| blob | ef5b8665e16b5f5fc33d0df718f0112af455f82f |
1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -x86-experimental-vector-widening-legalization -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
3 ; RUN: llc < %s -x86-experimental-vector-widening-legalization -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2 --check-prefix=AVX2-SLOW
4 ; RUN: llc < %s -x86-experimental-vector-widening-legalization -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefix=AVX --check-prefix=AVX2 --check-prefix=AVX2-FAST
5 ; RUN: llc < %s -x86-experimental-vector-widening-legalization -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512F
6 ; RUN: llc < %s -x86-experimental-vector-widening-legalization -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
7 ; RUN: llc < %s -x86-experimental-vector-widening-legalization -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
8 ; RUN: llc < %s -x86-experimental-vector-widening-legalization -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
9 ; RUN: llc < %s -x86-experimental-vector-widening-legalization -mtriple=x86_64-unknown-unknown -mattr=+avx512vbmi,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VBMIVL
11 ; PR31551
12 ; Pairs of shufflevector:trunc functions with functional equivalence.
13 ; Ideally, the shuffles should be lowered to code with the same quality as the truncates.
15 define void @shuffle_v32i8_to_v16i8(<32 x i8>* %L, <16 x i8>* %S) nounwind {
16 ; AVX-LABEL: shuffle_v32i8_to_v16i8:
17 ; AVX: # %bb.0:
18 ; AVX-NEXT: vmovdqa (%rdi), %xmm0
19 ; AVX-NEXT: vmovdqa 16(%rdi), %xmm1
20 ; AVX-NEXT: vmovdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
21 ; AVX-NEXT: vpshufb %xmm2, %xmm1, %xmm1
22 ; AVX-NEXT: vpshufb %xmm2, %xmm0, %xmm0
23 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
24 ; AVX-NEXT: vmovdqa %xmm0, (%rsi)
25 ; AVX-NEXT: retq
26 ;
27 ; AVX512F-LABEL: shuffle_v32i8_to_v16i8:
28 ; AVX512F: # %bb.0:
29 ; AVX512F-NEXT: vmovdqa (%rdi), %xmm0
30 ; AVX512F-NEXT: vmovdqa 16(%rdi), %xmm1
31 ; AVX512F-NEXT: vmovdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
32 ; AVX512F-NEXT: vpshufb %xmm2, %xmm1, %xmm1
33 ; AVX512F-NEXT: vpshufb %xmm2, %xmm0, %xmm0
34 ; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
35 ; AVX512F-NEXT: vmovdqa %xmm0, (%rsi)
36 ; AVX512F-NEXT: retq
37 ;
38 ; AVX512VL-LABEL: shuffle_v32i8_to_v16i8:
39 ; AVX512VL: # %bb.0:
40 ; AVX512VL-NEXT: vmovdqa (%rdi), %xmm0
41 ; AVX512VL-NEXT: vmovdqa 16(%rdi), %xmm1
42 ; AVX512VL-NEXT: vmovdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
43 ; AVX512VL-NEXT: vpshufb %xmm2, %xmm1, %xmm1
44 ; AVX512VL-NEXT: vpshufb %xmm2, %xmm0, %xmm0
45 ; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
46 ; AVX512VL-NEXT: vmovdqa %xmm0, (%rsi)
47 ; AVX512VL-NEXT: retq
48 ;
49 ; AVX512BW-LABEL: shuffle_v32i8_to_v16i8:
50 ; AVX512BW: # %bb.0:
51 ; AVX512BW-NEXT: vmovdqa (%rdi), %xmm0
52 ; AVX512BW-NEXT: vmovdqa 16(%rdi), %xmm1
53 ; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
54 ; AVX512BW-NEXT: vpshufb %xmm2, %xmm1, %xmm1
55 ; AVX512BW-NEXT: vpshufb %xmm2, %xmm0, %xmm0
56 ; AVX512BW-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
57 ; AVX512BW-NEXT: vmovdqa %xmm0, (%rsi)
58 ; AVX512BW-NEXT: retq
59 ;
60 ; AVX512BWVL-LABEL: shuffle_v32i8_to_v16i8:
61 ; AVX512BWVL: # %bb.0:
62 ; AVX512BWVL-NEXT: vmovdqa (%rdi), %xmm0
63 ; AVX512BWVL-NEXT: vmovdqa 16(%rdi), %xmm1
64 ; AVX512BWVL-NEXT: vmovdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
65 ; AVX512BWVL-NEXT: vpshufb %xmm2, %xmm1, %xmm1
66 ; AVX512BWVL-NEXT: vpshufb %xmm2, %xmm0, %xmm0
67 ; AVX512BWVL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
68 ; AVX512BWVL-NEXT: vmovdqa %xmm0, (%rsi)
69 ; AVX512BWVL-NEXT: retq
70 ;
71 ; AVX512VBMIVL-LABEL: shuffle_v32i8_to_v16i8:
72 ; AVX512VBMIVL: # %bb.0:
73 ; AVX512VBMIVL-NEXT: vmovdqa (%rdi), %xmm0
74 ; AVX512VBMIVL-NEXT: vmovdqa {{.*#+}} xmm1 = [0,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30]
75 ; AVX512VBMIVL-NEXT: vpermi2b 16(%rdi), %xmm0, %xmm1
76 ; AVX512VBMIVL-NEXT: vmovdqa %xmm1, (%rsi)
77 ; AVX512VBMIVL-NEXT: retq
78 %vec = load <32 x i8>, <32 x i8>* %L
79 %strided.vec = shufflevector <32 x i8> %vec, <32 x i8> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
80 store <16 x i8> %strided.vec, <16 x i8>* %S
81 ret void
82 }
84 define void @trunc_v16i16_to_v16i8(<32 x i8>* %L, <16 x i8>* %S) nounwind {
85 ; AVX1-LABEL: trunc_v16i16_to_v16i8:
86 ; AVX1: # %bb.0:
87 ; AVX1-NEXT: vmovaps (%rdi), %ymm0
88 ; AVX1-NEXT: vandps {{.*}}(%rip), %ymm0, %ymm0
89 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
90 ; AVX1-NEXT: vpackuswb %xmm1, %xmm0, %xmm0
91 ; AVX1-NEXT: vmovdqa %xmm0, (%rsi)
92 ; AVX1-NEXT: vzeroupper
93 ; AVX1-NEXT: retq
94 ;
95 ; AVX2-LABEL: trunc_v16i16_to_v16i8:
96 ; AVX2: # %bb.0:
97 ; AVX2-NEXT: vmovdqa (%rdi), %ymm0
98 ; AVX2-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0
99 ; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
100 ; AVX2-NEXT: vpackuswb %xmm1, %xmm0, %xmm0
101 ; AVX2-NEXT: vmovdqa %xmm0, (%rsi)
102 ; AVX2-NEXT: vzeroupper
103 ; AVX2-NEXT: retq
104 ;
105 ; AVX512F-LABEL: trunc_v16i16_to_v16i8:
106 ; AVX512F: # %bb.0:
107 ; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero,mem[8],zero,mem[9],zero,mem[10],zero,mem[11],zero,mem[12],zero,mem[13],zero,mem[14],zero,mem[15],zero
108 ; AVX512F-NEXT: vpmovdb %zmm0, (%rsi)
109 ; AVX512F-NEXT: vzeroupper
110 ; AVX512F-NEXT: retq
111 ;
112 ; AVX512VL-LABEL: trunc_v16i16_to_v16i8:
113 ; AVX512VL: # %bb.0:
114 ; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero,mem[8],zero,mem[9],zero,mem[10],zero,mem[11],zero,mem[12],zero,mem[13],zero,mem[14],zero,mem[15],zero
115 ; AVX512VL-NEXT: vpmovdb %zmm0, (%rsi)
116 ; AVX512VL-NEXT: vzeroupper
117 ; AVX512VL-NEXT: retq
118 ;
119 ; AVX512BW-LABEL: trunc_v16i16_to_v16i8:
120 ; AVX512BW: # %bb.0:
121 ; AVX512BW-NEXT: vmovdqa (%rdi), %ymm0
122 ; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0
123 ; AVX512BW-NEXT: vmovdqa %xmm0, (%rsi)
124 ; AVX512BW-NEXT: vzeroupper
125 ; AVX512BW-NEXT: retq
126 ;
127 ; AVX512BWVL-LABEL: trunc_v16i16_to_v16i8:
128 ; AVX512BWVL: # %bb.0:
129 ; AVX512BWVL-NEXT: vmovdqa (%rdi), %ymm0
130 ; AVX512BWVL-NEXT: vpmovwb %ymm0, (%rsi)
131 ; AVX512BWVL-NEXT: vzeroupper
132 ; AVX512BWVL-NEXT: retq
133 ;
134 ; AVX512VBMIVL-LABEL: trunc_v16i16_to_v16i8:
135 ; AVX512VBMIVL: # %bb.0:
136 ; AVX512VBMIVL-NEXT: vmovdqa (%rdi), %ymm0
137 ; AVX512VBMIVL-NEXT: vpmovwb %ymm0, (%rsi)
138 ; AVX512VBMIVL-NEXT: vzeroupper
139 ; AVX512VBMIVL-NEXT: retq
140 %vec = load <32 x i8>, <32 x i8>* %L
141 %bc = bitcast <32 x i8> %vec to <16 x i16>
142 %strided.vec = trunc <16 x i16> %bc to <16 x i8>
143 store <16 x i8> %strided.vec, <16 x i8>* %S
144 ret void
145 }
147 define void @shuffle_v16i16_to_v8i16(<16 x i16>* %L, <8 x i16>* %S) nounwind {
148 ; AVX-LABEL: shuffle_v16i16_to_v8i16:
149 ; AVX: # %bb.0:
150 ; AVX-NEXT: vmovdqa (%rdi), %xmm0
151 ; AVX-NEXT: vmovdqa 16(%rdi), %xmm1
152 ; AVX-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
153 ; AVX-NEXT: vpshufb %xmm2, %xmm1, %xmm1
154 ; AVX-NEXT: vpshufb %xmm2, %xmm0, %xmm0
155 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
156 ; AVX-NEXT: vmovdqa %xmm0, (%rsi)
157 ; AVX-NEXT: retq
158 ;
159 ; AVX512F-LABEL: shuffle_v16i16_to_v8i16:
160 ; AVX512F: # %bb.0:
161 ; AVX512F-NEXT: vmovdqa (%rdi), %xmm0
162 ; AVX512F-NEXT: vmovdqa 16(%rdi), %xmm1
163 ; AVX512F-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
164 ; AVX512F-NEXT: vpshufb %xmm2, %xmm1, %xmm1
165 ; AVX512F-NEXT: vpshufb %xmm2, %xmm0, %xmm0
166 ; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
167 ; AVX512F-NEXT: vmovdqa %xmm0, (%rsi)
168 ; AVX512F-NEXT: retq
169 ;
170 ; AVX512VL-LABEL: shuffle_v16i16_to_v8i16:
171 ; AVX512VL: # %bb.0:
172 ; AVX512VL-NEXT: vmovdqa (%rdi), %xmm0
173 ; AVX512VL-NEXT: vmovdqa 16(%rdi), %xmm1
