1 ; RUN: llc < %s -mcpu=x86-64 -mattr=+avx512f -x86-experimental-vector-shuffle-legality | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512F
2 ; RUN: llc < %s -mcpu=x86-64 -mattr=+avx512bw -x86-experimental-vector-shuffle-legality | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512BW
4 target triple = "x86_64-unknown-unknown"
6 define <8 x double> @shuffle_v8f64_00000000(<8 x double> %a, <8 x double> %b) {
7 ; ALL-LABEL: shuffle_v8f64_00000000:
9 ; ALL-NEXT: vbroadcastsd %xmm0, %zmm0
11 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
12 ret <8 x double> %shuffle
15 define <8 x double> @shuffle_v8f64_00000010(<8 x double> %a, <8 x double> %b) {
16 ; ALL-LABEL: shuffle_v8f64_00000010:
18 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm1
19 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,0]
20 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
22 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 0>
23 ret <8 x double> %shuffle
26 define <8 x double> @shuffle_v8f64_00000200(<8 x double> %a, <8 x double> %b) {
27 ; ALL-LABEL: shuffle_v8f64_00000200:
29 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm1
30 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,0]
31 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
33 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 2, i32 0, i32 0>
34 ret <8 x double> %shuffle
37 define <8 x double> @shuffle_v8f64_00003000(<8 x double> %a, <8 x double> %b) {
38 ; ALL-LABEL: shuffle_v8f64_00003000:
40 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm1
41 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[3,0,0,0]
42 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
44 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 3, i32 0, i32 0, i32 0>
45 ret <8 x double> %shuffle
48 define <8 x double> @shuffle_v8f64_00040000(<8 x double> %a, <8 x double> %b) {
49 ; ALL-LABEL: shuffle_v8f64_00040000:
51 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1
52 ; ALL-NEXT: vbroadcastsd %xmm1, %ymm1
53 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm0
54 ; ALL-NEXT: vblendpd {{.*#+}} ymm1 = ymm0[0,1,2],ymm1[3]
55 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
57 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 4, i32 0, i32 0, i32 0, i32 0>
58 ret <8 x double> %shuffle
61 define <8 x double> @shuffle_v8f64_00500000(<8 x double> %a, <8 x double> %b) {
62 ; ALL-LABEL: shuffle_v8f64_00500000:
64 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1
65 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[0,1,1,3]
66 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm0
67 ; ALL-NEXT: vblendpd {{.*#+}} ymm1 = ymm0[0,1],ymm1[2],ymm0[3]
68 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
70 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 5, i32 0, i32 0, i32 0, i32 0, i32 0>
71 ret <8 x double> %shuffle
74 define <8 x double> @shuffle_v8f64_06000000(<8 x double> %a, <8 x double> %b) {
75 ; ALL-LABEL: shuffle_v8f64_06000000:
77 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1
78 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[0,2,2,3]
79 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm0
80 ; ALL-NEXT: vblendpd {{.*#+}} ymm1 = ymm0[0],ymm1[1],ymm0[2,3]
81 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
83 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 6, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
84 ret <8 x double> %shuffle
87 define <8 x double> @shuffle_v8f64_70000000(<8 x double> %a, <8 x double> %b) {
88 ; ALL-LABEL: shuffle_v8f64_70000000:
90 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1
91 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[3,1,2,3]
92 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm0
93 ; ALL-NEXT: vblendpd {{.*#+}} ymm1 = ymm1[0],ymm0[1,2,3]
94 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
96 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 7, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
97 ret <8 x double> %shuffle
100 define <8 x double> @shuffle_v8f64_01014545(<8 x double> %a, <8 x double> %b) {
101 ; ALL-LABEL: shuffle_v8f64_01014545:
103 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1
104 ; ALL-NEXT: vinsertf128 $1, %xmm1, %ymm1, %ymm1
105 ; ALL-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
106 ; ALL-NEXT: vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
108 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 4, i32 5, i32 4, i32 5>
109 ret <8 x double> %shuffle
112 define <8 x double> @shuffle_v8f64_00112233(<8 x double> %a, <8 x double> %b) {
113 ; ALL-LABEL: shuffle_v8f64_00112233:
115 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,0,1,1]
116 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,2,3,3]
117 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
119 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 1, i32 1, i32 2, i32 2, i32 3, i32 3>
120 ret <8 x double> %shuffle
123 define <8 x double> @shuffle_v8f64_00001111(<8 x double> %a, <8 x double> %b) {
124 ; ALL-LABEL: shuffle_v8f64_00001111:
126 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm1
127 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[1,1,1,1]
128 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
130 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1>
131 ret <8 x double> %shuffle
134 define <8 x double> @shuffle_v8f64_81a3c5e7(<8 x double> %a, <8 x double> %b) {
135 ; ALL-LABEL: shuffle_v8f64_81a3c5e7:
137 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm2
138 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm3
139 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2],ymm2[3]
140 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3]
141 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
143 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 8, i32 1, i32 10, i32 3, i32 12, i32 5, i32 14, i32 7>
144 ret <8 x double> %shuffle
147 define <8 x double> @shuffle_v8f64_08080808(<8 x double> %a, <8 x double> %b) {
148 ; ALL-LABEL: shuffle_v8f64_08080808:
150 ; ALL-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
151 ; ALL-NEXT: vbroadcastsd %xmm1, %ymm1
152 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3]
153 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm0, %zmm0
155 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 8, i32 0, i32 8, i32 0, i32 8, i32 0, i32 8>
156 ret <8 x double> %shuffle
159 define <8 x double> @shuffle_v8f64_08084c4c(<8 x double> %a, <8 x double> %b) {
160 ; ALL-LABEL: shuffle_v8f64_08084c4c:
162 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm2
163 ; ALL-NEXT: vinsertf128 $1, %xmm2, %ymm2, %ymm2
164 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm3
165 ; ALL-NEXT: vbroadcastsd %xmm3, %ymm3
166 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm2[0],ymm3[1],ymm2[2],ymm3[3]
167 ; ALL-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
168 ; ALL-NEXT: vbroadcastsd %xmm1, %ymm1
169 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3]
170 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
172 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 8, i32 0, i32 8, i32 4, i32 12, i32 4, i32 12>
173 ret <8 x double> %shuffle
176 define <8 x double> @shuffle_v8f64_8823cc67(<8 x double> %a, <8 x double> %b) {
177 ; ALL-LABEL: shuffle_v8f64_8823cc67:
179 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm2
180 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm3
181 ; ALL-NEXT: vbroadcastsd %xmm3, %ymm3
182 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3]
183 ; ALL-NEXT: vbroadcastsd %xmm1, %ymm1
184 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3]
