• Home
  • Raw
  • Download

Lines Matching refs:xmm1

308 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0]
309 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
314 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0]
315 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
320 ; SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0]
321 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
326 ; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,2,0]
327 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5],xmm1[6,7]
332 ; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,1,2,0]
333 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5],xmm1[6,7]
338 ; AVX2-NEXT: vpbroadcastd %xmm1, %xmm1
339 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1,2],xmm1[3]
347 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0]
348 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
353 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0]
354 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
359 ; SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0]
360 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
365 ; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
367 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5],xmm0[6,7]
372 ; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
374 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5],xmm0[6,7]
379 ; AVX2-NEXT: vpbroadcastq %xmm1, %xmm1
381 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2],xmm0[3]
389 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,0]
390 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[1,2]
391 ; SSE2-NEXT: movaps %xmm1, %xmm0
396 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,0]
397 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[1,2]
398 ; SSE3-NEXT: movaps %xmm1, %xmm0
403 ; SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,0]
404 ; SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[1,2]
405 ; SSSE3-NEXT: movaps %xmm1, %xmm0
410 ; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
412 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7]
417 ; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
419 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7]
424 ; AVX2-NEXT: vpbroadcastd %xmm1, %xmm1
426 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
434 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,0]
435 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2],xmm0[1,2]
436 ; SSE2-NEXT: movaps %xmm1, %xmm0
441 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,0]
442 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2],xmm0[1,2]
443 ; SSE3-NEXT: movaps %xmm1, %xmm0
448 ; SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,0]
449 ; SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2],xmm0[1,2]
450 ; SSSE3-NEXT: movaps %xmm1, %xmm0
456 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3,4,5,6,7]
462 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3,4,5,6,7]
468 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3]
476 ; SSE-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
481 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
489 ; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
495 ; SSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
501 ; SSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
507 ; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
509 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5],xmm0[6,7]
514 ; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
516 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5],xmm0[6,7]
521 ; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
523 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[3]
531 ; SSE-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
532 ; SSE-NEXT: movdqa %xmm1, %xmm0
537 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
545 ; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
551 ; SSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
557 ; SSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
563 ; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
565 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3,4,5],xmm1[6,7]
570 ; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
572 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3,4,5],xmm1[6,7]
577 ; AVX2-NEXT: vpbroadcastq %xmm1, %xmm1
579 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0],xmm0[1,2],xmm1[3]
588 ; SSE2-NEXT: xorps %xmm1, %xmm1
589 ; SSE2-NEXT: movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
590 ; SSE2-NEXT: movaps %xmm1, %xmm0
595 ; SSE3-NEXT: xorps %xmm1, %xmm1
596 ; SSE3-NEXT: movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
597 ; SSE3-NEXT: movaps %xmm1, %xmm0
602 ; SSSE3-NEXT: xorps %xmm1, %xmm1
603 ; SSSE3-NEXT: movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
604 ; SSSE3-NEXT: movaps %xmm1, %xmm0
609 ; SSE41-NEXT: xorps %xmm1, %xmm1
610 ; SSE41-NEXT: blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]
615 ; AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1
616 ; AVX-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]
625 ; SSE2-NEXT: xorps %xmm1, %xmm1
626 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,0]
627 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
632 ; SSE3-NEXT: xorps %xmm1, %xmm1
633 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,0]
634 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
639 ; SSSE3-NEXT: xorps %xmm1, %xmm1
640 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,0]
641 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
660 ; SSE2-NEXT: xorps %xmm1, %xmm1
661 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[3,0]
662 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[0,2]
