Debian Codegraph @

llvm/test/Transforms/VectorCombine/AArch64/vecreduce-shuffle.ll

raw
1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt -passes=vector-combine -S %s | FileCheck %s
3
4 target triple = "aarch64"
5
6 define i32 @reducebase_v4i32(<4 x i32> %a, <4 x i32> %b) {
7 ; CHECK-LABEL: @reducebase_v4i32(
8 ; CHECK-NEXT: [[X:%.*]] = xor <4 x i32> [[A:%.*]], [[B:%.*]]
9 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[X]])
10 ; CHECK-NEXT: ret i32 [[R]]
11 ;
12 %x = xor <4 x i32> %a, %b
13 %r = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %x)
14 ret i32 %r
15 }
16
17 define i32 @reduceshuffle_onein_v4i32(<4 x i32> %a) {
18 ; CHECK-LABEL: @reduceshuffle_onein_v4i32(
19 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[X:%.*]])
20 ; CHECK-NEXT: ret i32 [[R]]
21 ;
22 %x = shufflevector <4 x i32> %a, <4 x i32> undef, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
23 %r = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %x)
24 ret i32 %r
25 }
26
27 define i32 @reduceshuffle_onein_const_v4i32(<4 x i32> %a) {
28 ; CHECK-LABEL: @reduceshuffle_onein_const_v4i32(
29 ; CHECK-NEXT: [[X:%.*]] = xor <4 x i32> [[S:%.*]], splat (i32 -1)
30 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[X]])
31 ; CHECK-NEXT: ret i32 [[R]]
32 ;
33 %s = shufflevector <4 x i32> %a, <4 x i32> undef, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
34 %x = xor <4 x i32> %s, <i32 -1, i32 -1, i32 -1, i32 -1>
35 %r = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %x)
36 ret i32 %r
37 }
38
39 define i32 @reduceshuffle_onein_argin_v4i32(<4 x i32> %a, <4 x i32> %b) {
40 ; CHECK-LABEL: @reduceshuffle_onein_argin_v4i32(
41 ; CHECK-NEXT: [[S:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> undef, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
42 ; CHECK-NEXT: [[X:%.*]] = xor <4 x i32> [[S]], [[B:%.*]]
43 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[X]])
44 ; CHECK-NEXT: ret i32 [[R]]
45 ;
46 %s = shufflevector <4 x i32> %a, <4 x i32> undef, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
47 %x = xor <4 x i32> %s, %b
48 %r = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %x)
49 ret i32 %r
50 }
51
52 define i32 @reduceshuffle_onein_constnonsplat_v4i32(<4 x i32> %a) {
53 ; CHECK-LABEL: @reduceshuffle_onein_constnonsplat_v4i32(
54 ; CHECK-NEXT: [[S:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> undef, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
55 ; CHECK-NEXT: [[X:%.*]] = xor <4 x i32> [[S]], <i32 -1, i32 1, i32 2, i32 3>
56 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[X]])
57 ; CHECK-NEXT: ret i32 [[R]]
58 ;
59 %s = shufflevector <4 x i32> %a, <4 x i32> undef, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
60 %x = xor <4 x i32> %s, <i32 -1, i32 1, i32 2, i32 3>
61 %r = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %x)
62 ret i32 %r
63 }
64
65 define i32 @reduceshuffle_twoin_OK_v4i32(<4 x i32> %a, <4 x i32> %b) {
66 ; CHECK-LABEL: @reduceshuffle_twoin_OK_v4i32(
67 ; CHECK-NEXT: [[X:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
68 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[X]])
69 ; CHECK-NEXT: ret i32 [[R]]
70 ;
71 %x = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 1, i32 4, i32 5>
