Lines Matching refs:SSSE3

3 …N: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3 | FileCheck %s --check-prefixes=SSE,SSSE3
89 ; SSSE3-LABEL: var_shuffle_v4f32_v4f32_xxxx_i32:
90 ; SSSE3:       # %bb.0:
91 ; SSSE3-NEXT:    # kill: def $ecx killed $ecx def $rcx
92 ; SSSE3-NEXT:    # kill: def $edx killed $edx def $rdx
93 ; SSSE3-NEXT:    # kill: def $esi killed $esi def $rsi
94 ; SSSE3-NEXT:    # kill: def $edi killed $edi def $rdi
95 ; SSSE3-NEXT:    andl $3, %edi
96 ; SSSE3-NEXT:    andl $3, %esi
97 ; SSSE3-NEXT:    andl $3, %edx
98 ; SSSE3-NEXT:    movaps %xmm0, -{{[0-9]+}}(%rsp)
99 ; SSSE3-NEXT:    andl $3, %ecx
100 ; SSSE3-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
101 ; SSSE3-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
102 ; SSSE3-NEXT:    unpcklps {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
103 ; SSSE3-NEXT:    movss {{.*#+}} xmm2 = mem[0],zero,zero,zero
104 ; SSSE3-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
105 ; SSSE3-NEXT:    unpcklps {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
106 ; SSSE3-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
107 ; SSSE3-NEXT:    retq
174 ; SSSE3-LABEL: var_shuffle_v4i32_v4i32_xxxx_i32:
175 ; SSSE3:       # %bb.0:
176 ; SSSE3-NEXT:    # kill: def $ecx killed $ecx def $rcx
177 ; SSSE3-NEXT:    # kill: def $edx killed $edx def $rdx
178 ; SSSE3-NEXT:    # kill: def $esi killed $esi def $rsi
179 ; SSSE3-NEXT:    # kill: def $edi killed $edi def $rdi
180 ; SSSE3-NEXT:    andl $3, %edi
181 ; SSSE3-NEXT:    andl $3, %esi
182 ; SSSE3-NEXT:    andl $3, %edx
183 ; SSSE3-NEXT:    movaps %xmm0, -{{[0-9]+}}(%rsp)
184 ; SSSE3-NEXT:    andl $3, %ecx
185 ; SSSE3-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
186 ; SSSE3-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
187 ; SSSE3-NEXT:    unpcklps {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
188 ; SSSE3-NEXT:    movss {{.*#+}} xmm2 = mem[0],zero,zero,zero
189 ; SSSE3-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
190 ; SSSE3-NEXT:    unpcklps {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
191 ; SSSE3-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
192 ; SSSE3-NEXT:    retq
283 ; SSSE3-LABEL: var_shuffle_v8i16_v8i16_xxxxxxxx_i16:
284 ; SSSE3:       # %bb.0:
285 ; SSSE3-NEXT:    # kill: def $r9d killed $r9d def $r9
286 ; SSSE3-NEXT:    # kill: def $r8d killed $r8d def $r8
287 ; SSSE3-NEXT:    # kill: def $ecx killed $ecx def $rcx
288 ; SSSE3-NEXT:    # kill: def $edx killed $edx def $rdx
289 ; SSSE3-NEXT:    # kill: def $esi killed $esi def $rsi
290 ; SSSE3-NEXT:    # kill: def $edi killed $edi def $rdi
291 ; SSSE3-NEXT:    movzwl {{[0-9]+}}(%rsp), %r10d
292 ; SSSE3-NEXT:    andl $7, %r10d
293 ; SSSE3-NEXT:    movzwl {{[0-9]+}}(%rsp), %eax
294 ; SSSE3-NEXT:    andl $7, %eax
295 ; SSSE3-NEXT:    andl $7, %edi
296 ; SSSE3-NEXT:    andl $7, %esi
