Lines Matching refs:vpshufb

28 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,2,3,0,1]
35 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,2,3,0,1]
47 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,4,5,0,1,0,1]
54 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,4,5,0,1,0,1]
66 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,6,7,0,1,0,1,0,1]
73 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,6,7,0,1,0,1,0,1]
85 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,8,9,0,1,0,1,0,1,0,1]
92 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,8,9,0,1,0,1,0,1,0,1]
104 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,10,11,0,1,0,1,0,1,0,1,0,1]
111 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,10,11,0,1,0,1,0,1,0,1,0,1]
123 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,12,13,0,1,0,1,0,1,0,1,0,1,0,1]
130 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,12,13,0,1,0,1,0,1,0,1,0,1,0,1]
142 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[14,15,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
149 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[14,15,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
161 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,0,1,0,1,0,1,0,1,0,1,0,1,2,3]
185 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,0,1,0,1,0,1,0,1,0,1,6,7,0,1]
196 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,0,1,0,1,0,1,0,1,2,3,0,1,16,17,16,17,16,17,16,1…
207 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,0,1,0,1,0,1,0,1,10,11,0,1,0,1]
217 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,0,1,0,1,0,1,4,5,0,1,0,1,16,17,16,17,16,17,16,1…
228 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,0,1,0,1,0,1,14,15,0,1,0,1,0,1]
238 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,0,1,0,1,6,7,0,1,0,1,0,1,16,17,16,17,16,17,16,1…
249 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,0,1,0,1,8,9,0,1,0,1,0,1,0,1]
259 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,0,1,8,9,0,1,0,1,0,1,0,1,16,17,16,17,16,17,16,1…
270 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,0,1,10,11,0,1,0,1,0,1,0,1,0,1]
280 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,10,11,0,1,0,1,0,1,0,1,0,1,16,17,16,17,16,17,16…
291 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,12,13,0,1,0,1,0,1,0,1,0,1,0,1]
301 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,12,13,0,1,0,1,0,1,0,1,0,1,0,1,16,17,16,17,16,17,16…
312 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[14,15,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
322 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[14,15,0,1,0,1,0,1,0,1,0,1,0,1,0,1,16,17,16,17,16,17,16…
451 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,2,3,0,1]
457 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,2,3,0,1]
467 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,4,5,0,1,0,1]
473 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,4,5,0,1,0,1]
483 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,6,7,0,1,0,1,0,1]
489 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,6,7,0,1,0,1,0,1]
499 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,8,9,0,1,0,1,0,1,0,1]
505 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,8,9,0,1,0,1,0,1,0,1]
515 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,10,11,0,1,0,1,0,1,0,1,0,1]
521 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,10,11,0,1,0,1,0,1,0,1,0,1]
531 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,12,13,0,1,0,1,0,1,0,1,0,1,0,1]
537 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,12,13,0,1,0,1,0,1,0,1,0,1,0,1]
547 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[14,15,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
553 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[14,15,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
804 ; AVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
806 ; AVX1-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
826 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
827 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
833 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,0,1,0,1,0,1,0,1,2,3,0,1,16,17,16,17,16,17,16,1…
844 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
845 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
851 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,0,1,0,1,0,1,4,5,0,1,0,1,16,17,16,17,16,17,16,1…
862 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
863 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
869 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,0,1,0,1,6,7,0,1,0,1,0,1,16,17,16,17,16,17,16,1…
880 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
881 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
887 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,0,1,8,9,0,1,0,1,0,1,0,1,16,17,16,17,16,17,24,2…
898 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
899 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