185 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
187 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 8, i32 8, i32 2, i32 3, i32 12, i32 12, i32 6, i32 7>
188 ret <8 x double> %shuffle
191 define <8 x double> @shuffle_v8f64_9832dc76(<8 x double> %a, <8 x double> %b) {
192 ; ALL-LABEL: shuffle_v8f64_9832dc76:
194 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm1[0,1],ymm0[2,3]
195 ; ALL-NEXT: vpermilpd {{.*#+}} ymm2 = ymm2[1,0,3,2]
196 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
197 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm1
198 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3]
199 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[1,0,3,2]
200 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm2, %zmm0
202 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 9, i32 8, i32 3, i32 2, i32 13, i32 12, i32 7, i32 6>
203 ret <8 x double> %shuffle
206 define <8 x double> @shuffle_v8f64_9810dc54(<8 x double> %a, <8 x double> %b) {
207 ; ALL-LABEL: shuffle_v8f64_9810dc54:
209 ; ALL-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm2
210 ; ALL-NEXT: vpermilpd {{.*#+}} ymm2 = ymm2[1,0,3,2]
211 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm1
212 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
213 ; ALL-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
214 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[1,0,3,2]
215 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm2, %zmm0
217 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 9, i32 8, i32 1, i32 0, i32 13, i32 12, i32 5, i32 4>
218 ret <8 x double> %shuffle
221 define <8 x double> @shuffle_v8f64_08194c5d(<8 x double> %a, <8 x double> %b) {
222 ; ALL-LABEL: shuffle_v8f64_08194c5d:
224 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm2
225 ; ALL-NEXT: vpermpd {{.*#+}} ymm2 = ymm2[0,0,2,1]
226 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm3
227 ; ALL-NEXT: vpermpd {{.*#+}} ymm3 = ymm3[0,1,1,3]
228 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2],ymm2[3]
229 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[0,0,2,1]
230 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,1,1,3]
231 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3]
232 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
234 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
235 ret <8 x double> %shuffle
238 define <8 x double> @shuffle_v8f64_2a3b6e7f(<8 x double> %a, <8 x double> %b) {
239 ; ALL-LABEL: shuffle_v8f64_2a3b6e7f:
241 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm2
242 ; ALL-NEXT: vpermpd {{.*#+}} ymm2 = ymm2[0,2,2,3]
243 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm3
244 ; ALL-NEXT: vpermpd {{.*#+}} ymm3 = ymm3[2,1,3,3]
245 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2],ymm2[3]
246 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[0,2,2,3]
247 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,1,3,3]
248 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3]
249 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
251 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
252 ret <8 x double> %shuffle
255 define <8 x double> @shuffle_v8f64_08192a3b(<8 x double> %a, <8 x double> %b) {
256 ; ALL-LABEL: shuffle_v8f64_08192a3b:
258 ; ALL-NEXT: vpermpd {{.*#+}} ymm2 = ymm1[0,2,2,3]
259 ; ALL-NEXT: vpermpd {{.*#+}} ymm3 = ymm0[2,1,3,3]
260 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2],ymm2[3]
261 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[0,0,2,1]
262 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,1,1,3]
263 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3]
264 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
266 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11>
267 ret <8 x double> %shuffle
270 define <8 x double> @shuffle_v8f64_08991abb(<8 x double> %a, <8 x double> %b) {
271 ; ALL-LABEL: shuffle_v8f64_08991abb:
273 ; ALL-NEXT: vpermilpd {{.*#+}} ymm2 = ymm0[1,0,2,2]
274 ; ALL-NEXT: vpermpd {{.*#+}} ymm3 = ymm1[0,2,3,3]
275 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm2[0],ymm3[1,2,3]
276 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[0,0,1,1]
277 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3]
278 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
280 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 8, i32 9, i32 9, i32 1, i32 10, i32 11, i32 11>
281 ret <8 x double> %shuffle
284 define <8 x double> @shuffle_v8f64_091b2d3f(<8 x double> %a, <8 x double> %b) {
285 ; ALL-LABEL: shuffle_v8f64_091b2d3f:
287 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm2
288 ; ALL-NEXT: vpermpd {{.*#+}} ymm3 = ymm0[2,1,3,3]
289 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2],ymm2[3]
290 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,1,1,3]
291 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3]
292 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
294 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 9, i32 1, i32 11, i32 2, i32 13, i32 3, i32 15>
295 ret <8 x double> %shuffle
298 define <8 x double> @shuffle_v8f64_09ab1def(<8 x double> %a, <8 x double> %b) {
299 ; ALL-LABEL: shuffle_v8f64_09ab1def:
301 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm2
302 ; ALL-NEXT: vpermilpd {{.*#+}} ymm3 = ymm0[1,0,2,2]
303 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0],ymm2[1,2,3]
304 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3]
305 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
307 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 9, i32 10, i32 11, i32 1, i32 13, i32 14, i32 15>
308 ret <8 x double> %shuffle
311 define <8 x double> @shuffle_v8f64_00014445(<8 x double> %a, <8 x double> %b) {
312 ; ALL-LABEL: shuffle_v8f64_00014445:
314 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,0,0,1]
315 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
316 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,0,0,1]
317 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
319 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 1, i32 4, i32 4, i32 4, i32 5>
320 ret <8 x double> %shuffle
323 define <8 x double> @shuffle_v8f64_00204464(<8 x double> %a, <8 x double> %b) {
324 ; ALL-LABEL: shuffle_v8f64_00204464:
326 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,0,2,0]
327 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
328 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,0,2,0]
329 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
331 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 0, i32 4, i32 4, i32 6, i32 4>
332 ret <8 x double> %shuffle
335 define <8 x double> @shuffle_v8f64_03004744(<8 x double> %a, <8 x double> %b) {
336 ; ALL-LABEL: shuffle_v8f64_03004744:
338 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,3,0,0]
339 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
340 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,3,0,0]
341 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
343 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 3, i32 0, i32 0, i32 4, i32 7, i32 4, i32 4>
344 ret <8 x double> %shuffle
347 define <8 x double> @shuffle_v8f64_10005444(<8 x double> %a, <8 x double> %b) {
348 ; ALL-LABEL: shuffle_v8f64_10005444:
350 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[1,0,0,0]
351 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
352 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[1,0,0,0]
353 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
355 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 0, i32 0, i32 0, i32 5, i32 4, i32 4, i32 4>
