Lines Matching refs:xmm1

307 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0]
308 ; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
313 ; SSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0]
314 ; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
319 ; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0]
320 ; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
325 ; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,2,0]
326 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5],xmm1[6,7]
331 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,2,0]
332 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5],xmm1[6,7]
337 ; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
338 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1,2],xmm1[3]
346 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0]
347 ; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
352 ; SSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0]
353 ; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
358 ; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0]
359 ; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
364 ; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
366 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5],xmm0[6,7]
371 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
373 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5],xmm0[6,7]
378 ; AVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
380 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2],xmm0[3]
388 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,0]
389 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[1,2]
390 ; SSE2-NEXT:    movaps %xmm1, %xmm0
395 ; SSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,0]
396 ; SSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[1,2]
397 ; SSE3-NEXT:    movaps %xmm1, %xmm0
402 ; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,0]
403 ; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[1,2]
404 ; SSSE3-NEXT:    movaps %xmm1, %xmm0
409 ; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
411 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7]
416 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
418 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7]
423 ; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
425 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
433 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,0]
434 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,2],xmm0[1,2]
435 ; SSE2-NEXT:    movaps %xmm1, %xmm0
440 ; SSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,0]
441 ; SSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,2],xmm0[1,2]
442 ; SSE3-NEXT:    movaps %xmm1, %xmm0
447 ; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,0]
448 ; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,2],xmm0[1,2]
449 ; SSSE3-NEXT:    movaps %xmm1, %xmm0
455 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3,4,5,6,7]
461 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3,4,5,6,7]
467 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3]
475 ; SSE-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
480 ; AVX-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
488 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
494 ; SSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
500 ; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
506 ; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
508 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5],xmm0[6,7]
513 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
515 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5],xmm0[6,7]
520 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
522 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[3]
530 ; SSE-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
531 ; SSE-NEXT:    movdqa %xmm1, %xmm0
536 ; AVX-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
544 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
550 ; SSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
556 ; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
562 ; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
564 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3,4,5],xmm1[6,7]
569 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
571 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3,4,5],xmm1[6,7]
576 ; AVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
578 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0],xmm0[1,2],xmm1[3]
587 ; SSE2-NEXT:    xorps %xmm1, %xmm1
588 ; SSE2-NEXT:    movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
589 ; SSE2-NEXT:    movaps %xmm1, %xmm0
594 ; SSE3-NEXT:    xorps %xmm1, %xmm1
595 ; SSE3-NEXT:    movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
596 ; SSE3-NEXT:    movaps %xmm1, %xmm0
601 ; SSSE3-NEXT:    xorps %xmm1, %xmm1
602 ; SSSE3-NEXT:    movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
603 ; SSSE3-NEXT:    movaps %xmm1, %xmm0
608 ; SSE41-NEXT:    xorps %xmm1, %xmm1
609 ; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]
614 ; AVX-NEXT:    vxorps %xmm1, %xmm1, %xmm1
615 ; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]
624 ; SSE2-NEXT:    xorps %xmm1, %xmm1
625 ; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,0]
626 ; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
631 ; SSE3-NEXT:    xorps %xmm1, %xmm1
632 ; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,0]
633 ; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
638 ; SSSE3-NEXT:    xorps %xmm1, %xmm1
639 ; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,0]