297 ; SSSE3-NEXT:    andl $7, %edx
298 ; SSSE3-NEXT:    andl $7, %ecx
299 ; SSSE3-NEXT:    andl $7, %r8d
300 ; SSSE3-NEXT:    movaps %xmm0, -{{[0-9]+}}(%rsp)
301 ; SSSE3-NEXT:    andl $7, %r9d
302 ; SSSE3-NEXT:    movzwl -24(%rsp,%rcx,2), %ecx
303 ; SSSE3-NEXT:    movd %ecx, %xmm0
304 ; SSSE3-NEXT:    movzwl -24(%rsp,%rdx,2), %ecx
305 ; SSSE3-NEXT:    movd %ecx, %xmm1
306 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],…
307 ; SSSE3-NEXT:    movzwl -24(%rsp,%rsi,2), %ecx
308 ; SSSE3-NEXT:    movd %ecx, %xmm2
309 ; SSSE3-NEXT:    movzwl -24(%rsp,%rdi,2), %ecx
310 ; SSSE3-NEXT:    movd %ecx, %xmm0
311 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],…
312 ; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
313 ; SSSE3-NEXT:    movzwl -24(%rsp,%r9,2), %ecx
314 ; SSSE3-NEXT:    movd %ecx, %xmm1
315 ; SSSE3-NEXT:    movzwl -24(%rsp,%r8,2), %ecx
316 ; SSSE3-NEXT:    movd %ecx, %xmm2
317 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],…
318 ; SSSE3-NEXT:    movzwl -24(%rsp,%rax,2), %eax
319 ; SSSE3-NEXT:    movd %eax, %xmm1
320 ; SSSE3-NEXT:    movzwl -24(%rsp,%r10,2), %eax
321 ; SSSE3-NEXT:    movd %eax, %xmm3
322 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1],xmm3[2],xmm1[2],xmm3[3],…
323 ; SSSE3-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
324 ; SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
325 ; SSSE3-NEXT:    retq
490 ; SSSE3-LABEL: var_shuffle_v16i8_v16i8_xxxxxxxxxxxxxxxx_i8:
491 ; SSSE3:       # %bb.0:
492 ; SSSE3-NEXT:    # kill: def $r9d killed $r9d def $r9
493 ; SSSE3-NEXT:    # kill: def $r8d killed $r8d def $r8
494 ; SSSE3-NEXT:    # kill: def $ecx killed $ecx def $rcx
495 ; SSSE3-NEXT:    # kill: def $edx killed $edx def $rdx
496 ; SSSE3-NEXT:    # kill: def $esi killed $esi def $rsi
497 ; SSSE3-NEXT:    # kill: def $edi killed $edi def $rdi
498 ; SSSE3-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
499 ; SSSE3-NEXT:    andl $15, %eax
500 ; SSSE3-NEXT:    movaps %xmm0, -{{[0-9]+}}(%rsp)
501 ; SSSE3-NEXT:    movzbl -24(%rsp,%rax), %eax
502 ; SSSE3-NEXT:    movd %eax, %xmm8
503 ; SSSE3-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
504 ; SSSE3-NEXT:    andl $15, %eax
505 ; SSSE3-NEXT:    movzbl -24(%rsp,%rax), %eax
506 ; SSSE3-NEXT:    movd %eax, %xmm15
507 ; SSSE3-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
508 ; SSSE3-NEXT:    andl $15, %eax
509 ; SSSE3-NEXT:    movzbl -24(%rsp,%rax), %eax
510 ; SSSE3-NEXT:    movd %eax, %xmm9
511 ; SSSE3-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
512 ; SSSE3-NEXT:    andl $15, %eax
513 ; SSSE3-NEXT:    movzbl -24(%rsp,%rax), %eax
514 ; SSSE3-NEXT:    movd %eax, %xmm3
515 ; SSSE3-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
516 ; SSSE3-NEXT:    andl $15, %eax
517 ; SSSE3-NEXT:    movzbl -24(%rsp,%rax), %eax
518 ; SSSE3-NEXT:    movd %eax, %xmm10
