Mercurial > libavcodec.hg

comparison i386/dsputil_mmx.c @ 954:13aec7e50c52 libavcodec

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
qpel in mmx2/3dnow qpel refinement quality parameter
author michaelni
date Sun, 05 Jan 2003 15:57:10 +0000
parents caa77cd960c0
children 9bb668034ecf
comparison
equal deleted inserted replaced
953:9eb066d6e0db 954:13aec7e50c52
50 50
51 /* pixel operations */ 51 /* pixel operations */
52 static const uint64_t mm_bone __attribute__ ((aligned(8))) = 0x0101010101010101ULL; 52 static const uint64_t mm_bone __attribute__ ((aligned(8))) = 0x0101010101010101ULL;
53 static const uint64_t mm_wone __attribute__ ((aligned(8))) = 0x0001000100010001ULL; 53 static const uint64_t mm_wone __attribute__ ((aligned(8))) = 0x0001000100010001ULL;
54 static const uint64_t mm_wtwo __attribute__ ((aligned(8))) = 0x0002000200020002ULL; 54 static const uint64_t mm_wtwo __attribute__ ((aligned(8))) = 0x0002000200020002ULL;
55
56 static const uint64_t ff_pw_20 __attribute__ ((aligned(8))) = 0x0014001400140014ULL;
57 static const uint64_t ff_pw_3 __attribute__ ((aligned(8))) = 0x0003000300030003ULL;
58 static const uint64_t ff_pw_16 __attribute__ ((aligned(8))) = 0x0010001000100010ULL;
59 static const uint64_t ff_pw_15 __attribute__ ((aligned(8))) = 0x000F000F000F000FULL;
55 60
56 #define JUMPALIGN() __asm __volatile (".balign 8"::) 61 #define JUMPALIGN() __asm __volatile (".balign 8"::)
57 #define MOVQ_ZERO(regd) __asm __volatile ("pxor %%" #regd ", %%" #regd ::) 62 #define MOVQ_ZERO(regd) __asm __volatile ("pxor %%" #regd ", %%" #regd ::)
58 63
59 #define MOVQ_WONE(regd) \ 64 #define MOVQ_WONE(regd) \
644 return sum&0xFFFF; 649 return sum&0xFFFF;
645 } 650 }
646 651
647 WARPER88_1616(hadamard8_diff_mmx, hadamard8_diff16_mmx) 652 WARPER88_1616(hadamard8_diff_mmx, hadamard8_diff16_mmx)
648 653
654 #define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\
655 "paddw " #m4 ", " #m3 " \n\t" /* x1 */\
656 "movq " #pw_20 ", %%mm4 \n\t" /* 20 */\
657 "pmullw " #m3 ", %%mm4 \n\t" /* 20x1 */\
658 "movq "#in7", " #m3 " \n\t" /* d */\
659 "movq "#in0", %%mm5 \n\t" /* D */\
660 "paddw " #m3 ", %%mm5 \n\t" /* x4 */\
661 "psubw %%mm5, %%mm4 \n\t" /* 20x1 - x4 */\
662 "movq "#in1", %%mm5 \n\t" /* C */\
663 "movq "#in2", %%mm6 \n\t" /* B */\
664 "paddw " #m6 ", %%mm5 \n\t" /* x3 */\
665 "paddw " #m5 ", %%mm6 \n\t" /* x2 */\
666 "paddw %%mm6, %%mm6 \n\t" /* 2x2 */\
667 "psubw %%mm6, %%mm5 \n\t" /* -2x2 + x3 */\
668 "pmullw " #pw_3 ", %%mm5 \n\t" /* -6x2 + 3x3 */\
669 "paddw " #rnd ", %%mm4 \n\t" /* x2 */\
670 "paddw %%mm4, %%mm5 \n\t" /* 20x1 - 6x2 + 3x3 - x4 */\
671 "psraw $5, %%mm5 \n\t"\
672 "packuswb %%mm5, %%mm5 \n\t"\
673 OP(%%mm5, out, %%mm7, d)
674
675 #define QPEL_BASE(OPNAME, ROUNDER, RND, OP)\
676 void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
677 uint64_t temp;\
678 \
679 asm volatile(\
680 "pxor %%mm7, %%mm7 \n\t"\
681 "1: \n\t"\
682 "movq (%0), %%mm0 \n\t" /* ABCDEFGH */\
683 "movq %%mm0, %%mm1 \n\t" /* ABCDEFGH */\
684 "movq %%mm0, %%mm2 \n\t" /* ABCDEFGH */\
685 "punpcklbw %%mm7, %%mm0 \n\t" /* 0A0B0C0D */\
686 "punpckhbw %%mm7, %%mm1 \n\t" /* 0E0F0G0H */\
687 "pshufw $0x90, %%mm0, %%mm5 \n\t" /* 0A0A0B0C */\
688 "pshufw $0x41, %%mm0, %%mm6 \n\t" /* 0B0A0A0B */\
689 "movq %%mm2, %%mm3 \n\t" /* ABCDEFGH */\
690 "movq %%mm2, %%mm4 \n\t" /* ABCDEFGH */\
691 "psllq $8, %%mm2 \n\t" /* 0ABCDEFG */\
692 "psllq $16, %%mm3 \n\t" /* 00ABCDEF */\
693 "psllq $24, %%mm4 \n\t" /* 000ABCDE */\
694 "punpckhbw %%mm7, %%mm2 \n\t" /* 0D0E0F0G */\
695 "punpckhbw %%mm7, %%mm3 \n\t" /* 0C0D0E0F */\
696 "punpckhbw %%mm7, %%mm4 \n\t" /* 0B0C0D0E */\
697 "paddw %%mm3, %%mm5 \n\t" /* b */\
698 "paddw %%mm2, %%mm6 \n\t" /* c */\
699 "paddw %%mm5, %%mm5 \n\t" /* 2b */\
700 "psubw %%mm5, %%mm6 \n\t" /* c - 2b */\
701 "pshufw $0x06, %%mm0, %%mm5 \n\t" /* 0C0B0A0A */\
702 "pmullw %6, %%mm6 \n\t" /* 3c - 6b */\
703 "paddw %%mm4, %%mm0 \n\t" /* a */\
704 "paddw %%mm1, %%mm5 \n\t" /* d */\
705 "pmullw %5, %%mm0 \n\t" /* 20a */\
706 "psubw %%mm5, %%mm0 \n\t" /* 20a - d */\
707 "paddw %8, %%mm6 \n\t"\
708 "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\
709 "psraw $5, %%mm0 \n\t"\
710 "movq %%mm0, %7 \n\t"\
711 /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
712 \
713 "movq 5(%0), %%mm0 \n\t" /* FGHIJKLM */\
714 "movq %%mm0, %%mm5 \n\t" /* FGHIJKLM */\
715 "movq %%mm0, %%mm6 \n\t" /* FGHIJKLM */\
716 "psrlq $8, %%mm0 \n\t" /* GHIJKLM0 */\
717 "psrlq $16, %%mm5 \n\t" /* HIJKLM00 */\
718 "punpcklbw %%mm7, %%mm0 \n\t" /* 0G0H0I0J */\
719 "punpcklbw %%mm7, %%mm5 \n\t" /* 0H0I0J0K */\
720 "paddw %%mm0, %%mm2 \n\t" /* b */\
721 "paddw %%mm5, %%mm3 \n\t" /* c */\
722 "paddw %%mm2, %%mm2 \n\t" /* 2b */\
723 "psubw %%mm2, %%mm3 \n\t" /* c - 2b */\
724 "movq %%mm6, %%mm2 \n\t" /* FGHIJKLM */\
725 "psrlq $24, %%mm6 \n\t" /* IJKLM000 */\
726 "punpcklbw %%mm7, %%mm2 \n\t" /* 0F0G0H0I */\
727 "punpcklbw %%mm7, %%mm6 \n\t" /* 0I0J0K0L */\
728 "pmullw %6, %%mm3 \n\t" /* 3c - 6b */\
