1 ; RUN: llc < %s -mcpu=x86-64 -mattr=+avx -x86-experimental-vector-shuffle-lowering | FileCheck %s --check-prefix=ALL --check-prefix=AVX1
2 ; RUN: llc < %s -mcpu=x86-64 -mattr=+avx2 -x86-experimental-vector-shuffle-lowering | FileCheck %s --check-prefix=ALL --check-prefix=AVX2
4 target triple = "x86_64-unknown-unknown"
6 define <8 x float> @shuffle_v8f32_00000000(<8 x float> %a, <8 x float> %b) {
7 ; AVX1-LABEL: shuffle_v8f32_00000000:
9 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
10 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
13 ; AVX2-LABEL: shuffle_v8f32_00000000:
15 ; AVX2-NEXT: vbroadcastss %xmm0, %ymm0
17 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
18 ret <8 x float> %shuffle
21 define <8 x float> @shuffle_v8f32_00000010(<8 x float> %a, <8 x float> %b) {
22 ; AVX1-LABEL: shuffle_v8f32_00000010:
24 ; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[0,0,0,0]
25 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,1,0]
26 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
29 ; AVX2-LABEL: shuffle_v8f32_00000010:
31 ; AVX2-NEXT: vmovaps {{.*#+}} ymm1 = [0,0,0,0,0,0,1,0]
32 ; AVX2-NEXT: vpermps %ymm0, %ymm1, %ymm0
34 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 0>
35 ret <8 x float> %shuffle
38 define <8 x float> @shuffle_v8f32_00000200(<8 x float> %a, <8 x float> %b) {
39 ; AVX1-LABEL: shuffle_v8f32_00000200:
41 ; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[0,0,0,0]
42 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,2,0,0]
43 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
46 ; AVX2-LABEL: shuffle_v8f32_00000200:
48 ; AVX2-NEXT: vmovaps {{.*#+}} ymm1 = [0,0,0,0,0,2,0,0]
49 ; AVX2-NEXT: vpermps %ymm0, %ymm1, %ymm0
51 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 2, i32 0, i32 0>
52 ret <8 x float> %shuffle
55 define <8 x float> @shuffle_v8f32_00003000(<8 x float> %a, <8 x float> %b) {
56 ; AVX1-LABEL: shuffle_v8f32_00003000:
58 ; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[0,0,0,0]
59 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,0,0,0]
60 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
63 ; AVX2-LABEL: shuffle_v8f32_00003000:
65 ; AVX2-NEXT: vmovaps {{.*#+}} ymm1 = [0,0,0,0,3,0,0,0]
66 ; AVX2-NEXT: vpermps %ymm0, %ymm1, %ymm0
68 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 3, i32 0, i32 0, i32 0>
69 ret <8 x float> %shuffle
72 define <8 x float> @shuffle_v8f32_00040000(<8 x float> %a, <8 x float> %b) {
73 ; AVX1-LABEL: shuffle_v8f32_00040000:
75 ; AVX1-NEXT: vperm2f128 {{.*#+}} ymm1 = ymm0[2,3,0,1]
76 ; AVX1-NEXT: vpermilps {{.*#+}} ymm1 = ymm1[0,0,0,0,4,4,4,4]
77 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,0,3,4,4,4,7]
78 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
81 ; AVX2-LABEL: shuffle_v8f32_00040000:
83 ; AVX2-NEXT: vmovaps {{.*#+}} ymm1 = [0,0,0,4,0,0,0,0]
84 ; AVX2-NEXT: vpermps %ymm0, %ymm1, %ymm0
86 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 4, i32 0, i32 0, i32 0, i32 0>
87 ret <8 x float> %shuffle
90 define <8 x float> @shuffle_v8f32_00500000(<8 x float> %a, <8 x float> %b) {
91 ; AVX1-LABEL: shuffle_v8f32_00500000:
93 ; AVX1-NEXT: vperm2f128 {{.*#+}} ymm1 = ymm0[2,3,0,1]
94 ; AVX1-NEXT: vpermilps {{.*#+}} ymm1 = ymm1[u,u,1,u,4,4,4,4]
95 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,2,0,4,4,6,4]
96 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3],ymm1[4,5,6,7]
99 ; AVX2-LABEL: shuffle_v8f32_00500000:
101 ; AVX2-NEXT: vmovaps {{.*#+}} ymm1 = [0,0,5,0,0,0,0,0]
102 ; AVX2-NEXT: vpermps %ymm0, %ymm1, %ymm0
104 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 5, i32 0, i32 0, i32 0, i32 0, i32 0>
105 ret <8 x float> %shuffle
108 define <8 x float> @shuffle_v8f32_06000000(<8 x float> %a, <8 x float> %b) {
109 ; AVX1-LABEL: shuffle_v8f32_06000000:
111 ; AVX1-NEXT: vperm2f128 {{.*#+}} ymm1 = ymm0[2,3,0,1]
112 ; AVX1-NEXT: vpermilps {{.*#+}} ymm1 = ymm1[u,2,u,u,4,4,4,4]
113 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,1,0,0,4,5,4,4]
114 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4,5,6,7]
117 ; AVX2-LABEL: shuffle_v8f32_06000000:
119 ; AVX2-NEXT: vmovaps {{.*#+}} ymm1 = [0,6,0,0,0,0,0,0]
120 ; AVX2-NEXT: vpermps %ymm0, %ymm1, %ymm0
122 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 6, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
123 ret <8 x float> %shuffle
126 define <8 x float> @shuffle_v8f32_70000000(<8 x float> %a, <8 x float> %b) {
127 ; AVX1-LABEL: shuffle_v8f32_70000000:
129 ; AVX1-NEXT: vperm2f128 {{.*#+}} ymm1 = ymm0[2,3,0,1]
130 ; AVX1-NEXT: vpermilps {{.*#+}} ymm1 = ymm1[3,u,u,u,4,4,4,4]
131 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,0,0,4,4,4,4]
132 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3],ymm1[4,5,6,7]
135 ; AVX2-LABEL: shuffle_v8f32_70000000:
137 ; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
138 ; AVX2-NEXT: movl $7, %eax
139 ; AVX2-NEXT: vpinsrd $0, %eax, %xmm1, %xmm1
140 ; AVX2-NEXT: vpxor %ymm2, %ymm2, %ymm2
141 ; AVX2-NEXT: vinserti128 $0, %xmm1, %ymm2, %ymm1
142 ; AVX2-NEXT: vpermps %ymm0, %ymm1, %ymm0
144 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 7, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
145 ret <8 x float> %shuffle
148 define <8 x float> @shuffle_v8f32_01014545(<8 x float> %a, <8 x float> %b) {
149 ; ALL-LABEL: shuffle_v8f32_01014545:
151 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[0,0,2,2]
153 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 4, i32 5, i32 4, i32 5>
154 ret <8 x float> %shuffle
157 define <8 x float> @shuffle_v8f32_00112233(<8 x float> %a, <8 x float> %b) {
158 ; AVX1-LABEL: shuffle_v8f32_00112233:
160 ; AVX1-NEXT: vunpcklps {{.*#+}} xmm1 = xmm0[0,0,1,1]
161 ; AVX1-NEXT: vunpckhps {{.*#+}} xmm0 = xmm0[2,2,3,3]
162 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
165 ; AVX2-LABEL: shuffle_v8f32_00112233:
167 ; AVX2-NEXT: vmovaps {{.*#+}} ymm1 = [0,0,1,1,2,2,3,3]
168 ; AVX2-NEXT: vpermps %ymm0, %ymm1, %ymm0
170 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 1, i32 1, i32 2, i32 2, i32 3, i32 3>
171 ret <8 x float> %shuffle
174 define <8 x float> @shuffle_v8f32_00001111(<8 x float> %a, <8 x float> %b) {
175 ; AVX1-LABEL: shuffle_v8f32_00001111:
177 ; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[0,0,0,0]
178 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[1,1,1,1]
179 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
182 ; AVX2-LABEL: shuffle_v8f32_00001111:
184 ; AVX2-NEXT: vmovaps {{.*#+}} ymm1 = [0,0,0,0,1,1,1,1]
185 ; AVX2-NEXT: vpermps %ymm0, %ymm1, %ymm0
187 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1>
188 ret <8 x float> %shuffle
191 define <8 x float> @shuffle_v8f32_81a3c5e7(<8 x float> %a, <8 x float> %b) {
192 ; ALL-LABEL: shuffle_v8f32_81a3c5e7:
194 ; ALL-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4],ymm0[5],ymm1[6],ymm0[7]
196 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 8, i32 1, i32 10, i32 3, i32 12, i32 5, i32 14, i32 7>
197 ret <8 x float> %shuffle
200 define <8 x float> @shuffle_v8f32_08080808(<8 x float> %a, <8 x float> %b) {
201 ; AVX1-LABEL: shuffle_v8f32_08080808:
203 ; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm1[0,0,2,0]
204 ; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm1, %ymm1
205 ; AVX1-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0,0]
206 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
207 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
210 ; AVX2-LABEL: shuffle_v8f32_08080808:
212 ; AVX2-NEXT: vbroadcastss %xmm1, %ymm1
213 ; AVX2-NEXT: vbroadcastsd %xmm0, %ymm0
214 ; AVX2-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
216 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 8, i32 0, i32 8, i32 0, i32 8, i32 0, i32 8>
217 ret <8 x float> %shuffle
220 define <8 x float> @shuffle_v8f32_08084c4c(<8 x float> %a, <8 x float> %b) {
221 ; ALL-LABEL: shuffle_v8f32_08084c4c:
223 ; ALL-NEXT: vshufps {{.*#+}} ymm0 = ymm0[0,0],ymm1[0,0],ymm0[4,4],ymm1[4,4]
224 ; ALL-NEXT: vshufps {{.*#+}} ymm0 = ymm0[0,2,1,3,4,6,5,7]
226 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 8, i32 0, i32 8, i32 4, i32 12, i32 4, i32 12>
227 ret <8 x float> %shuffle
230 define <8 x float> @shuffle_v8f32_8823cc67(<8 x float> %a, <8 x float> %b) {
231 ; ALL-LABEL: shuffle_v8f32_8823cc67:
233 ; ALL-NEXT: vshufps {{.*#+}} ymm0 = ymm1[0,0],ymm0[2,3],ymm1[4,4],ymm0[6,7]
235 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 8, i32 8, i32 2, i32 3, i32 12, i32 12, i32 6, i32 7>
236 ret <8 x float> %shuffle
239 define <8 x float> @shuffle_v8f32_9832dc76(<8 x float> %a, <8 x float> %b) {
240 ; ALL-LABEL: shuffle_v8f32_9832dc76:
242 ; ALL-NEXT: vshufps {{.*#+}} ymm0 = ymm1[1,0],ymm0[3,2],ymm1[5,4],ymm0[7,6]
244 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 9, i32 8, i32 3, i32 2, i32 13, i32 12, i32 7, i32 6>
245 ret <8 x float> %shuffle
248 define <8 x float> @shuffle_v8f32_9810dc54(<8 x float> %a, <8 x float> %b) {
249 ; ALL-LABEL: shuffle_v8f32_9810dc54:
251 ; ALL-NEXT: vshufps {{.*#+}} ymm0 = ymm1[1,0],ymm0[1,0],ymm1[5,4],ymm0[5,4]
253 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 9, i32 8, i32 1, i32 0, i32 13, i32 12, i32 5, i32 4>
254 ret <8 x float> %shuffle
257 define <8 x float> @shuffle_v8f32_08194c5d(<8 x float> %a, <8 x float> %b) {
258 ; ALL-LABEL: shuffle_v8f32_08194c5d:
260 ; ALL-NEXT: vunpcklps {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
262 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
263 ret <8 x float> %shuffle
266 define <8 x float> @shuffle_v8f32_2a3b6e7f(<8 x float> %a, <8 x float> %b) {
267 ; ALL-LABEL: shuffle_v8f32_2a3b6e7f:
269 ; ALL-NEXT: vunpckhps {{.*#+}} ymm0 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
271 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
272 ret <8 x float> %shuffle
275 define <8 x float> @shuffle_v8f32_08192a3b(<8 x float> %a, <8 x float> %b) {
276 ; AVX1-LABEL: shuffle_v8f32_08192a3b:
278 ; AVX1-NEXT: vunpckhps {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
279 ; AVX1-NEXT: vunpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
280 ; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
283 ; AVX2-LABEL: shuffle_v8f32_08192a3b:
285 ; AVX2-NEXT: vmovaps {{.*#+}} ymm2 = <u,0,u,1,u,2,u,3>
286 ; AVX2-NEXT: vpermps %ymm1, %ymm2, %ymm1
287 ; AVX2-NEXT: vmovaps {{.*#+}} ymm2 = <0,u,1,u,2,u,3,u>
288 ; AVX2-NEXT: vpermps %ymm0, %ymm2, %ymm0
289 ; AVX2-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
291 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11>
292 ret <8 x float> %shuffle
295 define <8 x float> @shuffle_v8f32_08991abb(<8 x float> %a, <8 x float> %b) {
296 ; AVX1-LABEL: shuffle_v8f32_08991abb:
298 ; AVX1-NEXT: vshufps {{.*#+}} xmm2 = xmm0[1,0],xmm1[2,0]
299 ; AVX1-NEXT: vshufps {{.*#+}} xmm2 = xmm2[0,2],xmm1[3,3]
300 ; AVX1-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,0]
301 ; AVX1-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,1]
302 ; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
305 ; AVX2-LABEL: shuffle_v8f32_08991abb:
307 ; AVX2-NEXT: vmovaps {{.*#+}} ymm2 = <0,u,u,u,1,u,u,u>
308 ; AVX2-NEXT: vpermps %ymm0, %ymm2, %ymm0
309 ; AVX2-NEXT: vmovaps {{.*#+}} ymm2 = <u,0,1,1,u,2,3,3>
310 ; AVX2-NEXT: vpermps %ymm1, %ymm2, %ymm1
311 ; AVX2-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
313 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 8, i32 9, i32 9, i32 1, i32 10, i32 11, i32 11>
314 ret <8 x float> %shuffle
317 define <8 x float> @shuffle_v8f32_091b2d3f(<8 x float> %a, <8 x float> %b) {
318 ; AVX1-LABEL: shuffle_v8f32_091b2d3f:
320 ; AVX1-NEXT: vpermilps {{.*#+}} xmm2 = xmm0[0,1,1,3]
321 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[2,1,3,3]
322 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm2, %ymm0
323 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
326 ; AVX2-LABEL: shuffle_v8f32_091b2d3f:
328 ; AVX2-NEXT: vmovaps {{.*#+}} ymm2 = <0,u,1,u,2,u,3,u>
329 ; AVX2-NEXT: vpermps %ymm0, %ymm2, %ymm0
330 ; AVX2-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
332 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 9, i32 1, i32 11, i32 2, i32 13, i32 3, i32 15>
333 ret <8 x float> %shuffle
336 define <8 x float> @shuffle_v8f32_09ab1def(<8 x float> %a, <8 x float> %b) {
337 ; AVX1-LABEL: shuffle_v8f32_09ab1def:
339 ; AVX1-NEXT: vpermilps {{.*#+}} xmm2 = xmm0[1,1,2,3]
340 ; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
341 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
344 ; AVX2-LABEL: shuffle_v8f32_09ab1def:
346 ; AVX2-NEXT: vmovaps {{.*#+}} ymm2 = <0,u,u,u,1,u,u,u>
347 ; AVX2-NEXT: vpermps %ymm0, %ymm2, %ymm0
348 ; AVX2-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
350 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 9, i32 10, i32 11, i32 1, i32 13, i32 14, i32 15>
351 ret <8 x float> %shuffle
354 define <8 x float> @shuffle_v8f32_00014445(<8 x float> %a, <8 x float> %b) {
355 ; ALL-LABEL: shuffle_v8f32_00014445:
357 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,0,1,4,4,4,5]
359 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 1, i32 4, i32 4, i32 4, i32 5>
360 ret <8 x float> %shuffle
363 define <8 x float> @shuffle_v8f32_00204464(<8 x float> %a, <8 x float> %b) {
364 ; ALL-LABEL: shuffle_v8f32_00204464:
366 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,2,0,4,4,6,4]
368 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 0, i32 4, i32 4, i32 6, i32 4>
369 ret <8 x float> %shuffle
372 define <8 x float> @shuffle_v8f32_03004744(<8 x float> %a, <8 x float> %b) {
373 ; ALL-LABEL: shuffle_v8f32_03004744:
375 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,3,0,0,4,7,4,4]
377 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 3, i32 0, i32 0, i32 4, i32 7, i32 4, i32 4>
378 ret <8 x float> %shuffle
381 define <8 x float> @shuffle_v8f32_10005444(<8 x float> %a, <8 x float> %b) {
382 ; ALL-LABEL: shuffle_v8f32_10005444:
384 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,0,0,0,5,4,4,4]
386 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 1, i32 0, i32 0, i32 0, i32 5, i32 4, i32 4, i32 4>
387 ret <8 x float> %shuffle
390 define <8 x float> @shuffle_v8f32_22006644(<8 x float> %a, <8 x float> %b) {
391 ; ALL-LABEL: shuffle_v8f32_22006644:
393 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[2,2,0,0,6,6,4,4]
395 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 2, i32 2, i32 0, i32 0, i32 6, i32 6, i32 4, i32 4>
396 ret <8 x float> %shuffle
399 define <8 x float> @shuffle_v8f32_33307774(<8 x float> %a, <8 x float> %b) {
400 ; ALL-LABEL: shuffle_v8f32_33307774:
402 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,3,3,0,7,7,7,4]
404 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 3, i32 3, i32 3, i32 0, i32 7, i32 7, i32 7, i32 4>
405 ret <8 x float> %shuffle
408 define <8 x float> @shuffle_v8f32_32107654(<8 x float> %a, <8 x float> %b) {
409 ; ALL-LABEL: shuffle_v8f32_32107654:
411 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
413 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4>
414 ret <8 x float> %shuffle
417 define <8 x float> @shuffle_v8f32_00234467(<8 x float> %a, <8 x float> %b) {
418 ; ALL-LABEL: shuffle_v8f32_00234467:
420 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,2,3,4,4,6,7]
422 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 3, i32 4, i32 4, i32 6, i32 7>
423 ret <8 x float> %shuffle
426 define <8 x float> @shuffle_v8f32_00224466(<8 x float> %a, <8 x float> %b) {
427 ; ALL-LABEL: shuffle_v8f32_00224466:
429 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,2,2,4,4,6,6]
431 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 2, i32 4, i32 4, i32 6, i32 6>
432 ret <8 x float> %shuffle
435 define <8 x float> @shuffle_v8f32_10325476(<8 x float> %a, <8 x float> %b) {
436 ; ALL-LABEL: shuffle_v8f32_10325476:
438 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,0,3,2,5,4,7,6]
440 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6>
441 ret <8 x float> %shuffle
444 define <8 x float> @shuffle_v8f32_11335577(<8 x float> %a, <8 x float> %b) {
445 ; ALL-LABEL: shuffle_v8f32_11335577:
447 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,1,3,3,5,5,7,7]
449 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7>
450 ret <8 x float> %shuffle
453 define <8 x float> @shuffle_v8f32_10235467(<8 x float> %a, <8 x float> %b) {
454 ; ALL-LABEL: shuffle_v8f32_10235467:
456 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,0,2,3,5,4,6,7]
458 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 1, i32 0, i32 2, i32 3, i32 5, i32 4, i32 6, i32 7>
