1 ; RUN: llc < %s -mcpu=x86-64 -mattr=+avx512f -x86-experimental-vector-shuffle-lowering | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512F
2 ; RUN: llc < %s -mcpu=x86-64 -mattr=+avx512bw -x86-experimental-vector-shuffle-lowering | 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: vextractf64x4 $1, %zmm0, %ymm2
195 ; ALL-NEXT: vpermilpd {{.*#+}} ymm2 = ymm2[0,0,3,2]
196 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm3
197 ; ALL-NEXT: vpermilpd {{.*#+}} ymm3 = ymm3[1,0,2,2]
198 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3]
199 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[0,0,3,2]
200 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm1[1,0,2,2]
201 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3]
202 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
204 %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>
205 ret <8 x double> %shuffle
208 define <8 x double> @shuffle_v8f64_9810dc54(<8 x double> %a, <8 x double> %b) {
209 ; ALL-LABEL: shuffle_v8f64_9810dc54:
211 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm2
212 ; ALL-NEXT: vpermpd {{.*#+}} ymm2 = ymm2[0,1,1,0]
213 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm3
214 ; ALL-NEXT: vpermilpd {{.*#+}} ymm3 = ymm3[1,0,2,2]
215 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3]
216 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,1,1,0]
217 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm1[1,0,2,2]
218 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3]
219 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
221 %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>
222 ret <8 x double> %shuffle
225 define <8 x double> @shuffle_v8f64_08194c5d(<8 x double> %a, <8 x double> %b) {
226 ; ALL-LABEL: shuffle_v8f64_08194c5d:
228 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm2
229 ; ALL-NEXT: vpermpd {{.*#+}} ymm2 = ymm2[0,0,2,1]
230 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm3
231 ; ALL-NEXT: vpermpd {{.*#+}} ymm3 = ymm3[0,1,1,3]
232 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2],ymm2[3]
233 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[0,0,2,1]
234 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,1,1,3]
235 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3]
236 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
238 %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>
239 ret <8 x double> %shuffle
242 define <8 x double> @shuffle_v8f64_2a3b6e7f(<8 x double> %a, <8 x double> %b) {
243 ; ALL-LABEL: shuffle_v8f64_2a3b6e7f:
245 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm2
246 ; ALL-NEXT: vpermpd {{.*#+}} ymm2 = ymm2[0,2,2,3]
247 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm3
248 ; ALL-NEXT: vpermpd {{.*#+}} ymm3 = ymm3[2,1,3,3]
249 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2],ymm2[3]
250 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[0,2,2,3]
251 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,1,3,3]
252 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3]
253 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
255 %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>
256 ret <8 x double> %shuffle
259 define <8 x double> @shuffle_v8f64_08192a3b(<8 x double> %a, <8 x double> %b) {
260 ; ALL-LABEL: shuffle_v8f64_08192a3b:
262 ; ALL-NEXT: vpermpd {{.*#+}} ymm2 = ymm1[0,2,2,3]
263 ; ALL-NEXT: vpermpd {{.*#+}} ymm3 = ymm0[2,1,3,3]
264 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2],ymm2[3]
265 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[0,0,2,1]
266 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,1,1,3]
267 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3]
268 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
270 %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>
271 ret <8 x double> %shuffle
274 define <8 x double> @shuffle_v8f64_08991abb(<8 x double> %a, <8 x double> %b) {
275 ; ALL-LABEL: shuffle_v8f64_08991abb:
277 ; ALL-NEXT: vpermilpd {{.*#+}} ymm2 = ymm0[1,0,2,2]
278 ; ALL-NEXT: vpermpd {{.*#+}} ymm3 = ymm1[0,2,3,3]
279 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm2[0],ymm3[1,2,3]
280 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[0,0,1,1]
281 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3]
282 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
284 %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>
285 ret <8 x double> %shuffle
288 define <8 x double> @shuffle_v8f64_091b2d3f(<8 x double> %a, <8 x double> %b) {
289 ; ALL-LABEL: shuffle_v8f64_091b2d3f:
291 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm2
292 ; ALL-NEXT: vpermpd {{.*#+}} ymm3 = ymm0[2,1,3,3]
293 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2],ymm2[3]
294 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,1,1,3]
295 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3]
296 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
298 %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>
299 ret <8 x double> %shuffle
302 define <8 x double> @shuffle_v8f64_09ab1def(<8 x double> %a, <8 x double> %b) {
303 ; ALL-LABEL: shuffle_v8f64_09ab1def:
305 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm2
306 ; ALL-NEXT: vpermilpd {{.*#+}} ymm3 = ymm0[1,0,2,2]
307 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0],ymm2[1,2,3]
308 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3]
309 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
311 %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>
312 ret <8 x double> %shuffle
315 define <8 x double> @shuffle_v8f64_00014445(<8 x double> %a, <8 x double> %b) {
316 ; ALL-LABEL: shuffle_v8f64_00014445:
318 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,0,0,1]
319 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
320 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,0,0,1]
321 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
323 %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>
324 ret <8 x double> %shuffle
327 define <8 x double> @shuffle_v8f64_00204464(<8 x double> %a, <8 x double> %b) {
328 ; ALL-LABEL: shuffle_v8f64_00204464:
330 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,0,2,0]
331 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
332 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,0,2,0]
333 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
335 %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>
336 ret <8 x double> %shuffle
339 define <8 x double> @shuffle_v8f64_03004744(<8 x double> %a, <8 x double> %b) {
340 ; ALL-LABEL: shuffle_v8f64_03004744:
342 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,3,0,0]
343 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
344 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,3,0,0]
345 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
347 %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>
348 ret <8 x double> %shuffle
351 define <8 x double> @shuffle_v8f64_10005444(<8 x double> %a, <8 x double> %b) {
352 ; ALL-LABEL: shuffle_v8f64_10005444:
354 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[1,0,0,0]
355 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
356 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[1,0,0,0]
357 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
359 %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>
360 ret <8 x double> %shuffle
363 define <8 x double> @shuffle_v8f64_22006644(<8 x double> %a, <8 x double> %b) {
364 ; ALL-LABEL: shuffle_v8f64_22006644:
366 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[2,2,0,0]
367 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
368 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,2,0,0]
369 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
371 %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>
372 ret <8 x double> %shuffle
375 define <8 x double> @shuffle_v8f64_33307774(<8 x double> %a, <8 x double> %b) {
376 ; ALL-LABEL: shuffle_v8f64_33307774:
378 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[3,3,3,0]
379 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
380 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[3,3,3,0]
381 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
383 %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>
384 ret <8 x double> %shuffle
387 define <8 x double> @shuffle_v8f64_32107654(<8 x double> %a, <8 x double> %b) {
388 ; ALL-LABEL: shuffle_v8f64_32107654:
390 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[3,2,1,0]
391 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
392 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[3,2,1,0]
393 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
395 %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>
396 ret <8 x double> %shuffle
399 define <8 x double> @shuffle_v8f64_00234467(<8 x double> %a, <8 x double> %b) {
400 ; ALL-LABEL: shuffle_v8f64_00234467:
402 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[0,0,2,3]
403 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
404 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[0,0,2,3]
405 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
407 %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>
408 ret <8 x double> %shuffle
411 define <8 x double> @shuffle_v8f64_00224466(<8 x double> %a, <8 x double> %b) {
412 ; ALL-LABEL: shuffle_v8f64_00224466:
414 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[0,0,2,2]
415 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
416 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[0,0,2,2]
417 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
419 %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>
420 ret <8 x double> %shuffle
423 define <8 x double> @shuffle_v8f64_10325476(<8 x double> %a, <8 x double> %b) {
424 ; ALL-LABEL: shuffle_v8f64_10325476:
426 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,0,3,2]
427 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
428 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[1,0,3,2]
429 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
431 %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>
432 ret <8 x double> %shuffle
435 define <8 x double> @shuffle_v8f64_11335577(<8 x double> %a, <8 x double> %b) {
436 ; ALL-LABEL: shuffle_v8f64_11335577:
438 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,1,3,3]
439 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
440 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[1,1,3,3]
441 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
443 %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>
444 ret <8 x double> %shuffle
447 define <8 x double> @shuffle_v8f64_10235467(<8 x double> %a, <8 x double> %b) {
448 ; ALL-LABEL: shuffle_v8f64_10235467:
450 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,0,2,3]
451 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
452 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[1,0,2,3]
453 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
455 %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>
456 ret <8 x double> %shuffle
459 define <8 x double> @shuffle_v8f64_10225466(<8 x double> %a, <8 x double> %b) {
460 ; ALL-LABEL: shuffle_v8f64_10225466:
462 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,0,2,2]
463 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
464 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[1,0,2,2]
465 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
467 %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>
468 ret <8 x double> %shuffle
471 define <8 x double> @shuffle_v8f64_00015444(<8 x double> %a, <8 x double> %b) {
472 ; ALL-LABEL: shuffle_v8f64_00015444:
474 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,0,0,1]
475 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
476 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[1,0,0,0]
477 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
479 %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>
480 ret <8 x double> %shuffle
483 define <8 x double> @shuffle_v8f64_00204644(<8 x double> %a, <8 x double> %b) {
484 ; ALL-LABEL: shuffle_v8f64_00204644:
486 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,0,2,0]
487 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