729 "paddw %%mm2, %%mm1 \n\t" /* a */\
730 "paddw %%mm6, %%mm4 \n\t" /* d */\
731 "pmullw %5, %%mm1 \n\t" /* 20a */\
732 "psubw %%mm4, %%mm3 \n\t" /* - 6b +3c - d */\
733 "paddw %8, %%mm1 \n\t"\
734 "paddw %%mm1, %%mm3 \n\t" /* 20a - 6b +3c - d */\
735 "psraw $5, %%mm3 \n\t"\
736 "movq %7, %%mm1 \n\t"\
737 "packuswb %%mm3, %%mm1 \n\t"\
738 OP(%%mm1, (%1),%%mm4, q)\
739 /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\
740 \
741 "movq 9(%0), %%mm1 \n\t" /* JKLMNOPQ */\
742 "movq %%mm1, %%mm4 \n\t" /* JKLMNOPQ */\
743 "movq %%mm1, %%mm3 \n\t" /* JKLMNOPQ */\
744 "psrlq $8, %%mm1 \n\t" /* KLMNOPQ0 */\
745 "psrlq $16, %%mm4 \n\t" /* LMNOPQ00 */\
746 "punpcklbw %%mm7, %%mm1 \n\t" /* 0K0L0M0N */\
747 "punpcklbw %%mm7, %%mm4 \n\t" /* 0L0M0N0O */\
748 "paddw %%mm1, %%mm5 \n\t" /* b */\
749 "paddw %%mm4, %%mm0 \n\t" /* c */\
750 "paddw %%mm5, %%mm5 \n\t" /* 2b */\
751 "psubw %%mm5, %%mm0 \n\t" /* c - 2b */\
752 "movq %%mm3, %%mm5 \n\t" /* JKLMNOPQ */\
753 "psrlq $24, %%mm3 \n\t" /* MNOPQ000 */\
754 "pmullw %6, %%mm0 \n\t" /* 3c - 6b */\
755 "punpcklbw %%mm7, %%mm3 \n\t" /* 0M0N0O0P */\
756 "paddw %%mm3, %%mm2 \n\t" /* d */\
757 "psubw %%mm2, %%mm0 \n\t" /* -6b + 3c - d */\
758 "movq %%mm5, %%mm2 \n\t" /* JKLMNOPQ */\
759 "punpcklbw %%mm7, %%mm2 \n\t" /* 0J0K0L0M */\
760 "punpckhbw %%mm7, %%mm5 \n\t" /* 0N0O0P0Q */\
761 "paddw %%mm2, %%mm6 \n\t" /* a */\
762 "pmullw %5, %%mm6 \n\t" /* 20a */\
763 "paddw %8, %%mm0 \n\t"\
764 "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\
765 "psraw $5, %%mm0 \n\t"\
766 /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\
767 \
768 "paddw %%mm5, %%mm3 \n\t" /* a */\
769 "pshufw $0xF9, %%mm5, %%mm6 \n\t" /* 0O0P0Q0Q */\
770 "paddw %%mm4, %%mm6 \n\t" /* b */\
771 "pshufw $0xBE, %%mm5, %%mm4 \n\t" /* 0P0Q0Q0P */\
772 "pshufw $0x6F, %%mm5, %%mm5 \n\t" /* 0Q0Q0P0O */\
773 "paddw %%mm1, %%mm4 \n\t" /* c */\
774 "paddw %%mm2, %%mm5 \n\t" /* d */\
775 "paddw %%mm6, %%mm6 \n\t" /* 2b */\
776 "psubw %%mm6, %%mm4 \n\t" /* c - 2b */\
777 "pmullw %5, %%mm3 \n\t" /* 20a */\
778 "pmullw %6, %%mm4 \n\t" /* 3c - 6b */\
779 "psubw %%mm5, %%mm3 \n\t" /* -6b + 3c - d */\
780 "paddw %8, %%mm4 \n\t"\
781 "paddw %%mm3, %%mm4 \n\t" /* 20a - 6b + 3c - d */\
782 "psraw $5, %%mm4 \n\t"\
783 "packuswb %%mm4, %%mm0 \n\t"\
784 OP(%%mm0, 8(%1), %%mm4, q)\
785 \
786 "addl %3, %0 \n\t"\
787 "addl %4, %1 \n\t"\
788 "decl %2 \n\t"\
789 " jnz 1b \n\t"\
790 : "+r"(src), "+r"(dst), "+g"(h)\
791 : "r"(srcStride), "r"(dstStride), "m"(ff_pw_20), "m"(ff_pw_3), "m"(temp), "m"(ROUNDER)\
792 );\
793 }\
794 \
795 static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
796 int i;\
797 int16_t temp[16];\
798 /* quick HACK, XXX FIXME MUST be optimized */\
799 for(i=0; i<h; i++)\
800 {\
801 temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
802 temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
803 temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
804 temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
805 temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
806 temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\
807 temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\
808 temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\
809 temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\
810 temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\
811 temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\
812 temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\
813 temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\
814 temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\
815 temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\
816 temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\
817 asm volatile(\
818 "movq (%0), %%mm0 \n\t"\
819 "movq 8(%0), %%mm1 \n\t"\
820 "paddw %2, %%mm0 \n\t"\
821 "paddw %2, %%mm1 \n\t"\
822 "psraw $5, %%mm0 \n\t"\
823 "psraw $5, %%mm1 \n\t"\
824 "packuswb %%mm1, %%mm0 \n\t"\
825 OP(%%mm0, (%1), %%mm1, q)\
826 "movq 16(%0), %%mm0 \n\t"\
827 "movq 24(%0), %%mm1 \n\t"\
828 "paddw %2, %%mm0 \n\t"\
829 "paddw %2, %%mm1 \n\t"\
830 "psraw $5, %%mm0 \n\t"\
831 "psraw $5, %%mm1 \n\t"\
832 "packuswb %%mm1, %%mm0 \n\t"\
833 OP(%%mm0, 8(%1), %%mm1, q)\
834 :: "r"(temp), "r"(dst), "m"(ROUNDER)\
835 );\
836 dst+=dstStride;\
837 src+=srcStride;\
838 }\
839 }\
840 \
841 void OPNAME ## mpeg4_qpel16_v_lowpass_mmx(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
842 uint64_t temp[17*4];\
843 uint64_t *temp_ptr= temp;\
844 int count= 17;\
845 \
846 /*FIXME unroll */\
847 asm volatile(\
848 "pxor %%mm7, %%mm7 \n\t"\
849 "1: \n\t"\
850 "movq (%0), %%mm0 \n\t"\
851 "movq (%0), %%mm1 \n\t"\
852 "movq 8(%0), %%mm2 \n\t"\
853 "movq 8(%0), %%mm3 \n\t"\
854 "punpcklbw %%mm7, %%mm0 \n\t"\
855 "punpckhbw %%mm7, %%mm1 \n\t"\
856 "punpcklbw %%mm7, %%mm2 \n\t"\
857 "punpckhbw %%mm7, %%mm3 \n\t"\
858 "movq %%mm0, (%1) \n\t"\
859 "movq %%mm1, 17*8(%1) \n\t"\
860 "movq %%mm2, (%1, %4) \n\t"\
861 "movq %%mm3, (%1, %5) \n\t"\
