX_5:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 5
95 ; CHECK-NEXT: [[TMP5:%.*]] = load i32, i32* [[ARRAYIDX_5]], align 4
96 ; CHECK-NEXT: [[MUL_512:%.*]] = mul i32 [[TMP5]], [[MUL_411]]
97 ; CHECK-NEXT: [[ARRAYIDX_6:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 6
98 ; CHECK-NEXT: [[TMP6:%.*]] = load i32, i32* [[ARRAYIDX_6]], align 4
99 ; CHECK-NEXT: [[MUL_613:%.*]] = mul i32 [[TMP6]], [[MUL_512]]
100 ; CHECK-NEXT: [[ARRAYIDX_7:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 7
101 ; CHECK-NEXT: [[TMP7:%.*]] = load i32, i32* [[ARRAYIDX_7]], align 4
102 ; CHECK-NEXT: [[MUL_714:%.*]] = mul i32 [[TMP7]], [[MUL_613]]
103 ; CHECK-NEXT: ret i32 [[MUL_714]]
106 %0 = load i32, i32* %p, align 4
107 %arrayidx.1 = getelementptr inbounds i32, i32* %p, i64 1
108 %1 = load i32, i32* %arrayidx.1, align 4
109 %mul.18 = mul i32 %1, %0
110 %arrayidx.2 = getelementptr inbounds i32, i32* %p, i64 2
111 %2 = load i32, i32* %arrayidx.2, align 4
112 %mul.29 = mul i32 %2, %mul.18
113 %arrayidx.3 = getelementptr inbounds i32, i32* %p, i64 3
114 %3 = load i32, i32* %arrayidx.3, align 4
115 %mul.310 = mul i32 %3, %mul.29
116 %arrayidx.4 = getelementptr inbounds i32, i32* %p, i64 4
117 %4 = load i32, i32* %arrayidx.4, align 4
118 %mul.411 = mul i32 %4, %mul.310
119 %arrayidx.5 = getelementptr inbounds i32, i32* %p, i64 5
120 %5 = load i32, i32* %arrayidx.5, align 4
121 %mul.512 = mul i32 %5, %mul.411
122 %arrayidx.6 = getelementptr inbounds i32, i32* %p, i64 6
123 %6 = load i32, i32* %arrayidx.6, align 4
124 %mul.613 = mul i32 %6, %mul.512
125 %arrayidx.7 = getelementptr inbounds i32, i32* %p, i64 7
126 %7 = load i32, i32* %arrayidx.7, align 4
127 %mul.714 = mul i32 %7, %mul.613
131 ; int test_and(unsigned int *p) {
133 ; for (int i = 0; i < 8; i++)
138 define i32 @test_and(i32* nocapture readonly %p) {
139 ; CHECK-LABEL: @test_and(
141 ; CHECK-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, i32* [[P:%.*]], i64 1
142 ; CHECK-NEXT: [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 2
143 ; CHECK-NEXT: [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 3
144 ; CHECK-NEXT: [[ARRAYIDX_4:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 4
145 ; CHECK-NEXT: [[ARRAYIDX_5:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 5
146 ; CHECK-NEXT: [[ARRAYIDX_6:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 6
147 ; CHECK-NEXT: [[ARRAYIDX_7:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 7
148 ; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[P]] to <8 x i32>*
149 ; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i32>, <8 x i32>* [[TMP0]], align 4
150 ; CHECK-NEXT: [[MUL_18:%.*]] = and i32 undef, undef
151 ; CHECK-NEXT: [[MUL_29:%.*]] = and i32 undef, [[MUL_18]]
152 ; CHECK-NEXT: [[MUL_310:%.*]] = and i32 undef, [[MUL_29]]
153 ; CHECK-NEXT: [[MUL_411:%.*]] = and i32 undef, [[MUL_310]]
154 ; CHECK-NEXT: [[MUL_512:%.*]] = and i32 undef, [[MUL_411]]
155 ; CHECK-NEXT: [[MUL_613:%.*]] = and i32 undef, [[MUL_512]]
156 ; CHECK-NEXT: [[RDX_SHUF:%.*]] = shufflevector <8 x i32> [[TMP1]], <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
157 ; CHECK-NEXT: [[BIN_RDX:%.*]] = and <8 x i32> [[TMP1]], [[RDX_SHUF]]
158 ; CHECK-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <8 x i32> [[BIN_RDX]], <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