356 ret <8 x double> %shuffle
359 define <8 x double> @shuffle_v8f64_22006644(<8 x double> %a, <8 x double> %b) {
360 ; ALL-LABEL: shuffle_v8f64_22006644:
362 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[2,2,0,0]
363 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
364 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,2,0,0]
365 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
367 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 2, i32 2, i32 0, i32 0, i32 6, i32 6, i32 4, i32 4>
368 ret <8 x double> %shuffle
371 define <8 x double> @shuffle_v8f64_33307774(<8 x double> %a, <8 x double> %b) {
372 ; ALL-LABEL: shuffle_v8f64_33307774:
374 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[3,3,3,0]
375 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
376 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[3,3,3,0]
377 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
379 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 3, i32 3, i32 3, i32 0, i32 7, i32 7, i32 7, i32 4>
380 ret <8 x double> %shuffle
383 define <8 x double> @shuffle_v8f64_32107654(<8 x double> %a, <8 x double> %b) {
384 ; ALL-LABEL: shuffle_v8f64_32107654:
386 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[3,2,1,0]
387 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
388 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[3,2,1,0]
389 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
391 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4>
392 ret <8 x double> %shuffle
395 define <8 x double> @shuffle_v8f64_00234467(<8 x double> %a, <8 x double> %b) {
396 ; ALL-LABEL: shuffle_v8f64_00234467:
398 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[0,0,2,3]
399 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
400 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[0,0,2,3]
401 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
403 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 3, i32 4, i32 4, i32 6, i32 7>
404 ret <8 x double> %shuffle
407 define <8 x double> @shuffle_v8f64_00224466(<8 x double> %a, <8 x double> %b) {
408 ; ALL-LABEL: shuffle_v8f64_00224466:
410 ; ALL-NEXT: vmovddup {{.*#+}} ymm1 = ymm0[0,0,2,2]
411 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
412 ; ALL-NEXT: vmovddup {{.*#+}} ymm0 = ymm0[0,0,2,2]
413 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
415 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 2, i32 4, i32 4, i32 6, i32 6>
416 ret <8 x double> %shuffle
419 define <8 x double> @shuffle_v8f64_10325476(<8 x double> %a, <8 x double> %b) {
420 ; ALL-LABEL: shuffle_v8f64_10325476:
422 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,0,3,2]
423 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
424 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[1,0,3,2]
425 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
427 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6>
428 ret <8 x double> %shuffle
431 define <8 x double> @shuffle_v8f64_11335577(<8 x double> %a, <8 x double> %b) {
432 ; ALL-LABEL: shuffle_v8f64_11335577:
434 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,1,3,3]
435 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
436 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[1,1,3,3]
437 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
439 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7>
440 ret <8 x double> %shuffle
443 define <8 x double> @shuffle_v8f64_10235467(<8 x double> %a, <8 x double> %b) {
444 ; ALL-LABEL: shuffle_v8f64_10235467:
446 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,0,2,3]
447 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
448 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[1,0,2,3]
449 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
451 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 0, i32 2, i32 3, i32 5, i32 4, i32 6, i32 7>
452 ret <8 x double> %shuffle
455 define <8 x double> @shuffle_v8f64_10225466(<8 x double> %a, <8 x double> %b) {
456 ; ALL-LABEL: shuffle_v8f64_10225466:
458 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,0,2,2]
459 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
460 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[1,0,2,2]
461 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
463 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 0, i32 2, i32 2, i32 5, i32 4, i32 6, i32 6>
464 ret <8 x double> %shuffle
467 define <8 x double> @shuffle_v8f64_00015444(<8 x double> %a, <8 x double> %b) {
468 ; ALL-LABEL: shuffle_v8f64_00015444:
470 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,0,0,1]
471 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
472 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[1,0,0,0]
473 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
475 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 1, i32 5, i32 4, i32 4, i32 4>
476 ret <8 x double> %shuffle
479 define <8 x double> @shuffle_v8f64_00204644(<8 x double> %a, <8 x double> %b) {
480 ; ALL-LABEL: shuffle_v8f64_00204644:
482 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,0,2,0]
483 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
484 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,0]
485 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
487 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 0, i32 4, i32 6, i32 4, i32 4>
488 ret <8 x double> %shuffle
491 define <8 x double> @shuffle_v8f64_03004474(<8 x double> %a, <8 x double> %b) {
492 ; ALL-LABEL: shuffle_v8f64_03004474:
494 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,3,0,0]
495 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
496 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,0,3,0]
497 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
499 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 3, i32 0, i32 0, i32 4, i32 4, i32 7, i32 4>
500 ret <8 x double> %shuffle
503 define <8 x double> @shuffle_v8f64_10004444(<8 x double> %a, <8 x double> %b) {
504 ; ALL-LABEL: shuffle_v8f64_10004444:
506 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[1,0,0,0]
507 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
508 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm0
509 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
511 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 0, i32 0, i32 0, i32 4, i32 4, i32 4, i32 4>
512 ret <8 x double> %shuffle
515 define <8 x double> @shuffle_v8f64_22006446(<8 x double> %a, <8 x double> %b) {
516 ; ALL-LABEL: shuffle_v8f64_22006446:
518 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[2,2,0,0]
519 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
520 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,0,0,2]
521 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
523 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 2, i32 2, i32 0, i32 0, i32 6, i32 4, i32 4, i32 6>
524 ret <8 x double> %shuffle
527 define <8 x double> @shuffle_v8f64_33307474(<8 x double> %a, <8 x double> %b) {
528 ; ALL-LABEL: shuffle_v8f64_33307474:
530 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[3,3,3,0]
531 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
532 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[3,0,3,0]
533 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
535 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 3, i32 3, i32 3, i32 0, i32 7, i32 4, i32 7, i32 4>
536 ret <8 x double> %shuffle
539 define <8 x double> @shuffle_v8f64_32104567(<8 x double> %a, <8 x double> %b) {
540 ; ALL-LABEL: shuffle_v8f64_32104567:
542 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[3,2,1,0]