174 ; AVX512VL-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
175 ; AVX512VL-NEXT: vpshufb %xmm2, %xmm1, %xmm1
176 ; AVX512VL-NEXT: vpshufb %xmm2, %xmm0, %xmm0
177 ; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
178 ; AVX512VL-NEXT: vmovdqa %xmm0, (%rsi)
179 ; AVX512VL-NEXT: retq
180 ;
181 ; AVX512BW-LABEL: shuffle_v16i16_to_v8i16:
182 ; AVX512BW: # %bb.0:
183 ; AVX512BW-NEXT: vmovdqa (%rdi), %xmm0
184 ; AVX512BW-NEXT: vmovdqa 16(%rdi), %xmm1
185 ; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
186 ; AVX512BW-NEXT: vpshufb %xmm2, %xmm1, %xmm1
187 ; AVX512BW-NEXT: vpshufb %xmm2, %xmm0, %xmm0
188 ; AVX512BW-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
189 ; AVX512BW-NEXT: vmovdqa %xmm0, (%rsi)
190 ; AVX512BW-NEXT: retq
191 ;
192 ; AVX512BWVL-LABEL: shuffle_v16i16_to_v8i16:
193 ; AVX512BWVL: # %bb.0:
194 ; AVX512BWVL-NEXT: vmovdqa (%rdi), %xmm0
195 ; AVX512BWVL-NEXT: vmovdqa {{.*#+}} xmm1 = [0,2,4,6,8,10,12,14]
196 ; AVX512BWVL-NEXT: vpermi2w 16(%rdi), %xmm0, %xmm1
197 ; AVX512BWVL-NEXT: vmovdqa %xmm1, (%rsi)
198 ; AVX512BWVL-NEXT: retq
199 ;
200 ; AVX512VBMIVL-LABEL: shuffle_v16i16_to_v8i16:
201 ; AVX512VBMIVL: # %bb.0:
202 ; AVX512VBMIVL-NEXT: vmovdqa (%rdi), %xmm0
203 ; AVX512VBMIVL-NEXT: vmovdqa {{.*#+}} xmm1 = [0,2,4,6,8,10,12,14]
204 ; AVX512VBMIVL-NEXT: vpermi2w 16(%rdi), %xmm0, %xmm1
205 ; AVX512VBMIVL-NEXT: vmovdqa %xmm1, (%rsi)
206 ; AVX512VBMIVL-NEXT: retq
207 %vec = load <16 x i16>, <16 x i16>* %L
208 %strided.vec = shufflevector <16 x i16> %vec, <16 x i16> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
209 store <8 x i16> %strided.vec, <8 x i16>* %S
210 ret void
211 }
213 define void @trunc_v8i32_to_v8i16(<16 x i16>* %L, <8 x i16>* %S) nounwind {
214 ; AVX1-LABEL: trunc_v8i32_to_v8i16:
215 ; AVX1: # %bb.0:
216 ; AVX1-NEXT: vmovdqa (%rdi), %xmm0
217 ; AVX1-NEXT: vmovdqa 16(%rdi), %xmm1
218 ; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
219 ; AVX1-NEXT: vpshufb %xmm2, %xmm1, %xmm1
220 ; AVX1-NEXT: vpshufb %xmm2, %xmm0, %xmm0
221 ; AVX1-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
222 ; AVX1-NEXT: vmovdqa %xmm0, (%rsi)
223 ; AVX1-NEXT: retq
224 ;
225 ; AVX2-LABEL: trunc_v8i32_to_v8i16:
226 ; AVX2: # %bb.0:
227 ; AVX2-NEXT: vmovdqa (%rdi), %ymm0
228 ; AVX2-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
229 ; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
230 ; AVX2-NEXT: vmovdqa %xmm0, (%rsi)
231 ; AVX2-NEXT: vzeroupper
232 ; AVX2-NEXT: retq
233 ;
234 ; AVX512F-LABEL: trunc_v8i32_to_v8i16:
235 ; AVX512F: # %bb.0:
236 ; AVX512F-NEXT: vmovdqa (%rdi), %ymm0
237 ; AVX512F-NEXT: vpmovdw %zmm0, %ymm0
238 ; AVX512F-NEXT: vmovdqa %xmm0, (%rsi)
239 ; AVX512F-NEXT: vzeroupper
240 ; AVX512F-NEXT: retq
241 ;
242 ; AVX512VL-LABEL: trunc_v8i32_to_v8i16:
243 ; AVX512VL: # %bb.0:
244 ; AVX512VL-NEXT: vmovdqa (%rdi), %ymm0
245 ; AVX512VL-NEXT: vpmovdw %ymm0, (%rsi)
246 ; AVX512VL-NEXT: vzeroupper
247 ; AVX512VL-NEXT: retq
248 ;
249 ; AVX512BW-LABEL: trunc_v8i32_to_v8i16:
250 ; AVX512BW: # %bb.0:
251 ; AVX512BW-NEXT: vmovdqa (%rdi), %ymm0
252 ; AVX512BW-NEXT: vpmovdw %zmm0, %ymm0
253 ; AVX512BW-NEXT: vmovdqa %xmm0, (%rsi)
254 ; AVX512BW-NEXT: vzeroupper
255 ; AVX512BW-NEXT: retq
256 ;
257 ; AVX512BWVL-LABEL: trunc_v8i32_to_v8i16:
258 ; AVX512BWVL: # %bb.0:
259 ; AVX512BWVL-NEXT: vmovdqa (%rdi), %ymm0
260 ; AVX512BWVL-NEXT: vpmovdw %ymm0, (%rsi)
261 ; AVX512BWVL-NEXT: vzeroupper
262 ; AVX512BWVL-NEXT: retq
263 ;
264 ; AVX512VBMIVL-LABEL: trunc_v8i32_to_v8i16:
265 ; AVX512VBMIVL: # %bb.0:
266 ; AVX512VBMIVL-NEXT: vmovdqa (%rdi), %ymm0
267 ; AVX512VBMIVL-NEXT: vpmovdw %ymm0, (%rsi)
268 ; AVX512VBMIVL-NEXT: vzeroupper
269 ; AVX512VBMIVL-NEXT: retq
270 %vec = load <16 x i16>, <16 x i16>* %L
271 %bc = bitcast <16 x i16> %vec to <8 x i32>
272 %strided.vec = trunc <8 x i32> %bc to <8 x i16>
273 store <8 x i16> %strided.vec, <8 x i16>* %S
274 ret void
275 }
277 define void @shuffle_v8i32_to_v4i32(<8 x i32>* %L, <4 x i32>* %S) nounwind {
278 ; AVX-LABEL: shuffle_v8i32_to_v4i32:
279 ; AVX: # %bb.0:
280 ; AVX-NEXT: vmovaps (%rdi), %xmm0
281 ; AVX-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],mem[0,2]
282 ; AVX-NEXT: vmovaps %xmm0, (%rsi)
283 ; AVX-NEXT: retq
284 ;
285 ; AVX512-LABEL: shuffle_v8i32_to_v4i32:
286 ; AVX512: # %bb.0:
287 ; AVX512-NEXT: vmovaps (%rdi), %xmm0
288 ; AVX512-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],mem[0,2]
289 ; AVX512-NEXT: vmovaps %xmm0, (%rsi)
290 ; AVX512-NEXT: retq
291 %vec = load <8 x i32>, <8 x i32>* %L
292 %strided.vec = shufflevector <8 x i32> %vec, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
293 store <4 x i32> %strided.vec, <4 x i32>* %S
294 ret void
295 }
297 define void @trunc_v4i64_to_v4i32(<8 x i32>* %L, <4 x i32>* %S) nounwind {
298 ; AVX1-LABEL: trunc_v4i64_to_v4i32:
299 ; AVX1: # %bb.0:
300 ; AVX1-NEXT: vmovaps (%rdi), %xmm0
301 ; AVX1-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],mem[0,2]
302 ; AVX1-NEXT: vmovaps %xmm0, (%rsi)
303 ; AVX1-NEXT: retq
304 ;
305 ; AVX2-SLOW-LABEL: trunc_v4i64_to_v4i32:
306 ; AVX2-SLOW: # %bb.0:
307 ; AVX2-SLOW-NEXT: vmovaps (%rdi), %xmm0
308 ; AVX2-SLOW-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],mem[0,2]
309 ; AVX2-SLOW-NEXT: vmovaps %xmm0, (%rsi)
310 ; AVX2-SLOW-NEXT: retq
311 ;
312 ; AVX2-FAST-LABEL: trunc_v4i64_to_v4i32:
313 ; AVX2-FAST: # %bb.0:
314 ; AVX2-FAST-NEXT: vmovaps {{.*#+}} ymm0 = [0,2,4,6,4,6,6,7]
315 ; AVX2-FAST-NEXT: vpermps (%rdi), %ymm0, %ymm0
316 ; AVX2-FAST-NEXT: vmovaps %xmm0, (%rsi)
317 ; AVX2-FAST-NEXT: vzeroupper
318 ; AVX2-FAST-NEXT: retq
319 ;
320 ; AVX512F-LABEL: trunc_v4i64_to_v4i32:
321 ; AVX512F: # %bb.0:
322 ; AVX512F-NEXT: vmovdqa (%rdi), %ymm0
323 ; AVX512F-NEXT: vpmovqd %zmm0, %ymm0
324 ; AVX512F-NEXT: vmovdqa %xmm0, (%rsi)
325 ; AVX512F-NEXT: vzeroupper
326 ; AVX512F-NEXT: retq
327 ;
328 ; AVX512VL-LABEL: trunc_v4i64_to_v4i32:
329 ; AVX512VL: # %bb.0:
330 ; AVX512VL-NEXT: vmovdqa (%rdi), %ymm0
331 ; AVX512VL-NEXT: vpmovqd %ymm0, (%rsi)
332 ; AVX512VL-NEXT: vzeroupper
333 ; AVX512VL-NEXT: retq
334 ;
335 ; AVX512BW-LABEL: trunc_v4i64_to_v4i32:
336 ; AVX512BW: # %bb.0:
337 ; AVX512BW-NEXT: vmovdqa (%rdi), %ymm0
338 ; AVX512BW-NEXT: vpmovqd %zmm0, %ymm0
339 ; AVX512BW-NEXT: vmovdqa %xmm0, (%rsi)
340 ; AVX512BW-NEXT: vzeroupper
341 ; AVX512BW-NEXT: retq
342 ;
343 ; AVX512BWVL-LABEL: trunc_v4i64_to_v4i32:
344 ; AVX512BWVL: # %bb.0:
345 ; AVX512BWVL-NEXT: vmovdqa (%rdi), %ymm0
346 ; AVX512BWVL-NEXT: vpmovqd %ymm0, (%rsi)
347 ; AVX512BWVL-NEXT: vzeroupper
348 ; AVX512BWVL-NEXT: retq
349 ;
350 ; AVX512VBMIVL-LABEL: trunc_v4i64_to_v4i32:
351 ; AVX512VBMIVL: # %bb.0:
352 ; AVX512VBMIVL-NEXT: vmovdqa (%rdi), %ymm0
353 ; AVX512VBMIVL-NEXT: vpmovqd %ymm0, (%rsi)
354 ; AVX512VBMIVL-NEXT: vzeroupper
355 ; AVX512VBMIVL-NEXT: retq
356 %vec = load <8 x i32>, <8 x i32>* %L
357 %bc = bitcast <8 x i32> %vec to <4 x i64>
358 %strided.vec = trunc <4 x i64> %bc to <4 x i32>
359 store <4 x i32> %strided.vec, <4 x i32>* %S
360 ret void
361 }
363 define void @shuffle_v32i8_to_v8i8(<32 x i8>* %L, <8 x i8>* %S) nounwind {
364 ; AVX-LABEL: shuffle_v32i8_to_v8i8:
365 ; AVX: # %bb.0:
366 ; AVX-NEXT: vmovdqa (%rdi), %xmm0
367 ; AVX-NEXT: vmovdqa 16(%rdi), %xmm1
368 ; AVX-NEXT: vmovdqa {{.*#+}} xmm2 = <0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u>
369 ; AVX-NEXT: vpshufb %xmm2, %xmm1, %xmm1
370 ; AVX-NEXT: vpshufb %xmm2, %xmm0, %xmm0
371 ; AVX-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
372 ; AVX-NEXT: vmovq %xmm0, (%rsi)
373 ; AVX-NEXT: retq
374 ;
375 ; AVX512F-LABEL: shuffle_v32i8_to_v8i8:
376 ; AVX512F: # %bb.0:
377 ; AVX512F-NEXT: vmovdqa (%rdi), %xmm0
378 ; AVX512F-NEXT: vmovdqa 16(%rdi), %xmm1
379 ; AVX512F-NEXT: vmovdqa {{.*#+}} xmm2 = <0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u>
380 ; AVX512F-NEXT: vpshufb %xmm2, %xmm1, %xmm1