488 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,0]
489 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
491 %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>
492 ret <8 x double> %shuffle
495 define <8 x double> @shuffle_v8f64_03004474(<8 x double> %a, <8 x double> %b) {
496 ; ALL-LABEL: shuffle_v8f64_03004474:
498 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,3,0,0]
499 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
500 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,0,3,0]
501 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
503 %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>
504 ret <8 x double> %shuffle
507 define <8 x double> @shuffle_v8f64_10004444(<8 x double> %a, <8 x double> %b) {
508 ; ALL-LABEL: shuffle_v8f64_10004444:
510 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[1,0,0,0]
511 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
512 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm0
513 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
515 %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>
516 ret <8 x double> %shuffle
519 define <8 x double> @shuffle_v8f64_22006446(<8 x double> %a, <8 x double> %b) {
520 ; ALL-LABEL: shuffle_v8f64_22006446:
522 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[2,2,0,0]
523 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
524 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,0,0,2]
525 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
527 %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>
528 ret <8 x double> %shuffle
531 define <8 x double> @shuffle_v8f64_33307474(<8 x double> %a, <8 x double> %b) {
532 ; ALL-LABEL: shuffle_v8f64_33307474:
534 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[3,3,3,0]
535 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
536 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[3,0,3,0]
537 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
539 %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>
540 ret <8 x double> %shuffle
543 define <8 x double> @shuffle_v8f64_32104567(<8 x double> %a, <8 x double> %b) {
544 ; ALL-LABEL: shuffle_v8f64_32104567:
546 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[3,2,1,0]
547 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
548 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
550 %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>
551 ret <8 x double> %shuffle
554 define <8 x double> @shuffle_v8f64_00236744(<8 x double> %a, <8 x double> %b) {
555 ; ALL-LABEL: shuffle_v8f64_00236744:
557 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[0,0,2,3]
558 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
559 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,3,0,0]
560 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
562 %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>
563 ret <8 x double> %shuffle
566 define <8 x double> @shuffle_v8f64_00226644(<8 x double> %a, <8 x double> %b) {
567 ; ALL-LABEL: shuffle_v8f64_00226644:
569 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[0,0,2,2]
570 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
571 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,2,0,0]
572 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
574 %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>
575 ret <8 x double> %shuffle
578 define <8 x double> @shuffle_v8f64_10324567(<8 x double> %a, <8 x double> %b) {
579 ; ALL-LABEL: shuffle_v8f64_10324567:
581 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,0,3,2]
582 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
583 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
585 %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>
586 ret <8 x double> %shuffle
589 define <8 x double> @shuffle_v8f64_11334567(<8 x double> %a, <8 x double> %b) {
590 ; ALL-LABEL: shuffle_v8f64_11334567:
592 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,1,3,3]
593 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
594 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
596 %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>
597 ret <8 x double> %shuffle
600 define <8 x double> @shuffle_v8f64_01235467(<8 x double> %a, <8 x double> %b) {
601 ; ALL-LABEL: shuffle_v8f64_01235467:
603 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1
604 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm1[1,0,2,3]
605 ; ALL-NEXT: vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
607 %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>
608 ret <8 x double> %shuffle
611 define <8 x double> @shuffle_v8f64_01235466(<8 x double> %a, <8 x double> %b) {
612 ; ALL-LABEL: shuffle_v8f64_01235466:
614 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1
615 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm1[1,0,2,2]
616 ; ALL-NEXT: vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
618 %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>
619 ret <8 x double> %shuffle
622 define <8 x double> @shuffle_v8f64_002u6u44(<8 x double> %a, <8 x double> %b) {
623 ; ALL-LABEL: shuffle_v8f64_002u6u44:
625 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[0,0,2,2]
626 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
627 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,1,0,0]
628 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
630 %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>
631 ret <8 x double> %shuffle
634 define <8 x double> @shuffle_v8f64_00uu66uu(<8 x double> %a, <8 x double> %b) {
635 ; ALL-LABEL: shuffle_v8f64_00uu66uu:
637 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm1
638 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
639 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,2,2,3]
640 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
642 %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>
643 ret <8 x double> %shuffle
646 define <8 x double> @shuffle_v8f64_103245uu(<8 x double> %a, <8 x double> %b) {
647 ; ALL-LABEL: shuffle_v8f64_103245uu:
649 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,0,3,2]