72 %r = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %x)
73 ret i32 %r
74 }
75
76 define i32 @reduceshuffle_twoin_concat_v4i32(<2 x i32> %a, <2 x i32> %b) {
77 ; CHECK-LABEL: @reduceshuffle_twoin_concat_v4i32(
78 ; CHECK-NEXT: [[X:%.*]] = shufflevector <2 x i32> [[A:%.*]], <2 x i32> [[B:%.*]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
79 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[X]])
80 ; CHECK-NEXT: ret i32 [[R]]
81 ;
82 %x = shufflevector <2 x i32> %a, <2 x i32> %b, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
83 %r = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %x)
84 ret i32 %r
85 }
86
87 define i32 @reduceshuffle_twoin_lowelts_v4i32(<4 x i32> %a, <4 x i32> %b) {
88 ; CHECK-LABEL: @reduceshuffle_twoin_lowelts_v4i32(
89 ; CHECK-NEXT: [[X:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
90 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[X]])
91 ; CHECK-NEXT: ret i32 [[R]]
92 ;
93 %x = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 5, i32 1, i32 4>
94 %r = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %x)
95 ret i32 %r
96 }
97
98 define i32 @reduceshuffle_twoin_notlowelts_v4i32(<4 x i32> %a, <4 x i32> %b) {
99 ; CHECK-LABEL: @reduceshuffle_twoin_notlowelts_v4i32(
100 ; CHECK-NEXT: [[X:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]], <4 x i32> <i32 0, i32 1, i32 4, i32 6>
101 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[X]])
102 ; CHECK-NEXT: ret i32 [[R]]
103 ;
104 %x = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 6, i32 1, i32 4>
105 %r = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %x)
106 ret i32 %r
107 }
108
109 define i32 @reduceshuffle_twoin_repeat_v4i32(<4 x i32> %a, <4 x i32> %b) {
110 ; CHECK-LABEL: @reduceshuffle_twoin_repeat_v4i32(
111 ; CHECK-NEXT: [[X:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]], <4 x i32> <i32 0, i32 4, i32 1, i32 4>
112 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[X]])
113 ; CHECK-NEXT: ret i32 [[R]]
114 ;
115 %x = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 4, i32 1, i32 4>
116 %r = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %x)
117 ret i32 %r
118 }
119
120 define i32 @reduceshuffle_twoin_uneven_v4i32(<4 x i32> %a, <4 x i32> %b) {
121 ; CHECK-LABEL: @reduceshuffle_twoin_uneven_v4i32(
122 ; CHECK-NEXT: [[X:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]], <4 x i32> <i32 0, i32 1, i32 2, i32 4>
123 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[X]])
124 ; CHECK-NEXT: ret i32 [[R]]
125 ;
126 %x = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 2, i32 1, i32 4>
127 %r = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %x)
128 ret i32 %r
129 }
130
131 define i32 @reduceshuffle_twoin_undef_v4i32(<4 x i32> %a, <4 x i32> %b) {
132 ; CHECK-LABEL: @reduceshuffle_twoin_undef_v4i32(
133 ; CHECK-NEXT: [[X:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]], <4 x i32> <i32 0, i32 poison, i32 1, i32 5>
134 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[X]])
135 ; CHECK-NEXT: ret i32 [[R]]
136 ;
137 %x = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 undef, i32 1, i32 5>
138 %r = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %x)
139 ret i32 %r
140 }
141
142 define i32 @reduceshuffle_twoin_undef2_v4i32(<4 x i32> %a, <4 x i32> %b) {