381 ; AVX512F-NEXT: vpshufb %xmm2, %xmm0, %xmm0
382 ; AVX512F-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
383 ; AVX512F-NEXT: vmovq %xmm0, (%rsi)
384 ; AVX512F-NEXT: retq
385 ;
386 ; AVX512VL-LABEL: shuffle_v32i8_to_v8i8:
387 ; AVX512VL: # %bb.0:
388 ; AVX512VL-NEXT: vmovdqa (%rdi), %xmm0
389 ; AVX512VL-NEXT: vmovdqa 16(%rdi), %xmm1
390 ; AVX512VL-NEXT: vmovdqa {{.*#+}} xmm2 = <0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u>
391 ; AVX512VL-NEXT: vpshufb %xmm2, %xmm1, %xmm1
392 ; AVX512VL-NEXT: vpshufb %xmm2, %xmm0, %xmm0
393 ; AVX512VL-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
394 ; AVX512VL-NEXT: vmovq %xmm0, (%rsi)
395 ; AVX512VL-NEXT: retq
396 ;
397 ; AVX512BW-LABEL: shuffle_v32i8_to_v8i8:
398 ; AVX512BW: # %bb.0:
399 ; AVX512BW-NEXT: vmovdqa (%rdi), %xmm0
400 ; AVX512BW-NEXT: vmovdqa 16(%rdi), %xmm1
401 ; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm2 = <0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u>
402 ; AVX512BW-NEXT: vpshufb %xmm2, %xmm1, %xmm1
403 ; AVX512BW-NEXT: vpshufb %xmm2, %xmm0, %xmm0
404 ; AVX512BW-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
405 ; AVX512BW-NEXT: vmovq %xmm0, (%rsi)
406 ; AVX512BW-NEXT: retq
407 ;
408 ; AVX512BWVL-LABEL: shuffle_v32i8_to_v8i8:
409 ; AVX512BWVL: # %bb.0:
410 ; AVX512BWVL-NEXT: vmovdqa (%rdi), %xmm0
411 ; AVX512BWVL-NEXT: vmovdqa 16(%rdi), %xmm1
412 ; AVX512BWVL-NEXT: vmovdqa {{.*#+}} xmm2 = <0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u>
413 ; AVX512BWVL-NEXT: vpshufb %xmm2, %xmm1, %xmm1
414 ; AVX512BWVL-NEXT: vpshufb %xmm2, %xmm0, %xmm0
415 ; AVX512BWVL-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
416 ; AVX512BWVL-NEXT: vmovq %xmm0, (%rsi)
417 ; AVX512BWVL-NEXT: retq
418 ;
419 ; AVX512VBMIVL-LABEL: shuffle_v32i8_to_v8i8:
420 ; AVX512VBMIVL: # %bb.0:
421 ; AVX512VBMIVL-NEXT: vmovdqa (%rdi), %xmm0
422 ; AVX512VBMIVL-NEXT: vpbroadcastq {{.*#+}} xmm1 = [2024390091656922112,2024390091656922112]
423 ; AVX512VBMIVL-NEXT: vpermi2b 16(%rdi), %xmm0, %xmm1
424 ; AVX512VBMIVL-NEXT: vmovq %xmm1, (%rsi)
425 ; AVX512VBMIVL-NEXT: retq
426 %vec = load <32 x i8>, <32 x i8>* %L
427 %strided.vec = shufflevector <32 x i8> %vec, <32 x i8> undef, <8 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28>
428 store <8 x i8> %strided.vec, <8 x i8>* %S
429 ret void
430 }
432 define void @trunc_v8i32_to_v8i8(<32 x i8>* %L, <8 x i8>* %S) nounwind {
433 ; AVX-LABEL: trunc_v8i32_to_v8i8:
434 ; AVX: # %bb.0:
435 ; AVX-NEXT: vmovdqa (%rdi), %xmm0
436 ; AVX-NEXT: vmovdqa 16(%rdi), %xmm1
437 ; AVX-NEXT: vmovdqa {{.*#+}} xmm2 = <0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u>
438 ; AVX-NEXT: vpshufb %xmm2, %xmm1, %xmm1
439 ; AVX-NEXT: vpshufb %xmm2, %xmm0, %xmm0
440 ; AVX-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
441 ; AVX-NEXT: vmovq %xmm0, (%rsi)
442 ; AVX-NEXT: retq
443 ;
444 ; AVX512F-LABEL: trunc_v8i32_to_v8i8:
445 ; AVX512F: # %bb.0:
446 ; AVX512F-NEXT: vmovdqa (%rdi), %ymm0
447 ; AVX512F-NEXT: vpmovdb %zmm0, %xmm0
448 ; AVX512F-NEXT: vmovq %xmm0, (%rsi)
449 ; AVX512F-NEXT: vzeroupper
450 ; AVX512F-NEXT: retq
451 ;
452 ; AVX512VL-LABEL: trunc_v8i32_to_v8i8:
453 ; AVX512VL: # %bb.0:
454 ; AVX512VL-NEXT: vmovdqa (%rdi), %ymm0
455 ; AVX512VL-NEXT: vpmovdb %ymm0, (%rsi)
456 ; AVX512VL-NEXT: vzeroupper
457 ; AVX512VL-NEXT: retq
458 ;
459 ; AVX512BW-LABEL: trunc_v8i32_to_v8i8:
460 ; AVX512BW: # %bb.0:
461 ; AVX512BW-NEXT: vmovdqa (%rdi), %ymm0
462 ; AVX512BW-NEXT: vpmovdb %zmm0, %xmm0
463 ; AVX512BW-NEXT: vmovq %xmm0, (%rsi)
464 ; AVX512BW-NEXT: vzeroupper
465 ; AVX512BW-NEXT: retq
466 ;
467 ; AVX512BWVL-LABEL: trunc_v8i32_to_v8i8:
468 ; AVX512BWVL: # %bb.0:
469 ; AVX512BWVL-NEXT: vmovdqa (%rdi), %ymm0
470 ; AVX512BWVL-NEXT: vpmovdb %ymm0, (%rsi)
471 ; AVX512BWVL-NEXT: vzeroupper
472 ; AVX512BWVL-NEXT: retq
473 ;
474 ; AVX512VBMIVL-LABEL: trunc_v8i32_to_v8i8:
475 ; AVX512VBMIVL: # %bb.0:
476 ; AVX512VBMIVL-NEXT: vmovdqa (%rdi), %ymm0
477 ; AVX512VBMIVL-NEXT: vpmovdb %ymm0, (%rsi)
478 ; AVX512VBMIVL-NEXT: vzeroupper
479 ; AVX512VBMIVL-NEXT: retq
480 %vec = load <32 x i8>, <32 x i8>* %L
481 %bc = bitcast <32 x i8> %vec to <8 x i32>
482 %strided.vec = trunc <8 x i32> %bc to <8 x i8>
483 store <8 x i8> %strided.vec, <8 x i8>* %S
484 ret void
485 }
487 define <2 x i64> @trunc_v8i32_to_v8i8_return_v2i64(<8 x i32> %vec) nounwind {
488 ; IR generated from:
489 ; return (__m128i) {(long long)__builtin_convertvector((__v8si)__A, __v8qi), 0};
490 ; AVX1-LABEL: trunc_v8i32_to_v8i8_return_v2i64:
491 ; AVX1: # %bb.0:
492 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
493 ; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = <0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u>
494 ; AVX1-NEXT: vpshufb %xmm2, %xmm1, %xmm1
495 ; AVX1-NEXT: vpshufb %xmm2, %xmm0, %xmm0
496 ; AVX1-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
497 ; AVX1-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
498 ; AVX1-NEXT: vzeroupper
499 ; AVX1-NEXT: retq
500 ;
501 ; AVX2-LABEL: trunc_v8i32_to_v8i8_return_v2i64:
502 ; AVX2: # %bb.0:
503 ; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
504 ; AVX2-NEXT: vmovdqa {{.*#+}} xmm2 = <0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u>
505 ; AVX2-NEXT: vpshufb %xmm2, %xmm1, %xmm1
506 ; AVX2-NEXT: vpshufb %xmm2, %xmm0, %xmm0
507 ; AVX2-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
508 ; AVX2-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
509 ; AVX2-NEXT: vzeroupper
510 ; AVX2-NEXT: retq
511 ;
512 ; AVX512F-LABEL: trunc_v8i32_to_v8i8_return_v2i64:
513 ; AVX512F: # %bb.0:
514 ; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
515 ; AVX512F-NEXT: vpmovdb %zmm0, %xmm0
516 ; AVX512F-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
517 ; AVX512F-NEXT: vzeroupper
518 ; AVX512F-NEXT: retq
519 ;
520 ; AVX512VL-LABEL: trunc_v8i32_to_v8i8_return_v2i64:
521 ; AVX512VL: # %bb.0:
522 ; AVX512VL-NEXT: vpmovdb %ymm0, %xmm0
523 ; AVX512VL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
524 ; AVX512VL-NEXT: vzeroupper
525 ; AVX512VL-NEXT: retq
526 ;
527 ; AVX512BW-LABEL: trunc_v8i32_to_v8i8_return_v2i64:
528 ; AVX512BW: # %bb.0:
529 ; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
530 ; AVX512BW-NEXT: vpmovdb %zmm0, %xmm0
531 ; AVX512BW-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
532 ; AVX512BW-NEXT: vzeroupper
533 ; AVX512BW-NEXT: retq
534 ;
535 ; AVX512BWVL-LABEL: trunc_v8i32_to_v8i8_return_v2i64:
536 ; AVX512BWVL: # %bb.0:
537 ; AVX512BWVL-NEXT: vpmovdb %ymm0, %xmm0
538 ; AVX512BWVL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
539 ; AVX512BWVL-NEXT: vzeroupper
540 ; AVX512BWVL-NEXT: retq
541 ;
542 ; AVX512VBMIVL-LABEL: trunc_v8i32_to_v8i8_return_v2i64:
543 ; AVX512VBMIVL: # %bb.0:
544 ; AVX512VBMIVL-NEXT: vpmovdb %ymm0, %xmm0
545 ; AVX512VBMIVL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
546 ; AVX512VBMIVL-NEXT: vzeroupper
547 ; AVX512VBMIVL-NEXT: retq
548 %truncated.vec = trunc <8 x i32> %vec to <8 x i8>
549 %bc = bitcast <8 x i8> %truncated.vec to i64
550 %result = insertelement <2 x i64> zeroinitializer, i64 %bc, i32 0
551 ret <2 x i64> %result
552 }
554 define <16 x i8> @trunc_v8i32_to_v8i8_with_zext_return_v16i8(<8 x i32> %vec) nounwind {
555 ; AVX1-LABEL: trunc_v8i32_to_v8i8_with_zext_return_v16i8:
556 ; AVX1: # %bb.0:
557 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
558 ; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
559 ; AVX1-NEXT: vpshufb %xmm2, %xmm1, %xmm1
560 ; AVX1-NEXT: vpshufb %xmm2, %xmm0, %xmm0
561 ; AVX1-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
562 ; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
563 ; AVX1-NEXT: vzeroupper
564 ; AVX1-NEXT: retq
565 ;
566 ; AVX2-LABEL: trunc_v8i32_to_v8i8_with_zext_return_v16i8:
567 ; AVX2: # %bb.0:
568 ; AVX2-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