543 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
544 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
546 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 4, i32 5, i32 6, i32 7>
547 ret <8 x double> %shuffle
550 define <8 x double> @shuffle_v8f64_00236744(<8 x double> %a, <8 x double> %b) {
551 ; ALL-LABEL: shuffle_v8f64_00236744:
553 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[0,0,2,3]
554 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
555 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,3,0,0]
556 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
558 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 3, i32 6, i32 7, i32 4, i32 4>
559 ret <8 x double> %shuffle
562 define <8 x double> @shuffle_v8f64_00226644(<8 x double> %a, <8 x double> %b) {
563 ; ALL-LABEL: shuffle_v8f64_00226644:
565 ; ALL-NEXT: vmovddup {{.*#+}} ymm1 = ymm0[0,0,2,2]
566 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
567 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,2,0,0]
568 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
570 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 2, i32 6, i32 6, i32 4, i32 4>
571 ret <8 x double> %shuffle
574 define <8 x double> @shuffle_v8f64_10324567(<8 x double> %a, <8 x double> %b) {
575 ; ALL-LABEL: shuffle_v8f64_10324567:
577 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,0,3,2]
578 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
579 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
581 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 4, i32 5, i32 6, i32 7>
582 ret <8 x double> %shuffle
585 define <8 x double> @shuffle_v8f64_11334567(<8 x double> %a, <8 x double> %b) {
586 ; ALL-LABEL: shuffle_v8f64_11334567:
588 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,1,3,3]
589 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
590 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
592 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 4, i32 5, i32 6, i32 7>
593 ret <8 x double> %shuffle
596 define <8 x double> @shuffle_v8f64_01235467(<8 x double> %a, <8 x double> %b) {
597 ; ALL-LABEL: shuffle_v8f64_01235467:
599 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1
600 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm1[1,0,2,3]
601 ; ALL-NEXT: vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
603 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 6, i32 7>
604 ret <8 x double> %shuffle
607 define <8 x double> @shuffle_v8f64_01235466(<8 x double> %a, <8 x double> %b) {
608 ; ALL-LABEL: shuffle_v8f64_01235466:
610 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1
611 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm1[1,0,2,2]
612 ; ALL-NEXT: vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
614 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 6, i32 6>
615 ret <8 x double> %shuffle
618 define <8 x double> @shuffle_v8f64_002u6u44(<8 x double> %a, <8 x double> %b) {
619 ; ALL-LABEL: shuffle_v8f64_002u6u44:
621 ; ALL-NEXT: vmovddup {{.*#+}} ymm1 = ymm0[0,0,2,2]
622 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
623 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,1,0,0]
624 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
626 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 undef, i32 6, i32 undef, i32 4, i32 4>
627 ret <8 x double> %shuffle
630 define <8 x double> @shuffle_v8f64_00uu66uu(<8 x double> %a, <8 x double> %b) {
631 ; ALL-LABEL: shuffle_v8f64_00uu66uu:
633 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm1
634 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
635 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,2,2,3]
636 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
638 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 undef, i32 undef, i32 6, i32 6, i32 undef, i32 undef>
639 ret <8 x double> %shuffle
642 define <8 x double> @shuffle_v8f64_103245uu(<8 x double> %a, <8 x double> %b) {
643 ; ALL-LABEL: shuffle_v8f64_103245uu:
645 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,0,3,2]
646 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
647 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
649 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 4, i32 5, i32 undef, i32 undef>
650 ret <8 x double> %shuffle
653 define <8 x double> @shuffle_v8f64_1133uu67(<8 x double> %a, <8 x double> %b) {
654 ; ALL-LABEL: shuffle_v8f64_1133uu67:
656 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,1,3,3]
657 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
658 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
660 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 undef, i32 undef, i32 6, i32 7>
661 ret <8 x double> %shuffle
664 define <8 x double> @shuffle_v8f64_0uu354uu(<8 x double> %a, <8 x double> %b) {
665 ; ALL-LABEL: shuffle_v8f64_0uu354uu:
667 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1
668 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm1[1,0,2,2]
669 ; ALL-NEXT: vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
671 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 undef, i32 undef, i32 3, i32 5, i32 4, i32 undef, i32 undef>
672 ret <8 x double> %shuffle
675 define <8 x double> @shuffle_v8f64_uuu3uu66(<8 x double> %a, <8 x double> %b) {
676 ; ALL-LABEL: shuffle_v8f64_uuu3uu66:
678 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1
679 ; ALL-NEXT: vmovddup {{.*#+}} ymm1 = ymm1[0,0,2,2]
680 ; ALL-NEXT: vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
682 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 undef, i32 undef, i32 undef, i32 3, i32 undef, i32 undef, i32 6, i32 6>
683 ret <8 x double> %shuffle
686 define <8 x double> @shuffle_v8f64_c348cda0(<8 x double> %a, <8 x double> %b) {
687 ; ALL-LABEL: shuffle_v8f64_c348cda0:
689 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm2
690 ; ALL-NEXT: vperm2f128 {{.*#+}} ymm2 = ymm0[0,1],ymm2[0,1]
691 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm3
692 ; ALL-NEXT: vbroadcastsd %xmm1, %ymm4
693 ; ALL-NEXT: vblendpd {{.*#+}} ymm4 = ymm3[0,1,2],ymm4[3]
694 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm4[0],ymm2[1,2],ymm4[3]
695 ; ALL-NEXT: vblendpd {{.*#+}} ymm1 = ymm3[0,1],ymm1[2],ymm3[3]
696 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm0
697 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3]
698 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm2, %zmm0
700 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 12, i32 3, i32 4, i32 8, i32 12, i32 13, i32 10, i32 0>
701 ret <8 x double> %shuffle
704 define <8 x double> @shuffle_v8f64_f511235a(<8 x double> %a, <8 x double> %b) {
705 ; ALL-LABEL: shuffle_v8f64_f511235a:
707 ; ALL-NEXT: vperm2f128 {{.*#+}} ymm2 = ymm0[2,3,0,1]
708 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm3
709 ; ALL-NEXT: vpermpd {{.*#+}} ymm4 = ymm3[0,1,1,3]
710 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm2[0,1],ymm4[2],ymm2[3]
711 ; ALL-NEXT: vmovddup {{.*#+}} ymm4 = ymm1[0,0,2,2]
712 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm2[0,1,2],ymm4[3]
713 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,1,1,1]
714 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm3[1],ymm0[2,3]
715 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm1
716 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[3,1,2,3]
717 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3]
718 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
720 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 15, i32 5, i32 1, i32 1, i32 2, i32 3, i32 5, i32 10>