663 ; SSE2-NEXT: movaps %xmm1, %xmm0
668 ; SSE3-NEXT: xorps %xmm1, %xmm1
669 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[3,0]
670 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[0,2]
671 ; SSE3-NEXT: movaps %xmm1, %xmm0
676 ; SSSE3-NEXT: xorps %xmm1, %xmm1
677 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[3,0]
678 ; SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[0,2]
679 ; SSSE3-NEXT: movaps %xmm1, %xmm0
698 ; SSE2-NEXT: xorps %xmm1, %xmm1
699 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
700 ; SSE2-NEXT: movaps %xmm1, %xmm0
705 ; SSE3-NEXT: xorps %xmm1, %xmm1
706 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
707 ; SSE3-NEXT: movaps %xmm1, %xmm0
712 ; SSSE3-NEXT: xorps %xmm1, %xmm1
713 ; SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
714 ; SSSE3-NEXT: movaps %xmm1, %xmm0
733 ; SSE2-NEXT: xorps %xmm1, %xmm1
734 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,0],xmm1[2,0]
735 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
736 ; SSE2-NEXT: movaps %xmm1, %xmm0
741 ; SSE3-NEXT: xorps %xmm1, %xmm1
742 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,0],xmm1[2,0]
743 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
744 ; SSE3-NEXT: movaps %xmm1, %xmm0
749 ; SSSE3-NEXT: xorps %xmm1, %xmm1
750 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,0],xmm1[2,0]
751 ; SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
752 ; SSSE3-NEXT: movaps %xmm1, %xmm0
757 ; SSE41-NEXT: xorps %xmm1, %xmm1
758 ; SSE41-NEXT: blendps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[3]
763 ; AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1
764 ; AVX-NEXT: vblendps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[3]
773 ; SSE2-NEXT: xorps %xmm1, %xmm1
774 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[0,0]
775 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
780 ; SSE3-NEXT: xorps %xmm1, %xmm1
781 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[0,0]
782 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
787 ; SSSE3-NEXT: xorps %xmm1, %xmm1
788 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[0,0]
789 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
808 ; SSE2-NEXT: xorps %xmm1, %xmm1
809 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,0],xmm0[0,0]
810 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[2,3]
811 ; SSE2-NEXT: movaps %xmm1, %xmm0
816 ; SSE3-NEXT: xorps %xmm1, %xmm1
817 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,0],xmm0[0,0]
818 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[2,3]
819 ; SSE3-NEXT: movaps %xmm1, %xmm0
824 ; SSSE3-NEXT: xorps %xmm1, %xmm1
825 ; SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,0],xmm0[0,0]
826 ; SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[2,3]
827 ; SSSE3-NEXT: movaps %xmm1, %xmm0
832 ; SSE41-NEXT: xorps %xmm1, %xmm1
833 ; SSE41-NEXT: blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
838 ; AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1
839 ; AVX-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
848 ; SSE2-NEXT: xorps %xmm1, %xmm1
849 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
850 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
855 ; SSE3-NEXT: xorps %xmm1, %xmm1
856 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
857 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
862 ; SSSE3-NEXT: xorps %xmm1, %xmm1
863 ; SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
864 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
869 ; SSE41-NEXT: xorps %xmm1, %xmm1
870 ; SSE41-NEXT: blendps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2],xmm0[3]
875 ; AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1
876 ; AVX-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2],xmm0[3]
885 ; SSE2-NEXT: xorps %xmm1, %xmm1
886 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,0],xmm0[2,0]
887 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
892 ; SSE3-NEXT: xorps %xmm1, %xmm1
893 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,0],xmm0[2,0]
894 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
899 ; SSSE3-NEXT: xorps %xmm1, %xmm1
900 ; SSSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,0],xmm0[2,0]
901 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
906 ; SSE41-NEXT: xorps %xmm1, %xmm1
907 ; SSE41-NEXT: blendps {{.*#+}} xmm0 = xmm0[0,1,2],xmm1[3]
912 ; AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1
913 ; AVX-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0,1,2],xmm1[3]
922 ; SSE2-NEXT: xorps %xmm1, %xmm1
923 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,3],xmm1[1,2]
929 ; SSE3-NEXT: xorps %xmm1, %xmm1
930 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,3],xmm1[1,2]
936 ; SSSE3-NEXT: xorps %xmm1, %xmm1
937 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,3],xmm1[1,2]
943 ; SSE41-NEXT: xorps %xmm1, %xmm1
944 ; SSE41-NEXT: blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[3]
949 ; AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1
950 ; AVX-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[3]
959 ; SSE2-NEXT: xorps %xmm1, %xmm1
960 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,0]
966 ; SSE3-NEXT: xorps %xmm1, %xmm1
967 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,0]
973 ; SSSE3-NEXT: xorps %xmm1, %xmm1
974 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,0]
980 ; SSE41-NEXT: xorps %xmm1, %xmm1
981 ; SSE41-NEXT: blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
986 ; AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1
987 ; AVX-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
996 ; SSE-NEXT: unpcklps {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
997 ; SSE-NEXT: movaps %xmm1, %xmm0
1002 ; AVX-NEXT: vunpcklps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