640 ; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
659 ; SSE2-NEXT:    xorps %xmm1, %xmm1
660 ; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[3,0]
661 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[0,2]
662 ; SSE2-NEXT:    movaps %xmm1, %xmm0
667 ; SSE3-NEXT:    xorps %xmm1, %xmm1
668 ; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[3,0]
669 ; SSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[0,2]
670 ; SSE3-NEXT:    movaps %xmm1, %xmm0
675 ; SSSE3-NEXT:    xorps %xmm1, %xmm1
676 ; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[3,0]
677 ; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[0,2]
678 ; SSSE3-NEXT:    movaps %xmm1, %xmm0
697 ; SSE2-NEXT:    xorps %xmm1, %xmm1
698 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
699 ; SSE2-NEXT:    movaps %xmm1, %xmm0
704 ; SSE3-NEXT:    xorps %xmm1, %xmm1
705 ; SSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
706 ; SSE3-NEXT:    movaps %xmm1, %xmm0
711 ; SSSE3-NEXT:    xorps %xmm1, %xmm1
712 ; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
713 ; SSSE3-NEXT:    movaps %xmm1, %xmm0
732 ; SSE2-NEXT:    xorps %xmm1, %xmm1
733 ; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,0],xmm1[2,0]
734 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
735 ; SSE2-NEXT:    movaps %xmm1, %xmm0
740 ; SSE3-NEXT:    xorps %xmm1, %xmm1
741 ; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,0],xmm1[2,0]
742 ; SSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
743 ; SSE3-NEXT:    movaps %xmm1, %xmm0
748 ; SSSE3-NEXT:    xorps %xmm1, %xmm1
749 ; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,0],xmm1[2,0]
750 ; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
751 ; SSSE3-NEXT:    movaps %xmm1, %xmm0
756 ; SSE41-NEXT:    xorps %xmm1, %xmm1
757 ; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[3]
762 ; AVX-NEXT:    vxorps %xmm1, %xmm1, %xmm1
763 ; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[3]
772 ; SSE2-NEXT:    xorps %xmm1, %xmm1
773 ; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[0,0]
774 ; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
779 ; SSE3-NEXT:    xorps %xmm1, %xmm1
780 ; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[0,0]
781 ; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
786 ; SSSE3-NEXT:    xorps %xmm1, %xmm1
787 ; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[0,0]
788 ; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
807 ; SSE2-NEXT:    xorps %xmm1, %xmm1
808 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,0],xmm0[0,0]
809 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[2,3]
810 ; SSE2-NEXT:    movaps %xmm1, %xmm0
815 ; SSE3-NEXT:    xorps %xmm1, %xmm1
816 ; SSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,0],xmm0[0,0]
817 ; SSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[2,3]
818 ; SSE3-NEXT:    movaps %xmm1, %xmm0
823 ; SSSE3-NEXT:    xorps %xmm1, %xmm1
824 ; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,0],xmm0[0,0]
825 ; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[2,3]
826 ; SSSE3-NEXT:    movaps %xmm1, %xmm0
831 ; SSE41-NEXT:    xorps %xmm1, %xmm1
832 ; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
837 ; AVX-NEXT:    vxorps %xmm1, %xmm1, %xmm1
838 ; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
847 ; SSE2-NEXT:    xorps %xmm1, %xmm1
848 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
849 ; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
854 ; SSE3-NEXT:    xorps %xmm1, %xmm1
855 ; SSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
856 ; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
861 ; SSSE3-NEXT:    xorps %xmm1, %xmm1
862 ; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
863 ; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
868 ; SSE41-NEXT:    xorps %xmm1, %xmm1
869 ; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2],xmm0[3]
874 ; AVX-NEXT:    vxorps %xmm1, %xmm1, %xmm1
875 ; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2],xmm0[3]
884 ; SSE2-NEXT:    xorps %xmm1, %xmm1
885 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[3,0],xmm0[2,0]
886 ; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
891 ; SSE3-NEXT:    xorps %xmm1, %xmm1
892 ; SSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[3,0],xmm0[2,0]
893 ; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
898 ; SSSE3-NEXT:    xorps %xmm1, %xmm1
899 ; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[3,0],xmm0[2,0]
900 ; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
905 ; SSE41-NEXT:    xorps %xmm1, %xmm1
906 ; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm0[0,1,2],xmm1[3]
911 ; AVX-NEXT:    vxorps %xmm1, %xmm1, %xmm1
912 ; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0,1,2],xmm1[3]
921 ; SSE2-NEXT:    xorps %xmm1, %xmm1
922 ; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,3],xmm1[1,2]
928 ; SSE3-NEXT:    xorps %xmm1, %xmm1
929 ; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,3],xmm1[1,2]
935 ; SSSE3-NEXT:    xorps %xmm1, %xmm1
936 ; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,3],xmm1[1,2]
942 ; SSE41-NEXT:    xorps %xmm1, %xmm1
943 ; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[3]
948 ; AVX-NEXT:    vxorps %xmm1, %xmm1, %xmm1
949 ; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[3]
958 ; SSE-NEXT:    unpcklps {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
959 ; SSE-NEXT:    movaps %xmm1, %xmm0
964 ; AVX-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
973 ; SSE2-NEXT:    xorps %xmm1, %xmm1
974 ; SSE2-NEXT:    movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
975 ; SSE2-NEXT:    movaps %xmm1, %xmm0
980 ; SSE3-NEXT:    xorps %xmm1, %xmm1