862 "addl $8, %1 \n\t"\
863 "addl %3, %0 \n\t"\
864 "decl %2 \n\t"\
865 " jnz 1b \n\t"\
866 : "+r" (src), "+r" (temp_ptr), "+r"(count)\
867 : "r" (srcStride), "r"(2*8*17), "r"(3*8*17)\
868 );\
869 \
870 temp_ptr= temp;\
871 count=4;\
872 \
873 /*FIXME reorder for speed */\
874 asm volatile(\
875 /*"pxor %%mm7, %%mm7 \n\t"*/\
876 "1: \n\t"\
877 "movq (%0), %%mm0 \n\t"\
878 "movq 8(%0), %%mm1 \n\t"\
879 "movq 16(%0), %%mm2 \n\t"\
880 "movq 24(%0), %%mm3 \n\t"\
881 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %7, 16(%0), 8(%0), (%0), 32(%0), (%1), OP)\
882 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %7, 8(%0), (%0), (%0), 40(%0), (%1, %3), OP)\
883 "addl %4, %1 \n\t"\
884 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %7, (%0), (%0), 8(%0), 48(%0), (%1), OP)\
885 \
886 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %7, (%0), 8(%0), 16(%0), 56(%0), (%1, %3), OP)\
887 "addl %4, %1 \n\t"\
888 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %7, 8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
889 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %7, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\
890 "addl %4, %1 \n\t"\
891 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %7, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\
892 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %7, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\
893 "addl %4, %1 \n\t"\
894 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %7, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\
895 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %7, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\
896 "addl %4, %1 \n\t"\
897 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %7, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\
898 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %7, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\
899 "addl %4, %1 \n\t"\
900 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %7, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\
901 \
902 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %7, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\
903 "addl %4, %1 \n\t" \
904 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %7, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\
905 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %7, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\
906 \
907 "addl $136, %0 \n\t"\
908 "addl %8, %1 \n\t"\
909 "decl %2 \n\t"\
910 " jnz 1b \n\t"\
911 \
912 : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
913 : "r"(dstStride), "r"(2*dstStride), "m"(ff_pw_20), "m"(ff_pw_3), "m"(ROUNDER), "g"(4-14*dstStride)\
914 );\
915 }\
916 void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
917 uint64_t temp;\
918 \
919 asm volatile(\
920 "pxor %%mm7, %%mm7 \n\t"\
921 "1: \n\t"\
922 "movq (%0), %%mm0 \n\t" /* ABCDEFGH */\
923 "movq %%mm0, %%mm1 \n\t" /* ABCDEFGH */\
924 "movq %%mm0, %%mm2 \n\t" /* ABCDEFGH */\
925 "punpcklbw %%mm7, %%mm0 \n\t" /* 0A0B0C0D */\
926 "punpckhbw %%mm7, %%mm1 \n\t" /* 0E0F0G0H */\
927 "pshufw $0x90, %%mm0, %%mm5 \n\t" /* 0A0A0B0C */\
928 "pshufw $0x41, %%mm0, %%mm6 \n\t" /* 0B0A0A0B */\
929 "movq %%mm2, %%mm3 \n\t" /* ABCDEFGH */\
930 "movq %%mm2, %%mm4 \n\t" /* ABCDEFGH */\
931 "psllq $8, %%mm2 \n\t" /* 0ABCDEFG */\
932 "psllq $16, %%mm3 \n\t" /* 00ABCDEF */\
933 "psllq $24, %%mm4 \n\t" /* 000ABCDE */\
934 "punpckhbw %%mm7, %%mm2 \n\t" /* 0D0E0F0G */\
935 "punpckhbw %%mm7, %%mm3 \n\t" /* 0C0D0E0F */\
936 "punpckhbw %%mm7, %%mm4 \n\t" /* 0B0C0D0E */\
937 "paddw %%mm3, %%mm5 \n\t" /* b */\
938 "paddw %%mm2, %%mm6 \n\t" /* c */\
939 "paddw %%mm5, %%mm5 \n\t" /* 2b */\
940 "psubw %%mm5, %%mm6 \n\t" /* c - 2b */\
941 "pshufw $0x06, %%mm0, %%mm5 \n\t" /* 0C0B0A0A */\
942 "pmullw %6, %%mm6 \n\t" /* 3c - 6b */\
943 "paddw %%mm4, %%mm0 \n\t" /* a */\
944 "paddw %%mm1, %%mm5 \n\t" /* d */\
945 "pmullw %5, %%mm0 \n\t" /* 20a */\
946 "psubw %%mm5, %%mm0 \n\t" /* 20a - d */\
947 "paddw %8, %%mm6 \n\t"\
948 "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\
949 "psraw $5, %%mm0 \n\t"\
950 /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
951 \
952 "movd 5(%0), %%mm5 \n\t" /* FGHI */\
953 "punpcklbw %%mm7, %%mm5 \n\t" /* 0F0G0H0I */\
954 "pshufw $0xF9, %%mm5, %%mm6 \n\t" /* 0G0H0I0I */\
955 "paddw %%mm5, %%mm1 \n\t" /* a */\
956 "paddw %%mm6, %%mm2 \n\t" /* b */\
957 "pshufw $0xBE, %%mm5, %%mm6 \n\t" /* 0H0I0I0H */\
958 "pshufw $0x6F, %%mm5, %%mm5 \n\t" /* 0I0I0H0G */\
959 "paddw %%mm6, %%mm3 \n\t" /* c */\
960 "paddw %%mm5, %%mm4 \n\t" /* d */\
961 "paddw %%mm2, %%mm2 \n\t" /* 2b */\
962 "psubw %%mm2, %%mm3 \n\t" /* c - 2b */\
963 "pmullw %5, %%mm1 \n\t" /* 20a */\
964 "pmullw %6, %%mm3 \n\t" /* 3c - 6b */\
965 "psubw %%mm4, %%mm3 \n\t" /* -6b + 3c - d */\
966 "paddw %8, %%mm1 \n\t"\
967 "paddw %%mm1, %%mm3 \n\t" /* 20a - 6b + 3c - d */\
968 "psraw $5, %%mm3 \n\t"\
969 "packuswb %%mm3, %%mm0 \n\t"\