1011 ; SSE2-NEXT: xorps %xmm1, %xmm1
1012 ; SSE2-NEXT: movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1013 ; SSE2-NEXT: movaps %xmm1, %xmm0
1018 ; SSE3-NEXT: xorps %xmm1, %xmm1
1019 ; SSE3-NEXT: movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1020 ; SSE3-NEXT: movaps %xmm1, %xmm0
1025 ; SSSE3-NEXT: xorps %xmm1, %xmm1
1026 ; SSSE3-NEXT: movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1027 ; SSSE3-NEXT: movaps %xmm1, %xmm0
1032 ; SSE41-NEXT: pxor %xmm1, %xmm1
1033 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
1038 ; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
1039 ; AVX-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
1048 ; SSE2-NEXT: xorps %xmm1, %xmm1
1049 ; SSE2-NEXT: movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1050 ; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,0,1,1]
1055 ; SSE3-NEXT: xorps %xmm1, %xmm1
1056 ; SSE3-NEXT: movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1057 ; SSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,0,1,1]
1062 ; SSSE3-NEXT: xorps %xmm1, %xmm1
1063 ; SSSE3-NEXT: movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1064 ; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,0,1,1]
1069 ; SSE41-NEXT: pxor %xmm1, %xmm1
1070 ; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3,4,5,6,7]
1071 ; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,0,1,1]
1076 ; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
1077 ; AVX-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
1087 ; SSE2-NEXT: xorps %xmm1, %xmm1
1088 ; SSE2-NEXT: movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1089 ; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,0,1]
1094 ; SSE3-NEXT: xorps %xmm1, %xmm1
1095 ; SSE3-NEXT: movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1096 ; SSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,0,1]
1101 ; SSSE3-NEXT: xorps %xmm1, %xmm1
1102 ; SSSE3-NEXT: movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1103 ; SSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,0,1]
1108 ; SSE41-NEXT: pxor %xmm1, %xmm1
1109 ; SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3,4,5,6,7]
1110 ; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,0,1]
1115 ; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
1116 ; AVX-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
1140 ; SSE2-NEXT: xorps %xmm1, %xmm1
1141 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[0,0]
1142 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
1147 ; SSE3-NEXT: xorps %xmm1, %xmm1
1148 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[0,0]
1149 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
1154 ; SSSE3-NEXT: xorps %xmm1, %xmm1
1155 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[0,0]
1156 ; SSSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
1161 ; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,2,3,3]
1163 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7]
1169 ; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
1170 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3],xmm1[4,5,6,7]
1176 ; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
1177 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
1186 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,0],xmm0[0,0]
1187 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2],xmm0[1,2]
1188 ; SSE2-NEXT: movaps %xmm1, %xmm0
1193 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,0],xmm0[0,0]
1194 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2],xmm0[1,2]
1195 ; SSE3-NEXT: movaps %xmm1, %xmm0
1200 ; SSSE3-NEXT: palignr {{.*#+}} xmm0 = xmm1[12,13,14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11]
1205 ; SSE41-NEXT: palignr {{.*#+}} xmm0 = xmm1[12,13,14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11]
1210 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm1[12,13,14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10,11]
1219 ; SSE2-NEXT: shufpd {{.*#+}} xmm1 = xmm1[1],xmm0[0]
1220 ; SSE2-NEXT: movapd %xmm1, %xmm0
1225 ; SSE3-NEXT: shufpd {{.*#+}} xmm1 = xmm1[1],xmm0[0]
1226 ; SSE3-NEXT: movapd %xmm1, %xmm0
1231 ; SSSE3-NEXT: palignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
1236 ; SSE41-NEXT: palignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
1241 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
1250 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[3,0]
1251 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,2],xmm0[2,0]
1252 ; SSE2-NEXT: movaps %xmm1, %xmm0
1257 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[3,0]
1258 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,2],xmm0[2,0]
1259 ; SSE3-NEXT: movaps %xmm1, %xmm0
1264 ; SSSE3-NEXT: palignr {{.*#+}} xmm0 = xmm1[4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2,3]
1269 ; SSE41-NEXT: palignr {{.*#+}} xmm0 = xmm1[4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2,3]
1274 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm1[4,5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2,3]
1283 ; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[3,0]
1284 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,2],xmm1[2,0]
1289 ; SSE3-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[3,0]
1290 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,2],xmm1[2,0]
1295 ; SSSE3-NEXT: palignr {{.*#+}} xmm1 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3]
1296 ; SSSE3-NEXT: movdqa %xmm1, %xmm0
1301 ; SSE41-NEXT: palignr {{.*#+}} xmm1 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3]
1302 ; SSE41-NEXT: movdqa %xmm1, %xmm0
1307 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3]
1316 ; SSE2-NEXT: shufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0]
1321 ; SSE3-NEXT: shufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0]