905 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,10,11,0,1,0,1,0,1,0,1,0,1,16,17,16,17,26,27,16…
916 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
917 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
923 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,12,13,0,1,0,1,0,1,0,1,0,1,0,1,16,17,28,29,16,17,16…
934 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
935 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
941 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[14,15,0,1,0,1,0,1,0,1,0,1,0,1,0,1,30,31,16,17,16,17,16…
995 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,0,1,u,u,2,3,u,u,4,5,u,u,6,7,u,u,24,25,u,u,26,27,u,…
996 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,u,u,2,3,u,u,4,5,u,u,6,7,u,u,24,25,u,u,26,27,u,u,28…
1015 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,8,9,u,u,10,11,u,u,12,13,u,u,14,15,u,u,16,17,u,u,18…
1016 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[8,9,u,u,10,11,u,u,12,13,u,u,14,15,u,u,16,17,u,u,18,19,…
1026 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,2,3,0,1]
1028 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,0,1,0,1,0,1,0,1,0,1,0,1]
1034 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,0,1,0,1,0,1,0,1,2,3,0,1,16,17,18,19,16,17,16,1…
1043 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[0,1,0,1,0,1,0,1,0,1,4,5,0,1,0,1]
1045 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,4,5,0,1,0,1,0,1,0,1,0,1]
1051 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,0,1,0,1,0,1,4,5,0,1,0,1,16,17,16,17,20,21,16,1…
1060 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[0,1,0,1,0,1,0,1,6,7,0,1,0,1,0,1]
1062 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,6,7,0,1,0,1,0,1,0,1]
1068 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,0,1,0,1,6,7,0,1,0,1,0,1,16,17,16,17,16,17,22,2…
1077 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[0,1,0,1,0,1,8,9,0,1,0,1,0,1,0,1]
1079 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,8,9,0,1,0,1,0,1]
1085 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,0,1,8,9,0,1,0,1,0,1,0,1,16,17,16,17,16,17,16,1…
1094 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[0,1,0,1,10,11,0,1,0,1,0,1,0,1,0,1]
1096 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,10,11,0,1,0,1]
1102 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,10,11,0,1,0,1,0,1,0,1,0,1,16,17,16,17,16,17,16…
1111 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[0,1,12,13,0,1,0,1,0,1,0,1,0,1,0,1]
1113 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,12,13,0,1]
1119 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,12,13,0,1,0,1,0,1,0,1,0,1,0,1,16,17,16,17,16,17,16…
1128 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[14,15,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
1130 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,14,15]
1136 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[14,15,0,1,0,1,0,1,0,1,0,1,0,1,0,1,16,17,16,17,16,17,16…
1148 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[12,13,12,13,8,9,8,9,4,5,4,5,0,1,0,1]
1154 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,4,5,4,5,8,9,8,9,12,13,12,13,28,29,28,29,24,25,…
1163 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[8,9,8,9,8,9,8,9,0,1,0,1,0,1,0,1]
1172 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[8,9,8,9,8,9,8,9,0,1,0,1,0,1,0,1,16,17,16,17,16,17,16,1…
1184 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,4,5,6,7,0,1,0,1,12,13,0,1]
1190 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,u,u,u,u,0,1,0,1,0,1,0,1,0,1,16,17,16,17,u,u,u,u,16…
1199 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[14,15,2,3,0,1,0,1,0,1,0,1,0,1,0,1]
1201 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,4,5,6,7,0,1,0,1,0,1,14,15]
1207 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[14,15,u,u,0,1,0,1,0,1,0,1,0,1,0,1,16,17,16,17,u,u,u,u,…
1219 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[12,13,12,13,12,13,8,9,4,5,4,5,0,1,0,1]
1225 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,u,u,u,u,4,5,8,9,8,9,u,u,12,13,28,29,28,29,u,u,24,2…
1244 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[8,9,8,9,8,9,8,9,u,u,u,u,u,u,u,u,16,17,16,17,16,17,u,u,…
1650 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,4,5,4,5,6,7,0,1,4,5,8,9,14,15]
1673 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,2,3]
1683 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,2,3]
1798 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8,9,8,9,8,9,8,9,0,1,0,1,0,1,14,15]
1800 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[8,9,8,9,8,9,8,9,0,1,0,1,0,1,0,1]
1808 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8,9,8,9,8,9,8,9,0,1,0,1,0,1,14,15]
1810 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[8,9,8,9,8,9,8,9,0,1,0,1,0,1,0,1]
1894 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm3
1896 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
1904 ; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm3
1906 ; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