159 ; CHECK-NEXT: [[BIN_RDX2:%.*]] = and <8 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
160 ; CHECK-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <8 x i32> [[BIN_RDX2]], <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
161 ; CHECK-NEXT: [[BIN_RDX4:%.*]] = and <8 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
162 ; CHECK-NEXT: [[TMP2:%.*]] = extractelement <8 x i32> [[BIN_RDX4]], i32 0
163 ; CHECK-NEXT: [[MUL_714:%.*]] = and i32 undef, [[MUL_613]]
164 ; CHECK-NEXT: ret i32 [[TMP2]]
167 %0 = load i32, i32* %p, align 4
168 %arrayidx.1 = getelementptr inbounds i32, i32* %p, i64 1
169 %1 = load i32, i32* %arrayidx.1, align 4
170 %mul.18 = and i32 %1, %0
171 %arrayidx.2 = getelementptr inbounds i32, i32* %p, i64 2
172 %2 = load i32, i32* %arrayidx.2, align 4
173 %mul.29 = and i32 %2, %mul.18
174 %arrayidx.3 = getelementptr inbounds i32, i32* %p, i64 3
175 %3 = load i32, i32* %arrayidx.3, align 4
176 %mul.310 = and i32 %3, %mul.29
177 %arrayidx.4 = getelementptr inbounds i32, i32* %p, i64 4
178 %4 = load i32, i32* %arrayidx.4, align 4
179 %mul.411 = and i32 %4, %mul.310
180 %arrayidx.5 = getelementptr inbounds i32, i32* %p, i64 5
181 %5 = load i32, i32* %arrayidx.5, align 4
182 %mul.512 = and i32 %5, %mul.411
183 %arrayidx.6 = getelementptr inbounds i32, i32* %p, i64 6
184 %6 = load i32, i32* %arrayidx.6, align 4
185 %mul.613 = and i32 %6, %mul.512
186 %arrayidx.7 = getelementptr inbounds i32, i32* %p, i64 7
187 %7 = load i32, i32* %arrayidx.7, align 4
188 %mul.714 = and i32 %7, %mul.613
192 ; int test_or(unsigned int *p) {
194 ; for (int i = 0; i < 8; i++)
199 define i32 @test_or(i32* nocapture readonly %p) {
200 ; CHECK-LABEL: @test_or(
202 ; CHECK-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, i32* [[P:%.*]], i64 1
203 ; CHECK-NEXT: [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 2
204 ; CHECK-NEXT: [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 3
205 ; CHECK-NEXT: [[ARRAYIDX_4:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 4
206 ; CHECK-NEXT: [[ARRAYIDX_5:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 5
207 ; CHECK-NEXT: [[ARRAYIDX_6:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 6
208 ; CHECK-NEXT: [[ARRAYIDX_7:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 7
209 ; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[P]] to <8 x i32>*
210 ; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i32>, <8 x i32>* [[TMP0]], align 4
211 ; CHECK-NEXT: [[MUL_18:%.*]] = or i32 undef, undef
212 ; CHECK-NEXT: [[MUL_29:%.*]] = or i32 undef, [[MUL_18]]
213 ; CHECK-NEXT: [[MUL_310:%.*]] = or i32 undef, [[MUL_29]]
214 ; CHECK-NEXT: [[MUL_411:%.*]] = or i32 undef, [[MUL_310]]
215 ; CHECK-NEXT: [[MUL_512:%.*]] = or i32 undef, [[MUL_411]]
216 ; CHECK-NEXT: [[MUL_613:%.*]] = or i32 undef, [[MUL_512]]
217 ; CHECK-NEXT: [[RDX_SHUF:%.*]] = shufflevector <8 x i32> [[TMP1]], <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
218 ; CHECK-NEXT: [[BIN_RDX:%.*]] = or <8 x i32> [[TMP1]], [[RDX_SHUF]]
219 ; CHECK-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <8 x i32> [[BIN_RDX]], <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
220 ; CHECK-NEXT: [[BIN_RDX2:%.*]] = or <8 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