459 ret <8 x float> %shuffle
462 define <8 x float> @shuffle_v8f32_10225466(<8 x float> %a, <8 x float> %b) {
463 ; ALL-LABEL: shuffle_v8f32_10225466:
465 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,0,2,2,5,4,6,6]
467 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 1, i32 0, i32 2, i32 2, i32 5, i32 4, i32 6, i32 6>
468 ret <8 x float> %shuffle
471 define <8 x float> @shuffle_v8f32_00015444(<8 x float> %a, <8 x float> %b) {
472 ; ALL-LABEL: shuffle_v8f32_00015444:
474 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,0,1,5,4,4,4]
476 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 1, i32 5, i32 4, i32 4, i32 4>
477 ret <8 x float> %shuffle
480 define <8 x float> @shuffle_v8f32_00204644(<8 x float> %a, <8 x float> %b) {
481 ; ALL-LABEL: shuffle_v8f32_00204644:
483 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,2,0,4,6,4,4]
485 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 0, i32 4, i32 6, i32 4, i32 4>
486 ret <8 x float> %shuffle
489 define <8 x float> @shuffle_v8f32_03004474(<8 x float> %a, <8 x float> %b) {
490 ; ALL-LABEL: shuffle_v8f32_03004474:
492 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,3,0,0,4,4,7,4]
494 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 3, i32 0, i32 0, i32 4, i32 4, i32 7, i32 4>
495 ret <8 x float> %shuffle
498 define <8 x float> @shuffle_v8f32_10004444(<8 x float> %a, <8 x float> %b) {
499 ; ALL-LABEL: shuffle_v8f32_10004444:
501 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,0,0,0,4,4,4,4]
503 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 1, i32 0, i32 0, i32 0, i32 4, i32 4, i32 4, i32 4>
504 ret <8 x float> %shuffle
507 define <8 x float> @shuffle_v8f32_22006446(<8 x float> %a, <8 x float> %b) {
508 ; ALL-LABEL: shuffle_v8f32_22006446:
510 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[2,2,0,0,6,4,4,6]
512 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 2, i32 2, i32 0, i32 0, i32 6, i32 4, i32 4, i32 6>
513 ret <8 x float> %shuffle
516 define <8 x float> @shuffle_v8f32_33307474(<8 x float> %a, <8 x float> %b) {
517 ; ALL-LABEL: shuffle_v8f32_33307474:
519 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,3,3,0,7,4,7,4]
521 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 3, i32 3, i32 3, i32 0, i32 7, i32 4, i32 7, i32 4>
522 ret <8 x float> %shuffle
525 define <8 x float> @shuffle_v8f32_32104567(<8 x float> %a, <8 x float> %b) {
526 ; ALL-LABEL: shuffle_v8f32_32104567:
528 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,4,5,6,7]
530 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 4, i32 5, i32 6, i32 7>
531 ret <8 x float> %shuffle
534 define <8 x float> @shuffle_v8f32_00236744(<8 x float> %a, <8 x float> %b) {
535 ; ALL-LABEL: shuffle_v8f32_00236744:
537 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,2,3,6,7,4,4]
539 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 3, i32 6, i32 7, i32 4, i32 4>
540 ret <8 x float> %shuffle
543 define <8 x float> @shuffle_v8f32_00226644(<8 x float> %a, <8 x float> %b) {
544 ; ALL-LABEL: shuffle_v8f32_00226644:
546 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,2,2,6,6,4,4]
548 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 2, i32 6, i32 6, i32 4, i32 4>
549 ret <8 x float> %shuffle
552 define <8 x float> @shuffle_v8f32_10324567(<8 x float> %a, <8 x float> %b) {
553 ; ALL-LABEL: shuffle_v8f32_10324567:
555 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,0,3,2,4,5,6,7]
557 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 4, i32 5, i32 6, i32 7>
558 ret <8 x float> %shuffle
561 define <8 x float> @shuffle_v8f32_11334567(<8 x float> %a, <8 x float> %b) {
562 ; ALL-LABEL: shuffle_v8f32_11334567:
564 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,1,3,3,4,5,6,7]
566 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 4, i32 5, i32 6, i32 7>
567 ret <8 x float> %shuffle
570 define <8 x float> @shuffle_v8f32_01235467(<8 x float> %a, <8 x float> %b) {
571 ; ALL-LABEL: shuffle_v8f32_01235467:
573 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,1,2,3,5,4,6,7]
575 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 6, i32 7>
576 ret <8 x float> %shuffle
579 define <8 x float> @shuffle_v8f32_01235466(<8 x float> %a, <8 x float> %b) {
580 ; ALL-LABEL: shuffle_v8f32_01235466:
582 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,1,2,3,5,4,6,6]
584 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 6, i32 6>
585 ret <8 x float> %shuffle
588 define <8 x float> @shuffle_v8f32_002u6u44(<8 x float> %a, <8 x float> %b) {
589 ; ALL-LABEL: shuffle_v8f32_002u6u44:
591 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,2,u,6,u,4,4]
593 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 undef, i32 6, i32 undef, i32 4, i32 4>
594 ret <8 x float> %shuffle
597 define <8 x float> @shuffle_v8f32_00uu66uu(<8 x float> %a, <8 x float> %b) {
598 ; ALL-LABEL: shuffle_v8f32_00uu66uu:
600 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,u,u,6,6,u,u]
602 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 undef, i32 undef, i32 6, i32 6, i32 undef, i32 undef>
603 ret <8 x float> %shuffle
606 define <8 x float> @shuffle_v8f32_103245uu(<8 x float> %a, <8 x float> %b) {
607 ; ALL-LABEL: shuffle_v8f32_103245uu:
609 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,0,3,2,4,5,u,u]
611 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 4, i32 5, i32 undef, i32 undef>
612 ret <8 x float> %shuffle
615 define <8 x float> @shuffle_v8f32_1133uu67(<8 x float> %a, <8 x float> %b) {
616 ; ALL-LABEL: shuffle_v8f32_1133uu67:
618 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,1,3,3,u,u,6,7]
620 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 undef, i32 undef, i32 6, i32 7>
621 ret <8 x float> %shuffle
624 define <8 x float> @shuffle_v8f32_0uu354uu(<8 x float> %a, <8 x float> %b) {
625 ; ALL-LABEL: shuffle_v8f32_0uu354uu:
627 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,u,u,3,5,4,u,u]
629 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 undef, i32 undef, i32 3, i32 5, i32 4, i32 undef, i32 undef>
630 ret <8 x float> %shuffle
633 define <8 x float> @shuffle_v8f32_uuu3uu66(<8 x float> %a, <8 x float> %b) {
634 ; ALL-LABEL: shuffle_v8f32_uuu3uu66:
636 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[u,u,u,3,u,u,6,6]
638 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 undef, i32 undef, i32 undef, i32 3, i32 undef, i32 undef, i32 6, i32 6>
639 ret <8 x float> %shuffle
642 define <8 x float> @shuffle_v8f32_c348cda0(<8 x float> %a, <8 x float> %b) {
643 ; AVX1-LABEL: shuffle_v8f32_c348cda0:
645 ; AVX1-NEXT: vperm2f128 {{.*#+}} ymm2 = ymm0[2,3,0,1]
646 ; AVX1-NEXT: vshufps {{.*#+}} ymm0 = ymm0[0,3],ymm2[0,0],ymm0[4,7],ymm2[4,4]
647 ; AVX1-NEXT: vperm2f128 {{.*#+}} ymm2 = ymm1[2,3,0,1]
648 ; AVX1-NEXT: vpermilps {{.*#+}} ymm1 = ymm1[0,1,2,0,4,5,6,4]
649 ; AVX1-NEXT: vblendpd {{.*#+}} ymm1 = ymm2[0],ymm1[1,2],ymm2[3]
650 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2],ymm1[3,4,5,6],ymm0[7]
653 ; AVX2-LABEL: shuffle_v8f32_c348cda0:
655 ; AVX2-NEXT: vmovaps {{.*#+}} ymm2 = <u,3,4,u,u,u,u,0>
656 ; AVX2-NEXT: vpermps %ymm0, %ymm2, %ymm0
657 ; AVX2-NEXT: vmovaps {{.*#+}} ymm2 = <4,u,u,0,4,5,2,u>
658 ; AVX2-NEXT: vpermps %ymm1, %ymm2, %ymm1
659 ; AVX2-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2],ymm1[3,4,5,6],ymm0[7]
661 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 12, i32 3, i32 4, i32 8, i32 12, i32 13, i32 10, i32 0>
662 ret <8 x float> %shuffle
665 define <8 x float> @shuffle_v8f32_f511235a(<8 x float> %a, <8 x float> %b) {
666 ; AVX1-LABEL: shuffle_v8f32_f511235a:
668 ; AVX1-NEXT: vperm2f128 {{.*#+}} ymm2 = ymm0[2,3,0,1]
669 ; AVX1-NEXT: vpermilpd {{.*#+}} ymm2 = ymm2[0,0,3,2]
670 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,1,1,1,4,5,5,5]
671 ; AVX1-NEXT: vblendpd {{.*#+}} ymm0 = ymm2[0],ymm0[1],ymm2[2],ymm0[3]
672 ; AVX1-NEXT: vperm2f128 {{.*#+}} ymm1 = ymm1[2,3,0,1]
673 ; AVX1-NEXT: vpermilps {{.*#+}} ymm1 = ymm1[3,1,2,2,7,5,6,6]
674 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6],ymm1[7]
677 ; AVX2-LABEL: shuffle_v8f32_f511235a:
679 ; AVX2-NEXT: vmovaps {{.*#+}} ymm2 = <7,u,u,u,u,u,u,2>
680 ; AVX2-NEXT: vpermps %ymm1, %ymm2, %ymm1
681 ; AVX2-NEXT: vmovaps {{.*#+}} ymm2 = <u,5,1,1,2,3,5,u>
682 ; AVX2-NEXT: vpermps %ymm0, %ymm2, %ymm0