519 ; SSSE3-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
520 ; SSSE3-NEXT:    andl $15, %eax
521 ; SSSE3-NEXT:    movzbl -24(%rsp,%rax), %eax
522 ; SSSE3-NEXT:    movd %eax, %xmm7
523 ; SSSE3-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
524 ; SSSE3-NEXT:    andl $15, %eax
525 ; SSSE3-NEXT:    movzbl -24(%rsp,%rax), %eax
526 ; SSSE3-NEXT:    movd %eax, %xmm11
527 ; SSSE3-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
528 ; SSSE3-NEXT:    andl $15, %eax
529 ; SSSE3-NEXT:    movzbl -24(%rsp,%rax), %eax
530 ; SSSE3-NEXT:    movd %eax, %xmm6
531 ; SSSE3-NEXT:    andl $15, %ecx
532 ; SSSE3-NEXT:    movzbl -24(%rsp,%rcx), %eax
533 ; SSSE3-NEXT:    movd %eax, %xmm12
534 ; SSSE3-NEXT:    andl $15, %edx
535 ; SSSE3-NEXT:    movzbl -24(%rsp,%rdx), %eax
536 ; SSSE3-NEXT:    movd %eax, %xmm5
537 ; SSSE3-NEXT:    andl $15, %esi
538 ; SSSE3-NEXT:    movzbl -24(%rsp,%rsi), %eax
539 ; SSSE3-NEXT:    movd %eax, %xmm13
540 ; SSSE3-NEXT:    andl $15, %edi
541 ; SSSE3-NEXT:    movzbl -24(%rsp,%rdi), %eax
542 ; SSSE3-NEXT:    movd %eax, %xmm0
543 ; SSSE3-NEXT:    andl $15, %r9d
544 ; SSSE3-NEXT:    movzbl -24(%rsp,%r9), %eax
545 ; SSSE3-NEXT:    movd %eax, %xmm14
546 ; SSSE3-NEXT:    andl $15, %r8d
547 ; SSSE3-NEXT:    movzbl -24(%rsp,%r8), %eax
548 ; SSSE3-NEXT:    movd %eax, %xmm1
549 ; SSSE3-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
550 ; SSSE3-NEXT:    andl $15, %eax
551 ; SSSE3-NEXT:    movzbl -24(%rsp,%rax), %eax
552 ; SSSE3-NEXT:    movd %eax, %xmm4
553 ; SSSE3-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
554 ; SSSE3-NEXT:    andl $15, %eax
555 ; SSSE3-NEXT:    movzbl -24(%rsp,%rax), %eax
556 ; SSSE3-NEXT:    movd %eax, %xmm2
557 ; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm15 = xmm15[0],xmm8[0],xmm15[1],xmm8[1],xmm15[2],xmm8[2],xmm1…
558 ; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm9[0],xmm3[1],xmm9[1],xmm3[2],xmm9[2],xmm3[3],…
559 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm15[0],xmm3[1],xmm15[1],xmm3[2],xmm15[2],xmm3[…
560 ; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm10[0],xmm7[1],xmm10[1],xmm7[2],xmm10[2],xmm7[…
561 ; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0],xmm11[0],xmm6[1],xmm11[1],xmm6[2],xmm11[2],xmm6[…
562 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm6 = xmm6[0],xmm7[0],xmm6[1],xmm7[1],xmm6[2],xmm7[2],xmm6[3],…
563 ; SSSE3-NEXT:    punpckldq {{.*#+}} xmm6 = xmm6[0],xmm3[0],xmm6[1],xmm3[1]
564 ; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm12[0],xmm5[1],xmm12[1],xmm5[2],xmm12[2],xmm5[…
565 ; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm13[0],xmm0[1],xmm13[1],xmm0[2],xmm13[2],xmm0[…
566 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm5[0],xmm0[1],xmm5[1],xmm0[2],xmm5[2],xmm0[3],…
567 ; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm14[0],xmm1[1],xmm14[1],xmm1[2],xmm14[2],xmm1[…