569 ; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
570 ; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
571 ; AVX2-NEXT: vzeroupper
572 ; AVX2-NEXT: retq
573 ;
574 ; AVX512F-LABEL: trunc_v8i32_to_v8i8_with_zext_return_v16i8:
575 ; AVX512F: # %bb.0:
576 ; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
577 ; AVX512F-NEXT: vpmovdw %zmm0, %ymm0
578 ; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
579 ; AVX512F-NEXT: vzeroupper
580 ; AVX512F-NEXT: retq
581 ;
582 ; AVX512VL-LABEL: trunc_v8i32_to_v8i8_with_zext_return_v16i8:
583 ; AVX512VL: # %bb.0:
584 ; AVX512VL-NEXT: vpmovdb %ymm0, %xmm0
585 ; AVX512VL-NEXT: vzeroupper
586 ; AVX512VL-NEXT: retq
587 ;
588 ; AVX512BW-LABEL: trunc_v8i32_to_v8i8_with_zext_return_v16i8:
589 ; AVX512BW: # %bb.0:
590 ; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
591 ; AVX512BW-NEXT: vpmovdw %zmm0, %ymm0
592 ; AVX512BW-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
593 ; AVX512BW-NEXT: vzeroupper
594 ; AVX512BW-NEXT: retq
595 ;
596 ; AVX512BWVL-LABEL: trunc_v8i32_to_v8i8_with_zext_return_v16i8:
597 ; AVX512BWVL: # %bb.0:
598 ; AVX512BWVL-NEXT: vpmovdb %ymm0, %xmm0
599 ; AVX512BWVL-NEXT: vzeroupper
600 ; AVX512BWVL-NEXT: retq
601 ;
602 ; AVX512VBMIVL-LABEL: trunc_v8i32_to_v8i8_with_zext_return_v16i8:
603 ; AVX512VBMIVL: # %bb.0:
604 ; AVX512VBMIVL-NEXT: vpmovdb %ymm0, %xmm0
605 ; AVX512VBMIVL-NEXT: vzeroupper
606 ; AVX512VBMIVL-NEXT: retq
607 %truncated = trunc <8 x i32> %vec to <8 x i8>
608 %truncated.ext = zext <8 x i8> %truncated to <8 x i16>
609 %bc = bitcast <8 x i16> %truncated.ext to <16 x i8>
610 %result = shufflevector <16 x i8> %bc, <16 x i8> zeroinitializer, <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>
611 ret <16 x i8> %result
612 }
614 define <16 x i8> @trunc_v8i32_to_v8i8_via_v8i16_return_v16i8(<8 x i32> %vec) nounwind {
615 ; AVX1-LABEL: trunc_v8i32_to_v8i8_via_v8i16_return_v16i8:
616 ; AVX1: # %bb.0:
617 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
618 ; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
619 ; AVX1-NEXT: vpshufb %xmm2, %xmm1, %xmm1
620 ; AVX1-NEXT: vpshufb %xmm2, %xmm0, %xmm0
621 ; AVX1-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
622 ; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
623 ; AVX1-NEXT: vzeroupper
624 ; AVX1-NEXT: retq
625 ;
626 ; AVX2-LABEL: trunc_v8i32_to_v8i8_via_v8i16_return_v16i8:
627 ; AVX2: # %bb.0:
628 ; AVX2-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
629 ; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
630 ; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
631 ; AVX2-NEXT: vzeroupper
632 ; AVX2-NEXT: retq
633 ;
634 ; AVX512F-LABEL: trunc_v8i32_to_v8i8_via_v8i16_return_v16i8:
635 ; AVX512F: # %bb.0:
636 ; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
637 ; AVX512F-NEXT: vpmovdw %zmm0, %ymm0
638 ; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
639 ; AVX512F-NEXT: vzeroupper
640 ; AVX512F-NEXT: retq
641 ;
642 ; AVX512VL-LABEL: trunc_v8i32_to_v8i8_via_v8i16_return_v16i8:
643 ; AVX512VL: # %bb.0:
644 ; AVX512VL-NEXT: vpmovdb %ymm0, %xmm0
645 ; AVX512VL-NEXT: vzeroupper
646 ; AVX512VL-NEXT: retq
647 ;
648 ; AVX512BW-LABEL: trunc_v8i32_to_v8i8_via_v8i16_return_v16i8:
649 ; AVX512BW: # %bb.0:
650 ; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
651 ; AVX512BW-NEXT: vpmovdw %zmm0, %ymm0
652 ; AVX512BW-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
653 ; AVX512BW-NEXT: vzeroupper
654 ; AVX512BW-NEXT: retq
655 ;
656 ; AVX512BWVL-LABEL: trunc_v8i32_to_v8i8_via_v8i16_return_v16i8:
657 ; AVX512BWVL: # %bb.0:
658 ; AVX512BWVL-NEXT: vpmovdb %ymm0, %xmm0
659 ; AVX512BWVL-NEXT: vzeroupper
660 ; AVX512BWVL-NEXT: retq
661 ;
662 ; AVX512VBMIVL-LABEL: trunc_v8i32_to_v8i8_via_v8i16_return_v16i8:
663 ; AVX512VBMIVL: # %bb.0:
664 ; AVX512VBMIVL-NEXT: vpmovdb %ymm0, %xmm0
665 ; AVX512VBMIVL-NEXT: vzeroupper
666 ; AVX512VBMIVL-NEXT: retq
667 %truncated = trunc <8 x i32> %vec to <8 x i16>
668 %bc = bitcast <8 x i16> %truncated to <16 x i8>
669 %result = shufflevector <16 x i8> %bc, <16 x i8> zeroinitializer, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 17, i32 20, i32 24, i32 22, i32 31, i32 28, i32 28, i32 29>
670 ret <16 x i8> %result
671 }
673 define <16 x i8> @trunc_v8i32_to_v8i8_return_v16i8(<8 x i32> %vec) nounwind {
674 ; AVX1-LABEL: trunc_v8i32_to_v8i8_return_v16i8:
675 ; AVX1: # %bb.0:
676 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
677 ; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = <0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u>
678 ; AVX1-NEXT: vpshufb %xmm2, %xmm1, %xmm1
679 ; AVX1-NEXT: vpshufb %xmm2, %xmm0, %xmm0
680 ; AVX1-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
681 ; AVX1-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
682 ; AVX1-NEXT: vzeroupper
683 ; AVX1-NEXT: retq
684 ;
685 ; AVX2-LABEL: trunc_v8i32_to_v8i8_return_v16i8:
686 ; AVX2: # %bb.0:
687 ; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
688 ; AVX2-NEXT: vmovdqa {{.*#+}} xmm2 = <0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u>
689 ; AVX2-NEXT: vpshufb %xmm2, %xmm1, %xmm1
690 ; AVX2-NEXT: vpshufb %xmm2, %xmm0, %xmm0
691 ; AVX2-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
692 ; AVX2-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
693 ; AVX2-NEXT: vzeroupper
694 ; AVX2-NEXT: retq
695 ;
696 ; AVX512F-LABEL: trunc_v8i32_to_v8i8_return_v16i8:
697 ; AVX512F: # %bb.0:
698 ; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
699 ; AVX512F-NEXT: vpmovdb %zmm0, %xmm0
700 ; AVX512F-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
701 ; AVX512F-NEXT: vzeroupper
702 ; AVX512F-NEXT: retq
703 ;
704 ; AVX512VL-LABEL: trunc_v8i32_to_v8i8_return_v16i8:
705 ; AVX512VL: # %bb.0:
706 ; AVX512VL-NEXT: vpmovdb %ymm0, %xmm0
707 ; AVX512VL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
708 ; AVX512VL-NEXT: vzeroupper
709 ; AVX512VL-NEXT: retq
710 ;
711 ; AVX512BW-LABEL: trunc_v8i32_to_v8i8_return_v16i8:
712 ; AVX512BW: # %bb.0:
713 ; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
714 ; AVX512BW-NEXT: vpmovdb %zmm0, %xmm0
715 ; AVX512BW-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
716 ; AVX512BW-NEXT: vzeroupper
717 ; AVX512BW-NEXT: retq
718 ;
719 ; AVX512BWVL-LABEL: trunc_v8i32_to_v8i8_return_v16i8:
720 ; AVX512BWVL: # %bb.0:
721 ; AVX512BWVL-NEXT: vpmovdb %ymm0, %xmm0
722 ; AVX512BWVL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
723 ; AVX512BWVL-NEXT: vzeroupper
724 ; AVX512BWVL-NEXT: retq
725 ;
726 ; AVX512VBMIVL-LABEL: trunc_v8i32_to_v8i8_return_v16i8:
727 ; AVX512VBMIVL: # %bb.0:
728 ; AVX512VBMIVL-NEXT: vpmovdb %ymm0, %xmm0
729 ; AVX512VBMIVL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
730 ; AVX512VBMIVL-NEXT: vzeroupper
731 ; AVX512VBMIVL-NEXT: retq
732 %truncated = trunc <8 x i32> %vec to <8 x i8>
733 %result = shufflevector <8 x i8> %truncated, <8 x i8> zeroinitializer, <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>
734 ret <16 x i8> %result
735 }
737 define <2 x i64> @trunc_v4i64_to_v4i16_return_v2i64(<4 x i64> %vec) nounwind {
738 ; IR generated from:
739 ; return (__m128i) {(long long)__builtin_convertvector((__v4di)x, __v4hi), 0};
740 ; AVX1-LABEL: trunc_v4i64_to_v4i16_return_v2i64:
741 ; AVX1: # %bb.0:
742 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
743 ; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
744 ; AVX1-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
745 ; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
746 ; AVX1-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
747 ; AVX1-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
748 ; AVX1-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
749 ; AVX1-NEXT: vzeroupper
750 ; AVX1-NEXT: retq
751 ;
752 ; AVX2-SLOW-LABEL: trunc_v4i64_to_v4i16_return_v2i64:
753 ; AVX2-SLOW: # %bb.0:
754 ; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm1