683 ; AVX2-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6],ymm1[7]
685 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 15, i32 5, i32 1, i32 1, i32 2, i32 3, i32 5, i32 10>
686 ret <8 x float> %shuffle
689 define <8 x float> @shuffle_v8f32_32103210(<8 x float> %a, <8 x float> %b) {
690 ; AVX1-LABEL: shuffle_v8f32_32103210:
692 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
693 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
696 ; AVX2-LABEL: shuffle_v8f32_32103210:
698 ; AVX2-NEXT: vmovaps {{.*#+}} ymm1 = [3,2,1,0,3,2,1,0]
699 ; AVX2-NEXT: vpermps %ymm0, %ymm1, %ymm0
701 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 3, i32 2, i32 1, i32 0>
702 ret <8 x float> %shuffle
705 define <8 x float> @shuffle_v8f32_76547654(<8 x float> %a, <8 x float> %b) {
706 ; AVX1-LABEL: shuffle_v8f32_76547654:
708 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
709 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
710 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
713 ; AVX2-LABEL: shuffle_v8f32_76547654:
715 ; AVX2-NEXT: vmovaps {{.*#+}} ymm1 = [7,6,5,4,7,6,5,4]
716 ; AVX2-NEXT: vpermps %ymm0, %ymm1, %ymm0
718 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 7, i32 6, i32 5, i32 4>
719 ret <8 x float> %shuffle
722 define <8 x float> @shuffle_v8f32_76543210(<8 x float> %a, <8 x float> %b) {
723 ; AVX1-LABEL: shuffle_v8f32_76543210:
725 ; AVX1-NEXT: vperm2f128 {{.*#+}} ymm0 = ymm0[2,3,0,1]
726 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
729 ; AVX2-LABEL: shuffle_v8f32_76543210:
731 ; AVX2-NEXT: vmovaps {{.*#+}} ymm1 = [7,6,5,4,3,2,1,0]
732 ; AVX2-NEXT: vpermps %ymm0, %ymm1, %ymm0
734 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
735 ret <8 x float> %shuffle
738 define <8 x float> @shuffle_v8f32_3210ba98(<8 x float> %a, <8 x float> %b) {
739 ; ALL-LABEL: shuffle_v8f32_3210ba98:
741 ; ALL-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
742 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
744 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 11, i32 10, i32 9, i32 8>
745 ret <8 x float> %shuffle
748 define <8 x float> @shuffle_v8f32_3210fedc(<8 x float> %a, <8 x float> %b) {
749 ; AVX1-LABEL: shuffle_v8f32_3210fedc:
751 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
752 ; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm1
753 ; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm1[3,2,1,0]
754 ; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
757 ; AVX2-LABEL: shuffle_v8f32_3210fedc:
759 ; AVX2-NEXT: vpermilps {{.*#+}} ymm1 = ymm1[3,2,1,0,7,6,5,4]
760 ; AVX2-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
761 ; AVX2-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3]
763 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 15, i32 14, i32 13, i32 12>
764 ret <8 x float> %shuffle
767 define <8 x float> @shuffle_v8f32_7654fedc(<8 x float> %a, <8 x float> %b) {
768 ; ALL-LABEL: shuffle_v8f32_7654fedc:
770 ; ALL-NEXT: vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[2,3]
771 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
773 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 15, i32 14, i32 13, i32 12>
774 ret <8 x float> %shuffle
777 define <8 x float> @shuffle_v8f32_fedc7654(<8 x float> %a, <8 x float> %b) {
778 ; ALL-LABEL: shuffle_v8f32_fedc7654:
780 ; ALL-NEXT: vperm2f128 {{.*#+}} ymm0 = ymm1[2,3],ymm0[2,3]
781 ; ALL-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
783 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 15, i32 14, i32 13, i32 12, i32 7, i32 6, i32 5, i32 4>
784 ret <8 x float> %shuffle
787 define <8 x float> @shuffle_v8f32_ba987654(<8 x float> %a, <8 x float> %b) {
788 ; AVX1-LABEL: shuffle_v8f32_ba987654:
790 ; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm1[3,2,1,0]
791 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
792 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
793 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
796 ; AVX2-LABEL: shuffle_v8f32_ba987654:
798 ; AVX2-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
799 ; AVX2-NEXT: vpermilps {{.*#+}} ymm1 = ymm1[3,2,1,0,7,6,5,4]
800 ; AVX2-NEXT: vblendpd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3]
802 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4>
803 ret <8 x float> %shuffle
806 define <8 x float> @shuffle_v8f32_ba983210(<8 x float> %a, <8 x float> %b) {
807 ; AVX1-LABEL: shuffle_v8f32_ba983210:
809 ; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm1[3,2,1,0]
810 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
811 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
812 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
815 ; AVX2-LABEL: shuffle_v8f32_ba983210:
817 ; AVX2-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
818 ; AVX2-NEXT: vpermilps {{.*#+}} ymm1 = ymm1[3,2,1,0,7,6,5,4]
819 ; AVX2-NEXT: vblendpd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3]
821 %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4>
822 ret <8 x float> %shuffle
825 define <8 x i32> @shuffle_v8i32_00000000(<8 x i32> %a, <8 x i32> %b) {
826 ; AVX1-LABEL: shuffle_v8i32_00000000:
828 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
829 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
832 ; AVX2-LABEL: shuffle_v8i32_00000000:
834 ; AVX2-NEXT: vbroadcastss %xmm0, %ymm0
836 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
837 ret <8 x i32> %shuffle
840 define <8 x i32> @shuffle_v8i32_00000010(<8 x i32> %a, <8 x i32> %b) {
841 ; AVX1-LABEL: shuffle_v8i32_00000010:
843 ; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[0,0,0,0]
844 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,1,0]
845 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
848 ; AVX2-LABEL: shuffle_v8i32_00000010:
850 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,1,0]
851 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
853 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 0>
854 ret <8 x i32> %shuffle
857 define <8 x i32> @shuffle_v8i32_00000200(<8 x i32> %a, <8 x i32> %b) {
858 ; AVX1-LABEL: shuffle_v8i32_00000200:
860 ; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[0,0,0,0]
861 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,2,0,0]
862 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
865 ; AVX2-LABEL: shuffle_v8i32_00000200:
867 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,2,0,0]
868 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
870 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 2, i32 0, i32 0>
871 ret <8 x i32> %shuffle
874 define <8 x i32> @shuffle_v8i32_00003000(<8 x i32> %a, <8 x i32> %b) {
875 ; AVX1-LABEL: shuffle_v8i32_00003000:
877 ; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[0,0,0,0]
878 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,0,0,0]
879 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
882 ; AVX2-LABEL: shuffle_v8i32_00003000:
884 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [0,0,0,0,3,0,0,0]
885 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
887 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 3, i32 0, i32 0, i32 0>
888 ret <8 x i32> %shuffle
891 define <8 x i32> @shuffle_v8i32_00040000(<8 x i32> %a, <8 x i32> %b) {
892 ; AVX1-LABEL: shuffle_v8i32_00040000:
894 ; AVX1-NEXT: vperm2f128 {{.*#+}} ymm1 = ymm0[2,3,0,1]
895 ; AVX1-NEXT: vpermilps {{.*#+}} ymm1 = ymm1[0,0,0,0,4,4,4,4]
896 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,0,3,4,4,4,7]
897 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
900 ; AVX2-LABEL: shuffle_v8i32_00040000:
902 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [0,0,0,4,0,0,0,0]
903 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
905 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 4, i32 0, i32 0, i32 0, i32 0>
906 ret <8 x i32> %shuffle
909 define <8 x i32> @shuffle_v8i32_00500000(<8 x i32> %a, <8 x i32> %b) {
910 ; AVX1-LABEL: shuffle_v8i32_00500000:
912 ; AVX1-NEXT: vperm2f128 {{.*#+}} ymm1 = ymm0[2,3,0,1]
913 ; AVX1-NEXT: vpermilps {{.*#+}} ymm1 = ymm1[u,u,1,u,4,4,4,4]
914 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,2,0,4,4,6,4]
915 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3],ymm1[4,5,6,7]
918 ; AVX2-LABEL: shuffle_v8i32_00500000:
920 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [0,0,5,0,0,0,0,0]
921 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
923 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 5, i32 0, i32 0, i32 0, i32 0, i32 0>
924 ret <8 x i32> %shuffle
927 define <8 x i32> @shuffle_v8i32_06000000(<8 x i32> %a, <8 x i32> %b) {
928 ; AVX1-LABEL: shuffle_v8i32_06000000:
930 ; AVX1-NEXT: vperm2f128 {{.*#+}} ymm1 = ymm0[2,3,0,1]
931 ; AVX1-NEXT: vpermilps {{.*#+}} ymm1 = ymm1[u,2,u,u,4,4,4,4]
932 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,1,0,0,4,5,4,4]
933 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4,5,6,7]
936 ; AVX2-LABEL: shuffle_v8i32_06000000:
938 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [0,6,0,0,0,0,0,0]
939 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
941 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 6, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
942 ret <8 x i32> %shuffle
945 define <8 x i32> @shuffle_v8i32_70000000(<8 x i32> %a, <8 x i32> %b) {
946 ; AVX1-LABEL: shuffle_v8i32_70000000:
948 ; AVX1-NEXT: vperm2f128 {{.*#+}} ymm1 = ymm0[2,3,0,1]
949 ; AVX1-NEXT: vpermilps {{.*#+}} ymm1 = ymm1[3,u,u,u,4,4,4,4]
950 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,0,0,4,4,4,4]
951 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3],ymm1[4,5,6,7]
954 ; AVX2-LABEL: shuffle_v8i32_70000000:
956 ; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
957 ; AVX2-NEXT: movl $7, %eax
958 ; AVX2-NEXT: vpinsrd $0, %eax, %xmm1, %xmm1
959 ; AVX2-NEXT: vpxor %ymm2, %ymm2, %ymm2
960 ; AVX2-NEXT: vinserti128 $0, %xmm1, %ymm2, %ymm1
961 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
963 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 7, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
964 ret <8 x i32> %shuffle
967 define <8 x i32> @shuffle_v8i32_01014545(<8 x i32> %a, <8 x i32> %b) {
968 ; AVX1-LABEL: shuffle_v8i32_01014545:
970 ; AVX1-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[0,0,2,2]
973 ; AVX2-LABEL: shuffle_v8i32_01014545:
975 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[0,1,0,1,4,5,4,5]
977 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 4, i32 5, i32 4, i32 5>
978 ret <8 x i32> %shuffle
981 define <8 x i32> @shuffle_v8i32_00112233(<8 x i32> %a, <8 x i32> %b) {
982 ; AVX1-LABEL: shuffle_v8i32_00112233:
984 ; AVX1-NEXT: vunpcklps {{.*#+}} xmm1 = xmm0[0,0,1,1]
985 ; AVX1-NEXT: vunpckhps {{.*#+}} xmm0 = xmm0[2,2,3,3]
986 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
989 ; AVX2-LABEL: shuffle_v8i32_00112233:
991 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [0,0,1,1,2,2,3,3]
992 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
994 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 1, i32 1, i32 2, i32 2, i32 3, i32 3>
995 ret <8 x i32> %shuffle
998 define <8 x i32> @shuffle_v8i32_00001111(<8 x i32> %a, <8 x i32> %b) {
999 ; AVX1-LABEL: shuffle_v8i32_00001111:
1001 ; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[0,0,0,0]
1002 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[1,1,1,1]
1003 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
1006 ; AVX2-LABEL: shuffle_v8i32_00001111:
1008 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [0,0,0,0,1,1,1,1]
1009 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1011 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1>
1012 ret <8 x i32> %shuffle
1015 define <8 x i32> @shuffle_v8i32_81a3c5e7(<8 x i32> %a, <8 x i32> %b) {
1016 ; AVX1-LABEL: shuffle_v8i32_81a3c5e7:
1018 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4],ymm0[5],ymm1[6],ymm0[7]
1021 ; AVX2-LABEL: shuffle_v8i32_81a3c5e7:
1023 ; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4],ymm0[5],ymm1[6],ymm0[7]
1025 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 8, i32 1, i32 10, i32 3, i32 12, i32 5, i32 14, i32 7>
1026 ret <8 x i32> %shuffle
1029 define <8 x i32> @shuffle_v8i32_08080808(<8 x i32> %a, <8 x i32> %b) {
1030 ; AVX1-LABEL: shuffle_v8i32_08080808:
1032 ; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm1[0,0,2,0]
1033 ; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm1, %ymm1
1034 ; AVX1-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0,0]
1035 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
1036 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
1039 ; AVX2-LABEL: shuffle_v8i32_08080808:
1041 ; AVX2-NEXT: vpbroadcastd %xmm1, %ymm1
1042 ; AVX2-NEXT: vpbroadcastq %xmm0, %ymm0
1043 ; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
1045 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 8, i32 0, i32 8, i32 0, i32 8, i32 0, i32 8>
1046 ret <8 x i32> %shuffle
1049 define <8 x i32> @shuffle_v8i32_08084c4c(<8 x i32> %a, <8 x i32> %b) {
1050 ; AVX1-LABEL: shuffle_v8i32_08084c4c:
1052 ; AVX1-NEXT: vshufps {{.*#+}} ymm0 = ymm0[0,0],ymm1[0,0],ymm0[4,4],ymm1[4,4]
1053 ; AVX1-NEXT: vshufps {{.*#+}} ymm0 = ymm0[0,2,1,3,4,6,5,7]
1056 ; AVX2-LABEL: shuffle_v8i32_08084c4c:
1058 ; AVX2-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
1059 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[0,1,0,1,4,5,4,5]
1060 ; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
1062 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 8, i32 0, i32 8, i32 4, i32 12, i32 4, i32 12>
1063 ret <8 x i32> %shuffle
1066 define <8 x i32> @shuffle_v8i32_8823cc67(<8 x i32> %a, <8 x i32> %b) {
1067 ; AVX1-LABEL: shuffle_v8i32_8823cc67:
1069 ; AVX1-NEXT: vshufps {{.*#+}} ymm0 = ymm1[0,0],ymm0[2,3],ymm1[4,4],ymm0[6,7]
1072 ; AVX2-LABEL: shuffle_v8i32_8823cc67:
1074 ; AVX2-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[0,0,2,3,4,4,6,7]
1075 ; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
1077 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 8, i32 8, i32 2, i32 3, i32 12, i32 12, i32 6, i32 7>
1078 ret <8 x i32> %shuffle
1081 define <8 x i32> @shuffle_v8i32_9832dc76(<8 x i32> %a, <8 x i32> %b) {
1082 ; AVX1-LABEL: shuffle_v8i32_9832dc76:
1084 ; AVX1-NEXT: vshufps {{.*#+}} ymm0 = ymm1[1,0],ymm0[3,2],ymm1[5,4],ymm0[7,6]
1087 ; AVX2-LABEL: shuffle_v8i32_9832dc76:
1089 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[0,1,3,2,4,5,7,6]
1090 ; AVX2-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[1,0,2,3,5,4,6,7]
1091 ; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
1093 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 9, i32 8, i32 3, i32 2, i32 13, i32 12, i32 7, i32 6>
1094 ret <8 x i32> %shuffle
1097 define <8 x i32> @shuffle_v8i32_9810dc54(<8 x i32> %a, <8 x i32> %b) {
1098 ; AVX1-LABEL: shuffle_v8i32_9810dc54:
1100 ; AVX1-NEXT: vshufps {{.*#+}} ymm0 = ymm1[1,0],ymm0[1,0],ymm1[5,4],ymm0[5,4]
1103 ; AVX2-LABEL: shuffle_v8i32_9810dc54:
1105 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[0,1,1,0,4,5,5,4]
1106 ; AVX2-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[1,0,2,3,5,4,6,7]
1107 ; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
1109 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 9, i32 8, i32 1, i32 0, i32 13, i32 12, i32 5, i32 4>
1110 ret <8 x i32> %shuffle
1113 define <8 x i32> @shuffle_v8i32_08194c5d(<8 x i32> %a, <8 x i32> %b) {
1114 ; AVX1-LABEL: shuffle_v8i32_08194c5d:
1116 ; AVX1-NEXT: vunpcklps {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
1119 ; AVX2-LABEL: shuffle_v8i32_08194c5d:
1121 ; AVX2-NEXT: vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
1123 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
1124 ret <8 x i32> %shuffle
1127 define <8 x i32> @shuffle_v8i32_2a3b6e7f(<8 x i32> %a, <8 x i32> %b) {
1128 ; AVX1-LABEL: shuffle_v8i32_2a3b6e7f:
1130 ; AVX1-NEXT: vunpckhps {{.*#+}} ymm0 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
1133 ; AVX2-LABEL: shuffle_v8i32_2a3b6e7f:
1135 ; AVX2-NEXT: vpunpckhdq {{.*#+}} ymm0 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
1137 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
1138 ret <8 x i32> %shuffle
1141 define <8 x i32> @shuffle_v8i32_08192a3b(<8 x i32> %a, <8 x i32> %b) {
1142 ; AVX1-LABEL: shuffle_v8i32_08192a3b:
1144 ; AVX1-NEXT: vunpckhps {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
1145 ; AVX1-NEXT: vunpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
1146 ; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
1149 ; AVX2-LABEL: shuffle_v8i32_08192a3b:
1151 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = <u,0,u,1,u,2,u,3>
1152 ; AVX2-NEXT: vpermd %ymm1, %ymm2, %ymm1
1153 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = <0,u,1,u,2,u,3,u>
1154 ; AVX2-NEXT: vpermd %ymm0, %ymm2, %ymm0
1155 ; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
1157 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11>
1158 ret <8 x i32> %shuffle
1161 define <8 x i32> @shuffle_v8i32_08991abb(<8 x i32> %a, <8 x i32> %b) {
1162 ; AVX1-LABEL: shuffle_v8i32_08991abb:
1164 ; AVX1-NEXT: vshufps {{.*#+}} xmm2 = xmm0[1,0],xmm1[2,0]
1165 ; AVX1-NEXT: vshufps {{.*#+}} xmm2 = xmm2[0,2],xmm1[3,3]
1166 ; AVX1-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,0]
1167 ; AVX1-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,1]
1168 ; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
1171 ; AVX2-LABEL: shuffle_v8i32_08991abb:
1173 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = <0,u,u,u,1,u,u,u>
1174 ; AVX2-NEXT: vpermd %ymm0, %ymm2, %ymm0
1175 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = <u,0,1,1,u,2,3,3>
1176 ; AVX2-NEXT: vpermd %ymm1, %ymm2, %ymm1
1177 ; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
1179 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 8, i32 9, i32 9, i32 1, i32 10, i32 11, i32 11>
1180 ret <8 x i32> %shuffle
1183 define <8 x i32> @shuffle_v8i32_091b2d3f(<8 x i32> %a, <8 x i32> %b) {
1184 ; AVX1-LABEL: shuffle_v8i32_091b2d3f:
1186 ; AVX1-NEXT: vpermilps {{.*#+}} xmm2 = xmm0[0,1,1,3]
1187 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[2,1,3,3]
1188 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm2, %ymm0
1189 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
1192 ; AVX2-LABEL: shuffle_v8i32_091b2d3f:
1194 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = <0,u,1,u,2,u,3,u>
1195 ; AVX2-NEXT: vpermd %ymm0, %ymm2, %ymm0
1196 ; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
1198 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 9, i32 1, i32 11, i32 2, i32 13, i32 3, i32 15>
1199 ret <8 x i32> %shuffle
1202 define <8 x i32> @shuffle_v8i32_09ab1def(<8 x i32> %a, <8 x i32> %b) {
1203 ; AVX1-LABEL: shuffle_v8i32_09ab1def:
1205 ; AVX1-NEXT: vpermilps {{.*#+}} xmm2 = xmm0[1,1,2,3]
1206 ; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
1207 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
1210 ; AVX2-LABEL: shuffle_v8i32_09ab1def:
1212 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = <0,u,u,u,1,u,u,u>
1213 ; AVX2-NEXT: vpermd %ymm0, %ymm2, %ymm0
1214 ; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
1216 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 9, i32 10, i32 11, i32 1, i32 13, i32 14, i32 15>
1217 ret <8 x i32> %shuffle
1220 define <8 x i32> @shuffle_v8i32_00014445(<8 x i32> %a, <8 x i32> %b) {
1221 ; AVX1-LABEL: shuffle_v8i32_00014445:
1223 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,0,1,4,4,4,5]
1226 ; AVX2-LABEL: shuffle_v8i32_00014445:
1228 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[0,0,0,1,4,4,4,5]
1230 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 1, i32 4, i32 4, i32 4, i32 5>
1231 ret <8 x i32> %shuffle
1234 define <8 x i32> @shuffle_v8i32_00204464(<8 x i32> %a, <8 x i32> %b) {
1235 ; AVX1-LABEL: shuffle_v8i32_00204464:
1237 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,2,0,4,4,6,4]
1240 ; AVX2-LABEL: shuffle_v8i32_00204464:
1242 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[0,0,2,0,4,4,6,4]
1244 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 0, i32 4, i32 4, i32 6, i32 4>
1245 ret <8 x i32> %shuffle
1248 define <8 x i32> @shuffle_v8i32_03004744(<8 x i32> %a, <8 x i32> %b) {
1249 ; AVX1-LABEL: shuffle_v8i32_03004744:
1251 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,3,0,0,4,7,4,4]
1254 ; AVX2-LABEL: shuffle_v8i32_03004744:
1256 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[0,3,0,0,4,7,4,4]
1258 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 3, i32 0, i32 0, i32 4, i32 7, i32 4, i32 4>
1259 ret <8 x i32> %shuffle
1262 define <8 x i32> @shuffle_v8i32_10005444(<8 x i32> %a, <8 x i32> %b) {
1263 ; AVX1-LABEL: shuffle_v8i32_10005444:
1265 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,0,0,0,5,4,4,4]
1268 ; AVX2-LABEL: shuffle_v8i32_10005444:
1270 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[1,0,0,0,5,4,4,4]
1272 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 1, i32 0, i32 0, i32 0, i32 5, i32 4, i32 4, i32 4>
1273 ret <8 x i32> %shuffle
1276 define <8 x i32> @shuffle_v8i32_22006644(<8 x i32> %a, <8 x i32> %b) {
1277 ; AVX1-LABEL: shuffle_v8i32_22006644:
1279 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[2,2,0,0,6,6,4,4]
1282 ; AVX2-LABEL: shuffle_v8i32_22006644:
1284 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[2,2,0,0,6,6,4,4]
1286 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 2, i32 2, i32 0, i32 0, i32 6, i32 6, i32 4, i32 4>
1287 ret <8 x i32> %shuffle
1290 define <8 x i32> @shuffle_v8i32_33307774(<8 x i32> %a, <8 x i32> %b) {
1291 ; AVX1-LABEL: shuffle_v8i32_33307774:
1293 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,3,3,0,7,7,7,4]
1296 ; AVX2-LABEL: shuffle_v8i32_33307774:
1298 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[3,3,3,0,7,7,7,4]
1300 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 3, i32 3, i32 3, i32 0, i32 7, i32 7, i32 7, i32 4>
1301 ret <8 x i32> %shuffle
1304 define <8 x i32> @shuffle_v8i32_32107654(<8 x i32> %a, <8 x i32> %b) {
1305 ; AVX1-LABEL: shuffle_v8i32_32107654:
1307 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
1310 ; AVX2-LABEL: shuffle_v8i32_32107654:
1312 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
1314 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4>
1315 ret <8 x i32> %shuffle
1318 define <8 x i32> @shuffle_v8i32_00234467(<8 x i32> %a, <8 x i32> %b) {
1319 ; AVX1-LABEL: shuffle_v8i32_00234467:
1321 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,2,3,4,4,6,7]
1324 ; AVX2-LABEL: shuffle_v8i32_00234467:
1326 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[0,0,2,3,4,4,6,7]
1328 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 3, i32 4, i32 4, i32 6, i32 7>
1329 ret <8 x i32> %shuffle
1332 define <8 x i32> @shuffle_v8i32_00224466(<8 x i32> %a, <8 x i32> %b) {
1333 ; AVX1-LABEL: shuffle_v8i32_00224466:
1335 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,2,2,4,4,6,6]
1338 ; AVX2-LABEL: shuffle_v8i32_00224466:
1340 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[0,0,2,2,4,4,6,6]
1342 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 2, i32 4, i32 4, i32 6, i32 6>
1343 ret <8 x i32> %shuffle
1346 define <8 x i32> @shuffle_v8i32_10325476(<8 x i32> %a, <8 x i32> %b) {
1347 ; AVX1-LABEL: shuffle_v8i32_10325476:
1349 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,0,3,2,5,4,7,6]
1352 ; AVX2-LABEL: shuffle_v8i32_10325476:
1354 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[1,0,3,2,5,4,7,6]
1356 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6>
1357 ret <8 x i32> %shuffle
1360 define <8 x i32> @shuffle_v8i32_11335577(<8 x i32> %a, <8 x i32> %b) {
1361 ; AVX1-LABEL: shuffle_v8i32_11335577:
1363 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,1,3,3,5,5,7,7]
1366 ; AVX2-LABEL: shuffle_v8i32_11335577:
1368 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[1,1,3,3,5,5,7,7]
1370 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7>
1371 ret <8 x i32> %shuffle
1374 define <8 x i32> @shuffle_v8i32_10235467(<8 x i32> %a, <8 x i32> %b) {
1375 ; AVX1-LABEL: shuffle_v8i32_10235467:
1377 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,0,2,3,5,4,6,7]
1380 ; AVX2-LABEL: shuffle_v8i32_10235467:
1382 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[1,0,2,3,5,4,6,7]
1384 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 1, i32 0, i32 2, i32 3, i32 5, i32 4, i32 6, i32 7>
1385 ret <8 x i32> %shuffle
1388 define <8 x i32> @shuffle_v8i32_10225466(<8 x i32> %a, <8 x i32> %b) {
1389 ; AVX1-LABEL: shuffle_v8i32_10225466:
1391 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,0,2,2,5,4,6,6]
1394 ; AVX2-LABEL: shuffle_v8i32_10225466:
1396 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[1,0,2,2,5,4,6,6]
1398 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 1, i32 0, i32 2, i32 2, i32 5, i32 4, i32 6, i32 6>