1918 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[2,3,0,1,10,11,8,9,10,11,8,9,2,3,2,3]
1920 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[2,3,0,1,10,11,8,9,10,11,8,9,2,3,0,1]
1928 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[2,3,0,1,10,11,8,9,10,11,8,9,2,3,2,3]
1930 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[2,3,0,1,10,11,8,9,10,11,8,9,2,3,0,1]
1942 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[10,11,8,9,2,3,0,1,10,11,8,9,2,3,2,3]
1944 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[10,11,8,9,2,3,0,1,10,11,8,9,2,3,0,1]
1952 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[10,11,8,9,2,3,0,1,10,11,8,9,2,3,2,3]
1954 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[10,11,8,9,2,3,0,1,10,11,8,9,2,3,0,1]
1966 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[10,11,8,9,2,3,0,1,2,3,0,1,10,11,2,3]
1968 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[10,11,8,9,2,3,0,1,2,3,0,1,10,11,8,9]
1976 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[10,11,8,9,2,3,0,1,2,3,0,1,10,11,2,3]
1978 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[10,11,8,9,2,3,0,1,2,3,0,1,10,11,8,9]
1990 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,8,9,8,9,0,1,0,1,8,9,8,9,2,3]
1992 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,8,9,8,9,0,1,0,1,8,9,8,9,0,1]
2000 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,8,9,8,9,0,1,0,1,8,9,8,9,2,3]
2002 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,8,9,8,9,0,1,0,1,8,9,8,9,0,1]
2014 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8,9,0,1,0,1,8,9,8,9,0,1,0,1,2,3]
2016 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[8,9,0,1,0,1,8,9,8,9,0,1,0,1,8,9]
2024 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8,9,0,1,0,1,8,9,8,9,0,1,0,1,2,3]
2026 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[8,9,0,1,0,1,8,9,8,9,0,1,0,1,8,9]
2038 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm3
2040 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
2048 ; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm3
2050 ; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
2062 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm3
2064 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
2072 ; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm3
2074 ; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
2086 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm3
2088 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
2096 ; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm3
2098 ; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
2110 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm3
2112 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
2120 ; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm3
2122 ; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
2134 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,8,9,8,9,8,9,8,9,8,9,14,15]
2136 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,0,1,8,9,8,9,8,9,8,9,8,9,8,9]
2144 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,8,9,8,9,8,9,8,9,8,9,14,15]
2146 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,0,1,8,9,8,9,8,9,8,9,8,9,8,9]
2158 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8,9,8,9,0,1,0,1,8,9,8,9,8,9,14,15]
2160 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[8,9,8,9,0,1,0,1,8,9,8,9,8,9,8,9]
2168 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8,9,8,9,0,1,0,1,8,9,8,9,8,9,14,15]
2170 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[8,9,8,9,0,1,0,1,8,9,8,9,8,9,8,9]
2182 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,8,9,8,9,0,1,8,9,8,9,8,9,14,15]
2184 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,8,9,8,9,0,1,8,9,8,9,8,9,8,9]
2192 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,8,9,8,9,0,1,8,9,8,9,8,9,14,15]
2194 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,8,9,8,9,0,1,8,9,8,9,8,9,8,9]
2206 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,8,9,8,9,0,1,0,1,0,1,0,1,14,15]
2208 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,8,9,8,9,0,1,0,1,0,1,0,1,0,1]
2216 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,8,9,8,9,0,1,0,1,0,1,0,1,14,15]
2218 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,8,9,8,9,0,1,0,1,0,1,0,1,0,1]
2256 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,8,9,8,9,8,9,8,9,8,9,14,15]
2258 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,2,3,8,9,8,9,8,9,8,9,8,9,8,9]
2266 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,8,9,8,9,8,9,8,9,8,9,14,15]
2268 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,2,3,8,9,8,9,8,9,8,9,8,9,8,9]
2280 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8,9,8,9,8,9,0,1,8,9,8,9,8,9,14,15]
2282 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[8,9,8,9,8,9,0,1,8,9,8,9,8,9,8,9]
2290 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8,9,8,9,8,9,0,1,8,9,8,9,8,9,14,15]
2292 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[8,9,8,9,8,9,0,1,8,9,8,9,8,9,8,9]