755 ; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
756 ; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
757 ; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
758 ; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
759 ; AVX2-SLOW-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
760 ; AVX2-SLOW-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
761 ; AVX2-SLOW-NEXT: vzeroupper
762 ; AVX2-SLOW-NEXT: retq
763 ;
764 ; AVX2-FAST-LABEL: trunc_v4i64_to_v4i16_return_v2i64:
765 ; AVX2-FAST: # %bb.0:
766 ; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm1
767 ; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,8,9,8,9,10,11,8,9,10,11,12,13,14,15]
768 ; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm1, %xmm1
769 ; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm0, %xmm0
770 ; AVX2-FAST-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
771 ; AVX2-FAST-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
772 ; AVX2-FAST-NEXT: vzeroupper
773 ; AVX2-FAST-NEXT: retq
774 ;
775 ; AVX512F-LABEL: trunc_v4i64_to_v4i16_return_v2i64:
776 ; AVX512F: # %bb.0:
777 ; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
778 ; AVX512F-NEXT: vpmovqw %zmm0, %xmm0
779 ; AVX512F-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
780 ; AVX512F-NEXT: vzeroupper
781 ; AVX512F-NEXT: retq
782 ;
783 ; AVX512VL-LABEL: trunc_v4i64_to_v4i16_return_v2i64:
784 ; AVX512VL: # %bb.0:
785 ; AVX512VL-NEXT: vpmovqw %ymm0, %xmm0
786 ; AVX512VL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
787 ; AVX512VL-NEXT: vzeroupper
788 ; AVX512VL-NEXT: retq
789 ;
790 ; AVX512BW-LABEL: trunc_v4i64_to_v4i16_return_v2i64:
791 ; AVX512BW: # %bb.0:
792 ; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
793 ; AVX512BW-NEXT: vpmovqw %zmm0, %xmm0
794 ; AVX512BW-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
795 ; AVX512BW-NEXT: vzeroupper
796 ; AVX512BW-NEXT: retq
797 ;
798 ; AVX512BWVL-LABEL: trunc_v4i64_to_v4i16_return_v2i64:
799 ; AVX512BWVL: # %bb.0:
800 ; AVX512BWVL-NEXT: vpmovqw %ymm0, %xmm0
801 ; AVX512BWVL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
802 ; AVX512BWVL-NEXT: vzeroupper
803 ; AVX512BWVL-NEXT: retq
804 ;
805 ; AVX512VBMIVL-LABEL: trunc_v4i64_to_v4i16_return_v2i64:
806 ; AVX512VBMIVL: # %bb.0:
807 ; AVX512VBMIVL-NEXT: vpmovqw %ymm0, %xmm0
808 ; AVX512VBMIVL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
809 ; AVX512VBMIVL-NEXT: vzeroupper
810 ; AVX512VBMIVL-NEXT: retq
811 %truncated = trunc <4 x i64> %vec to <4 x i16>
812 %bc = bitcast <4 x i16> %truncated to i64
813 %result = insertelement <2 x i64> zeroinitializer, i64 %bc, i32 0
814 ret <2 x i64> %result
815 }
817 define <8 x i16> @trunc_v4i64_to_v4i16_with_zext_return_v8i16(<4 x i64> %vec) nounwind {
818 ; AVX1-LABEL: trunc_v4i64_to_v4i16_with_zext_return_v8i16:
819 ; AVX1: # %bb.0:
820 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
821 ; AVX1-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
822 ; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
823 ; AVX1-NEXT: vzeroupper
824 ; AVX1-NEXT: retq
825 ;
826 ; AVX2-SLOW-LABEL: trunc_v4i64_to_v4i16_with_zext_return_v8i16:
827 ; AVX2-SLOW: # %bb.0:
828 ; AVX2-SLOW-NEXT: vextractf128 $1, %ymm0, %xmm1
829 ; AVX2-SLOW-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
830 ; AVX2-SLOW-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
831 ; AVX2-SLOW-NEXT: vzeroupper
832 ; AVX2-SLOW-NEXT: retq
833 ;
834 ; AVX2-FAST-LABEL: trunc_v4i64_to_v4i16_with_zext_return_v8i16:
835 ; AVX2-FAST: # %bb.0:
836 ; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm1 = [0,2,4,6,4,6,6,7]
837 ; AVX2-FAST-NEXT: vpermd %ymm0, %ymm1, %ymm0
838 ; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
839 ; AVX2-FAST-NEXT: vzeroupper
840 ; AVX2-FAST-NEXT: retq
841 ;
842 ; AVX512F-LABEL: trunc_v4i64_to_v4i16_with_zext_return_v8i16:
843 ; AVX512F: # %bb.0:
844 ; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
845 ; AVX512F-NEXT: vpmovqd %zmm0, %ymm0
846 ; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
847 ; AVX512F-NEXT: vzeroupper
848 ; AVX512F-NEXT: retq
849 ;
850 ; AVX512VL-LABEL: trunc_v4i64_to_v4i16_with_zext_return_v8i16:
851 ; AVX512VL: # %bb.0:
852 ; AVX512VL-NEXT: vpmovqw %ymm0, %xmm0
853 ; AVX512VL-NEXT: vzeroupper
854 ; AVX512VL-NEXT: retq
855 ;
856 ; AVX512BW-LABEL: trunc_v4i64_to_v4i16_with_zext_return_v8i16:
857 ; AVX512BW: # %bb.0:
858 ; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
859 ; AVX512BW-NEXT: vpmovqd %zmm0, %ymm0
860 ; AVX512BW-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
861 ; AVX512BW-NEXT: vzeroupper
862 ; AVX512BW-NEXT: retq
863 ;
864 ; AVX512BWVL-LABEL: trunc_v4i64_to_v4i16_with_zext_return_v8i16:
865 ; AVX512BWVL: # %bb.0:
866 ; AVX512BWVL-NEXT: vpmovqw %ymm0, %xmm0
867 ; AVX512BWVL-NEXT: vzeroupper
868 ; AVX512BWVL-NEXT: retq
869 ;
870 ; AVX512VBMIVL-LABEL: trunc_v4i64_to_v4i16_with_zext_return_v8i16:
871 ; AVX512VBMIVL: # %bb.0:
872 ; AVX512VBMIVL-NEXT: vpmovqw %ymm0, %xmm0
873 ; AVX512VBMIVL-NEXT: vzeroupper
874 ; AVX512VBMIVL-NEXT: retq
875 %truncated = trunc <4 x i64> %vec to <4 x i16>
876 %truncated.ext = zext <4 x i16> %truncated to <4 x i32>
877 %bc = bitcast <4 x i32> %truncated.ext to <8 x i16>
878 %result = shufflevector <8 x i16> %bc, <8 x i16> zeroinitializer, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 1, i32 3, i32 5, i32 7>
879 ret <8 x i16> %result
880 }
882 define <8 x i16> @trunc_v4i64_to_v4i16_via_v4i32_return_v8i16(<4 x i64> %vec) nounwind {
883 ; AVX1-LABEL: trunc_v4i64_to_v4i16_via_v4i32_return_v8i16:
884 ; AVX1: # %bb.0:
885 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
886 ; AVX1-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
887 ; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
888 ; AVX1-NEXT: vzeroupper
889 ; AVX1-NEXT: retq
890 ;
891 ; AVX2-SLOW-LABEL: trunc_v4i64_to_v4i16_via_v4i32_return_v8i16:
892 ; AVX2-SLOW: # %bb.0:
893 ; AVX2-SLOW-NEXT: vextractf128 $1, %ymm0, %xmm1
894 ; AVX2-SLOW-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
895 ; AVX2-SLOW-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
896 ; AVX2-SLOW-NEXT: vzeroupper
897 ; AVX2-SLOW-NEXT: retq
898 ;
899 ; AVX2-FAST-LABEL: trunc_v4i64_to_v4i16_via_v4i32_return_v8i16:
900 ; AVX2-FAST: # %bb.0:
901 ; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm1 = [0,2,4,6,4,6,6,7]
902 ; AVX2-FAST-NEXT: vpermd %ymm0, %ymm1, %ymm0
903 ; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
904 ; AVX2-FAST-NEXT: vzeroupper
905 ; AVX2-FAST-NEXT: retq
906 ;
907 ; AVX512F-LABEL: trunc_v4i64_to_v4i16_via_v4i32_return_v8i16:
908 ; AVX512F: # %bb.0:
909 ; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
910 ; AVX512F-NEXT: vpmovqd %zmm0, %ymm0
911 ; AVX512F-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
912 ; AVX512F-NEXT: vzeroupper
913 ; AVX512F-NEXT: retq
914 ;
915 ; AVX512VL-LABEL: trunc_v4i64_to_v4i16_via_v4i32_return_v8i16:
916 ; AVX512VL: # %bb.0:
917 ; AVX512VL-NEXT: vpmovqw %ymm0, %xmm0
918 ; AVX512VL-NEXT: vzeroupper
919 ; AVX512VL-NEXT: retq
920 ;
921 ; AVX512BW-LABEL: trunc_v4i64_to_v4i16_via_v4i32_return_v8i16:
922 ; AVX512BW: # %bb.0:
923 ; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
924 ; AVX512BW-NEXT: vpmovqd %zmm0, %ymm0
925 ; AVX512BW-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
926 ; AVX512BW-NEXT: vzeroupper
927 ; AVX512BW-NEXT: retq
928 ;
929 ; AVX512BWVL-LABEL: trunc_v4i64_to_v4i16_via_v4i32_return_v8i16:
930 ; AVX512BWVL: # %bb.0:
931 ; AVX512BWVL-NEXT: vpmovqw %ymm0, %xmm0
932 ; AVX512BWVL-NEXT: vzeroupper
933 ; AVX512BWVL-NEXT: retq
934 ;
935 ; AVX512VBMIVL-LABEL: trunc_v4i64_to_v4i16_via_v4i32_return_v8i16:
936 ; AVX512VBMIVL: # %bb.0:
937 ; AVX512VBMIVL-NEXT: vpmovqw %ymm0, %xmm0
938 ; AVX512VBMIVL-NEXT: vzeroupper
939 ; AVX512VBMIVL-NEXT: retq
940 %truncated = trunc <4 x i64> %vec to <4 x i32>