1326 ; SSSE3-NEXT: palignr {{.*#+}} xmm1 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
1327 ; SSSE3-NEXT: movdqa %xmm1, %xmm0
1332 ; SSE41-NEXT: palignr {{.*#+}} xmm1 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
1333 ; SSE41-NEXT: movdqa %xmm1, %xmm0
1338 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7]
1347 ; SSE-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
1348 ; SSE-NEXT: movdqa %xmm1, %xmm0
1353 ; AVX-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
1362 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,0],xmm1[0,0]
1363 ; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,2]
1368 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,0],xmm1[0,0]
1369 ; SSE3-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,2]
1374 ; SSSE3-NEXT: palignr {{.*#+}} xmm1 = xmm0[12,13,14,15],xmm1[0,1,2,3,4,5,6,7,8,9,10,11]
1375 ; SSSE3-NEXT: movdqa %xmm1, %xmm0
1380 ; SSE41-NEXT: palignr {{.*#+}} xmm1 = xmm0[12,13,14,15],xmm1[0,1,2,3,4,5,6,7,8,9,10,11]
1381 ; SSE41-NEXT: movdqa %xmm1, %xmm0
1386 ; AVX-NEXT: vpalignr {{.*#+}} xmm0 = xmm0[12,13,14,15],xmm1[0,1,2,3,4,5,6,7,8,9,10,11]
1424 ; SSE2-NEXT: pxor %xmm1, %xmm1
1425 ; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
1430 ; SSE3-NEXT: pxor %xmm1, %xmm1
1431 ; SSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
1436 ; SSSE3-NEXT: pxor %xmm1, %xmm1
1437 ; SSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
1485 ; SSE41-NEXT: pxor %xmm1, %xmm1
1486 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7]
1491 ; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
1492 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7]
1497 ; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
1498 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
1522 ; SSE41-NEXT: pxor %xmm1, %xmm1
1523 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5],xmm0[6,7]
1528 ; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
1529 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4,5],xmm0[6,7]
1534 ; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
1535 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2],xmm0[3]
1559 ; SSE41-NEXT: pxor %xmm1, %xmm1
1560 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5],xmm1[6,7]
1565 ; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
1566 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5],xmm1[6,7]
1571 ; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
1572 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1,2],xmm1[3]
1596 ; SSE41-NEXT: pxor %xmm1, %xmm1
1597 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5],xmm0[6,7]
1602 ; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
1603 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5],xmm0[6,7]
1608 ; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
1609 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[3]
1618 ; SSE-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
1623 ; AVX-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
1635 ; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
1636 ; SSE-NEXT: unpcklpd {{.*#+}} xmm1 = xmm1[0],xmm0[0]
1637 ; SSE-NEXT: movapd %xmm1, %xmm0
1642 ; AVX-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
1643 ; AVX-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
1656 ; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,1]
1661 ; AVX-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,1]
1703 ; SSE2-NEXT: xorps %xmm1, %xmm1
1704 ; SSE2-NEXT: movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1705 ; SSE2-NEXT: movaps %xmm1, %xmm0
1710 ; SSE3-NEXT: xorps %xmm1, %xmm1
1711 ; SSE3-NEXT: movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1712 ; SSE3-NEXT: movaps %xmm1, %xmm0
1717 ; SSSE3-NEXT: xorps %xmm1, %xmm1
1718 ; SSSE3-NEXT: movss {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
1719 ; SSSE3-NEXT: movaps %xmm1, %xmm0
1724 ; SSE41-NEXT: xorps %xmm1, %xmm1
1725 ; SSE41-NEXT: blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]
1730 ; AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1
1731 ; AVX-NEXT: vmovss {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]
1757 ; SSE2-NEXT: movd %rdi, %xmm1
1758 ; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
1763 ; SSE3-NEXT: movd %rdi, %xmm1
1764 ; SSE3-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
1769 ; SSSE3-NEXT: movd %rdi, %xmm1
1770 ; SSSE3-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
1775 ; SSE41-NEXT: movd %rdi, %xmm1
1776 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
1781 ; AVX1-NEXT: vmovq %rdi, %xmm1
1782 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
1787 ; AVX2-NEXT: vmovq %rdi, %xmm1
1788 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
1814 ; SSE41-NEXT: movq {{.*#+}} xmm1 = mem[0],zero
1815 ; SSE41-NEXT: pblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
1820 ; AVX1-NEXT: vmovq {{.*#+}} xmm1 = mem[0],zero
1821 ; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
1826 ; AVX2-NEXT: vmovq {{.*#+}} xmm1 = mem[0],zero
1827 ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
1838 ; SSE-NEXT: movd %rdi, %xmm1
1839 ; SSE-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
1844 ; AVX-NEXT: vmovq %rdi, %xmm1
1845 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
1856 ; SSE-NEXT: movq {{.*#+}} xmm1 = mem[0],zero
1857 ; SSE-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
1862 ; AVX-NEXT: vmovq {{.*#+}} xmm1 = mem[0],zero
1863 ; AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
1874 ; SSE-NEXT: movsd {{.*#+}} xmm1 = xmm0[0],xmm1[1]
1875 ; SSE-NEXT: movapd %xmm1, %xmm0
1880 ; AVX-NEXT: vmovsd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
1907 ; SSE-NEXT: unpcklpd {{.*#+}} xmm1 = xmm1[0],xmm0[0]
1908 ; SSE-NEXT: movapd %xmm1, %xmm0
1913 ; AVX-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]