143 ; CHECK-LABEL: @reduceshuffle_twoin_undef2_v4i32(
144 ; CHECK-NEXT: [[X:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]], <4 x i32> <i32 0, i32 1, i32 4, i32 poison>
145 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[X]])
146 ; CHECK-NEXT: ret i32 [[R]]
147 ;
148 %x = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 4, i32 undef, i32 1, i32 0>
149 %r = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %x)
150 ret i32 %r
151 }
152
153 define i32 @reduceshuffle_twoin_multiundef_v4i32(<4 x i32> %a, <4 x i32> %b) {
154 ; CHECK-LABEL: @reduceshuffle_twoin_multiundef_v4i32(
155 ; CHECK-NEXT: [[X:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]], <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
156 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[X]])
157 ; CHECK-NEXT: ret i32 [[R]]
158 ;
159 %x = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 undef, i32 undef, i32 1>
160 %r = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %x)
161 ret i32 %r
162 }
163
164 define i32 @reduceshuffle_twoin_extrashuffleuse_v4i32(<4 x i32> %a, <4 x i32> %b) {
165 ; CHECK-LABEL: @reduceshuffle_twoin_extrashuffleuse_v4i32(
166 ; CHECK-NEXT: [[X:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]], <4 x i32> <i32 0, i32 5, i32 1, i32 4>
167 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[X]])
168 ; CHECK-NEXT: call void @use(<4 x i32> [[X]])
169 ; CHECK-NEXT: ret i32 [[R]]
170 ;
171 %x = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 5, i32 1, i32 4>
172 %r = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %x)
173 call void @use(<4 x i32> %x)
174 ret i32 %r
175 }
176
177 define i32 @reduceshuffle_twoin_extraotheruse_v4i32(<4 x i32> %a, <4 x i32> %b) {
178 ; CHECK-LABEL: @reduceshuffle_twoin_extraotheruse_v4i32(
179 ; CHECK-NEXT: [[S:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]], <4 x i32> <i32 0, i32 5, i32 1, i32 4>
180 ; CHECK-NEXT: [[X:%.*]] = xor <4 x i32> [[S]], splat (i32 -1)
181 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[X]])
182 ; CHECK-NEXT: call void @use(<4 x i32> [[X]])
183 ; CHECK-NEXT: ret i32 [[R]]
184 ;
185 %s = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 5, i32 1, i32 4>
186 %x = xor <4 x i32> %s, <i32 -1, i32 -1, i32 -1, i32 -1>
187 %r = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %x)
188 call void @use(<4 x i32> %x)
189 ret i32 %r
190 }
191
192 define i32 @reduceshuffle_twoin_splat_v4i32(<4 x i32> %a, <4 x i32> %b, i32 %c) {
193 ; CHECK-LABEL: @reduceshuffle_twoin_splat_v4i32(
194 ; CHECK-NEXT: [[S:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
195 ; CHECK-NEXT: [[INSERT:%.*]] = insertelement <4 x i32> poison, i32 [[C:%.*]], i32 0
196 ; CHECK-NEXT: [[SPLAT:%.*]] = shufflevector <4 x i32> [[INSERT]], <4 x i32> poison, <4 x i32> zeroinitializer
197 ; CHECK-NEXT: [[X:%.*]] = xor <4 x i32> [[S]], [[SPLAT]]
198 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[X]])
199 ; CHECK-NEXT: ret i32 [[R]]
200 ;
201 %s = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 5, i32 1, i32 4>
202 %insert = insertelement <4 x i32> poison, i32 %c, i32 0
203 %splat = shufflevector <4 x i32> %insert, <4 x i32> poison, <4 x i32> zeroinitializer
204 %x = xor <4 x i32> %s, %splat
205 %r = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %x)