941 %bc = bitcast <4 x i32> %truncated to <8 x i16>
942 %result = shufflevector <8 x i16> %bc, <8 x i16> zeroinitializer, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 8, i32 undef, i32 13>
943 ret <8 x i16> %result
944 }
946 define <8 x i16> @trunc_v4i64_to_v4i16_return_v8i16(<4 x i64> %vec) nounwind {
947 ; AVX1-LABEL: trunc_v4i64_to_v4i16_return_v8i16:
948 ; AVX1: # %bb.0:
949 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
950 ; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
951 ; AVX1-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
952 ; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
953 ; AVX1-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
954 ; AVX1-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
955 ; AVX1-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
956 ; AVX1-NEXT: vzeroupper
957 ; AVX1-NEXT: retq
958 ;
959 ; AVX2-SLOW-LABEL: trunc_v4i64_to_v4i16_return_v8i16:
960 ; AVX2-SLOW: # %bb.0:
961 ; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm1
962 ; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
963 ; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
964 ; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
965 ; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
966 ; AVX2-SLOW-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
967 ; AVX2-SLOW-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
968 ; AVX2-SLOW-NEXT: vzeroupper
969 ; AVX2-SLOW-NEXT: retq
970 ;
971 ; AVX2-FAST-LABEL: trunc_v4i64_to_v4i16_return_v8i16:
972 ; AVX2-FAST: # %bb.0:
973 ; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm1
974 ; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,8,9,8,9,10,11,8,9,10,11,12,13,14,15]
975 ; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm1, %xmm1
976 ; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm0, %xmm0
977 ; AVX2-FAST-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
978 ; AVX2-FAST-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
979 ; AVX2-FAST-NEXT: vzeroupper
980 ; AVX2-FAST-NEXT: retq
981 ;
982 ; AVX512F-LABEL: trunc_v4i64_to_v4i16_return_v8i16:
983 ; AVX512F: # %bb.0:
984 ; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
985 ; AVX512F-NEXT: vpmovqw %zmm0, %xmm0
986 ; AVX512F-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
987 ; AVX512F-NEXT: vzeroupper
988 ; AVX512F-NEXT: retq
989 ;
990 ; AVX512VL-LABEL: trunc_v4i64_to_v4i16_return_v8i16:
991 ; AVX512VL: # %bb.0:
992 ; AVX512VL-NEXT: vpmovqw %ymm0, %xmm0
993 ; AVX512VL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
994 ; AVX512VL-NEXT: vzeroupper
995 ; AVX512VL-NEXT: retq
996 ;
997 ; AVX512BW-LABEL: trunc_v4i64_to_v4i16_return_v8i16:
998 ; AVX512BW: # %bb.0:
999 ; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
1000 ; AVX512BW-NEXT: vpmovqw %zmm0, %xmm0
1001 ; AVX512BW-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
1002 ; AVX512BW-NEXT: vzeroupper
1003 ; AVX512BW-NEXT: retq
1004 ;
1005 ; AVX512BWVL-LABEL: trunc_v4i64_to_v4i16_return_v8i16:
1006 ; AVX512BWVL: # %bb.0:
1007 ; AVX512BWVL-NEXT: vpmovqw %ymm0, %xmm0
1008 ; AVX512BWVL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
1009 ; AVX512BWVL-NEXT: vzeroupper
1010 ; AVX512BWVL-NEXT: retq
1011 ;
1012 ; AVX512VBMIVL-LABEL: trunc_v4i64_to_v4i16_return_v8i16:
1013 ; AVX512VBMIVL: # %bb.0:
1014 ; AVX512VBMIVL-NEXT: vpmovqw %ymm0, %xmm0
1015 ; AVX512VBMIVL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
1016 ; AVX512VBMIVL-NEXT: vzeroupper
1017 ; AVX512VBMIVL-NEXT: retq
1018 %truncated = trunc <4 x i64> %vec to <4 x i16>
1019 %result = shufflevector <4 x i16> %truncated, <4 x i16> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
1020 ret <8 x i16> %result
1021 }
1023 define <16 x i8> @trunc_v4i64_to_v4i8_return_v16i8(<4 x i64> %vec) nounwind {
1024 ; AVX1-LABEL: trunc_v4i64_to_v4i8_return_v16i8:
1025 ; AVX1: # %bb.0:
1026 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
1027 ; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = <0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
1028 ; AVX1-NEXT: vpshufb %xmm2, %xmm1, %xmm1
1029 ; AVX1-NEXT: vpshufb %xmm2, %xmm0, %xmm0
1030 ; AVX1-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
1031 ; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
1032 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
1033 ; AVX1-NEXT: vzeroupper
1034 ; AVX1-NEXT: retq
1035 ;
1036 ; AVX2-LABEL: trunc_v4i64_to_v4i8_return_v16i8:
1037 ; AVX2: # %bb.0:
1038 ; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
1039 ; AVX2-NEXT: vmovdqa {{.*#+}} xmm2 = <0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
1040 ; AVX2-NEXT: vpshufb %xmm2, %xmm1, %xmm1
1041 ; AVX2-NEXT: vpshufb %xmm2, %xmm0, %xmm0
1042 ; AVX2-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
1043 ; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
1044 ; AVX2-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
1045 ; AVX2-NEXT: vzeroupper
1046 ; AVX2-NEXT: retq
1047 ;
1048 ; AVX512F-LABEL: trunc_v4i64_to_v4i8_return_v16i8:
1049 ; AVX512F: # %bb.0:
1050 ; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
1051 ; AVX512F-NEXT: vpmovqb %zmm0, %xmm0
1052 ; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
1053 ; AVX512F-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
1054 ; AVX512F-NEXT: vzeroupper
1055 ; AVX512F-NEXT: retq
1056 ;
1057 ; AVX512VL-LABEL: trunc_v4i64_to_v4i8_return_v16i8:
1058 ; AVX512VL: # %bb.0:
1059 ; AVX512VL-NEXT: vpmovqb %ymm0, %xmm0
1060 ; AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1
1061 ; AVX512VL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
1062 ; AVX512VL-NEXT: vzeroupper
1063 ; AVX512VL-NEXT: retq
1064 ;
1065 ; AVX512BW-LABEL: trunc_v4i64_to_v4i8_return_v16i8:
1066 ; AVX512BW: # %bb.0:
1067 ; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
1068 ; AVX512BW-NEXT: vpmovqb %zmm0, %xmm0
1069 ; AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
1070 ; AVX512BW-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
1071 ; AVX512BW-NEXT: vzeroupper
1072 ; AVX512BW-NEXT: retq
1073 ;
1074 ; AVX512BWVL-LABEL: trunc_v4i64_to_v4i8_return_v16i8:
1075 ; AVX512BWVL: # %bb.0:
1076 ; AVX512BWVL-NEXT: vpmovqb %ymm0, %xmm0
1077 ; AVX512BWVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
1078 ; AVX512BWVL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
1079 ; AVX512BWVL-NEXT: vzeroupper
1080 ; AVX512BWVL-NEXT: retq
1081 ;
1082 ; AVX512VBMIVL-LABEL: trunc_v4i64_to_v4i8_return_v16i8:
1083 ; AVX512VBMIVL: # %bb.0:
1084 ; AVX512VBMIVL-NEXT: vpmovqb %ymm0, %xmm0
1085 ; AVX512VBMIVL-NEXT: vpxor %xmm1, %xmm1, %xmm1
1086 ; AVX512VBMIVL-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
1087 ; AVX512VBMIVL-NEXT: vzeroupper
1088 ; AVX512VBMIVL-NEXT: retq
1089 %truncated = trunc <4 x i64> %vec to <4 x i8>
1090 %result = shufflevector <4 x i8> %truncated, <4 x i8> zeroinitializer, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7, i32 5, i32 5, i32 undef, i32 7>
1091 ret <16 x i8> %result
1092 }
1094 define void @shuffle_v16i16_to_v4i16(<16 x i16>* %L, <4 x i16>* %S) nounwind {
1095 ; AVX1-LABEL: shuffle_v16i16_to_v4i16:
1096 ; AVX1: # %bb.0:
1097 ; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = mem[0,2,2,3]
1098 ; AVX1-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
1099 ; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = mem[0,2,2,3]
1100 ; AVX1-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
1101 ; AVX1-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
1102 ; AVX1-NEXT: vmovq %xmm0, (%rsi)
1103 ; AVX1-NEXT: retq
1104 ;
1105 ; AVX2-SLOW-LABEL: shuffle_v16i16_to_v4i16:
1106 ; AVX2-SLOW: # %bb.0:
1107 ; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm0 = mem[0,2,2,3]
1108 ; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
1109 ; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = mem[0,2,2,3]