970 OP(%%mm0, (%1), %%mm4, q)\
971 \
972 "addl %3, %0 \n\t"\
973 "addl %4, %1 \n\t"\
974 "decl %2 \n\t"\
975 " jnz 1b \n\t"\
976 : "+r"(src), "+r"(dst), "+g"(h)\
977 : "r"(srcStride), "r"(dstStride), "m"(ff_pw_20), "m"(ff_pw_3), "m"(temp), "m"(ROUNDER)\
978 );\
979 }\
980 \
981 static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
982 int i;\
983 int16_t temp[8];\
984 /* quick HACK, XXX FIXME MUST be optimized */\
985 for(i=0; i<h; i++)\
986 {\
987 temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
988 temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
989 temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
990 temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
991 temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
992 temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\
993 temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\
994 temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\
995 asm volatile(\
996 "movq (%0), %%mm0 \n\t"\
997 "movq 8(%0), %%mm1 \n\t"\
998 "paddw %2, %%mm0 \n\t"\
999 "paddw %2, %%mm1 \n\t"\
1000 "psraw $5, %%mm0 \n\t"\
1001 "psraw $5, %%mm1 \n\t"\
1002 "packuswb %%mm1, %%mm0 \n\t"\
1003 OP(%%mm0, (%1), %%mm1, q)\
1004 :: "r"(temp), "r"(dst), "m"(ROUNDER)\
1005 );\
1006 dst+=dstStride;\
1007 src+=srcStride;\
1008 }\
1009 }\
1010 \
1011 void OPNAME ## mpeg4_qpel8_v_lowpass_mmx(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
1012 uint64_t temp[9*4];\
1013 uint64_t *temp_ptr= temp;\
1014 int count= 9;\
1015 \
1016 /*FIXME unroll */\
1017 asm volatile(\
1018 "pxor %%mm7, %%mm7 \n\t"\
1019 "1: \n\t"\
1020 "movq (%0), %%mm0 \n\t"\
1021 "movq (%0), %%mm1 \n\t"\
1022 "punpcklbw %%mm7, %%mm0 \n\t"\
1023 "punpckhbw %%mm7, %%mm1 \n\t"\
1024 "movq %%mm0, (%1) \n\t"\
1025 "movq %%mm1, 9*8(%1) \n\t"\
1026 "addl $8, %1 \n\t"\
1027 "addl %3, %0 \n\t"\
1028 "decl %2 \n\t"\
1029 " jnz 1b \n\t"\
1030 : "+r" (src), "+r" (temp_ptr), "+r"(count)\
1031 : "r" (srcStride)\
1032 );\
1033 \
1034 temp_ptr= temp;\
1035 count=2;\
1036 \
1037 /*FIXME reorder for speed */\
1038 asm volatile(\
1039 /*"pxor %%mm7, %%mm7 \n\t"*/\
1040 "1: \n\t"\
1041 "movq (%0), %%mm0 \n\t"\
1042 "movq 8(%0), %%mm1 \n\t"\
1043 "movq 16(%0), %%mm2 \n\t"\
1044 "movq 24(%0), %%mm3 \n\t"\
1045 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %7, 16(%0), 8(%0), (%0), 32(%0), (%1), OP)\
1046 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %7, 8(%0), (%0), (%0), 40(%0), (%1, %3), OP)\
1047 "addl %4, %1 \n\t"\
1048 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %7, (%0), (%0), 8(%0), 48(%0), (%1), OP)\
1049 \
1050 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %7, (%0), 8(%0), 16(%0), 56(%0), (%1, %3), OP)\
1051 "addl %4, %1 \n\t"\
1052 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %7, 8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
1053 \
1054 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %7, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\
1055 "addl %4, %1 \n\t"\
1056 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %7, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\
1057 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %7, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\
1058 \
1059 "addl $72, %0 \n\t"\
1060 "addl %8, %1 \n\t"\
1061 "decl %2 \n\t"\
1062 " jnz 1b \n\t"\
1063 \
1064 : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
1065 : "r"(dstStride), "r"(2*dstStride), "m"(ff_pw_20), "m"(ff_pw_3), "m"(ROUNDER), "g"(4-6*dstStride)\
1066 );\
1067 }
1068
1069 #define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\
1070 \
1071 static void OPNAME ## qpel8_mc00_ ## MMX (UINT8 *dst, UINT8 *src, int stride){\
1072 put_pixels8_mmx(dst, src, stride, 8);\
1073 }\
1074 \
1075 static void OPNAME ## qpel8_mc10_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1076 uint64_t temp[32];\
1077 uint8_t * const half= (uint8_t*)temp;\
1078 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
1079 OPNAME ## pixels8_l2_mmx(dst, src, half, stride, stride, 8);\
1080 }\
1081 \
1082 static void OPNAME ## qpel8_mc20_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1083 OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\
1084 }\
1085 \
1086 static void OPNAME ## qpel8_mc30_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1087 uint64_t temp[32];\
1088 uint8_t * const half= (uint8_t*)temp;\
1089 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
1090 OPNAME ## pixels8_l2_mmx(dst, src+1, half, stride, stride, 8);\
1091 }\
1092 \
1093 static void OPNAME ## qpel8_mc01_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1094 uint64_t temp[32];\
1095 uint8_t * const half= (uint8_t*)temp;\
1096 put ## RND ## mpeg4_qpel8_v_lowpass_mmx(half, src, 8, stride);\
1097 OPNAME ## pixels8_l2_mmx(dst, src, half, stride, stride, 8);\
1098 }\
1099 \
1100 static void OPNAME ## qpel8_mc02_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1101 OPNAME ## mpeg4_qpel8_v_lowpass_mmx(dst, src, stride, stride);\