1399 ret <8 x i32> %shuffle
1402 define <8 x i32> @shuffle_v8i32_00015444(<8 x i32> %a, <8 x i32> %b) {
1403 ; AVX1-LABEL: shuffle_v8i32_00015444:
1405 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,0,1,5,4,4,4]
1408 ; AVX2-LABEL: shuffle_v8i32_00015444:
1410 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [0,0,0,1,5,4,4,4]
1411 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1413 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 1, i32 5, i32 4, i32 4, i32 4>
1414 ret <8 x i32> %shuffle
1417 define <8 x i32> @shuffle_v8i32_00204644(<8 x i32> %a, <8 x i32> %b) {
1418 ; AVX1-LABEL: shuffle_v8i32_00204644:
1420 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,2,0,4,6,4,4]
1423 ; AVX2-LABEL: shuffle_v8i32_00204644:
1425 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [0,0,2,0,4,6,4,4]
1426 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1428 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 0, i32 4, i32 6, i32 4, i32 4>
1429 ret <8 x i32> %shuffle
1432 define <8 x i32> @shuffle_v8i32_03004474(<8 x i32> %a, <8 x i32> %b) {
1433 ; AVX1-LABEL: shuffle_v8i32_03004474:
1435 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,3,0,0,4,4,7,4]
1438 ; AVX2-LABEL: shuffle_v8i32_03004474:
1440 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [0,3,0,0,4,4,7,4]
1441 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1443 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 3, i32 0, i32 0, i32 4, i32 4, i32 7, i32 4>
1444 ret <8 x i32> %shuffle
1447 define <8 x i32> @shuffle_v8i32_10004444(<8 x i32> %a, <8 x i32> %b) {
1448 ; AVX1-LABEL: shuffle_v8i32_10004444:
1450 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,0,0,0,4,4,4,4]
1453 ; AVX2-LABEL: shuffle_v8i32_10004444:
1455 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [1,0,0,0,4,4,4,4]
1456 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1458 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 1, i32 0, i32 0, i32 0, i32 4, i32 4, i32 4, i32 4>
1459 ret <8 x i32> %shuffle
1462 define <8 x i32> @shuffle_v8i32_22006446(<8 x i32> %a, <8 x i32> %b) {
1463 ; AVX1-LABEL: shuffle_v8i32_22006446:
1465 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[2,2,0,0,6,4,4,6]
1468 ; AVX2-LABEL: shuffle_v8i32_22006446:
1470 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [2,2,0,0,6,4,4,6]
1471 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1473 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 2, i32 2, i32 0, i32 0, i32 6, i32 4, i32 4, i32 6>
1474 ret <8 x i32> %shuffle
1477 define <8 x i32> @shuffle_v8i32_33307474(<8 x i32> %a, <8 x i32> %b) {
1478 ; AVX1-LABEL: shuffle_v8i32_33307474:
1480 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,3,3,0,7,4,7,4]
1483 ; AVX2-LABEL: shuffle_v8i32_33307474:
1485 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [3,3,3,0,7,4,7,4]
1486 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1488 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 3, i32 3, i32 3, i32 0, i32 7, i32 4, i32 7, i32 4>
1489 ret <8 x i32> %shuffle
1492 define <8 x i32> @shuffle_v8i32_32104567(<8 x i32> %a, <8 x i32> %b) {
1493 ; AVX1-LABEL: shuffle_v8i32_32104567:
1495 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,4,5,6,7]
1498 ; AVX2-LABEL: shuffle_v8i32_32104567:
1500 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [3,2,1,0,4,5,6,7]
1501 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1503 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 4, i32 5, i32 6, i32 7>
1504 ret <8 x i32> %shuffle
1507 define <8 x i32> @shuffle_v8i32_00236744(<8 x i32> %a, <8 x i32> %b) {
1508 ; AVX1-LABEL: shuffle_v8i32_00236744:
1510 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,2,3,6,7,4,4]
1513 ; AVX2-LABEL: shuffle_v8i32_00236744:
1515 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [0,0,2,3,6,7,4,4]
1516 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1518 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 3, i32 6, i32 7, i32 4, i32 4>
1519 ret <8 x i32> %shuffle
1522 define <8 x i32> @shuffle_v8i32_00226644(<8 x i32> %a, <8 x i32> %b) {
1523 ; AVX1-LABEL: shuffle_v8i32_00226644:
1525 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,2,2,6,6,4,4]
1528 ; AVX2-LABEL: shuffle_v8i32_00226644:
1530 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [0,0,2,2,6,6,4,4]
1531 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1533 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 2, i32 6, i32 6, i32 4, i32 4>
1534 ret <8 x i32> %shuffle
1537 define <8 x i32> @shuffle_v8i32_10324567(<8 x i32> %a, <8 x i32> %b) {
1538 ; AVX1-LABEL: shuffle_v8i32_10324567:
1540 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,0,3,2,4,5,6,7]
1543 ; AVX2-LABEL: shuffle_v8i32_10324567:
1545 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [1,0,3,2,4,5,6,7]
1546 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1548 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 4, i32 5, i32 6, i32 7>
1549 ret <8 x i32> %shuffle
1552 define <8 x i32> @shuffle_v8i32_11334567(<8 x i32> %a, <8 x i32> %b) {
1553 ; AVX1-LABEL: shuffle_v8i32_11334567:
1555 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,1,3,3,4,5,6,7]
1558 ; AVX2-LABEL: shuffle_v8i32_11334567:
1560 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [1,1,3,3,4,5,6,7]
1561 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1563 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 4, i32 5, i32 6, i32 7>
1564 ret <8 x i32> %shuffle
1567 define <8 x i32> @shuffle_v8i32_01235467(<8 x i32> %a, <8 x i32> %b) {
1568 ; AVX1-LABEL: shuffle_v8i32_01235467:
1570 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,1,2,3,5,4,6,7]
1573 ; AVX2-LABEL: shuffle_v8i32_01235467:
1575 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [0,1,2,3,5,4,6,7]
1576 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1578 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 6, i32 7>
1579 ret <8 x i32> %shuffle
1582 define <8 x i32> @shuffle_v8i32_01235466(<8 x i32> %a, <8 x i32> %b) {
1583 ; AVX1-LABEL: shuffle_v8i32_01235466:
1585 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,1,2,3,5,4,6,6]
1588 ; AVX2-LABEL: shuffle_v8i32_01235466:
1590 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [0,1,2,3,5,4,6,6]
1591 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1593 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 6, i32 6>
1594 ret <8 x i32> %shuffle
1597 define <8 x i32> @shuffle_v8i32_002u6u44(<8 x i32> %a, <8 x i32> %b) {
1598 ; AVX1-LABEL: shuffle_v8i32_002u6u44:
1600 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,2,u,6,u,4,4]
1603 ; AVX2-LABEL: shuffle_v8i32_002u6u44:
1605 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = <0,0,2,u,6,u,4,4>
1606 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1608 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 undef, i32 6, i32 undef, i32 4, i32 4>
1609 ret <8 x i32> %shuffle
1612 define <8 x i32> @shuffle_v8i32_00uu66uu(<8 x i32> %a, <8 x i32> %b) {
1613 ; AVX1-LABEL: shuffle_v8i32_00uu66uu:
1615 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,0,u,u,6,6,u,u]
1618 ; AVX2-LABEL: shuffle_v8i32_00uu66uu:
1620 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = <0,0,u,u,6,6,u,u>
1621 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1623 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 undef, i32 undef, i32 6, i32 6, i32 undef, i32 undef>
1624 ret <8 x i32> %shuffle
1627 define <8 x i32> @shuffle_v8i32_103245uu(<8 x i32> %a, <8 x i32> %b) {
1628 ; AVX1-LABEL: shuffle_v8i32_103245uu:
1630 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,0,3,2,4,5,u,u]
1633 ; AVX2-LABEL: shuffle_v8i32_103245uu:
1635 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = <1,0,3,2,4,5,u,u>
1636 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1638 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 4, i32 5, i32 undef, i32 undef>
1639 ret <8 x i32> %shuffle
1642 define <8 x i32> @shuffle_v8i32_1133uu67(<8 x i32> %a, <8 x i32> %b) {
1643 ; AVX1-LABEL: shuffle_v8i32_1133uu67:
1645 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,1,3,3,u,u,6,7]
1648 ; AVX2-LABEL: shuffle_v8i32_1133uu67:
1650 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = <1,1,3,3,u,u,6,7>