221 ; CHECK-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <8 x i32> [[BIN_RDX2]], <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
222 ; CHECK-NEXT: [[BIN_RDX4:%.*]] = or <8 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
223 ; CHECK-NEXT: [[TMP2:%.*]] = extractelement <8 x i32> [[BIN_RDX4]], i32 0
224 ; CHECK-NEXT: [[MUL_714:%.*]] = or i32 undef, [[MUL_613]]
225 ; CHECK-NEXT: ret i32 [[TMP2]]
228 %0 = load i32, i32* %p, align 4
229 %arrayidx.1 = getelementptr inbounds i32, i32* %p, i64 1
230 %1 = load i32, i32* %arrayidx.1, align 4
231 %mul.18 = or i32 %1, %0
232 %arrayidx.2 = getelementptr inbounds i32, i32* %p, i64 2
233 %2 = load i32, i32* %arrayidx.2, align 4
234 %mul.29 = or i32 %2, %mul.18
235 %arrayidx.3 = getelementptr inbounds i32, i32* %p, i64 3
236 %3 = load i32, i32* %arrayidx.3, align 4
237 %mul.310 = or i32 %3, %mul.29
238 %arrayidx.4 = getelementptr inbounds i32, i32* %p, i64 4
239 %4 = load i32, i32* %arrayidx.4, align 4
240 %mul.411 = or i32 %4, %mul.310
241 %arrayidx.5 = getelementptr inbounds i32, i32* %p, i64 5
242 %5 = load i32, i32* %arrayidx.5, align 4
243 %mul.512 = or i32 %5, %mul.411
244 %arrayidx.6 = getelementptr inbounds i32, i32* %p, i64 6
245 %6 = load i32, i32* %arrayidx.6, align 4
246 %mul.613 = or i32 %6, %mul.512
247 %arrayidx.7 = getelementptr inbounds i32, i32* %p, i64 7
248 %7 = load i32, i32* %arrayidx.7, align 4
249 %mul.714 = or i32 %7, %mul.613
253 ; int test_xor(unsigned int *p) {
255 ; for (int i = 0; i < 8; i++)
260 define i32 @test_xor(i32* nocapture readonly %p) {
261 ; CHECK-LABEL: @test_xor(
263 ; CHECK-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, i32* [[P:%.*]], i64 1
264 ; CHECK-NEXT: [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 2
265 ; CHECK-NEXT: [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 3
266 ; CHECK-NEXT: [[ARRAYIDX_4:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 4
267 ; CHECK-NEXT: [[ARRAYIDX_5:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 5
268 ; CHECK-NEXT: [[ARRAYIDX_6:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 6
269 ; CHECK-NEXT: [[ARRAYIDX_7:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 7
270 ; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[P]] to <8 x i32>*
271 ; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i32>, <8 x i32>* [[TMP0]], align 4
272 ; CHECK-NEXT: [[MUL_18:%.*]] = xor i32 undef, undef
273 ; CHECK-NEXT: [[MUL_29:%.*]] = xor i32 undef, [[MUL_18]]
274 ; CHECK-NEXT: [[MUL_310:%.*]] = xor i32 undef, [[MUL_29]]
275 ; CHECK-NEXT: [[MUL_411:%.*]] = xor i32 undef, [[MUL_310]]
276 ; CHECK-NEXT: [[MUL_512:%.*]] = xor i32 undef, [[MUL_411]]
277 ; CHECK-NEXT: [[MUL_613:%.*]] = xor i32 undef, [[MUL_512]]
278 ; CHECK-NEXT: [[RDX_SHUF:%.*]] = shufflevector <8 x i32> [[TMP1]], <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
279 ; CHECK-NEXT: [[BIN_RDX:%.*]] = xor <8 x i32> [[TMP1]], [[RDX_SHUF]]
280 ; CHECK-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <8 x i32> [[BIN_RDX]], <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
281 ; CHECK-NEXT: [[BIN_RDX2:%.*]] = xor <8 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
282 ; CHECK-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <8 x i32> [[BIN_RDX2]], <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
283 ; CHECK-NEXT: [[BIN_RDX4:%.*]] = xor <8 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