721 ret <8 x double> %shuffle
724 define <8 x i64> @shuffle_v8i64_00000000(<8 x i64> %a, <8 x i64> %b) {
725 ; ALL-LABEL: shuffle_v8i64_00000000:
727 ; ALL-NEXT: vpbroadcastq %xmm0, %zmm0
729 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
730 ret <8 x i64> %shuffle
733 define <8 x i64> @shuffle_v8i64_00000010(<8 x i64> %a, <8 x i64> %b) {
734 ; ALL-LABEL: shuffle_v8i64_00000010:
736 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm1
737 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0]
738 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
740 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 0>
741 ret <8 x i64> %shuffle
744 define <8 x i64> @shuffle_v8i64_00000200(<8 x i64> %a, <8 x i64> %b) {
745 ; ALL-LABEL: shuffle_v8i64_00000200:
747 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm1
748 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,0,0]
749 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
751 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 2, i32 0, i32 0>
752 ret <8 x i64> %shuffle
755 define <8 x i64> @shuffle_v8i64_00003000(<8 x i64> %a, <8 x i64> %b) {
756 ; ALL-LABEL: shuffle_v8i64_00003000:
758 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm1
759 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[3,0,0,0]
760 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
762 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 3, i32 0, i32 0, i32 0>
763 ret <8 x i64> %shuffle
766 define <8 x i64> @shuffle_v8i64_00040000(<8 x i64> %a, <8 x i64> %b) {
767 ; ALL-LABEL: shuffle_v8i64_00040000:
769 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
770 ; ALL-NEXT: vpbroadcastq %xmm1, %ymm1
771 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm0
772 ; ALL-NEXT: vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3,4,5],ymm1[6,7]
773 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
775 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 4, i32 0, i32 0, i32 0, i32 0>
776 ret <8 x i64> %shuffle
779 define <8 x i64> @shuffle_v8i64_00500000(<8 x i64> %a, <8 x i64> %b) {
780 ; ALL-LABEL: shuffle_v8i64_00500000:
782 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
783 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,1,1,3]
784 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm0
785 ; ALL-NEXT: vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5],ymm0[6,7]
786 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
788 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 5, i32 0, i32 0, i32 0, i32 0, i32 0>
789 ret <8 x i64> %shuffle
792 define <8 x i64> @shuffle_v8i64_06000000(<8 x i64> %a, <8 x i64> %b) {
793 ; ALL-LABEL: shuffle_v8i64_06000000:
795 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
796 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
797 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm0
798 ; ALL-NEXT: vpblendd {{.*#+}} ymm1 = ymm0[0,1],ymm1[2,3],ymm0[4,5,6,7]
799 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
801 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 6, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
802 ret <8 x i64> %shuffle
805 define <8 x i64> @shuffle_v8i64_70000000(<8 x i64> %a, <8 x i64> %b) {
806 ; ALL-LABEL: shuffle_v8i64_70000000:
808 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
809 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[3,1,2,3]
810 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm0
811 ; ALL-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm0[2,3,4,5,6,7]
812 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
814 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 7, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
815 ret <8 x i64> %shuffle
818 define <8 x i64> @shuffle_v8i64_01014545(<8 x i64> %a, <8 x i64> %b) {
819 ; ALL-LABEL: shuffle_v8i64_01014545:
821 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
822 ; ALL-NEXT: vinserti128 $1, %xmm1, %ymm1, %ymm1
823 ; ALL-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
824 ; ALL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
826 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 4, i32 5, i32 4, i32 5>
827 ret <8 x i64> %shuffle
830 define <8 x i64> @shuffle_v8i64_00112233(<8 x i64> %a, <8 x i64> %b) {
831 ; ALL-LABEL: shuffle_v8i64_00112233:
833 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,1,1]
834 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,2,3,3]
835 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
837 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 1, i32 1, i32 2, i32 2, i32 3, i32 3>
838 ret <8 x i64> %shuffle
841 define <8 x i64> @shuffle_v8i64_00001111(<8 x i64> %a, <8 x i64> %b) {
842 ; ALL-LABEL: shuffle_v8i64_00001111:
844 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm1
845 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[1,1,1,1]
846 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
848 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1>
849 ret <8 x i64> %shuffle
852 define <8 x i64> @shuffle_v8i64_81a3c5e7(<8 x i64> %a, <8 x i64> %b) {
853 ; ALL-LABEL: shuffle_v8i64_81a3c5e7:
855 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm2
856 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm3
857 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7]
858 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
859 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
861 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 8, i32 1, i32 10, i32 3, i32 12, i32 5, i32 14, i32 7>
862 ret <8 x i64> %shuffle
865 define <8 x i64> @shuffle_v8i64_08080808(<8 x i64> %a, <8 x i64> %b) {
866 ; ALL-LABEL: shuffle_v8i64_08080808:
868 ; ALL-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
869 ; ALL-NEXT: vpbroadcastq %xmm1, %ymm1
870 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
871 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0
873 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 8, i32 0, i32 8, i32 0, i32 8, i32 0, i32 8>
874 ret <8 x i64> %shuffle
877 define <8 x i64> @shuffle_v8i64_08084c4c(<8 x i64> %a, <8 x i64> %b) {
878 ; ALL-LABEL: shuffle_v8i64_08084c4c:
880 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm2
881 ; ALL-NEXT: vinserti128 $1, %xmm2, %ymm2, %ymm2
882 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm3
883 ; ALL-NEXT: vpbroadcastq %xmm3, %ymm3
884 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1],ymm3[2,3],ymm2[4,5],ymm3[6,7]
885 ; ALL-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
886 ; ALL-NEXT: vpbroadcastq %xmm1, %ymm1
887 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
888 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
890 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 8, i32 0, i32 8, i32 4, i32 12, i32 4, i32 12>
891 ret <8 x i64> %shuffle
894 define <8 x i64> @shuffle_v8i64_8823cc67(<8 x i64> %a, <8 x i64> %b) {
895 ; ALL-LABEL: shuffle_v8i64_8823cc67:
897 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm2
898 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm3
899 ; ALL-NEXT: vpbroadcastq %xmm3, %ymm3
900 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
901 ; ALL-NEXT: vpbroadcastq %xmm1, %ymm1
902 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
903 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
905 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 8, i32 8, i32 2, i32 3, i32 12, i32 12, i32 6, i32 7>
906 ret <8 x i64> %shuffle
909 define <8 x i64> @shuffle_v8i64_9832dc76(<8 x i64> %a, <8 x i64> %b) {
910 ; ALL-LABEL: shuffle_v8i64_9832dc76:
912 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm0[4,5,6,7]