1102 }\
1103 \
1104 static void OPNAME ## qpel8_mc03_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1105 uint64_t temp[32];\
1106 uint8_t * const half= (uint8_t*)temp;\
1107 put ## RND ## mpeg4_qpel8_v_lowpass_mmx(half, src, 8, stride);\
1108 OPNAME ## pixels8_l2_mmx(dst, src+stride, half, stride, stride, 8);\
1109 }\
1110 static void OPNAME ## qpel8_mc11_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1111 uint64_t half[8*2 + 8*2 + 18*2];\
1112 uint8_t * const halfH= ((uint8_t*)half) + 2*64 + 8;\
1113 uint8_t * const halfV= ((uint8_t*)half);\
1114 uint8_t * const halfHV= ((uint8_t*)half) + 64;\
1115 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1116 put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfV, src, 8, stride);\
1117 put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfHV, halfH, 8, 8);\
1118 OPNAME ## pixels8_l4_mmx(dst, src, (uint8_t*)half, stride, 8);\
1119 }\
1120 static void OPNAME ## qpel8_mc31_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1121 uint64_t half[8*2 + 8*2 + 18*2];\
1122 uint8_t * const halfH= ((uint8_t*)half) + 2*64 + 8;\
1123 uint8_t * const halfV= ((uint8_t*)half);\
1124 uint8_t * const halfHV= ((uint8_t*)half) + 64;\
1125 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1126 put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfV, src+1, 8, stride);\
1127 put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfHV, halfH, 8, 8);\
1128 OPNAME ## pixels8_l4_mmx(dst, src+1, (uint8_t*)half, stride, 8);\
1129 }\
1130 static void OPNAME ## qpel8_mc13_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1131 uint64_t half[8*2 + 8*2 + 9*2];\
1132 uint8_t * const halfH= ((uint8_t*)half) + 2*64;\
1133 uint8_t * const halfV= ((uint8_t*)half);\
1134 uint8_t * const halfHV= ((uint8_t*)half) + 64;\
1135 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1136 put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfV, src, 8, stride);\
1137 put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfHV, halfH, 8, 8);\
1138 OPNAME ## pixels8_l4_mmx(dst, src+stride, (uint8_t*)half, stride, 8);\
1139 }\
1140 static void OPNAME ## qpel8_mc33_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1141 uint64_t half[8*2 + 8*2 + 9*2];\
1142 uint8_t * const halfH= ((uint8_t*)half) + 2*64;\
1143 uint8_t * const halfV= ((uint8_t*)half);\
1144 uint8_t * const halfHV= ((uint8_t*)half) + 64;\
1145 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src , 8, stride, 9);\
1146 put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfV, src+1, 8, stride);\
1147 put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfHV, halfH, 8, 8);\
1148 OPNAME ## pixels8_l4_mmx(dst, src+stride+1, (uint8_t*)half, stride, 8);\
1149 }\
1150 static void OPNAME ## qpel8_mc21_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1151 uint64_t half[8*2 + 9*2];\
1152 uint8_t * const halfH= ((uint8_t*)half) + 64;\
1153 uint8_t * const halfHV= ((uint8_t*)half);\
1154 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1155 put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfHV, halfH, 8, 8);\
1156 OPNAME ## pixels8_l2_mmx(dst, halfH, halfHV, stride, 8, 8);\
1157 }\
1158 static void OPNAME ## qpel8_mc23_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1159 uint64_t half[8*2 + 9*2];\
1160 uint8_t * const halfH= ((uint8_t*)half) + 64;\
1161 uint8_t * const halfHV= ((uint8_t*)half);\
1162 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1163 put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfHV, halfH, 8, 8);\
1164 OPNAME ## pixels8_l2_mmx(dst, halfH+8, halfHV, stride, 8, 8);\
1165 }\
1166 static void OPNAME ## qpel8_mc12_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1167 uint64_t half[8*2 + 8*2 + 9*2];\
1168 uint8_t * const halfH= ((uint8_t*)half) + 2*64;\
1169 uint8_t * const halfV= ((uint8_t*)half);\
1170 uint8_t * const halfHV= ((uint8_t*)half) + 64;\
1171 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1172 put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfV, src, 8, stride);\
1173 put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfHV, halfH, 8, 8);\
1174 OPNAME ## pixels8_l2_mmx(dst, halfV, halfHV, stride, 8, 8);\
1175 }\
1176 static void OPNAME ## qpel8_mc32_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1177 uint64_t half[8*2 + 8*2 + 9*2];\
1178 uint8_t * const halfH= ((uint8_t*)half) + 2*64;\
1179 uint8_t * const halfV= ((uint8_t*)half);\
1180 uint8_t * const halfHV= ((uint8_t*)half) + 64;\
1181 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1182 put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfV, src+1, 8, stride);\
1183 put ## RND ## mpeg4_qpel8_v_lowpass_mmx(halfHV, halfH, 8, 8);\
1184 OPNAME ## pixels8_l2_mmx(dst, halfV, halfHV, stride, 8, 8);\
1185 }\
1186 static void OPNAME ## qpel8_mc22_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1187 uint64_t half[9*2];\
1188 uint8_t * const halfH= ((uint8_t*)half);\
1189 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