568 ; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1],xmm2[2],xmm4[2],xmm2[3],…
569 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],…
570 ; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
571 ; SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm6[0]
572 ; SSSE3-NEXT:    retq
741 ; SSSE3-LABEL: mem_shuffle_v4i32_v4i32_xxxx_i32:
742 ; SSSE3:       # %bb.0:
743 ; SSSE3-NEXT:    movl (%rdi), %eax
744 ; SSSE3-NEXT:    movl 4(%rdi), %ecx
745 ; SSSE3-NEXT:    andl $3, %eax
746 ; SSSE3-NEXT:    andl $3, %ecx
747 ; SSSE3-NEXT:    movl 8(%rdi), %edx
748 ; SSSE3-NEXT:    andl $3, %edx
749 ; SSSE3-NEXT:    movl 12(%rdi), %esi
750 ; SSSE3-NEXT:    andl $3, %esi
751 ; SSSE3-NEXT:    movaps %xmm0, -{{[0-9]+}}(%rsp)
752 ; SSSE3-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
753 ; SSSE3-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
754 ; SSSE3-NEXT:    unpcklps {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
755 ; SSSE3-NEXT:    movss {{.*#+}} xmm2 = mem[0],zero,zero,zero
756 ; SSSE3-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
757 ; SSSE3-NEXT:    unpcklps {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
758 ; SSSE3-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
759 ; SSSE3-NEXT:    retq
912 ; SSSE3-LABEL: mem_shuffle_v16i8_v16i8_xxxxxxxxxxxxxxxx_i8:
913 ; SSSE3:       # %bb.0:
914 ; SSSE3-NEXT:    pushq %rbp
915 ; SSSE3-NEXT:    pushq %r15
916 ; SSSE3-NEXT:    pushq %r14
917 ; SSSE3-NEXT:    pushq %r13
918 ; SSSE3-NEXT:    pushq %r12
919 ; SSSE3-NEXT:    pushq %rbx
920 ; SSSE3-NEXT:    movzbl (%rdi), %eax
921 ; SSSE3-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
922 ; SSSE3-NEXT:    movzbl 1(%rdi), %r9d
923 ; SSSE3-NEXT:    movzbl 2(%rdi), %r10d
924 ; SSSE3-NEXT:    movzbl 3(%rdi), %r11d
925 ; SSSE3-NEXT:    movzbl 4(%rdi), %r14d
926 ; SSSE3-NEXT:    movzbl 5(%rdi), %r15d
927 ; SSSE3-NEXT:    movzbl 6(%rdi), %r12d
928 ; SSSE3-NEXT:    movzbl 7(%rdi), %r13d
929 ; SSSE3-NEXT:    movzbl 8(%rdi), %ebx
930 ; SSSE3-NEXT:    movzbl 9(%rdi), %r8d
931 ; SSSE3-NEXT:    movzbl 10(%rdi), %ecx
932 ; SSSE3-NEXT:    movzbl 11(%rdi), %edx
933 ; SSSE3-NEXT:    movzbl 12(%rdi), %esi
934 ; SSSE3-NEXT:    movzbl 13(%rdi), %ebp
935 ; SSSE3-NEXT:    movzbl 14(%rdi), %eax
936 ; SSSE3-NEXT:    movzbl 15(%rdi), %edi
937 ; SSSE3-NEXT:    andl $15, %edi
938 ; SSSE3-NEXT:    movaps %xmm0, -{{[0-9]+}}(%rsp)
939 ; SSSE3-NEXT:    movzbl -24(%rsp,%rdi), %edi
940 ; SSSE3-NEXT:    movd %edi, %xmm8
941 ; SSSE3-NEXT:    andl $15, %eax
942 ; SSSE3-NEXT:    movzbl -24(%rsp,%rax), %eax
943 ; SSSE3-NEXT:    movd %eax, %xmm15
944 ; SSSE3-NEXT:    andl $15, %ebp
945 ; SSSE3-NEXT:    movzbl -24(%rsp,%rbp), %eax
946 ; SSSE3-NEXT:    movd %eax, %xmm9
947 ; SSSE3-NEXT:    andl $15, %esi
948 ; SSSE3-NEXT:    movzbl -24(%rsp,%rsi), %eax
949 ; SSSE3-NEXT:    movd %eax, %xmm3
950 ; SSSE3-NEXT:    andl $15, %edx