2304 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,8,9,8,9,0,1,8,9,8,9,8,9,14,15]
2306 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,8,9,8,9,0,1,8,9,8,9,8,9,8,9]
2314 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,8,9,8,9,0,1,8,9,8,9,8,9,14,15]
2316 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,8,9,8,9,0,1,8,9,8,9,8,9,8,9]
2328 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
2329 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
2335 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,14,15,4,5,14,15,12,13,14,15,16,17,18,19,20…
2347 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[12,13,6,7,4,5,6,7,8,9,10,11,12,13,6,7]
2356 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[12,13,6,7,4,5,6,7,8,9,10,11,12,13,6,7]
2368 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
2369 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
2375 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[8,9,10,11,12,13,6,7,8,9,10,11,0,1,2,3,24,25,26,27,28,2…
2393 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[2,3],zero,zero,ymm0[4,5],zero,zero,ymm0[8,9,u,u,12,13,…
2404 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm3
2406 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
2414 ; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm3
2416 ; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
2427 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,6,7,0,1,2,3,4,5,2,3]
2429 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[8,9,10,11,12,13,6,7,0,1,2,3,4,5,14,15]
2436 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,6,7,0,1,2,3,4,5,2,3]
2438 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[8,9,10,11,12,13,6,7,0,1,2,3,4,5,14,15]
2450 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm3
2452 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
2460 ; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm3
2462 ; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
2476 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,4,5,8,9,14,15,14,15,8,9,12,13,14,15]
2504 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,4,5,8,9,14,15,14,15,8,9,12,13,14,15]
2513 ; AVX2-NEXT:    vpshufb %xmm1, %xmm0, %xmm2
2516 ; AVX2-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
2530 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[4,5,0,1,4,5,4,5,0,1,4,5,8,9,14,15]
2558 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[4,5,0,1,4,5,4,5,0,1,4,5,8,9,14,15]
2569 ; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
2570 ; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
2588 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15]
2620 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[8,9,10,11,0,1,2,3,2,3,0,1,12,13,2,3]
2630 ; AVX2-NEXT:    vpshufb %xmm3, %xmm2, %xmm4
2632 ; AVX2-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
2647 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[4,5,0,1,12,13,10,11,8,9,10,11,12,13,10,11]
2663 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm2 = xmm2[0,1,2,3,2,3,0,1,8,9,10,11,6,7,4,5]
2665 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[2,3,0,1,4,5,6,7,6,7,4,5,4,5,6,7,18,19,16,17,20,21,22,2…
2679 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,14,15,14,15,8,9,12,13,14,15]
2706 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4,5,0,1,4,5,4,5,0,1,4,5,8,9,14,15]
2736 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,8,9,4,5,14,15,0,1,4,5,4,5,6,7]
2788 ; AVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
2790 ; AVX1-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
2797 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,6,7,4,5,10,11,0,1,10,11,0,1,2,3,16,17,22,23,20,21,…
2888 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,10,11,8,9,10,11,12,13,6,7]
2898 ; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
2899 ; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
2958 ; AVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
2962 ; AVX1-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
2970 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[0,1,4,5,4,5,6,7,0,1,4,5,8,9,4,5,16,17,20,21,20,21,22,2…
3085 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,4,5,6,7,8,9,0,1,4,5,10,11]
3185 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,4,5,4,5,6,7,8,9,0,1,4,5,10,11]
3283 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[14,15,14,15,6,7,6,7,8,9,8,9,10,11,14,15,30,31,30,31,22…
3428 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm6 = xmm0[2,3,2,3,4,5,6,7,8,9,8,9,0,1,2,3]
3434 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[6,7,4,5,4,5,10,11,4,5,14,15,12,13,0,1]
3442 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[6,7,4,5,0,1,10,11,4,5,10,11,4,5,6,7,22,23,20,21,16,17,…
3443 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[8,9,10,11,4,5,8,9,0,1,14,15,12,13,0,1,24,25,26,27,20,2…
3447 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,6,7,u,u,18,19,u,u,u,u,u,u,u,u,…
3526 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[2,3,2,3,2,3,2,3,2,3,2,3,2,3,2,3]