1190 OPNAME ## mpeg4_qpel8_v_lowpass_mmx(dst, halfH, stride, 8);\
1191 }\
1192 static void OPNAME ## qpel16_mc00_ ## MMX (UINT8 *dst, UINT8 *src, int stride){\
1193 put_pixels16_mmx(dst, src, stride, 16);\
1194 }\
1195 \
1196 static void OPNAME ## qpel16_mc10_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1197 uint64_t temp[32];\
1198 uint8_t * const half= (uint8_t*)temp;\
1199 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
1200 OPNAME ## pixels16_l2_mmx(dst, src, half, stride, stride, 16);\
1201 }\
1202 \
1203 static void OPNAME ## qpel16_mc20_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1204 OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\
1205 }\
1206 \
1207 static void OPNAME ## qpel16_mc30_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1208 uint64_t temp[32];\
1209 uint8_t * const half= (uint8_t*)temp;\
1210 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
1211 OPNAME ## pixels16_l2_mmx(dst, src+1, half, stride, stride, 16);\
1212 }\
1213 \
1214 static void OPNAME ## qpel16_mc01_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1215 uint64_t temp[32];\
1216 uint8_t * const half= (uint8_t*)temp;\
1217 put ## RND ## mpeg4_qpel16_v_lowpass_mmx(half, src, 16, stride);\
1218 OPNAME ## pixels16_l2_mmx(dst, src, half, stride, stride, 16);\
1219 }\
1220 \
1221 static void OPNAME ## qpel16_mc02_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1222 OPNAME ## mpeg4_qpel16_v_lowpass_mmx(dst, src, stride, stride);\
1223 }\
1224 \
1225 static void OPNAME ## qpel16_mc03_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1226 uint64_t temp[32];\
1227 uint8_t * const half= (uint8_t*)temp;\
1228 put ## RND ## mpeg4_qpel16_v_lowpass_mmx(half, src, 16, stride);\
1229 OPNAME ## pixels16_l2_mmx(dst, src+stride, half, stride, stride, 16);\
1230 }\
1231 static void OPNAME ## qpel16_mc11_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1232 uint64_t half[16*2 + 16*2 + 18*2];\
1233 uint8_t * const halfH= ((uint8_t*)half) + 2*256 + 16;\
1234 uint8_t * const halfV= ((uint8_t*)half);\
1235 uint8_t * const halfHV= ((uint8_t*)half) + 256;\
1236 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1237 put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfV, src, 16, stride);\
1238 put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfHV, halfH, 16, 16);\
1239 OPNAME ## pixels16_l4_mmx(dst, src, (uint8_t*)half, stride, 16);\
1240 }\
1241 static void OPNAME ## qpel16_mc31_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1242 uint64_t half[16*2 + 16*2 + 18*2];\
1243 uint8_t * const halfH= ((uint8_t*)half) + 2*256 + 16;\
1244 uint8_t * const halfV= ((uint8_t*)half);\
1245 uint8_t * const halfHV= ((uint8_t*)half) + 256;\
1246 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1247 put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfV, src+1, 16, stride);\
1248 put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfHV, halfH, 16, 16);\
1249 OPNAME ## pixels16_l4_mmx(dst, src+1, (uint8_t*)half, stride, 16);\
1250 }\
1251 static void OPNAME ## qpel16_mc13_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1252 uint64_t half[16*2 + 16*2 + 17*2];\
1253 uint8_t * const halfH= ((uint8_t*)half) + 2*256;\
1254 uint8_t * const halfV= ((uint8_t*)half);\
1255 uint8_t * const halfHV= ((uint8_t*)half) + 256;\
1256 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1257 put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfV, src, 16, stride);\
1258 put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfHV, halfH, 16, 16);\
1259 OPNAME ## pixels16_l4_mmx(dst, src+stride, (uint8_t*)half, stride, 16);\
1260 }\
1261 static void OPNAME ## qpel16_mc33_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1262 uint64_t half[16*2 + 16*2 + 17*2];\
1263 uint8_t * const halfH= ((uint8_t*)half) + 2*256;\
1264 uint8_t * const halfV= ((uint8_t*)half);\
1265 uint8_t * const halfHV= ((uint8_t*)half) + 256;\
1266 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src , 16, stride, 17);\
1267 put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfV, src+1, 16, stride);\
1268 put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfHV, halfH, 16, 16);\
1269 OPNAME ## pixels16_l4_mmx(dst, src+stride+1, (uint8_t*)half, stride, 16);\
1270 }\
1271 static void OPNAME ## qpel16_mc21_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1272 uint64_t half[16*2 + 17*2];\
1273 uint8_t * const halfH= ((uint8_t*)half) + 256;\
1274 uint8_t * const halfHV= ((uint8_t*)half);\
1275 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1276 put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfHV, halfH, 16, 16);\
1277 OPNAME ## pixels16_l2_mmx(dst, halfH, halfHV, stride, 16, 16);\
1278 }\
1279 static void OPNAME ## qpel16_mc23_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1280 uint64_t half[16*2 + 17*2];\
1281 uint8_t * const halfH= ((uint8_t*)half) + 256;\
1282 uint8_t * const halfHV= ((uint8_t*)half);\
1283 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1284 put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfHV, halfH, 16, 16);\