913 ; ALL-NEXT: vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
914 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
915 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm1
916 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
917 ; ALL-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5]
918 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm2, %zmm0
920 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 9, i32 8, i32 3, i32 2, i32 13, i32 12, i32 7, i32 6>
921 ret <8 x i64> %shuffle
924 define <8 x i64> @shuffle_v8i64_9810dc54(<8 x i64> %a, <8 x i64> %b) {
925 ; ALL-LABEL: shuffle_v8i64_9810dc54:
927 ; ALL-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm2
928 ; ALL-NEXT: vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
929 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm1
930 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
931 ; ALL-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
932 ; ALL-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5]
933 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm2, %zmm0
935 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 9, i32 8, i32 1, i32 0, i32 13, i32 12, i32 5, i32 4>
936 ret <8 x i64> %shuffle
939 define <8 x i64> @shuffle_v8i64_08194c5d(<8 x i64> %a, <8 x i64> %b) {
940 ; ALL-LABEL: shuffle_v8i64_08194c5d:
942 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm2
943 ; ALL-NEXT: vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
944 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm3
945 ; ALL-NEXT: vpermq {{.*#+}} ymm3 = ymm3[0,1,1,3]
946 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7]
947 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
948 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,1,1,3]
949 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
950 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
952 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
953 ret <8 x i64> %shuffle
956 define <8 x i64> @shuffle_v8i64_2a3b6e7f(<8 x i64> %a, <8 x i64> %b) {
957 ; ALL-LABEL: shuffle_v8i64_2a3b6e7f:
959 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm2
960 ; ALL-NEXT: vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
961 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm3
962 ; ALL-NEXT: vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
963 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7]
964 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
965 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
966 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
967 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
969 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
970 ret <8 x i64> %shuffle
973 define <8 x i64> @shuffle_v8i64_08192a3b(<8 x i64> %a, <8 x i64> %b) {
974 ; ALL-LABEL: shuffle_v8i64_08192a3b:
976 ; ALL-NEXT: vpermq {{.*#+}} ymm2 = ymm1[0,2,2,3]
977 ; ALL-NEXT: vpermq {{.*#+}} ymm3 = ymm0[2,1,3,3]
978 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7]
979 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
980 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,1,1,3]
981 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
982 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
984 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11>
985 ret <8 x i64> %shuffle
988 define <8 x i64> @shuffle_v8i64_08991abb(<8 x i64> %a, <8 x i64> %b) {
989 ; ALL-LABEL: shuffle_v8i64_08991abb:
991 ; ALL-NEXT: vpshufd {{.*#+}} ymm2 = ymm0[2,3,2,3,6,7,6,7]
992 ; ALL-NEXT: vpermq {{.*#+}} ymm3 = ymm1[0,2,3,3]
993 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1],ymm3[2,3,4,5,6,7]
994 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,0,1,1]
995 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
996 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
998 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 8, i32 9, i32 9, i32 1, i32 10, i32 11, i32 11>
999 ret <8 x i64> %shuffle
1002 define <8 x i64> @shuffle_v8i64_091b2d3f(<8 x i64> %a, <8 x i64> %b) {
1003 ; ALL-LABEL: shuffle_v8i64_091b2d3f:
1005 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm2
1006 ; ALL-NEXT: vpermq {{.*#+}} ymm3 = ymm0[2,1,3,3]
1007 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7]
1008 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,1,1,3]
1009 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
1010 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
1012 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 9, i32 1, i32 11, i32 2, i32 13, i32 3, i32 15>
1013 ret <8 x i64> %shuffle
1016 define <8 x i64> @shuffle_v8i64_09ab1def(<8 x i64> %a, <8 x i64> %b) {
1017 ; ALL-LABEL: shuffle_v8i64_09ab1def:
1019 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm2
1020 ; ALL-NEXT: vpshufd {{.*#+}} ymm3 = ymm0[2,3,2,3,6,7,6,7]
1021 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3,4,5,6,7]
1022 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
1023 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
1025 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 9, i32 10, i32 11, i32 1, i32 13, i32 14, i32 15>
1026 ret <8 x i64> %shuffle
1029 define <8 x i64> @shuffle_v8i64_00014445(<8 x i64> %a, <8 x i64> %b) {
1030 ; ALL-LABEL: shuffle_v8i64_00014445:
1032 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,0,1]
1033 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1034 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
1035 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1037 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 1, i32 4, i32 4, i32 4, i32 5>
1038 ret <8 x i64> %shuffle
1041 define <8 x i64> @shuffle_v8i64_00204464(<8 x i64> %a, <8 x i64> %b) {
1042 ; ALL-LABEL: shuffle_v8i64_00204464:
1044 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,2,0]
1045 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1046 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,0,2,0]
1047 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1049 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 0, i32 4, i32 4, i32 6, i32 4>
1050 ret <8 x i64> %shuffle
1053 define <8 x i64> @shuffle_v8i64_03004744(<8 x i64> %a, <8 x i64> %b) {
1054 ; ALL-LABEL: shuffle_v8i64_03004744:
1056 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,3,0,0]
1057 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1058 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,0,0]
1059 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1061 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 3, i32 0, i32 0, i32 4, i32 7, i32 4, i32 4>
1062 ret <8 x i64> %shuffle
1065 define <8 x i64> @shuffle_v8i64_10005444(<8 x i64> %a, <8 x i64> %b) {
1066 ; ALL-LABEL: shuffle_v8i64_10005444:
1068 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[1,0,0,0]
1069 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1070 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[1,0,0,0]
1071 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1073 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 0, i32 0, i32 0, i32 5, i32 4, i32 4, i32 4>
1074 ret <8 x i64> %shuffle
1077 define <8 x i64> @shuffle_v8i64_22006644(<8 x i64> %a, <8 x i64> %b) {
1078 ; ALL-LABEL: shuffle_v8i64_22006644:
1080 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,2,0,0]
1081 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1082 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,2,0,0]
1083 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1085 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 2, i32 2, i32 0, i32 0, i32 6, i32 6, i32 4, i32 4>
1086 ret <8 x i64> %shuffle