206 ret i32 %r
207 }
208
209
210 define i32 @reducebase_v16i32(<16 x i32> %a, <16 x i32> %b) {
211 ; CHECK-LABEL: @reducebase_v16i32(
212 ; CHECK-NEXT: [[X:%.*]] = xor <16 x i32> [[A:%.*]], [[B:%.*]]
213 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[X]])
214 ; CHECK-NEXT: ret i32 [[R]]
215 ;
216 %x = xor <16 x i32> %a, %b
217 %r = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> %x)
218 ret i32 %r
219 }
220
221 define i32 @reduceshuffle_onein_v16i32(<16 x i32> %a) {
222 ; CHECK-LABEL: @reduceshuffle_onein_v16i32(
223 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[X:%.*]])
224 ; CHECK-NEXT: ret i32 [[R]]
225 ;
226 %x = shufflevector <16 x i32> %a, <16 x i32> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
227 %r = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> %x)
228 ret i32 %r
229 }
230
231 define i32 @reduceshuffle_onein_ext_v16i32(<16 x i32> %a) {
232 ; CHECK-LABEL: @reduceshuffle_onein_ext_v16i32(
233 ; CHECK-NEXT: [[X:%.*]] = xor <16 x i32> [[S:%.*]], splat (i32 -1)
234 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[X]])
235 ; CHECK-NEXT: ret i32 [[R]]
236 ;
237 %s = shufflevector <16 x i32> %a, <16 x i32> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
238 %x = xor <16 x i32> %s, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
239 %r = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> %x)
240 ret i32 %r
241 }
242
243 define i32 @reduceshuffle_twoin_concat_v16i32(<8 x i32> %a, <8 x i32> %b) {
244 ; CHECK-LABEL: @reduceshuffle_twoin_concat_v16i32(
245 ; CHECK-NEXT: [[S:%.*]] = shufflevector <8 x i32> [[A:%.*]], <8 x i32> [[B:%.*]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
246 ; CHECK-NEXT: [[X:%.*]] = xor <16 x i32> [[S]], splat (i32 -1)
247 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[X]])
248 ; CHECK-NEXT: ret i32 [[R]]
249 ;
250 %s = shufflevector <8 x i32> %a, <8 x i32> %b, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
251 %x = xor <16 x i32> %s, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
252 %r = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> %x)
253 ret i32 %r
254 }
255
256 define i32 @reduceshuffle_twoin_lowelt_v16i32(<16 x i32> %a, <16 x i32> %b) {
257 ; CHECK-LABEL: @reduceshuffle_twoin_lowelt_v16i32(
258 ; CHECK-NEXT: [[S:%.*]] = shufflevector <16 x i32> [[A:%.*]], <16 x i32> [[B:%.*]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
259 ; CHECK-NEXT: [[X:%.*]] = xor <16 x i32> [[S]], splat (i32 -1)
260 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[X]])
261 ; CHECK-NEXT: ret i32 [[R]]
262 ;
263 %s = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23>
264 %x = xor <16 x i32> %s, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
265 %r = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> %x)
266 ret i32 %r
267 }
268
269 define i32 @reduceshuffle_twoin_notlowelt_v16i32(<16 x i32> %a, <16 x i32> %b) {
270 ; CHECK-LABEL: @reduceshuffle_twoin_notlowelt_v16i32(
271 ; CHECK-NEXT: [[S:%.*]] = shufflevector <16 x i32> [[A:%.*]], <16 x i32> [[B:%.*]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 21, i32 22, i32 7, i32 24, i32 25, i32 10, i32 27, i32 28, i32 13, i32 30, i32 31>
272 ; CHECK-NEXT: [[X:%.*]] = xor <16 x i32> [[S]], splat (i32 -1)