284 ; CHECK-NEXT: [[TMP2:%.*]] = extractelement <8 x i32> [[BIN_RDX4]], i32 0
285 ; CHECK-NEXT: [[MUL_714:%.*]] = xor i32 undef, [[MUL_613]]
286 ; CHECK-NEXT: ret i32 [[TMP2]]
289 %0 = load i32, i32* %p, align 4
290 %arrayidx.1 = getelementptr inbounds i32, i32* %p, i64 1
291 %1 = load i32, i32* %arrayidx.1, align 4
292 %mul.18 = xor i32 %1, %0
293 %arrayidx.2 = getelementptr inbounds i32, i32* %p, i64 2
294 %2 = load i32, i32* %arrayidx.2, align 4
295 %mul.29 = xor i32 %2, %mul.18
296 %arrayidx.3 = getelementptr inbounds i32, i32* %p, i64 3
297 %3 = load i32, i32* %arrayidx.3, align 4
298 %mul.310 = xor i32 %3, %mul.29
299 %arrayidx.4 = getelementptr inbounds i32, i32* %p, i64 4
300 %4 = load i32, i32* %arrayidx.4, align 4
301 %mul.411 = xor i32 %4, %mul.310
302 %arrayidx.5 = getelementptr inbounds i32, i32* %p, i64 5
303 %5 = load i32, i32* %arrayidx.5, align 4
304 %mul.512 = xor i32 %5, %mul.411
305 %arrayidx.6 = getelementptr inbounds i32, i32* %p, i64 6
306 %6 = load i32, i32* %arrayidx.6, align 4
307 %mul.613 = xor i32 %6, %mul.512
308 %arrayidx.7 = getelementptr inbounds i32, i32* %p, i64 7
309 %7 = load i32, i32* %arrayidx.7, align 4
310 %mul.714 = xor i32 %7, %mul.613
314 define i32 @PR37731(<4 x i32>* noalias nocapture dereferenceable(16) %self) unnamed_addr #0 {
315 ; CHECK-LABEL: @PR37731(
317 ; CHECK-NEXT: [[TMP0:%.*]] = load <4 x i32>, <4 x i32>* [[SELF:%.*]], align 16
318 ; CHECK-NEXT: [[TMP1:%.*]] = shl <4 x i32> [[TMP0]], <i32 6, i32 2, i32 13, i32 3>
319 ; CHECK-NEXT: [[TMP2:%.*]] = xor <4 x i32> [[TMP1]], [[TMP0]]
320 ; CHECK-NEXT: [[TMP3:%.*]] = lshr <4 x i32> [[TMP2]], <i32 13, i32 27, i32 21, i32 12>
321 ; CHECK-NEXT: [[TMP4:%.*]] = and <4 x i32> [[TMP0]], <i32 -2, i32 -8, i32 -16, i32 -128>
322 ; CHECK-NEXT: [[TMP5:%.*]] = shl <4 x i32> [[TMP4]], <i32 18, i32 2, i32 7, i32 13>
323 ; CHECK-NEXT: [[TMP6:%.*]] = xor <4 x i32> [[TMP3]], [[TMP5]]
324 ; CHECK-NEXT: store <4 x i32> [[TMP6]], <4 x i32>* [[SELF]], align 16
325 ; CHECK-NEXT: [[TMP7:%.*]] = xor i32 undef, undef
326 ; CHECK-NEXT: [[TMP8:%.*]] = xor i32 [[TMP7]], undef
327 ; CHECK-NEXT: [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[TMP6]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
328 ; CHECK-NEXT: [[BIN_RDX:%.*]] = xor <4 x i32> [[TMP6]], [[RDX_SHUF]]
329 ; CHECK-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[BIN_RDX]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
330 ; CHECK-NEXT: [[BIN_RDX2:%.*]] = xor <4 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
331 ; CHECK-NEXT: [[TMP9:%.*]] = extractelement <4 x i32> [[BIN_RDX2]], i32 0
332 ; CHECK-NEXT: [[TMP10:%.*]] = xor i32 [[TMP8]], undef
333 ; CHECK-NEXT: ret i32 [[TMP9]]
336 %0 = load <4 x i32>, <4 x i32>* %self, align 16
337 %1 = shl <4 x i32> %0, <i32 6, i32 2, i32 13, i32 3>
338 %2 = xor <4 x i32> %1, %0
339 %3 = lshr <4 x i32> %2, <i32 13, i32 27, i32 21, i32 12>
340 %4 = and <4 x i32> %0, <i32 -2, i32 -8, i32 -16, i32 -128>
341 %5 = shl <4 x i32> %4, <i32 18, i32 2, i32 7, i32 13>
342 %6 = xor <4 x i32> %3, %5
343 store <4 x i32> %6, <4 x i32>* %self, align 16
344 %7 = extractelement <4 x i32> %6, i32 0
345 %8 = extractelement <4 x i32> %6, i32 1
347 %10 = extractelement <4 x i32> %6, i32 2
348 %11 = xor i32 %9, %10
349 %12 = extractelement <4 x i32> %6, i32 3
350 %13 = xor i32 %11, %12