1089 define <8 x i64> @shuffle_v8i64_33307774(<8 x i64> %a, <8 x i64> %b) {
1090 ; ALL-LABEL: shuffle_v8i64_33307774:
1092 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[3,3,3,0]
1093 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1094 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[3,3,3,0]
1095 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1097 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 3, i32 3, i32 3, i32 0, i32 7, i32 7, i32 7, i32 4>
1098 ret <8 x i64> %shuffle
1101 define <8 x i64> @shuffle_v8i64_32107654(<8 x i64> %a, <8 x i64> %b) {
1102 ; ALL-LABEL: shuffle_v8i64_32107654:
1104 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[3,2,1,0]
1105 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1106 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[3,2,1,0]
1107 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1109 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4>
1110 ret <8 x i64> %shuffle
1113 define <8 x i64> @shuffle_v8i64_00234467(<8 x i64> %a, <8 x i64> %b) {
1114 ; ALL-LABEL: shuffle_v8i64_00234467:
1116 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,2,3]
1117 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1118 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,0,2,3]
1119 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1121 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 3, i32 4, i32 4, i32 6, i32 7>
1122 ret <8 x i64> %shuffle
1125 define <8 x i64> @shuffle_v8i64_00224466(<8 x i64> %a, <8 x i64> %b) {
1126 ; ALL-LABEL: shuffle_v8i64_00224466:
1128 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[0,1,0,1,4,5,4,5]
1129 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1130 ; ALL-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[0,1,0,1,4,5,4,5]
1131 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1133 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 2, i32 4, i32 4, i32 6, i32 6>
1134 ret <8 x i64> %shuffle
1137 define <8 x i64> @shuffle_v8i64_10325476(<8 x i64> %a, <8 x i64> %b) {
1138 ; ALL-LABEL: shuffle_v8i64_10325476:
1140 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[2,3,0,1,6,7,4,5]
1141 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1142 ; ALL-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5]
1143 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1145 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6>
1146 ret <8 x i64> %shuffle
1149 define <8 x i64> @shuffle_v8i64_11335577(<8 x i64> %a, <8 x i64> %b) {
1150 ; ALL-LABEL: shuffle_v8i64_11335577:
1152 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[2,3,2,3,6,7,6,7]
1153 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1154 ; ALL-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[2,3,2,3,6,7,6,7]
1155 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1157 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7>
1158 ret <8 x i64> %shuffle
1161 define <8 x i64> @shuffle_v8i64_10235467(<8 x i64> %a, <8 x i64> %b) {
1162 ; ALL-LABEL: shuffle_v8i64_10235467:
1164 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[1,0,2,3]
1165 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1166 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[1,0,2,3]
1167 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1169 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 0, i32 2, i32 3, i32 5, i32 4, i32 6, i32 7>
1170 ret <8 x i64> %shuffle
1173 define <8 x i64> @shuffle_v8i64_10225466(<8 x i64> %a, <8 x i64> %b) {
1174 ; ALL-LABEL: shuffle_v8i64_10225466:
1176 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[1,0,2,2]
1177 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1178 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[1,0,2,2]
1179 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1181 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 0, i32 2, i32 2, i32 5, i32 4, i32 6, i32 6>
1182 ret <8 x i64> %shuffle
1185 define <8 x i64> @shuffle_v8i64_00015444(<8 x i64> %a, <8 x i64> %b) {
1186 ; ALL-LABEL: shuffle_v8i64_00015444:
1188 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,0,1]
1189 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1190 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[1,0,0,0]
1191 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1193 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 1, i32 5, i32 4, i32 4, i32 4>
1194 ret <8 x i64> %shuffle
1197 define <8 x i64> @shuffle_v8i64_00204644(<8 x i64> %a, <8 x i64> %b) {
1198 ; ALL-LABEL: shuffle_v8i64_00204644:
1200 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,2,0]
1201 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1202 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,0,0]
1203 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1205 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 0, i32 4, i32 6, i32 4, i32 4>
1206 ret <8 x i64> %shuffle
1209 define <8 x i64> @shuffle_v8i64_03004474(<8 x i64> %a, <8 x i64> %b) {
1210 ; ALL-LABEL: shuffle_v8i64_03004474:
1212 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,3,0,0]
1213 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1214 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,0,3,0]
1215 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1217 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 3, i32 0, i32 0, i32 4, i32 4, i32 7, i32 4>
1218 ret <8 x i64> %shuffle
1221 define <8 x i64> @shuffle_v8i64_10004444(<8 x i64> %a, <8 x i64> %b) {
1222 ; ALL-LABEL: shuffle_v8i64_10004444:
1224 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[1,0,0,0]
1225 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1226 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm0
1227 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1229 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 0, i32 0, i32 0, i32 4, i32 4, i32 4, i32 4>
1230 ret <8 x i64> %shuffle
1233 define <8 x i64> @shuffle_v8i64_22006446(<8 x i64> %a, <8 x i64> %b) {
1234 ; ALL-LABEL: shuffle_v8i64_22006446:
1236 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,2,0,0]
1237 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1238 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,0,0,2]
1239 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1241 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 2, i32 2, i32 0, i32 0, i32 6, i32 4, i32 4, i32 6>
1242 ret <8 x i64> %shuffle
1245 define <8 x i64> @shuffle_v8i64_33307474(<8 x i64> %a, <8 x i64> %b) {
1246 ; ALL-LABEL: shuffle_v8i64_33307474:
1248 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[3,3,3,0]
1249 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1250 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[3,0,3,0]
1251 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1253 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 3, i32 3, i32 3, i32 0, i32 7, i32 4, i32 7, i32 4>
1254 ret <8 x i64> %shuffle
1257 define <8 x i64> @shuffle_v8i64_32104567(<8 x i64> %a, <8 x i64> %b) {
1258 ; ALL-LABEL: shuffle_v8i64_32104567:
1260 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[3,2,1,0]
1261 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1262 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1264 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 4, i32 5, i32 6, i32 7>
1265 ret <8 x i64> %shuffle
1268 define <8 x i64> @shuffle_v8i64_00236744(<8 x i64> %a, <8 x i64> %b) {
1269 ; ALL-LABEL: shuffle_v8i64_00236744:
1271 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,2,3]
1272 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1273 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,3,0,0]