1651 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1653 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 undef, i32 undef, i32 6, i32 7>
1654 ret <8 x i32> %shuffle
1657 define <8 x i32> @shuffle_v8i32_0uu354uu(<8 x i32> %a, <8 x i32> %b) {
1658 ; AVX1-LABEL: shuffle_v8i32_0uu354uu:
1660 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[0,u,u,3,5,4,u,u]
1663 ; AVX2-LABEL: shuffle_v8i32_0uu354uu:
1665 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = <0,u,u,3,5,4,u,u>
1666 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1668 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 undef, i32 undef, i32 3, i32 5, i32 4, i32 undef, i32 undef>
1669 ret <8 x i32> %shuffle
1672 define <8 x i32> @shuffle_v8i32_uuu3uu66(<8 x i32> %a, <8 x i32> %b) {
1673 ; AVX1-LABEL: shuffle_v8i32_uuu3uu66:
1675 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[u,u,u,3,u,u,6,6]
1678 ; AVX2-LABEL: shuffle_v8i32_uuu3uu66:
1680 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = <u,u,u,3,u,u,6,6>
1681 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1683 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 undef, i32 undef, i32 undef, i32 3, i32 undef, i32 undef, i32 6, i32 6>
1684 ret <8 x i32> %shuffle
1687 define <8 x i32> @shuffle_v8i32_6caa87e5(<8 x i32> %a, <8 x i32> %b) {
1688 ; AVX1-LABEL: shuffle_v8i32_6caa87e5:
1690 ; AVX1-NEXT: vperm2f128 {{.*#+}} ymm2 = ymm1[2,3,0,1]
1691 ; AVX1-NEXT: vshufps {{.*#+}} ymm1 = ymm2[0,0],ymm1[2,2],ymm2[4,4],ymm1[6,6]
1692 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
1693 ; AVX1-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
1694 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
1695 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3,4],ymm0[5],ymm1[6],ymm0[7]
1698 ; AVX2-LABEL: shuffle_v8i32_6caa87e5:
1700 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = <u,4,2,2,0,u,6,u>
1701 ; AVX2-NEXT: vpermd %ymm1, %ymm2, %ymm1
1702 ; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[3,1,3,2]
1703 ; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3,4],ymm0[5],ymm1[6],ymm0[7]
1705 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 6, i32 12, i32 10, i32 10, i32 8, i32 7, i32 14, i32 5>
1706 ret <8 x i32> %shuffle
1709 define <8 x i32> @shuffle_v8i32_32103210(<8 x i32> %a, <8 x i32> %b) {
1710 ; AVX1-LABEL: shuffle_v8i32_32103210:
1712 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
1713 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
1716 ; AVX2-LABEL: shuffle_v8i32_32103210:
1718 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [3,2,1,0,3,2,1,0]
1719 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1721 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 3, i32 2, i32 1, i32 0>
1722 ret <8 x i32> %shuffle
1725 define <8 x i32> @shuffle_v8i32_76547654(<8 x i32> %a, <8 x i32> %b) {
1726 ; AVX1-LABEL: shuffle_v8i32_76547654:
1728 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
1729 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
1730 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
1733 ; AVX2-LABEL: shuffle_v8i32_76547654:
1735 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [7,6,5,4,7,6,5,4]
1736 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1738 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 7, i32 6, i32 5, i32 4>
1739 ret <8 x i32> %shuffle
1742 define <8 x i32> @shuffle_v8i32_76543210(<8 x i32> %a, <8 x i32> %b) {
1743 ; AVX1-LABEL: shuffle_v8i32_76543210:
1745 ; AVX1-NEXT: vperm2f128 {{.*#+}} ymm0 = ymm0[2,3,0,1]
1746 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
1749 ; AVX2-LABEL: shuffle_v8i32_76543210:
1751 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [7,6,5,4,3,2,1,0]
1752 ; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
1754 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
1755 ret <8 x i32> %shuffle
1758 define <8 x i32> @shuffle_v8i32_3210ba98(<8 x i32> %a, <8 x i32> %b) {
1759 ; AVX1-LABEL: shuffle_v8i32_3210ba98:
1761 ; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
1762 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
1765 ; AVX2-LABEL: shuffle_v8i32_3210ba98:
1767 ; AVX2-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0
1768 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
1770 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 11, i32 10, i32 9, i32 8>
1771 ret <8 x i32> %shuffle
1774 define <8 x i32> @shuffle_v8i32_3210fedc(<8 x i32> %a, <8 x i32> %b) {
1775 ; AVX1-LABEL: shuffle_v8i32_3210fedc:
1777 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
1778 ; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm1
1779 ; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm1[3,2,1,0]
1780 ; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
1783 ; AVX2-LABEL: shuffle_v8i32_3210fedc:
1785 ; AVX2-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[3,2,1,0,7,6,5,4]
1786 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
1787 ; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
1789 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 15, i32 14, i32 13, i32 12>
1790 ret <8 x i32> %shuffle
1793 define <8 x i32> @shuffle_v8i32_7654fedc(<8 x i32> %a, <8 x i32> %b) {
1794 ; AVX1-LABEL: shuffle_v8i32_7654fedc:
1796 ; AVX1-NEXT: vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[2,3]
1797 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
1800 ; AVX2-LABEL: shuffle_v8i32_7654fedc:
1802 ; AVX2-NEXT: vperm2i128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[2,3]
1803 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
1805 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 15, i32 14, i32 13, i32 12>
1806 ret <8 x i32> %shuffle
1809 define <8 x i32> @shuffle_v8i32_fedc7654(<8 x i32> %a, <8 x i32> %b) {
1810 ; AVX1-LABEL: shuffle_v8i32_fedc7654:
1812 ; AVX1-NEXT: vperm2f128 {{.*#+}} ymm0 = ymm1[2,3],ymm0[2,3]
1813 ; AVX1-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
1816 ; AVX2-LABEL: shuffle_v8i32_fedc7654:
1818 ; AVX2-NEXT: vperm2i128 {{.*#+}} ymm0 = ymm1[2,3],ymm0[2,3]
1819 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
1821 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 15, i32 14, i32 13, i32 12, i32 7, i32 6, i32 5, i32 4>
1822 ret <8 x i32> %shuffle
1825 define <8 x i32> @shuffle_v8i32_ba987654(<8 x i32> %a, <8 x i32> %b) {
1826 ; AVX1-LABEL: shuffle_v8i32_ba987654:
1828 ; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm1[3,2,1,0]
1829 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
1830 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
1831 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
1834 ; AVX2-LABEL: shuffle_v8i32_ba987654:
1836 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
1837 ; AVX2-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[3,2,1,0,7,6,5,4]
1838 ; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
1840 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4>
1841 ret <8 x i32> %shuffle
1844 define <8 x i32> @shuffle_v8i32_ba983210(<8 x i32> %a, <8 x i32> %b) {
1845 ; AVX1-LABEL: shuffle_v8i32_ba983210:
1847 ; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm1[3,2,1,0]
1848 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
1849 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
1850 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
1853 ; AVX2-LABEL: shuffle_v8i32_ba983210:
1855 ; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
1856 ; AVX2-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[3,2,1,0,7,6,5,4]
1857 ; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
1859 %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4>
1860 ret <8 x i32> %shuffle
1863 define <8 x float> @splat_mem_v8f32_2(float* %p) {
1864 ; ALL-LABEL: splat_mem_v8f32_2:
1866 ; ALL-NEXT: vbroadcastss (%rdi), %ymm0
1869 %2 = insertelement <4 x float> undef, float %1, i32 0
1870 %3 = shufflevector <4 x float> %2, <4 x float> undef, <8 x i32> zeroinitializer
1874 define <8 x float> @splat_v8f32(<4 x float> %r) {
1875 ; AVX1-LABEL: splat_v8f32:
1877 ; AVX1-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
1878 ; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
1881 ; AVX2-LABEL: splat_v8f32:
1883 ; AVX2-NEXT: vbroadcastss %xmm0, %ymm0
1885 %1 = shufflevector <4 x float> %r, <4 x float> undef, <8 x i32> zeroinitializer