951 ; SSSE3-NEXT:    movzbl -24(%rsp,%rdx), %eax
952 ; SSSE3-NEXT:    movd %eax, %xmm10
953 ; SSSE3-NEXT:    andl $15, %ecx
954 ; SSSE3-NEXT:    movzbl -24(%rsp,%rcx), %eax
955 ; SSSE3-NEXT:    movd %eax, %xmm7
956 ; SSSE3-NEXT:    andl $15, %r8d
957 ; SSSE3-NEXT:    movzbl -24(%rsp,%r8), %eax
958 ; SSSE3-NEXT:    movd %eax, %xmm11
959 ; SSSE3-NEXT:    andl $15, %ebx
960 ; SSSE3-NEXT:    movzbl -24(%rsp,%rbx), %eax
961 ; SSSE3-NEXT:    movd %eax, %xmm6
962 ; SSSE3-NEXT:    andl $15, %r13d
963 ; SSSE3-NEXT:    movzbl -24(%rsp,%r13), %eax
964 ; SSSE3-NEXT:    movd %eax, %xmm12
965 ; SSSE3-NEXT:    andl $15, %r12d
966 ; SSSE3-NEXT:    movzbl -24(%rsp,%r12), %eax
967 ; SSSE3-NEXT:    movd %eax, %xmm5
968 ; SSSE3-NEXT:    andl $15, %r15d
969 ; SSSE3-NEXT:    movzbl -24(%rsp,%r15), %eax
970 ; SSSE3-NEXT:    movd %eax, %xmm13
971 ; SSSE3-NEXT:    andl $15, %r14d
972 ; SSSE3-NEXT:    movzbl -24(%rsp,%r14), %eax
973 ; SSSE3-NEXT:    movd %eax, %xmm4
974 ; SSSE3-NEXT:    andl $15, %r11d
975 ; SSSE3-NEXT:    movzbl -24(%rsp,%r11), %eax
976 ; SSSE3-NEXT:    movd %eax, %xmm14
977 ; SSSE3-NEXT:    andl $15, %r10d
978 ; SSSE3-NEXT:    movzbl -24(%rsp,%r10), %eax
979 ; SSSE3-NEXT:    movd %eax, %xmm1
980 ; SSSE3-NEXT:    andl $15, %r9d
981 ; SSSE3-NEXT:    movzbl -24(%rsp,%r9), %eax
982 ; SSSE3-NEXT:    movd %eax, %xmm2
983 ; SSSE3-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
984 ; SSSE3-NEXT:    andl $15, %eax
985 ; SSSE3-NEXT:    movzbl -24(%rsp,%rax), %eax
986 ; SSSE3-NEXT:    movd %eax, %xmm0
987 ; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm15 = xmm15[0],xmm8[0],xmm15[1],xmm8[1],xmm15[2],xmm8[2],xmm1…
988 ; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm9[0],xmm3[1],xmm9[1],xmm3[2],xmm9[2],xmm3[3],…
989 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm15[0],xmm3[1],xmm15[1],xmm3[2],xmm15[2],xmm3[…
990 ; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm10[0],xmm7[1],xmm10[1],xmm7[2],xmm10[2],xmm7[…
991 ; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0],xmm11[0],xmm6[1],xmm11[1],xmm6[2],xmm11[2],xmm6[…
992 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm6 = xmm6[0],xmm7[0],xmm6[1],xmm7[1],xmm6[2],xmm7[2],xmm6[3],…
993 ; SSSE3-NEXT:    punpckldq {{.*#+}} xmm6 = xmm6[0],xmm3[0],xmm6[1],xmm3[1]
994 ; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm12[0],xmm5[1],xmm12[1],xmm5[2],xmm12[2],xmm5[…
995 ; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm13[0],xmm4[1],xmm13[1],xmm4[2],xmm13[2],xmm4[…
996 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1],xmm4[2],xmm5[2],xmm4[3],…
997 ; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm14[0],xmm1[1],xmm14[1],xmm1[2],xmm14[2],xmm1[…
998 ; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],…
999 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],…
1000 ; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1]
1001 ; SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm6[0]