981 ; SSE3-NEXT:    movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
982 ; SSE3-NEXT:    movaps %xmm1, %xmm0
987 ; SSSE3-NEXT:    xorps %xmm1, %xmm1
988 ; SSSE3-NEXT:    movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
989 ; SSSE3-NEXT:    movaps %xmm1, %xmm0
994 ; SSE41-NEXT:    pxor %xmm1, %xmm1
995 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
1000 ; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
1001 ; AVX-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
1010 ; SSE2-NEXT:    xorps %xmm1, %xmm1
1011 ; SSE2-NEXT:    movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1012 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,0,1,1]
1017 ; SSE3-NEXT:    xorps %xmm1, %xmm1
1018 ; SSE3-NEXT:    movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1019 ; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,0,1,1]
1024 ; SSSE3-NEXT:    xorps %xmm1, %xmm1
1025 ; SSSE3-NEXT:    movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1026 ; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,0,1,1]
1031 ; SSE41-NEXT:    pxor %xmm1, %xmm1
1032 ; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3,4,5,6,7]
1033 ; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,0,1,1]
1038 ; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
1039 ; AVX-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
1049 ; SSE2-NEXT:    xorps %xmm1, %xmm1
1050 ; SSE2-NEXT:    movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1051 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,0,1]
1056 ; SSE3-NEXT:    xorps %xmm1, %xmm1
1057 ; SSE3-NEXT:    movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1058 ; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,0,1]
1063 ; SSSE3-NEXT:    xorps %xmm1, %xmm1
1064 ; SSSE3-NEXT:    movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1065 ; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,0,1]
1070 ; SSE41-NEXT:    pxor %xmm1, %xmm1
1071 ; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3,4,5,6,7]
1072 ; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,0,1]
1077 ; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
1078 ; AVX-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
1102 ; SSE2-NEXT:    xorps %xmm1, %xmm1
1103 ; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[0,0]
1104 ; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
1109 ; SSE3-NEXT:    xorps %xmm1, %xmm1
1110 ; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[0,0]
1111 ; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
1116 ; SSSE3-NEXT:    xorps %xmm1, %xmm1
1117 ; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[0,0]
1118 ; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
1123 ; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,2,3,3]
1125 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7]
1131 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
1132 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3],xmm1[4,5,6,7]
1138 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
1139 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
1148 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[3,0],xmm0[0,0]
1149 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,2],xmm0[1,2]
1150 ; SSE2-NEXT:    movaps %xmm1, %xmm0
1155 ; SSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[3,0],xmm0[0,0]
1156 ; SSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,2],xmm0[1,2]
1157 ; SSE3-NEXT:    movaps %xmm1, %xmm0
1162 ; SSSE3-NEXT:    palignr {{.*#+}} xmm0 = xmm1[12,13,14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11]
1167 ; SSE41-NEXT:    palignr {{.*#+}} xmm0 = xmm1[12,13,14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11]
1172 ; AVX-NEXT:    vpalignr {{.*#+}} xmm0 = xmm1[12,13,14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11]
1181 ; SSE2-NEXT:    shufpd {{.*#+}} xmm1 = xmm1[1],xmm0[0]
1182 ; SSE2-NEXT:    movapd %xmm1, %xmm0
1187 ; SSE3-NEXT:    shufpd {{.*#+}} xmm1 = xmm1[1],xmm0[0]
1188 ; SSE3-NEXT:    movapd %xmm1, %xmm0
1193 ; SSSE3-NEXT:    palignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
1198 ; SSE41-NEXT:    palignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
1203 ; AVX-NEXT:    vpalignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
1212 ; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[3,0]
1213 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,2],xmm0[2,0]
1214 ; SSE2-NEXT:    movaps %xmm1, %xmm0
1219 ; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[3,0]
1220 ; SSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,2],xmm0[2,0]
1221 ; SSE3-NEXT:    movaps %xmm1, %xmm0
1226 ; SSSE3-NEXT:    palignr {{.*#+}} xmm0 = xmm1[4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2,3]
1231 ; SSE41-NEXT:    palignr {{.*#+}} xmm0 = xmm1[4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2,3]
1236 ; AVX-NEXT:    vpalignr {{.*#+}} xmm0 = xmm1[4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2,3]
1245 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[3,0]
1246 ; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,2],xmm1[2,0]
1251 ; SSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[3,0]
1252 ; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,2],xmm1[2,0]
1257 ; SSSE3-NEXT:    palignr {{.*#+}} xmm1 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3]
1258 ; SSSE3-NEXT:    movdqa %xmm1, %xmm0
1263 ; SSE41-NEXT:    palignr {{.*#+}} xmm1 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3]
1264 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
1269 ; AVX-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3]
1278 ; SSE2-NEXT:    shufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0]
1283 ; SSE3-NEXT:    shufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0]