650 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
651 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
653 %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>
654 ret <8 x double> %shuffle
657 define <8 x double> @shuffle_v8f64_1133uu67(<8 x double> %a, <8 x double> %b) {
658 ; ALL-LABEL: shuffle_v8f64_1133uu67:
660 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,1,3,3]
661 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0
662 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
664 %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>
665 ret <8 x double> %shuffle
668 define <8 x double> @shuffle_v8f64_0uu354uu(<8 x double> %a, <8 x double> %b) {
669 ; ALL-LABEL: shuffle_v8f64_0uu354uu:
671 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1
672 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm1[1,0,2,2]
673 ; ALL-NEXT: vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
675 %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>
676 ret <8 x double> %shuffle
679 define <8 x double> @shuffle_v8f64_uuu3uu66(<8 x double> %a, <8 x double> %b) {
680 ; ALL-LABEL: shuffle_v8f64_uuu3uu66:
682 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1
683 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm1[0,0,2,2]
684 ; ALL-NEXT: vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
686 %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>
687 ret <8 x double> %shuffle
690 define <8 x double> @shuffle_v8f64_c348cda0(<8 x double> %a, <8 x double> %b) {
691 ; ALL-LABEL: shuffle_v8f64_c348cda0:
693 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm2
694 ; ALL-NEXT: vperm2f128 {{.*#+}} ymm2 = ymm0[0,1],ymm2[0,1]
695 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm3
696 ; ALL-NEXT: vbroadcastsd %xmm1, %ymm4
697 ; ALL-NEXT: vblendpd {{.*#+}} ymm4 = ymm3[0,1,2],ymm4[3]
698 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm4[0],ymm2[1,2],ymm4[3]
699 ; ALL-NEXT: vblendpd {{.*#+}} ymm1 = ymm3[0,1],ymm1[2],ymm3[3]
700 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm0
701 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3]
702 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm2, %zmm0
704 %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>
705 ret <8 x double> %shuffle
708 define <8 x double> @shuffle_v8f64_f511235a(<8 x double> %a, <8 x double> %b) {
709 ; ALL-LABEL: shuffle_v8f64_f511235a:
711 ; ALL-NEXT: vperm2f128 {{.*#+}} ymm2 = ymm0[2,3,0,1]
712 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm3
713 ; ALL-NEXT: vpermpd {{.*#+}} ymm4 = ymm3[0,1,1,3]
714 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm2[0,1],ymm4[2],ymm2[3]
715 ; ALL-NEXT: vpermilpd {{.*#+}} ymm4 = ymm1[0,0,2,2]
716 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm2[0,1,2],ymm4[3]
717 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,1,1,1]
718 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm3[1],ymm0[2,3]
719 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm1
720 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[3,1,2,3]
721 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3]
722 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
724 %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>
725 ret <8 x double> %shuffle
728 define <8 x i64> @shuffle_v8i64_00000000(<8 x i64> %a, <8 x i64> %b) {
729 ; ALL-LABEL: shuffle_v8i64_00000000:
731 ; ALL-NEXT: vpbroadcastq %xmm0, %zmm0
733 %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>
734 ret <8 x i64> %shuffle
737 define <8 x i64> @shuffle_v8i64_00000010(<8 x i64> %a, <8 x i64> %b) {
738 ; ALL-LABEL: shuffle_v8i64_00000010:
740 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm1
741 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0]
742 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
744 %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>
745 ret <8 x i64> %shuffle
748 define <8 x i64> @shuffle_v8i64_00000200(<8 x i64> %a, <8 x i64> %b) {
749 ; ALL-LABEL: shuffle_v8i64_00000200:
751 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm1
752 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,0,0]
753 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
755 %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>
756 ret <8 x i64> %shuffle
759 define <8 x i64> @shuffle_v8i64_00003000(<8 x i64> %a, <8 x i64> %b) {
760 ; ALL-LABEL: shuffle_v8i64_00003000:
762 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm1
763 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[3,0,0,0]
764 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
766 %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>
767 ret <8 x i64> %shuffle
770 define <8 x i64> @shuffle_v8i64_00040000(<8 x i64> %a, <8 x i64> %b) {
771 ; ALL-LABEL: shuffle_v8i64_00040000:
773 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
774 ; ALL-NEXT: vpbroadcastq %xmm1, %ymm1
775 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm0
776 ; ALL-NEXT: vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3,4,5],ymm1[6,7]
777 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
779 %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>
780 ret <8 x i64> %shuffle
783 define <8 x i64> @shuffle_v8i64_00500000(<8 x i64> %a, <8 x i64> %b) {
784 ; ALL-LABEL: shuffle_v8i64_00500000:
786 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
787 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,1,1,3]
788 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm0
789 ; ALL-NEXT: vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5],ymm0[6,7]
790 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
792 %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>
793 ret <8 x i64> %shuffle
796 define <8 x i64> @shuffle_v8i64_06000000(<8 x i64> %a, <8 x i64> %b) {
797 ; ALL-LABEL: shuffle_v8i64_06000000:
799 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
800 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
801 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm0
802 ; ALL-NEXT: vpblendd {{.*#+}} ymm1 = ymm0[0,1],ymm1[2,3],ymm0[4,5,6,7]
803 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