1285 OPNAME ## pixels16_l2_mmx(dst, halfH+16, halfHV, stride, 16, 16);\
1286 }\
1287 static void OPNAME ## qpel16_mc12_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1288 uint64_t half[16*2 + 16*2 + 17*2];\
1289 uint8_t * const halfH= ((uint8_t*)half) + 2*256;\
1290 uint8_t * const halfV= ((uint8_t*)half);\
1291 uint8_t * const halfHV= ((uint8_t*)half) + 256;\
1292 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1293 put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfV, src, 16, stride);\
1294 put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfHV, halfH, 16, 16);\
1295 OPNAME ## pixels16_l2_mmx(dst, halfV, halfHV, stride, 16, 16);\
1296 }\
1297 static void OPNAME ## qpel16_mc32_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1298 uint64_t half[16*2 + 16*2 + 17*2];\
1299 uint8_t * const halfH= ((uint8_t*)half) + 2*256;\
1300 uint8_t * const halfV= ((uint8_t*)half);\
1301 uint8_t * const halfHV= ((uint8_t*)half) + 256;\
1302 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1303 put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfV, src+1, 16, stride);\
1304 put ## RND ## mpeg4_qpel16_v_lowpass_mmx(halfHV, halfH, 16, 16);\
1305 OPNAME ## pixels16_l2_mmx(dst, halfV, halfHV, stride, 16, 16);\
1306 }\
1307 static void OPNAME ## qpel16_mc22_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
1308 uint64_t half[17*2];\
1309 uint8_t * const halfH= ((uint8_t*)half);\
1310 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
1311 OPNAME ## mpeg4_qpel16_v_lowpass_mmx(dst, halfH, stride, 16);\
1312 }
1313
1314
1315 #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b " \n\t"
1316 #define AVG_OP(a,b,temp, size) \
1317 "mov" #size " " #b ", " #temp " \n\t"\
1318 "pavgusb " #temp ", " #a " \n\t"\
1319 "mov" #size " " #a ", " #b " \n\t"
1320
1321 QPEL_BASE(put_ , ff_pw_16, _ , PUT_OP)
1322 QPEL_BASE(avg_ , ff_pw_16, _ , AVG_OP)
1323 QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP)
1324 QPEL_OP(put_ , ff_pw_16, _ , PUT_OP, 3dnow)
1325 QPEL_OP(avg_ , ff_pw_16, _ , AVG_OP, 3dnow)
1326 QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)
1327
1328 #undef AVG_OP
1329 #define AVG_OP(a,b,temp, size) \
1330 "mov" #size " " #b ", " #temp " \n\t"\
1331 "pavgb " #temp ", " #a " \n\t"\
1332 "mov" #size " " #a ", " #b " \n\t"
1333 QPEL_OP(put_ , ff_pw_16, _ , PUT_OP, mmx2)
1334 QPEL_OP(avg_ , ff_pw_16, _ , AVG_OP, mmx2)
1335 QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)
1336
649 #if 0 1337 #if 0
650 static void just_return() { return; } 1338 static void just_return() { return; }
651 #endif 1339 #endif
652 1340
1341 #define SET_QPEL_FUNC(postfix1, postfix2) \
1342 c->put_ ## postfix1 = put_ ## postfix2;\
1343 c->put_no_rnd_ ## postfix1 = put_no_rnd_ ## postfix2;\
1344 c->avg_ ## postfix1 = avg_ ## postfix2;
1345
653 void dsputil_init_mmx(DSPContext* c, unsigned mask) 1346 void dsputil_init_mmx(DSPContext* c, unsigned mask)
654 { 1347 {
655 mm_flags = mm_support(); 1348 mm_flags = mm_support();
656 #if 0 1349 #if 0
657 fprintf(stderr, "libavcodec: CPU flags:"); 1350 fprintf(stderr, "libavcodec: CPU flags:");
722 1415
723 c->avg_no_rnd_pixels_tab[1][0] = avg_no_rnd_pixels8_mmx; 1416 c->avg_no_rnd_pixels_tab[1][0] = avg_no_rnd_pixels8_mmx;
724 c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_mmx; 1417 c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_mmx;
725 c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_mmx; 1418 c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_mmx;
726 c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_mmx; 1419 c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_mmx;
727 1420
728 c->add_bytes= add_bytes_mmx; 1421 c->add_bytes= add_bytes_mmx;
729 c->diff_bytes= diff_bytes_mmx; 1422 c->diff_bytes= diff_bytes_mmx;
730 1423
731 c->hadamard8_diff[0]= hadamard8_diff16_mmx; 1424 c->hadamard8_diff[0]= hadamard8_diff16_mmx;
732 c->hadamard8_diff[1]= hadamard8_diff_mmx; 1425 c->hadamard8_diff[1]= hadamard8_diff_mmx;
765 1458
766 c->avg_pixels_tab[1][0] = avg_pixels8_mmx2; 1459 c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;
767 c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2; 1460 c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;
768 c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2; 1461 c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;
769 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2; 1462 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;
1463 SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_mmx2)
1464 SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_mmx2)
1465 SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_mmx2)
1466 SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_mmx2)
1467 SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_mmx2)
1468 SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_mmx2)
1469 SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_mmx2)