1288 ; SSSE3-NEXT:    palignr {{.*#+}} xmm1 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
1289 ; SSSE3-NEXT:    movdqa %xmm1, %xmm0
1294 ; SSE41-NEXT:    palignr {{.*#+}} xmm1 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
1295 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
1300 ; AVX-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
1309 ; SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
1310 ; SSE-NEXT:    movdqa %xmm1, %xmm0
1315 ; AVX-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
1324 ; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,0],xmm1[0,0]
1325 ; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,2]
1330 ; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,0],xmm1[0,0]
1331 ; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,2]
1336 ; SSSE3-NEXT:    palignr {{.*#+}} xmm1 = xmm0[12,13,14,15],xmm1[0,1,2,3,4,5,6,7,8,9,10,11]
1337 ; SSSE3-NEXT:    movdqa %xmm1, %xmm0
1342 ; SSE41-NEXT:    palignr {{.*#+}} xmm1 = xmm0[12,13,14,15],xmm1[0,1,2,3,4,5,6,7,8,9,10,11]
1343 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
1348 ; AVX-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[12,13,14,15],xmm1[0,1,2,3,4,5,6,7,8,9,10,11]
1386 ; SSE2-NEXT:    pxor %xmm1, %xmm1
1387 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
1392 ; SSE3-NEXT:    pxor %xmm1, %xmm1
1393 ; SSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
1398 ; SSSE3-NEXT:    pxor %xmm1, %xmm1
1399 ; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
1447 ; SSE41-NEXT:    pxor %xmm1, %xmm1
1448 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7]
1453 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
1454 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7]
1459 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
1460 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
1484 ; SSE41-NEXT:    pxor %xmm1, %xmm1
1485 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5],xmm0[6,7]
1490 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
1491 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5],xmm0[6,7]
1496 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
1497 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2],xmm0[3]
1521 ; SSE41-NEXT:    pxor %xmm1, %xmm1
1522 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5],xmm1[6,7]
1527 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
1528 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5],xmm1[6,7]
1533 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
1534 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1,2],xmm1[3]
1558 ; SSE41-NEXT:    pxor %xmm1, %xmm1
1559 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5],xmm0[6,7]
1564 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
1565 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5],xmm0[6,7]
1570 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
1571 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[3]
1580 ; SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
1585 ; AVX-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
1628 ; SSE2-NEXT:    xorps %xmm1, %xmm1
1629 ; SSE2-NEXT:    movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1630 ; SSE2-NEXT:    movaps %xmm1, %xmm0
1635 ; SSE3-NEXT:    xorps %xmm1, %xmm1
1636 ; SSE3-NEXT:    movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1637 ; SSE3-NEXT:    movaps %xmm1, %xmm0
1642 ; SSSE3-NEXT:    xorps %xmm1, %xmm1
1643 ; SSSE3-NEXT:    movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1644 ; SSSE3-NEXT:    movaps %xmm1, %xmm0
1649 ; SSE41-NEXT:    xorps %xmm1, %xmm1
1650 ; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]
1655 ; AVX-NEXT:    vxorps %xmm1, %xmm1, %xmm1
1656 ; AVX-NEXT:    vmovss {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]
1682 ; SSE2-NEXT:    movd %rdi, %xmm1
1683 ; SSE2-NEXT:    movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
1688 ; SSE3-NEXT:    movd %rdi, %xmm1
1689 ; SSE3-NEXT:    movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
1694 ; SSSE3-NEXT:    movd %rdi, %xmm1
1695 ; SSSE3-NEXT:    movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
1700 ; SSE41-NEXT:    movd %rdi, %xmm1
1701 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
1706 ; AVX1-NEXT:    vmovq %rdi, %xmm1
1707 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
1712 ; AVX2-NEXT:    vmovq %rdi, %xmm1
1713 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
1739 ; SSE41-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero
1740 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
1745 ; AVX1-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
1746 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
1751 ; AVX2-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
1752 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
1763 ; SSE-NEXT:    movd %rdi, %xmm1
1764 ; SSE-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
1769 ; AVX-NEXT:    vmovq %rdi, %xmm1
1770 ; AVX-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
1781 ; SSE-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero
1782 ; SSE-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
1787 ; AVX-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
1788 ; AVX-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
1799 ; SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm0[0],xmm1[1]
1800 ; SSE-NEXT:    movapd %xmm1, %xmm0
1805 ; AVX-NEXT:    vmovsd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
1832 ; SSE-NEXT:    unpcklpd {{.*#+}} xmm1 = xmm1[0],xmm0[0]
1833 ; SSE-NEXT:    movapd %xmm1, %xmm0
1838 ; AVX-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]