1110 ; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
1111 ; AVX2-SLOW-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
1112 ; AVX2-SLOW-NEXT: vmovq %xmm0, (%rsi)
1113 ; AVX2-SLOW-NEXT: retq
1114 ;
1115 ; AVX2-FAST-LABEL: shuffle_v16i16_to_v4i16:
1116 ; AVX2-FAST: # %bb.0:
1117 ; AVX2-FAST-NEXT: vmovdqa (%rdi), %xmm0
1118 ; AVX2-FAST-NEXT: vmovdqa 16(%rdi), %xmm1
1119 ; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,8,9,8,9,10,11,8,9,10,11,12,13,14,15]
1120 ; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm1, %xmm1
1121 ; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm0, %xmm0
1122 ; AVX2-FAST-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
1123 ; AVX2-FAST-NEXT: vmovq %xmm0, (%rsi)
1124 ; AVX2-FAST-NEXT: retq
1125 ;
1126 ; AVX512F-LABEL: shuffle_v16i16_to_v4i16:
1127 ; AVX512F: # %bb.0:
1128 ; AVX512F-NEXT: vpshufd {{.*#+}} xmm0 = mem[0,2,2,3]
1129 ; AVX512F-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
1130 ; AVX512F-NEXT: vpshufd {{.*#+}} xmm1 = mem[0,2,2,3]
1131 ; AVX512F-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
1132 ; AVX512F-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
1133 ; AVX512F-NEXT: vmovq %xmm0, (%rsi)
1134 ; AVX512F-NEXT: retq
1135 ;
1136 ; AVX512VL-LABEL: shuffle_v16i16_to_v4i16:
1137 ; AVX512VL: # %bb.0:
1138 ; AVX512VL-NEXT: vmovdqa (%rdi), %xmm0
1139 ; AVX512VL-NEXT: vmovdqa 16(%rdi), %xmm1
1140 ; AVX512VL-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,8,9,8,9,10,11,8,9,10,11,12,13,14,15]
1141 ; AVX512VL-NEXT: vpshufb %xmm2, %xmm1, %xmm1
1142 ; AVX512VL-NEXT: vpshufb %xmm2, %xmm0, %xmm0
1143 ; AVX512VL-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
1144 ; AVX512VL-NEXT: vmovq %xmm0, (%rsi)
1145 ; AVX512VL-NEXT: retq
1146 ;
1147 ; AVX512BW-LABEL: shuffle_v16i16_to_v4i16:
1148 ; AVX512BW: # %bb.0:
1149 ; AVX512BW-NEXT: vmovdqa (%rdi), %xmm0
1150 ; AVX512BW-NEXT: vmovdqa 16(%rdi), %xmm1
1151 ; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,8,9,8,9,10,11,8,9,10,11,12,13,14,15]
1152 ; AVX512BW-NEXT: vpshufb %xmm2, %xmm1, %xmm1
1153 ; AVX512BW-NEXT: vpshufb %xmm2, %xmm0, %xmm0
1154 ; AVX512BW-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
1155 ; AVX512BW-NEXT: vmovq %xmm0, (%rsi)
1156 ; AVX512BW-NEXT: retq
1157 ;
1158 ; AVX512BWVL-LABEL: shuffle_v16i16_to_v4i16:
1159 ; AVX512BWVL: # %bb.0:
1160 ; AVX512BWVL-NEXT: vmovdqa (%rdi), %xmm0
1161 ; AVX512BWVL-NEXT: vmovdqa {{.*#+}} xmm1 = [0,4,8,12,4,5,12,13]
1162 ; AVX512BWVL-NEXT: vpermi2w 16(%rdi), %xmm0, %xmm1
1163 ; AVX512BWVL-NEXT: vmovq %xmm1, (%rsi)
1164 ; AVX512BWVL-NEXT: retq
1165 ;
1166 ; AVX512VBMIVL-LABEL: shuffle_v16i16_to_v4i16:
1167 ; AVX512VBMIVL: # %bb.0:
1168 ; AVX512VBMIVL-NEXT: vmovdqa (%rdi), %xmm0
1169 ; AVX512VBMIVL-NEXT: vmovdqa {{.*#+}} xmm1 = [0,4,8,12,4,5,12,13]
1170 ; AVX512VBMIVL-NEXT: vpermi2w 16(%rdi), %xmm0, %xmm1
1171 ; AVX512VBMIVL-NEXT: vmovq %xmm1, (%rsi)
1172 ; AVX512VBMIVL-NEXT: retq
1173 %vec = load <16 x i16>, <16 x i16>* %L
1174 %strided.vec = shufflevector <16 x i16> %vec, <16 x i16> undef, <4 x i32> <i32 0, i32 4, i32 8, i32 12>
1175 store <4 x i16> %strided.vec, <4 x i16>* %S
1176 ret void
1177 }
1179 define void @trunc_v4i64_to_v4i16(<16 x i16>* %L, <4 x i16>* %S) nounwind {
1180 ; AVX1-LABEL: trunc_v4i64_to_v4i16:
1181 ; AVX1: # %bb.0:
1182 ; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = mem[0,2,2,3]
1183 ; AVX1-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
1184 ; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = mem[0,2,2,3]
1185 ; AVX1-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
1186 ; AVX1-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
1187 ; AVX1-NEXT: vmovq %xmm0, (%rsi)
1188 ; AVX1-NEXT: retq
1189 ;
1190 ; AVX2-SLOW-LABEL: trunc_v4i64_to_v4i16:
1191 ; AVX2-SLOW: # %bb.0:
1192 ; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm0 = mem[0,2,2,3]
1193 ; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
1194 ; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = mem[0,2,2,3]
1195 ; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
1196 ; AVX2-SLOW-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
1197 ; AVX2-SLOW-NEXT: vmovq %xmm0, (%rsi)
1198 ; AVX2-SLOW-NEXT: retq
1199 ;
1200 ; AVX2-FAST-LABEL: trunc_v4i64_to_v4i16:
1201 ; AVX2-FAST: # %bb.0:
1202 ; AVX2-FAST-NEXT: vmovdqa (%rdi), %xmm0
1203 ; AVX2-FAST-NEXT: vmovdqa 16(%rdi), %xmm1
1204 ; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,8,9,8,9,10,11,8,9,10,11,12,13,14,15]
1205 ; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm1, %xmm1
1206 ; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm0, %xmm0
1207 ; AVX2-FAST-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
1208 ; AVX2-FAST-NEXT: vmovq %xmm0, (%rsi)
1209 ; AVX2-FAST-NEXT: retq
1210 ;
1211 ; AVX512F-LABEL: trunc_v4i64_to_v4i16:
1212 ; AVX512F: # %bb.0:
1213 ; AVX512F-NEXT: vmovdqa (%rdi), %ymm0
1214 ; AVX512F-NEXT: vpmovqw %zmm0, %xmm0
1215 ; AVX512F-NEXT: vmovq %xmm0, (%rsi)
1216 ; AVX512F-NEXT: vzeroupper
1217 ; AVX512F-NEXT: retq
1218 ;
1219 ; AVX512VL-LABEL: trunc_v4i64_to_v4i16:
1220 ; AVX512VL: # %bb.0:
1221 ; AVX512VL-NEXT: vmovdqa (%rdi), %ymm0
1222 ; AVX512VL-NEXT: vpmovqw %ymm0, (%rsi)
1223 ; AVX512VL-NEXT: vzeroupper
1224 ; AVX512VL-NEXT: retq
1225 ;
1226 ; AVX512BW-LABEL: trunc_v4i64_to_v4i16:
1227 ; AVX512BW: # %bb.0:
1228 ; AVX512BW-NEXT: vmovdqa (%rdi), %ymm0
1229 ; AVX512BW-NEXT: vpmovqw %zmm0, %xmm0
1230 ; AVX512BW-NEXT: vmovq %xmm0, (%rsi)
1231 ; AVX512BW-NEXT: vzeroupper
1232 ; AVX512BW-NEXT: retq
1233 ;
1234 ; AVX512BWVL-LABEL: trunc_v4i64_to_v4i16:
1235 ; AVX512BWVL: # %bb.0:
1236 ; AVX512BWVL-NEXT: vmovdqa (%rdi), %ymm0
1237 ; AVX512BWVL-NEXT: vpmovqw %ymm0, (%rsi)
1238 ; AVX512BWVL-NEXT: vzeroupper
1239 ; AVX512BWVL-NEXT: retq
1240 ;
1241 ; AVX512VBMIVL-LABEL: trunc_v4i64_to_v4i16:
1242 ; AVX512VBMIVL: # %bb.0:
1243 ; AVX512VBMIVL-NEXT: vmovdqa (%rdi), %ymm0
1244 ; AVX512VBMIVL-NEXT: vpmovqw %ymm0, (%rsi)
1245 ; AVX512VBMIVL-NEXT: vzeroupper
1246 ; AVX512VBMIVL-NEXT: retq
1247 %vec = load <16 x i16>, <16 x i16>* %L
1248 %bc = bitcast <16 x i16> %vec to <4 x i64>
1249 %strided.vec = trunc <4 x i64> %bc to <4 x i16>
1250 store <4 x i16> %strided.vec, <4 x i16>* %S
1251 ret void
1252 }
1254 define void @shuffle_v32i8_to_v4i8(<32 x i8>* %L, <4 x i8>* %S) nounwind {
1255 ; AVX-LABEL: shuffle_v32i8_to_v4i8:
1256 ; AVX: # %bb.0:
1257 ; AVX-NEXT: vmovdqa (%rdi), %xmm0
1258 ; AVX-NEXT: vmovdqa 16(%rdi), %xmm1
1259 ; AVX-NEXT: vmovdqa {{.*#+}} xmm2 = <0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
1260 ; AVX-NEXT: vpshufb %xmm2, %xmm1, %xmm1
1261 ; AVX-NEXT: vpshufb %xmm2, %xmm0, %xmm0
1262 ; AVX-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
1263 ; AVX-NEXT: vmovd %xmm0, (%rsi)
1264 ; AVX-NEXT: retq
1265 ;
1266 ; AVX512F-LABEL: shuffle_v32i8_to_v4i8:
1267 ; AVX512F: # %bb.0:
1268 ; AVX512F-NEXT: vmovdqa (%rdi), %xmm0
1269 ; AVX512F-NEXT: vmovdqa 16(%rdi), %xmm1
1270 ; AVX512F-NEXT: vmovdqa {{.*#+}} xmm2 = <0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
1271 ; AVX512F-NEXT: vpshufb %xmm2, %xmm1, %xmm1
1272 ; AVX512F-NEXT: vpshufb %xmm2, %xmm0, %xmm0
1273 ; AVX512F-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
1274 ; AVX512F-NEXT: vmovd %xmm0, (%rsi)
1275 ; AVX512F-NEXT: retq
1276 ;
1277 ; AVX512VL-LABEL: shuffle_v32i8_to_v4i8:
1278 ; AVX512VL: # %bb.0:
1279 ; AVX512VL-NEXT: vmovdqa (%rdi), %xmm0
1280 ; AVX512VL-NEXT: vmovdqa 16(%rdi), %xmm1
1281 ; AVX512VL-NEXT: vmovdqa {{.*#+}} xmm2 = <0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
1282 ; AVX512VL-NEXT: vpshufb %xmm2, %xmm1, %xmm1
1283 ; AVX512VL-NEXT: vpshufb %xmm2, %xmm0, %xmm0
1284 ; AVX512VL-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
1285 ; AVX512VL-NEXT: vmovd %xmm0, (%rsi)
1286 ; AVX512VL-NEXT: retq
1287 ;
1288 ; AVX512BW-LABEL: shuffle_v32i8_to_v4i8:
1289 ; AVX512BW: # %bb.0:
1290 ; AVX512BW-NEXT: vmovdqa (%rdi), %xmm0
1291 ; AVX512BW-NEXT: vmovdqa 16(%rdi), %xmm1
1292 ; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm2 = <0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
1293 ; AVX512BW-NEXT: vpshufb %xmm2, %xmm1, %xmm1
1294 ; AVX512BW-NEXT: vpshufb %xmm2, %xmm0, %xmm0
1295 ; AVX512BW-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
1296 ; AVX512BW-NEXT: vmovd %xmm0, (%rsi)
1297 ; AVX512BW-NEXT: retq
1298 ;
1299 ; AVX512BWVL-LABEL: shuffle_v32i8_to_v4i8:
1300 ; AVX512BWVL: # %bb.0:
1301 ; AVX512BWVL-NEXT: vmovdqa (%rdi), %xmm0
1302 ; AVX512BWVL-NEXT: vmovdqa 16(%rdi), %xmm1
1303 ; AVX512BWVL-NEXT: vmovdqa {{.*#+}} xmm2 = <0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
1304 ; AVX512BWVL-NEXT: vpshufb %xmm2, %xmm1, %xmm1
1305 ; AVX512BWVL-NEXT: vpshufb %xmm2, %xmm0, %xmm0
1306 ; AVX512BWVL-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
1307 ; AVX512BWVL-NEXT: vmovd %xmm0, (%rsi)
1308 ; AVX512BWVL-NEXT: retq
1309 ;
1310 ; AVX512VBMIVL-LABEL: shuffle_v32i8_to_v4i8:
1311 ; AVX512VBMIVL: # %bb.0:
1312 ; AVX512VBMIVL-NEXT: vmovdqa (%rdi), %xmm0
1313 ; AVX512VBMIVL-NEXT: vpbroadcastd {{.*#+}} xmm1 = [403703808,403703808,403703808,403703808]
1314 ; AVX512VBMIVL-NEXT: vpermi2b 16(%rdi), %xmm0, %xmm1
1315 ; AVX512VBMIVL-NEXT: vmovd %xmm1, (%rsi)
1316 ; AVX512VBMIVL-NEXT: retq
1317 %vec = load <32 x i8>, <32 x i8>* %L
1318 %strided.vec = shufflevector <32 x i8> %vec, <32 x i8> undef, <4 x i32> <i32 0, i32 8, i32 16, i32 24>
1319 store <4 x i8> %strided.vec, <4 x i8>* %S
1320 ret void
1321 }
1323 define void @trunc_v4i64_to_v4i8(<32 x i8>* %L, <4 x i8>* %S) nounwind {
1324 ; AVX-LABEL: trunc_v4i64_to_v4i8:
1325 ; AVX: # %bb.0:
1326 ; AVX-NEXT: vmovdqa (%rdi), %xmm0
1327 ; AVX-NEXT: vmovdqa 16(%rdi), %xmm1
1328 ; AVX-NEXT: vmovdqa {{.*#+}} xmm2 = <0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
1329 ; AVX-NEXT: vpshufb %xmm2, %xmm1, %xmm1
1330 ; AVX-NEXT: vpshufb %xmm2, %xmm0, %xmm0
1331 ; AVX-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
1332 ; AVX-NEXT: vmovd %xmm0, (%rsi)
1333 ; AVX-NEXT: retq
1334 ;
1335 ; AVX512F-LABEL: trunc_v4i64_to_v4i8:
1336 ; AVX512F: # %bb.0:
1337 ; AVX512F-NEXT: vmovdqa (%rdi), %ymm0
1338 ; AVX512F-NEXT: vpmovqb %zmm0, %xmm0
1339 ; AVX512F-NEXT: vmovd %xmm0, (%rsi)
1340 ; AVX512F-NEXT: vzeroupper
1341 ; AVX512F-NEXT: retq
1342 ;
1343 ; AVX512VL-LABEL: trunc_v4i64_to_v4i8:
1344 ; AVX512VL: # %bb.0:
1345 ; AVX512VL-NEXT: vmovdqa (%rdi), %ymm0
1346 ; AVX512VL-NEXT: vpmovqb %ymm0, (%rsi)
1347 ; AVX512VL-NEXT: vzeroupper
1348 ; AVX512VL-NEXT: retq
1349 ;
1350 ; AVX512BW-LABEL: trunc_v4i64_to_v4i8:
1351 ; AVX512BW: # %bb.0:
1352 ; AVX512BW-NEXT: vmovdqa (%rdi), %ymm0
1353 ; AVX512BW-NEXT: vpmovqb %zmm0, %xmm0
1354 ; AVX512BW-NEXT: vmovd %xmm0, (%rsi)
1355 ; AVX512BW-NEXT: vzeroupper
1356 ; AVX512BW-NEXT: retq
1357 ;
1358 ; AVX512BWVL-LABEL: trunc_v4i64_to_v4i8:
1359 ; AVX512BWVL: # %bb.0:
1360 ; AVX512BWVL-NEXT: vmovdqa (%rdi), %ymm0
1361 ; AVX512BWVL-NEXT: vpmovqb %ymm0, (%rsi)
1362 ; AVX512BWVL-NEXT: vzeroupper
1363 ; AVX512BWVL-NEXT: retq
1364 ;
1365 ; AVX512VBMIVL-LABEL: trunc_v4i64_to_v4i8:
1366 ; AVX512VBMIVL: # %bb.0:
1367 ; AVX512VBMIVL-NEXT: vmovdqa (%rdi), %ymm0
1368 ; AVX512VBMIVL-NEXT: vpmovqb %ymm0, (%rsi)
1369 ; AVX512VBMIVL-NEXT: vzeroupper
1370 ; AVX512VBMIVL-NEXT: retq
1371 %vec = load <32 x i8>, <32 x i8>* %L
1372 %bc = bitcast <32 x i8> %vec to <4 x i64>
1373 %strided.vec = trunc <4 x i64> %bc to <4 x i8>
1374 store <4 x i8> %strided.vec, <4 x i8>* %S
1375 ret void
1376 }
1378 ; In this case not all elements are collected from the same source vector, so
1379 ; the resulting BUILD_VECTOR should not be combined to a truncate.
1380 define <16 x i8> @negative(<32 x i8> %v, <32 x i8> %w) nounwind {
1381 ; AVX1-LABEL: negative:
1382 ; AVX1: # %bb.0:
1383 ; AVX1-NEXT: vpshufb {{.*#+}} xmm2 = xmm0[u,2,4,6,8,10,12,14],zero,zero,zero,zero,zero,zero,zero,zero
1384 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
1385 ; AVX1-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[u],zero,zero,zero,zero,zero,zero,zero,xmm0[0,2,4,6,8,10,12,14]
1386 ; AVX1-NEXT: vpor %xmm2, %xmm0, %xmm0
1387 ; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
1388 ; AVX1-NEXT: vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
1389 ; AVX1-NEXT: vzeroupper
1390 ; AVX1-NEXT: retq
1391 ;
1392 ; AVX2-LABEL: negative:
1393 ; AVX2: # %bb.0:
1394 ; AVX2-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[u,2,4,6,8,10,12,14,0,2,4,6,8,10,12,14,u,18,20,22,24,26,28,30,16,18,20,22,24,26,28,30]
1395 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
1396 ; AVX2-NEXT: vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
1397 ; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,2,3]
1398 ; AVX2-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0
1399 ; AVX2-NEXT: vzeroupper
1400 ; AVX2-NEXT: retq
1401 ;
1402 ; AVX512F-LABEL: negative:
1403 ; AVX512F: # %bb.0:
1404 ; AVX512F-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[u,2,4,6,8,10,12,14,0,2,4,6,8,10,12,14,u,18,20,22,24,26,28,30,16,18,20,22,24,26,28,30]
1405 ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
1406 ; AVX512F-NEXT: vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
1407 ; AVX512F-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,2,3]
1408 ; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0
1409 ; AVX512F-NEXT: vzeroupper
1410 ; AVX512F-NEXT: retq
1411 ;
1412 ; AVX512VL-LABEL: negative:
1413 ; AVX512VL: # %bb.0:
1414 ; AVX512VL-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[u,2,4,6,8,10,12,14,0,2,4,6,8,10,12,14,u,18,20,22,24,26,28,30,16,18,20,22,24,26,28,30]
1415 ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
1416 ; AVX512VL-NEXT: vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
1417 ; AVX512VL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,2,3]
1418 ; AVX512VL-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0
1419 ; AVX512VL-NEXT: vzeroupper
1420 ; AVX512VL-NEXT: retq
1421 ;
1422 ; AVX512BW-LABEL: negative:
1423 ; AVX512BW: # %bb.0:
1424 ; AVX512BW-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[u,2,4,6,8,10,12,14,0,2,4,6,8,10,12,14,u,18,20,22,24,26,28,30,16,18,20,22,24,26,28,30]
1425 ; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
1426 ; AVX512BW-NEXT: vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
1427 ; AVX512BW-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,2,3]
1428 ; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0
1429 ; AVX512BW-NEXT: vzeroupper
1430 ; AVX512BW-NEXT: retq
1431 ;
1432 ; AVX512BWVL-LABEL: negative:
1433 ; AVX512BWVL: # %bb.0:
1434 ; AVX512BWVL-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[u,2,4,6,8,10,12,14,0,2,4,6,8,10,12,14,u,18,20,22,24,26,28,30,16,18,20,22,24,26,28,30]
1435 ; AVX512BWVL-NEXT: movl $65537, %eax # imm = 0x10001
1436 ; AVX512BWVL-NEXT: kmovd %eax, %k1
1437 ; AVX512BWVL-NEXT: vmovdqu8 %ymm1, %ymm0 {%k1}
1438 ; AVX512BWVL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,2,3]
1439 ; AVX512BWVL-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0
1440 ; AVX512BWVL-NEXT: vzeroupper
1441 ; AVX512BWVL-NEXT: retq
1442 ;
1443 ; AVX512VBMIVL-LABEL: negative:
1444 ; AVX512VBMIVL: # %bb.0:
1445 ; AVX512VBMIVL-NEXT: vmovdqa {{.*#+}} ymm2 = [32,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30,48,18,20,22,24,26,28,30,16,18,20,22,24,26,28,30]
1446 ; AVX512VBMIVL-NEXT: vpermt2b %ymm1, %ymm2, %ymm0
1447 ; AVX512VBMIVL-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0
1448 ; AVX512VBMIVL-NEXT: vzeroupper
1449 ; AVX512VBMIVL-NEXT: retq
1450 %strided.vec = shufflevector <32 x i8> %v, <32 x i8> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
1451 %w0 = extractelement <32 x i8> %w, i32 0
1452 %merged = insertelement <16 x i8> %strided.vec, i8 %w0, i32 0
1453 ret <16 x i8> %merged
1454 }