805 %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>
806 ret <8 x i64> %shuffle
809 define <8 x i64> @shuffle_v8i64_70000000(<8 x i64> %a, <8 x i64> %b) {
810 ; ALL-LABEL: shuffle_v8i64_70000000:
812 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
813 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[3,1,2,3]
814 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm0
815 ; ALL-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm0[2,3,4,5,6,7]
816 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
818 %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>
819 ret <8 x i64> %shuffle
822 define <8 x i64> @shuffle_v8i64_01014545(<8 x i64> %a, <8 x i64> %b) {
823 ; ALL-LABEL: shuffle_v8i64_01014545:
825 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
826 ; ALL-NEXT: vinserti128 $1, %xmm1, %ymm1, %ymm1
827 ; ALL-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
828 ; ALL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
830 %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>
831 ret <8 x i64> %shuffle
834 define <8 x i64> @shuffle_v8i64_00112233(<8 x i64> %a, <8 x i64> %b) {
835 ; ALL-LABEL: shuffle_v8i64_00112233:
837 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,1,1]
838 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,2,3,3]
839 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
841 %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>
842 ret <8 x i64> %shuffle
845 define <8 x i64> @shuffle_v8i64_00001111(<8 x i64> %a, <8 x i64> %b) {
846 ; ALL-LABEL: shuffle_v8i64_00001111:
848 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm1
849 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[1,1,1,1]
850 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
852 %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>
853 ret <8 x i64> %shuffle
856 define <8 x i64> @shuffle_v8i64_81a3c5e7(<8 x i64> %a, <8 x i64> %b) {
857 ; ALL-LABEL: shuffle_v8i64_81a3c5e7:
859 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm2
860 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm3
861 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7]
862 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
863 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
865 %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>
866 ret <8 x i64> %shuffle
869 define <8 x i64> @shuffle_v8i64_08080808(<8 x i64> %a, <8 x i64> %b) {
870 ; ALL-LABEL: shuffle_v8i64_08080808:
872 ; ALL-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
873 ; ALL-NEXT: vpbroadcastq %xmm1, %ymm1
874 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
875 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0
877 %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>
878 ret <8 x i64> %shuffle
881 define <8 x i64> @shuffle_v8i64_08084c4c(<8 x i64> %a, <8 x i64> %b) {
882 ; ALL-LABEL: shuffle_v8i64_08084c4c:
884 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm2
885 ; ALL-NEXT: vinserti128 $1, %xmm2, %ymm2, %ymm2
886 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm3
887 ; ALL-NEXT: vpbroadcastq %xmm3, %ymm3
888 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1],ymm3[2,3],ymm2[4,5],ymm3[6,7]
889 ; ALL-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
890 ; ALL-NEXT: vpbroadcastq %xmm1, %ymm1
891 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
892 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
894 %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>
895 ret <8 x i64> %shuffle
898 define <8 x i64> @shuffle_v8i64_8823cc67(<8 x i64> %a, <8 x i64> %b) {
899 ; ALL-LABEL: shuffle_v8i64_8823cc67:
901 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm2
902 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm3
903 ; ALL-NEXT: vpbroadcastq %xmm3, %ymm3
904 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
905 ; ALL-NEXT: vpbroadcastq %xmm1, %ymm1
906 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
907 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
909 %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>
910 ret <8 x i64> %shuffle
913 define <8 x i64> @shuffle_v8i64_9832dc76(<8 x i64> %a, <8 x i64> %b) {
914 ; ALL-LABEL: shuffle_v8i64_9832dc76:
916 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm2
917 ; ALL-NEXT: vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
918 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm3
919 ; ALL-NEXT: vpshufd {{.*#+}} ymm3 = ymm3[2,3,0,1,6,7,4,5]
920 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
921 ; ALL-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5]
922 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[2,3,0,1,6,7,4,5]
923 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
924 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
926 %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>
927 ret <8 x i64> %shuffle
930 define <8 x i64> @shuffle_v8i64_9810dc54(<8 x i64> %a, <8 x i64> %b) {
931 ; ALL-LABEL: shuffle_v8i64_9810dc54:
933 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm2
934 ; ALL-NEXT: vpermq {{.*#+}} ymm2 = ymm2[0,1,1,0]
935 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm3
936 ; ALL-NEXT: vpshufd {{.*#+}} ymm3 = ymm3[2,3,0,1,6,7,4,5]
937 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
938 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,1,1,0]
939 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[2,3,0,1,6,7,4,5]
940 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
941 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
943 %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>
944 ret <8 x i64> %shuffle
947 define <8 x i64> @shuffle_v8i64_08194c5d(<8 x i64> %a, <8 x i64> %b) {
948 ; ALL-LABEL: shuffle_v8i64_08194c5d:
950 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm2
951 ; ALL-NEXT: vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
952 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm3
953 ; ALL-NEXT: vpermq {{.*#+}} ymm3 = ymm3[0,1,1,3]
954 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7]
955 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