1470 SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_mmx2)
1471 SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_mmx2)
1472 SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_mmx2)
1473 SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_mmx2)
1474 SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_mmx2)
1475 SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_mmx2)
1476 SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_mmx2)
1477 SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_mmx2)
1478 SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_mmx2)
1479 SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_mmx2)
1480 SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_mmx2)
1481 SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_mmx2)
1482 SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_mmx2)
1483 SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_mmx2)
1484 SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_mmx2)
1485 SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_mmx2)
1486 SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_mmx2)
1487 SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_mmx2)
1488 SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_mmx2)
1489 SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_mmx2)
1490 SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_mmx2)
1491 SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_mmx2)
1492 SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_mmx2)
1493 SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_mmx2)
1494 SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_mmx2)
770 } else if (mm_flags & MM_3DNOW) { 1495 } else if (mm_flags & MM_3DNOW) {
771 c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow; 1496 c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;
772 c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow; 1497 c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;
773 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow; 1498 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;
774 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow; 1499 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;
785 1510
786 c->avg_pixels_tab[1][0] = avg_pixels8_3dnow; 1511 c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;
787 c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow; 1512 c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;
788 c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow; 1513 c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;
789 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow; 1514 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;
1515
1516 SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_3dnow)
1517 SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_3dnow)
1518 SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_3dnow)
1519 SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_3dnow)
1520 SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_3dnow)
1521 SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_3dnow)
1522 SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_3dnow)
1523 SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_3dnow)
1524 SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_3dnow)
1525 SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_3dnow)
1526 SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_3dnow)
1527 SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_3dnow)
1528 SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_3dnow)
1529 SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_3dnow)
1530 SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_3dnow)
1531 SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_3dnow)
1532 SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_3dnow)
1533 SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_3dnow)
1534 SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_3dnow)
1535 SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_3dnow)
1536 SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_3dnow)
1537 SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_3dnow)
1538 SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_3dnow)
1539 SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_3dnow)
1540 SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_3dnow)
1541 SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_3dnow)
1542 SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_3dnow)
1543 SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_3dnow)
1544 SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_3dnow)
1545 SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_3dnow)
1546 SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_3dnow)
1547 SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_3dnow)
790 } 1548 }
791 } 1549 }
792 1550
793 #if 0 1551 #if 0
794 // for speed testing 1552 // for speed testing