1274 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1276 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 3, i32 6, i32 7, i32 4, i32 4>
1277 ret <8 x i64> %shuffle
1280 define <8 x i64> @shuffle_v8i64_00226644(<8 x i64> %a, <8 x i64> %b) {
1281 ; ALL-LABEL: shuffle_v8i64_00226644:
1283 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[0,1,0,1,4,5,4,5]
1284 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1285 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,2,0,0]
1286 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1288 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 2, i32 6, i32 6, i32 4, i32 4>
1289 ret <8 x i64> %shuffle
1292 define <8 x i64> @shuffle_v8i64_10324567(<8 x i64> %a, <8 x i64> %b) {
1293 ; ALL-LABEL: shuffle_v8i64_10324567:
1295 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[2,3,0,1,6,7,4,5]
1296 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1297 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1299 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 4, i32 5, i32 6, i32 7>
1300 ret <8 x i64> %shuffle
1303 define <8 x i64> @shuffle_v8i64_11334567(<8 x i64> %a, <8 x i64> %b) {
1304 ; ALL-LABEL: shuffle_v8i64_11334567:
1306 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[2,3,2,3,6,7,6,7]
1307 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1308 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1310 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 4, i32 5, i32 6, i32 7>
1311 ret <8 x i64> %shuffle
1314 define <8 x i64> @shuffle_v8i64_01235467(<8 x i64> %a, <8 x i64> %b) {
1315 ; ALL-LABEL: shuffle_v8i64_01235467:
1317 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
1318 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[1,0,2,3]
1319 ; ALL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
1321 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 6, i32 7>
1322 ret <8 x i64> %shuffle
1325 define <8 x i64> @shuffle_v8i64_01235466(<8 x i64> %a, <8 x i64> %b) {
1326 ; ALL-LABEL: shuffle_v8i64_01235466:
1328 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
1329 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[1,0,2,2]
1330 ; ALL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
1332 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 6, i32 6>
1333 ret <8 x i64> %shuffle
1336 define <8 x i64> @shuffle_v8i64_002u6u44(<8 x i64> %a, <8 x i64> %b) {
1337 ; ALL-LABEL: shuffle_v8i64_002u6u44:
1339 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[0,1,0,1,4,5,4,5]
1340 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1341 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,1,0,0]
1342 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1344 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 undef, i32 6, i32 undef, i32 4, i32 4>
1345 ret <8 x i64> %shuffle
1348 define <8 x i64> @shuffle_v8i64_00uu66uu(<8 x i64> %a, <8 x i64> %b) {
1349 ; ALL-LABEL: shuffle_v8i64_00uu66uu:
1351 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm1
1352 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1353 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
1354 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1356 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 undef, i32 undef, i32 6, i32 6, i32 undef, i32 undef>
1357 ret <8 x i64> %shuffle
1360 define <8 x i64> @shuffle_v8i64_103245uu(<8 x i64> %a, <8 x i64> %b) {
1361 ; ALL-LABEL: shuffle_v8i64_103245uu:
1363 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[2,3,0,1,6,7,4,5]
1364 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1365 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1367 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 4, i32 5, i32 undef, i32 undef>
1368 ret <8 x i64> %shuffle
1371 define <8 x i64> @shuffle_v8i64_1133uu67(<8 x i64> %a, <8 x i64> %b) {
1372 ; ALL-LABEL: shuffle_v8i64_1133uu67:
1374 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[2,3,2,3,6,7,6,7]
1375 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1376 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1378 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 undef, i32 undef, i32 6, i32 7>
1379 ret <8 x i64> %shuffle
1382 define <8 x i64> @shuffle_v8i64_0uu354uu(<8 x i64> %a, <8 x i64> %b) {
1383 ; ALL-LABEL: shuffle_v8i64_0uu354uu:
1385 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
1386 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[2,3,0,1,6,7,4,5]
1387 ; ALL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
1389 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 undef, i32 undef, i32 3, i32 5, i32 4, i32 undef, i32 undef>
1390 ret <8 x i64> %shuffle
1393 define <8 x i64> @shuffle_v8i64_uuu3uu66(<8 x i64> %a, <8 x i64> %b) {
1394 ; ALL-LABEL: shuffle_v8i64_uuu3uu66:
1396 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
1397 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[0,1,0,1,4,5,4,5]
1398 ; ALL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
1400 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 undef, i32 undef, i32 undef, i32 3, i32 undef, i32 undef, i32 6, i32 6>
1401 ret <8 x i64> %shuffle
1404 define <8 x i64> @shuffle_v8i64_6caa87e5(<8 x i64> %a, <8 x i64> %b) {
1405 ; ALL-LABEL: shuffle_v8i64_6caa87e5:
1407 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1408 ; ALL-NEXT: vperm2i128 {{.*#+}} ymm2 = ymm0[0,1,0,1]
1409 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm3
1410 ; ALL-NEXT: vpblendd {{.*#+}} ymm4 = ymm1[0,1,2,3],ymm3[4,5],ymm1[6,7]
1411 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm2[2,3],ymm4[4,5],ymm2[6,7]
1412 ; ALL-NEXT: vperm2i128 {{.*#+}} ymm0 = ymm0[2,3,0,1]
1413 ; ALL-NEXT: vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
1414 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[0,1,0,1,4,5,4,5]
1415 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
1416 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
1418 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 6, i32 12, i32 10, i32 10, i32 8, i32 7, i32 14, i32 5>
1419 ret <8 x i64> %shuffle
1422 define <8 x double> @shuffle_v8f64_082a4c6e(<8 x double> %a, <8 x double> %b) {
1423 ; ALL-LABEL: shuffle_v8f64_082a4c6e:
1425 ; ALL-NEXT: vunpcklpd {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[2],zmm1[2],zmm0[4],zmm1[4],zmm0[6],zmm1[6]
1427 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32><i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
1428 ret <8 x double> %shuffle
1431 define <8 x i64> @shuffle_v8i64_082a4c6e(<8 x i64> %a, <8 x i64> %b) {
1432 ; ALL-LABEL: shuffle_v8i64_082a4c6e:
1434 ; ALL-NEXT: vpunpcklqdq {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[2],zmm1[2],zmm0[4],zmm1[4],zmm0[6],zmm1[6]
1436 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32><i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
1437 ret <8 x i64> %shuffle
1440 define <8 x double> @shuffle_v8f64_193b5d7f(<8 x double> %a, <8 x double> %b) {
1441 ; ALL-LABEL: shuffle_v8f64_193b5d7f:
1443 ; ALL-NEXT: vunpckhpd {{.*#+}} zmm0 = zmm0[1],zmm1[1],zmm0[3],zmm1[3],zmm0[5],zmm1[5],zmm0[7],zmm1[7]
1445 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32><i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
1446 ret <8 x double> %shuffle
1449 define <8 x i64> @shuffle_v8i64_193b5d7f(<8 x i64> %a, <8 x i64> %b) {
1450 ; ALL-LABEL: shuffle_v8i64_193b5d7f:
1452 ; ALL-NEXT: vpunpckhqdq {{.*#+}} zmm0 = zmm0[1],zmm1[1],zmm0[3],zmm1[3],zmm0[5],zmm1[5],zmm0[7],zmm1[7]
1454 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32><i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
1455 ret <8 x i64> %shuffle