956 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,1,1,3]
957 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
958 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
960 %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>
961 ret <8 x i64> %shuffle
964 define <8 x i64> @shuffle_v8i64_2a3b6e7f(<8 x i64> %a, <8 x i64> %b) {
965 ; ALL-LABEL: shuffle_v8i64_2a3b6e7f:
967 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm2
968 ; ALL-NEXT: vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
969 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm3
970 ; ALL-NEXT: vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
971 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7]
972 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
973 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
974 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
975 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
977 %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>
978 ret <8 x i64> %shuffle
981 define <8 x i64> @shuffle_v8i64_08192a3b(<8 x i64> %a, <8 x i64> %b) {
982 ; ALL-LABEL: shuffle_v8i64_08192a3b:
984 ; ALL-NEXT: vpermq {{.*#+}} ymm2 = ymm1[0,2,2,3]
985 ; ALL-NEXT: vpermq {{.*#+}} ymm3 = ymm0[2,1,3,3]
986 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7]
987 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
988 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,1,1,3]
989 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
990 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
992 %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>
993 ret <8 x i64> %shuffle
996 define <8 x i64> @shuffle_v8i64_08991abb(<8 x i64> %a, <8 x i64> %b) {
997 ; ALL-LABEL: shuffle_v8i64_08991abb:
999 ; ALL-NEXT: vpshufd {{.*#+}} ymm2 = ymm0[2,3,2,3,6,7,6,7]
1000 ; ALL-NEXT: vpermq {{.*#+}} ymm3 = ymm1[0,2,3,3]
1001 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1],ymm3[2,3,4,5,6,7]
1002 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,0,1,1]
1003 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
1004 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
1006 %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>
1007 ret <8 x i64> %shuffle
1010 define <8 x i64> @shuffle_v8i64_091b2d3f(<8 x i64> %a, <8 x i64> %b) {
1011 ; ALL-LABEL: shuffle_v8i64_091b2d3f:
1013 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm2
1014 ; ALL-NEXT: vpermq {{.*#+}} ymm3 = ymm0[2,1,3,3]
1015 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7]
1016 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,1,1,3]
1017 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
1018 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
1020 %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>
1021 ret <8 x i64> %shuffle
1024 define <8 x i64> @shuffle_v8i64_09ab1def(<8 x i64> %a, <8 x i64> %b) {
1025 ; ALL-LABEL: shuffle_v8i64_09ab1def:
1027 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm2
1028 ; ALL-NEXT: vpshufd {{.*#+}} ymm3 = ymm0[2,3,2,3,6,7,6,7]
1029 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3,4,5,6,7]
1030 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
1031 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
1033 %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>
1034 ret <8 x i64> %shuffle
1037 define <8 x i64> @shuffle_v8i64_00014445(<8 x i64> %a, <8 x i64> %b) {
1038 ; ALL-LABEL: shuffle_v8i64_00014445:
1040 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,0,1]
1041 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1042 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
1043 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1045 %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>
1046 ret <8 x i64> %shuffle
1049 define <8 x i64> @shuffle_v8i64_00204464(<8 x i64> %a, <8 x i64> %b) {
1050 ; ALL-LABEL: shuffle_v8i64_00204464:
1052 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,2,0]
1053 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1054 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,0,2,0]
1055 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1057 %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>
1058 ret <8 x i64> %shuffle
1061 define <8 x i64> @shuffle_v8i64_03004744(<8 x i64> %a, <8 x i64> %b) {
1062 ; ALL-LABEL: shuffle_v8i64_03004744:
1064 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,3,0,0]
1065 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1066 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,0,0]
1067 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1069 %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>
1070 ret <8 x i64> %shuffle
1073 define <8 x i64> @shuffle_v8i64_10005444(<8 x i64> %a, <8 x i64> %b) {
1074 ; ALL-LABEL: shuffle_v8i64_10005444:
1076 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[1,0,0,0]
1077 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1078 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[1,0,0,0]
1079 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1081 %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>
1082 ret <8 x i64> %shuffle
1085 define <8 x i64> @shuffle_v8i64_22006644(<8 x i64> %a, <8 x i64> %b) {
1086 ; ALL-LABEL: shuffle_v8i64_22006644:
1088 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,2,0,0]
1089 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1090 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,2,0,0]
1091 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1093 %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>
1094 ret <8 x i64> %shuffle
1097 define <8 x i64> @shuffle_v8i64_33307774(<8 x i64> %a, <8 x i64> %b) {
1098 ; ALL-LABEL: shuffle_v8i64_33307774:
1100 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[3,3,3,0]
1101 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1102 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[3,3,3,0]
1103 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1105 %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>
1106 ret <8 x i64> %shuffle
1109 define <8 x i64> @shuffle_v8i64_32107654(<8 x i64> %a, <8 x i64> %b) {
1110 ; ALL-LABEL: shuffle_v8i64_32107654:
1112 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[3,2,1,0]
1113 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1114 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[3,2,1,0]
1115 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1117 %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>
1118 ret <8 x i64> %shuffle
1121 define <8 x i64> @shuffle_v8i64_00234467(<8 x i64> %a, <8 x i64> %b) {
1122 ; ALL-LABEL: shuffle_v8i64_00234467:
1124 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,2,3]
1125 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1126 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,0,2,3]
1127 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1129 %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>
1130 ret <8 x i64> %shuffle
1133 define <8 x i64> @shuffle_v8i64_00224466(<8 x i64> %a, <8 x i64> %b) {
1134 ; ALL-LABEL: shuffle_v8i64_00224466:
1136 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[0,1,0,1,4,5,4,5]
1137 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1138 ; ALL-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[0,1,0,1,4,5,4,5]
1139 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1141 %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>
1142 ret <8 x i64> %shuffle
1145 define <8 x i64> @shuffle_v8i64_10325476(<8 x i64> %a, <8 x i64> %b) {
1146 ; ALL-LABEL: shuffle_v8i64_10325476:
1148 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[2,3,0,1,6,7,4,5]
1149 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1150 ; ALL-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5]
1151 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1153 %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>
1154 ret <8 x i64> %shuffle
1157 define <8 x i64> @shuffle_v8i64_11335577(<8 x i64> %a, <8 x i64> %b) {
1158 ; ALL-LABEL: shuffle_v8i64_11335577:
1160 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[2,3,2,3,6,7,6,7]
1161 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1162 ; ALL-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[2,3,2,3,6,7,6,7]
1163 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1165 %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>
1166 ret <8 x i64> %shuffle
1169 define <8 x i64> @shuffle_v8i64_10235467(<8 x i64> %a, <8 x i64> %b) {
1170 ; ALL-LABEL: shuffle_v8i64_10235467:
1172 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[1,0,2,3]
1173 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1174 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[1,0,2,3]
1175 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1177 %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>
1178 ret <8 x i64> %shuffle
1181 define <8 x i64> @shuffle_v8i64_10225466(<8 x i64> %a, <8 x i64> %b) {
1182 ; ALL-LABEL: shuffle_v8i64_10225466:
1184 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[1,0,2,2]
1185 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1186 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[1,0,2,2]
1187 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1189 %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>
1190 ret <8 x i64> %shuffle
1193 define <8 x i64> @shuffle_v8i64_00015444(<8 x i64> %a, <8 x i64> %b) {
1194 ; ALL-LABEL: shuffle_v8i64_00015444:
1196 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,0,1]
1197 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1198 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[1,0,0,0]
1199 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1201 %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>
1202 ret <8 x i64> %shuffle
1205 define <8 x i64> @shuffle_v8i64_00204644(<8 x i64> %a, <8 x i64> %b) {
1206 ; ALL-LABEL: shuffle_v8i64_00204644:
1208 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,2,0]
1209 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1210 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,0,0]
1211 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1213 %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>
1214 ret <8 x i64> %shuffle
1217 define <8 x i64> @shuffle_v8i64_03004474(<8 x i64> %a, <8 x i64> %b) {
1218 ; ALL-LABEL: shuffle_v8i64_03004474:
1220 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,3,0,0]
1221 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1222 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,0,3,0]
1223 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1225 %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>
1226 ret <8 x i64> %shuffle
1229 define <8 x i64> @shuffle_v8i64_10004444(<8 x i64> %a, <8 x i64> %b) {
1230 ; ALL-LABEL: shuffle_v8i64_10004444:
1232 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[1,0,0,0]
1233 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1234 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm0
1235 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1237 %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>
1238 ret <8 x i64> %shuffle
1241 define <8 x i64> @shuffle_v8i64_22006446(<8 x i64> %a, <8 x i64> %b) {
1242 ; ALL-LABEL: shuffle_v8i64_22006446:
1244 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,2,0,0]
1245 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1246 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,0,0,2]
1247 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1249 %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>
1250 ret <8 x i64> %shuffle
1253 define <8 x i64> @shuffle_v8i64_33307474(<8 x i64> %a, <8 x i64> %b) {
1254 ; ALL-LABEL: shuffle_v8i64_33307474:
1256 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[3,3,3,0]
1257 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1258 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[3,0,3,0]
1259 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1261 %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>
1262 ret <8 x i64> %shuffle
1265 define <8 x i64> @shuffle_v8i64_32104567(<8 x i64> %a, <8 x i64> %b) {
1266 ; ALL-LABEL: shuffle_v8i64_32104567:
1268 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[3,2,1,0]