273 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[X]])
274 ; CHECK-NEXT: ret i32 [[R]]
275 ;
276 %s = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 21, i32 22, i32 7, i32 24, i32 25, i32 10, i32 27, i32 28, i32 13, i32 30, i32 31>
277 %x = xor <16 x i32> %s, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
278 %r = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> %x)
279 ret i32 %r
280 }
281
282 define i32 @reduceshuffle_twoin_uneven_v16i32(<16 x i32> %a, <16 x i32> %b) {
283 ; CHECK-LABEL: @reduceshuffle_twoin_uneven_v16i32(
284 ; CHECK-NEXT: [[S:%.*]] = shufflevector <16 x i32> [[A:%.*]], <16 x i32> [[B:%.*]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22>
285 ; CHECK-NEXT: [[X:%.*]] = xor <16 x i32> [[S]], splat (i32 -1)
286 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[X]])
287 ; CHECK-NEXT: ret i32 [[R]]
288 ;
289 %s = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 8>
290 %x = xor <16 x i32> %s, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
291 %r = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> %x)
292 ret i32 %r
293 }
294
295 define i32 @reduceshuffle_twoin_shr1_v16i32(<16 x i32> %a, <16 x i32> %b) {
296 ; CHECK-LABEL: @reduceshuffle_twoin_shr1_v16i32(
297 ; CHECK-NEXT: [[S:%.*]] = shufflevector <16 x i32> [[A:%.*]], <16 x i32> [[B:%.*]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
298 ; CHECK-NEXT: [[A1:%.*]] = lshr <16 x i32> [[S]], splat (i32 15)
299 ; CHECK-NEXT: [[A2:%.*]] = and <16 x i32> [[A1]], splat (i32 65537)
300 ; CHECK-NEXT: [[A3:%.*]] = mul nuw <16 x i32> [[A2]], splat (i32 65535)
301 ; CHECK-NEXT: [[A4:%.*]] = add <16 x i32> [[A3]], [[S]]
302 ; CHECK-NEXT: [[A5:%.*]] = xor <16 x i32> [[A4]], [[A3]]
303 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[A5]])
304 ; CHECK-NEXT: ret i32 [[R]]
305 ;
306 %s = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 16, i32 17, i32 5, i32 18, i32 19, i32 6, i32 20, i32 21, i32 7, i32 22, i32 23>
307 %a1 = lshr <16 x i32> %s, <i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
308 %a2 = and <16 x i32> %a1, <i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537>
309 %a3 = mul nuw <16 x i32> %a2, <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
310 %a4 = add <16 x i32> %a3, %s
311 %a5 = xor <16 x i32> %a4, %a3
312 %r = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> %a5)
313 ret i32 %r
314 }
315
316 define i32 @reduceshuffle_twoin_shr2_v16i32(<16 x i32> %a, <16 x i32> %b) {
317 ; CHECK-LABEL: @reduceshuffle_twoin_shr2_v16i32(
318 ; CHECK-NEXT: [[S:%.*]] = shufflevector <16 x i32> [[A:%.*]], <16 x i32> [[B:%.*]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
319 ; CHECK-NEXT: [[A1:%.*]] = lshr <16 x i32> splat (i32 15), [[S]]
320 ; CHECK-NEXT: [[A2:%.*]] = and <16 x i32> [[A1]], splat (i32 65537)
321 ; CHECK-NEXT: [[A3:%.*]] = mul nuw <16 x i32> [[A2]], splat (i32 65535)
322 ; CHECK-NEXT: [[A4:%.*]] = add <16 x i32> [[A3]], [[S]]
323 ; CHECK-NEXT: [[A5:%.*]] = xor <16 x i32> [[A4]], [[A3]]
324 ; CHECK-NEXT: [[R:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[A5]])
325 ; CHECK-NEXT: ret i32 [[R]]
326 ;
327 %s = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 16, i32 17, i32 5, i32 18, i32 19, i32 6, i32 20, i32 21, i32 7, i32 22, i32 23>