1002 ; SSSE3-NEXT:    popq %rbx
1003 ; SSSE3-NEXT:    popq %r12
1004 ; SSSE3-NEXT:    popq %r13
1005 ; SSSE3-NEXT:    popq %r14
1006 ; SSSE3-NEXT:    popq %r15
1007 ; SSSE3-NEXT:    popq %rbp
1008 ; SSSE3-NEXT:    retq
1234 ; SSSE3-LABEL: var_shuffle_v4f32_v4f32_x0yx_i32:
1235 ; SSSE3:       # %bb.0:
1236 ; SSSE3-NEXT:    # kill: def $ecx killed $ecx def $rcx
1237 ; SSSE3-NEXT:    # kill: def $edx killed $edx def $rdx
1238 ; SSSE3-NEXT:    # kill: def $edi killed $edi def $rdi
1239 ; SSSE3-NEXT:    andl $3, %edi
1240 ; SSSE3-NEXT:    movaps %xmm1, -{{[0-9]+}}(%rsp)
1241 ; SSSE3-NEXT:    andl $3, %edx
1242 ; SSSE3-NEXT:    movaps %xmm0, -{{[0-9]+}}(%rsp)
1243 ; SSSE3-NEXT:    andl $3, %ecx
1244 ; SSSE3-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
1245 ; SSSE3-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
1246 ; SSSE3-NEXT:    movss {{.*#+}} xmm2 = mem[0],zero,zero,zero
1247 ; SSSE3-NEXT:    unpcklps {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
1248 ; SSSE3-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
1249 ; SSSE3-NEXT:    retq
1331 ; SSSE3-LABEL: var_shuffle_v8i16_v8i16_xyxyxy00_i16:
1332 ; SSSE3:       # %bb.0:
1333 ; SSSE3-NEXT:    # kill: def $r9d killed $r9d def $r9
1334 ; SSSE3-NEXT:    # kill: def $r8d killed $r8d def $r8
1335 ; SSSE3-NEXT:    # kill: def $ecx killed $ecx def $rcx
1336 ; SSSE3-NEXT:    # kill: def $edx killed $edx def $rdx
1337 ; SSSE3-NEXT:    # kill: def $esi killed $esi def $rsi
1338 ; SSSE3-NEXT:    # kill: def $edi killed $edi def $rdi
1339 ; SSSE3-NEXT:    andl $7, %edi
1340 ; SSSE3-NEXT:    andl $7, %esi
1341 ; SSSE3-NEXT:    andl $7, %edx
1342 ; SSSE3-NEXT:    andl $7, %ecx
1343 ; SSSE3-NEXT:    movaps %xmm0, -{{[0-9]+}}(%rsp)
1344 ; SSSE3-NEXT:    andl $7, %r8d
1345 ; SSSE3-NEXT:    movaps %xmm1, -{{[0-9]+}}(%rsp)
1346 ; SSSE3-NEXT:    andl $7, %r9d
1347 ; SSSE3-NEXT:    movzwl -24(%rsp,%rcx,2), %eax
1348 ; SSSE3-NEXT:    movd %eax, %xmm0
1349 ; SSSE3-NEXT:    movzwl -40(%rsp,%rdx,2), %eax
1350 ; SSSE3-NEXT:    movd %eax, %xmm1
1351 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],…
1352 ; SSSE3-NEXT:    movzwl -24(%rsp,%rsi,2), %eax
1353 ; SSSE3-NEXT:    movd %eax, %xmm2
1354 ; SSSE3-NEXT:    movzwl -40(%rsp,%rdi,2), %eax
1355 ; SSSE3-NEXT:    movd %eax, %xmm0
1356 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],…
1357 ; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
1358 ; SSSE3-NEXT:    movzwl -24(%rsp,%r9,2), %eax
1359 ; SSSE3-NEXT:    movd %eax, %xmm1
1360 ; SSSE3-NEXT:    movzwl -40(%rsp,%r8,2), %eax
1361 ; SSSE3-NEXT:    movd %eax, %xmm2
1362 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],…
1363 ; SSSE3-NEXT:    pxor %xmm1, %xmm1
1364 ; SSSE3-NEXT:    movss {{.*#+}} xmm1 = xmm2[0],xmm1[1,2,3]
1365 ; SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
1366 ; SSSE3-NEXT:    retq