1269 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1270 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1272 %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>
1273 ret <8 x i64> %shuffle
1276 define <8 x i64> @shuffle_v8i64_00236744(<8 x i64> %a, <8 x i64> %b) {
1277 ; ALL-LABEL: shuffle_v8i64_00236744:
1279 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,2,3]
1280 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1281 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,3,0,0]
1282 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1284 %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>
1285 ret <8 x i64> %shuffle
1288 define <8 x i64> @shuffle_v8i64_00226644(<8 x i64> %a, <8 x i64> %b) {
1289 ; ALL-LABEL: shuffle_v8i64_00226644:
1291 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[0,1,0,1,4,5,4,5]
1292 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1293 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,2,0,0]
1294 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1296 %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>
1297 ret <8 x i64> %shuffle
1300 define <8 x i64> @shuffle_v8i64_10324567(<8 x i64> %a, <8 x i64> %b) {
1301 ; ALL-LABEL: shuffle_v8i64_10324567:
1303 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[2,3,0,1,6,7,4,5]
1304 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1305 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1307 %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>
1308 ret <8 x i64> %shuffle
1311 define <8 x i64> @shuffle_v8i64_11334567(<8 x i64> %a, <8 x i64> %b) {
1312 ; ALL-LABEL: shuffle_v8i64_11334567:
1314 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[2,3,2,3,6,7,6,7]
1315 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1316 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1318 %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>
1319 ret <8 x i64> %shuffle
1322 define <8 x i64> @shuffle_v8i64_01235467(<8 x i64> %a, <8 x i64> %b) {
1323 ; ALL-LABEL: shuffle_v8i64_01235467:
1325 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
1326 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[1,0,2,3]
1327 ; ALL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
1329 %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>
1330 ret <8 x i64> %shuffle
1333 define <8 x i64> @shuffle_v8i64_01235466(<8 x i64> %a, <8 x i64> %b) {
1334 ; ALL-LABEL: shuffle_v8i64_01235466:
1336 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
1337 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[1,0,2,2]
1338 ; ALL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
1340 %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>
1341 ret <8 x i64> %shuffle
1344 define <8 x i64> @shuffle_v8i64_002u6u44(<8 x i64> %a, <8 x i64> %b) {
1345 ; ALL-LABEL: shuffle_v8i64_002u6u44:
1347 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[0,1,0,1,4,5,4,5]
1348 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1349 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,1,0,0]
1350 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1352 %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>
1353 ret <8 x i64> %shuffle
1356 define <8 x i64> @shuffle_v8i64_00uu66uu(<8 x i64> %a, <8 x i64> %b) {
1357 ; ALL-LABEL: shuffle_v8i64_00uu66uu:
1359 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm1
1360 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1361 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
1362 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1364 %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>
1365 ret <8 x i64> %shuffle
1368 define <8 x i64> @shuffle_v8i64_103245uu(<8 x i64> %a, <8 x i64> %b) {
1369 ; ALL-LABEL: shuffle_v8i64_103245uu:
1371 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[2,3,0,1,6,7,4,5]
1372 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1373 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1375 %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>
1376 ret <8 x i64> %shuffle
1379 define <8 x i64> @shuffle_v8i64_1133uu67(<8 x i64> %a, <8 x i64> %b) {
1380 ; ALL-LABEL: shuffle_v8i64_1133uu67:
1382 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[2,3,2,3,6,7,6,7]
1383 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1384 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
1386 %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>
1387 ret <8 x i64> %shuffle
1390 define <8 x i64> @shuffle_v8i64_0uu354uu(<8 x i64> %a, <8 x i64> %b) {
1391 ; ALL-LABEL: shuffle_v8i64_0uu354uu:
1393 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
1394 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[2,3,0,1,6,7,4,5]
1395 ; ALL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
1397 %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>
1398 ret <8 x i64> %shuffle
1401 define <8 x i64> @shuffle_v8i64_uuu3uu66(<8 x i64> %a, <8 x i64> %b) {
1402 ; ALL-LABEL: shuffle_v8i64_uuu3uu66:
1404 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
1405 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[0,1,0,1,4,5,4,5]
1406 ; ALL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
1408 %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>
1409 ret <8 x i64> %shuffle
1412 define <8 x i64> @shuffle_v8i64_6caa87e5(<8 x i64> %a, <8 x i64> %b) {
1413 ; ALL-LABEL: shuffle_v8i64_6caa87e5:
1415 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
1416 ; ALL-NEXT: vperm2i128 {{.*#+}} ymm2 = ymm0[0,1,0,1]
1417 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm3
1418 ; ALL-NEXT: vpblendd {{.*#+}} ymm4 = ymm1[0,1,2,3],ymm3[4,5],ymm1[6,7]
1419 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm2[2,3],ymm4[4,5],ymm2[6,7]
1420 ; ALL-NEXT: vperm2i128 {{.*#+}} ymm0 = ymm0[2,3,0,1]
1421 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[0,1,0,1,4,5,4,5]
1422 ; ALL-NEXT: vpbroadcastq %xmm3, %ymm3
1423 ; ALL-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm3[2,3],ymm1[4,5,6,7]
1424 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
1425 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
1427 %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>
1428 ret <8 x i64> %shuffle