328 %a1 = lshr <16 x i32> <i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>, %s
329 %a2 = and <16 x i32> %a1, <i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537>
330 %a3 = mul nuw <16 x i32> %a2, <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
331 %a4 = add <16 x i32> %a3, %s
332 %a5 = xor <16 x i32> %a4, %a3
333 %r = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> %a5)
334 ret i32 %r
335 }
336
337 ; i16
338
339 define i16 @reducebase_v16i16(<16 x i16> %a, <16 x i16> %b) {
340 ; CHECK-LABEL: @reducebase_v16i16(
341 ; CHECK-NEXT: [[X:%.*]] = xor <16 x i16> [[A:%.*]], [[B:%.*]]
342 ; CHECK-NEXT: [[R:%.*]] = call i16 @llvm.vector.reduce.add.v16i16(<16 x i16> [[X]])
343 ; CHECK-NEXT: ret i16 [[R]]
344 ;
345 %x = xor <16 x i16> %a, %b
346 %r = call i16 @llvm.vector.reduce.add.v16i16(<16 x i16> %x)
347 ret i16 %r
348 }
349
350 define i16 @reduceshuffle_onein_v16i16(<16 x i16> %a) {
351 ; CHECK-LABEL: @reduceshuffle_onein_v16i16(
352 ; CHECK-NEXT: [[R:%.*]] = call i16 @llvm.vector.reduce.add.v16i16(<16 x i16> [[X:%.*]])
353 ; CHECK-NEXT: ret i16 [[R]]
354 ;
355 %x = shufflevector <16 x i16> %a, <16 x i16> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
356 %r = call i16 @llvm.vector.reduce.add.v16i16(<16 x i16> %x)
357 ret i16 %r
358 }
359
360 define i16 @reduceshuffle_onein_ext_v16i16(<16 x i16> %a) {
361 ; CHECK-LABEL: @reduceshuffle_onein_ext_v16i16(
362 ; CHECK-NEXT: [[X:%.*]] = xor <16 x i16> [[S:%.*]], splat (i16 -1)
363 ; CHECK-NEXT: [[R:%.*]] = call i16 @llvm.vector.reduce.add.v16i16(<16 x i16> [[X]])
364 ; CHECK-NEXT: ret i16 [[R]]
365 ;
366 %s = shufflevector <16 x i16> %a, <16 x i16> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
367 %x = xor <16 x i16> %s, <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>
368 %r = call i16 @llvm.vector.reduce.add.v16i16(<16 x i16> %x)
369 ret i16 %r
370 }
371
372 define i16 @reduceshuffle_twoin_concat_v16i16(<8 x i16> %a, <8 x i16> %b) {
373 ; CHECK-LABEL: @reduceshuffle_twoin_concat_v16i16(
374 ; CHECK-NEXT: [[S:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> [[B:%.*]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
375 ; CHECK-NEXT: [[X:%.*]] = xor <16 x i16> [[S]], splat (i16 -1)
376 ; CHECK-NEXT: [[R:%.*]] = call i16 @llvm.vector.reduce.add.v16i16(<16 x i16> [[X]])
377 ; CHECK-NEXT: ret i16 [[R]]
378 ;
379 %s = shufflevector <8 x i16> %a, <8 x i16> %b, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
380 %x = xor <16 x i16> %s, <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>
381 %r = call i16 @llvm.vector.reduce.add.v16i16(<16 x i16> %x)
382 ret i16 %r
383 }
384
385 define i16 @reduceshuffle_twoin_lowelt_v16i16(<16 x i16> %a, <16 x i16> %b) {
386 ; CHECK-LABEL: @reduceshuffle_twoin_lowelt_v16i16(
387 ; CHECK-NEXT: [[S:%.*]] = shufflevector <16 x i16> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
388 ; CHECK-NEXT: [[X:%.*]] = xor <16 x i16> [[S]], splat (i16 -1)
389 ; CHECK-NEXT: [[R:%.*]] = call i16 @llvm.vector.reduce.add.v16i16(<16 x i16> [[X]])
390 ; CHECK-NEXT: ret i16 [[R]]
391 ;
392 %s = shufflevector <16 x i16> %a, <16 x i16> %b, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23>
393 %x = xor <16 x i16> %s, <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>
394 %r = call i16 @llvm.vector.reduce.add.v16i16(<16 x i16> %x)
395 ret i16 %r
396 }
397
398 define i16 @reduceshuffle_twoin_notlowelt_v16i16(<16 x i16> %a, <16 x i16> %b) {
399 ; CHECK-LABEL: @reduceshuffle_twoin_notlowelt_v16i16(
400 ; CHECK-NEXT: [[S:%.*]] = shufflevector <16 x i16> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 21, i32 22, i32 7, i32 24, i32 25, i32 10, i32 27, i32 28, i32 13, i32 30, i32 31>
401 ; CHECK-NEXT: [[X:%.*]] = xor <16 x i16> [[S]], splat (i16 -1)
402 ; CHECK-NEXT: [[R:%.*]] = call i16 @llvm.vector.reduce.add.v16i16(<16 x i16> [[X]])
403 ; CHECK-NEXT: ret i16 [[R]]
404 ;
405 %s = shufflevector <16 x i16> %a, <16 x i16> %b, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 21, i32 22, i32 7, i32 24, i32 25, i32 10, i32 27, i32 28, i32 13, i32 30, i32 31>
406 %x = xor <16 x i16> %s, <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>
407 %r = call i16 @llvm.vector.reduce.add.v16i16(<16 x i16> %x)
408 ret i16 %r
409 }
410
411 define i16 @reduceshuffle_twoin_uneven_v16i16(<16 x i16> %a, <16 x i16> %b) {
412 ; CHECK-LABEL: @reduceshuffle_twoin_uneven_v16i16(
413 ; CHECK-NEXT: [[S:%.*]] = shufflevector <16 x i16> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 8>
414 ; CHECK-NEXT: [[X:%.*]] = xor <16 x i16> [[S]], splat (i16 -1)
415 ; CHECK-NEXT: [[R:%.*]] = call i16 @llvm.vector.reduce.add.v16i16(<16 x i16> [[X]])
416 ; CHECK-NEXT: ret i16 [[R]]
417 ;
418 %s = shufflevector <16 x i16> %a, <16 x i16> %b, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 8>
419 %x = xor <16 x i16> %s, <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>
420 %r = call i16 @llvm.vector.reduce.add.v16i16(<16 x i16> %x)
421 ret i16 %r
422 }
423
424 define i16 @reduceshuffle_twoin_ext_v16i16(<16 x i16> %a, <16 x i16> %b) {
425 ; CHECK-LABEL: @reduceshuffle_twoin_ext_v16i16(
426 ; CHECK-NEXT: [[S:%.*]] = shufflevector <16 x i16> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
427 ; CHECK-NEXT: [[A1:%.*]] = lshr <16 x i16> [[S]], splat (i16 7)
428 ; CHECK-NEXT: [[A2:%.*]] = and <16 x i16> [[A1]], splat (i16 257)
429 ; CHECK-NEXT: [[A3:%.*]] = mul nuw <16 x i16> [[A2]], splat (i16 255)
430 ; CHECK-NEXT: [[A4:%.*]] = add <16 x i16> [[A3]], [[S]]
431 ; CHECK-NEXT: [[A5:%.*]] = xor <16 x i16> [[A4]], [[A3]]
432 ; CHECK-NEXT: [[R:%.*]] = call i16 @llvm.vector.reduce.add.v16i16(<16 x i16> [[A5]])
433 ; CHECK-NEXT: ret i16 [[R]]
434 ;
435 %s = shufflevector <16 x i16> %a, <16 x i16> %b, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 16, i32 17, i32 5, i32 18, i32 19, i32 6, i32 20, i32 21, i32 7, i32 22, i32 23>
436 %a1 = lshr <16 x i16> %s, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
437 %a2 = and <16 x i16> %a1, <i16 257, i16 257, i16 257, i16 257, i16 257, i16 257, i16 257, i16 257, i16 257, i16 257, i16 257, i16 257, i16 257, i16 257, i16 257, i16 257>
438 %a3 = mul nuw <16 x i16> %a2, <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
439 %a4 = add <16 x i16> %a3, %s
440 %a5 = xor <16 x i16> %a4, %a3
441 %r = call i16 @llvm.vector.reduce.add.v16i16(<16 x i16> %a5)
442 ret i16 %r
443 }
444
445 declare void @use(<4 x i32>)
446 declare i32 @llvm.vector.reduce.add.v4i32(<4 x i32>)
447 declare i32 @llvm.vector.reduce.add.v16i32(<16 x i32>)
448 declare i16 @llvm.vector.reduce.add.v16i16(<16 x i16>)