movq            %2, %1    punpcklbw       %1, %7    punpckhbw       %2, %7    movq            %3, %6    movq            %4, %3    punpcklbw       %3, %7    punpckhbw       %4, %7    psubw           %1, %3    psubw           %2, %4%endmacro %macro MMX_LOADSUMSUB 4     ; returns %1=%3+%4, %2=%3-%4    movq            %2, %3    movq            %1, %4    MMX_SUMSUB_BA   %1, %2%endmacrocglobal x264_pixel_sub_8x8_mmxcglobal x264_pixel_add_8x8_mmxcglobal x264_transpose_8x8_mmxcglobal x264_ydct8_mmxcglobal x264_yidct8_mmxALIGN 16;-----------------------------------------------------------------------------;   void __cdecl x264_pixel_sub_8x8_mmx( int16_t *diff, uint8_t *pix1, uint8_t *pix2 );;-----------------------------------------------------------------------------x264_pixel_sub_8x8_mmx:    mov         edx, [esp+ 4]           ; diff    mov         eax, [esp+ 8]           ; pix1    mov         ecx, [esp+12]           ; pix2    MMX_ZERO    mm7    %assign disp 0    %rep  8    MMX_LOAD_DIFF_8P mm0, mm1, mm2, mm3, [eax], [ecx], mm7    movq        [edx+disp], mm0    movq        [edx+disp+8], mm1    add         eax, FENC_STRIDE    add         ecx, FDEC_STRIDE    %assign disp disp+16    %endrep    retALIGN 16;-----------------------------------------------------------------------------;   void __cdecl x264_ydct8_mmx( int16_t dest[8][8] );;-----------------------------------------------------------------------------x264_ydct8_mmx:    mov         eax, [esp+04]           ; dest    ;-------------------------------------------------------------------------    ; vertical dct ( compute 4 columns at a time -> 2 loops )    ;-------------------------------------------------------------------------    %assign disp 0    %rep 2        MMX_LOADSUMSUB  mm2, mm3, [eax+disp+0*16], [eax+disp+7*16] ; mm2 = s07, mm3 = d07    MMX_LOADSUMSUB  mm1, mm5, [eax+disp+1*16], [eax+disp+6*16] ; mm1 = s16, mm5 = d16    MMX_LOADSUMSUB  mm0, mm6, [eax+disp+2*16], [eax+disp+5*16] ; mm0 = s25, mm6 = d25    MMX_LOADSUMSUB  mm4, mm7, [eax+disp+3*16], [eax+disp+4*16] ; mm4 = s34, mm7 = d34    MMX_SUMSUB_BA   mm4, mm2        ; mm4 = a0, mm2 = a2    MMX_SUMSUB_BA   mm0, mm1        ; mm0 = a1, mm1 = a3    MMX_SUMSUB_BA   mm0, mm4        ; mm0 = dst0, mm1 = dst4    movq    [eax+disp+0*16], mm0    movq    [eax+disp+4*16], mm4    movq    mm0, mm1         ; a3    psraw   mm0, 1           ; a3>>1    paddw   mm0, mm2         ; a2 + (a3>>1)    psraw   mm2, 1           ; a2>>1    psubw   mm2, mm1         ; (a2>>1) - a3    movq    [eax+disp+2*16], mm0    movq    [eax+disp+6*16], mm2    movq    mm0, mm6    psraw   mm0, 1    paddw   mm0, mm6         ; d25+(d25>>1)    movq    mm1, mm3    psubw   mm1, mm7         ; a5 = d07-d34-(d25+(d25>>1))    psubw   mm1, mm0    movq    mm0, mm5    psraw   mm0, 1    paddw   mm0, mm5         ; d16+(d16>>1)    movq    mm2, mm3    paddw   mm2, mm7         ; a6 = d07+d34-(d16+(d16>>1))    psubw   mm2, mm0    movq    mm0, mm3    psraw   mm0, 1    paddw   mm0, mm3         ; d07+(d07>>1)    paddw   mm0, mm5    paddw   mm0, mm6         ; a4 = d16+d25+(d07+(d07>>1))    movq    mm3, mm7    psraw   mm3, 1    paddw   mm3, mm7         ; d34+(d34>>1)    paddw   mm3, mm5    psubw   mm3, mm6         ; a7 = d16-d25+(d34+(d34>>1))    movq    mm7, mm3    psraw   mm7, 2    paddw   mm7, mm0         ; a4 + (a7>>2)    movq    mm6, mm2    psraw   mm6, 2    paddw   mm6, mm1         ; a5 + (a6>>2)    psraw   mm0, 2    psraw   mm1, 2    psubw   mm0, mm3         ; (a4>>2) - a7    psubw   mm2, mm1         ; a6 - (a5>>2)    movq    [eax+disp+1*16], mm7    movq    [eax+disp+3*16], mm6    movq    [eax+disp+5*16], mm2    movq    [eax+disp+7*16], mm0    %assign disp disp+8    %endrep    retALIGN 16;-----------------------------------------------------------------------------;   void __cdecl x264_yidct8_mmx( int16_t dest[8][8] );;-----------------------------------------------------------------------------x264_yidct8_mmx:    mov         eax, [esp+04]           ; dest    ;-------------------------------------------------------------------------    ; vertical idct ( compute 4 columns at a time -> 2 loops )    ;-------------------------------------------------------------------------    %assign disp 0    %rep 2    movq        mm1, [eax+disp+1*16]    ; mm1 = d1    movq        mm3, [eax+disp+3*16]    ; mm3 = d3    movq        mm5, [eax+disp+5*16]    ; mm5 = d5    movq        mm7, [eax+disp+7*16]    ; mm7 = d7    movq        mm4, mm7    psraw       mm4, 1    movq        mm0, mm5    psubw       mm0, mm7    psubw       mm0, mm4    psubw       mm0, mm3                ; mm0 = e1    movq        mm6, mm3    psraw       mm6, 1    movq        mm2, mm7    psubw       mm2, mm6    psubw       mm2, mm3    paddw       mm2, mm1                ; mm2 = e3    movq        mm4, mm5    psraw       mm4, 1    paddw       mm4, mm5    paddw       mm4, mm7    psubw       mm4, mm1                ; mm4 = e5    movq        mm6, mm1    psraw       mm6, 1    paddw       mm6, mm1    paddw       mm6, mm5    paddw       mm6, mm3                ; mm6 = e7    movq        mm1, mm0    movq        mm3, mm4    movq        mm5, mm2    movq        mm7, mm6    psraw       mm6, 2    psraw       mm3, 2    psraw       mm5, 2    psraw       mm0, 2    paddw       mm1, mm6                ; mm1 = f1    paddw       mm3, mm2                ; mm3 = f3    psubw       mm5, mm4                ; mm5 = f5    psubw       mm7, mm0                ; mm7 = f7    movq        mm2, [eax+disp+2*16]    ; mm2 = d2    movq        mm6, [eax+disp+6*16]    ; mm6 = d6    movq        mm4, mm2    movq        mm0, mm6    psraw       mm4, 1    psraw       mm6, 1    psubw       mm4, mm0                ; mm4 = a4    paddw       mm6, mm2                ; mm6 = a6    movq        mm2, [eax+disp+0*16]    ; mm2 = d0    movq        mm0, [eax+disp+4*16]    ; mm0 = d4    MMX_SUMSUB_BA   mm0, mm2                ; mm0 = a0, mm2 = a2    MMX_SUMSUB_BA   mm6, mm0                ; mm6 = f0, mm0 = f6    MMX_SUMSUB_BA   mm4, mm2                ; mm4 = f2, mm2 = f4    MMX_SUMSUB_BA   mm7, mm6                ; mm7 = g0, mm6 = g7    MMX_SUMSUB_BA   mm5, mm4                ; mm5 = g1, mm4 = g6    MMX_SUMSUB_BA   mm3, mm2                ; mm3 = g2, mm2 = g5    MMX_SUMSUB_BA   mm1, mm0                ; mm1 = g3, mm0 = g4    movq        [eax+disp+0*16], mm7    movq        [eax+disp+1*16], mm5    movq        [eax+disp+2*16], mm3    movq        [eax+disp+3*16], mm1    movq        [eax+disp+4*16], mm0    movq        [eax+disp+5*16], mm2    movq        [eax+disp+6*16], mm4    movq        [eax+disp+7*16], mm6    %assign disp disp+8    %endrep    retALIGN 16;-----------------------------------------------------------------------------;   void __cdecl x264_pixel_add_8x8_mmx( uint8_t *dst, int16_t src[8][8] );;-----------------------------------------------------------------------------x264_pixel_add_8x8_mmx:    mov         eax, [esp+4]       ; dst    mov         edx, [esp+8]       ; src    MMX_ZERO    mm7    %assign disp 0    %rep 8    movq        mm0, [eax]    movq        mm2, [edx+disp]    movq        mm3, [edx+disp+8]    movq        mm1, mm0    psraw       mm2, 6    psraw       mm3, 6    punpcklbw   mm0, mm7    punpckhbw   mm1, mm7    paddw       mm0, mm2    paddw       mm1, mm3    packuswb    mm0, mm1    movq      [eax], mm0    add         eax, FDEC_STRIDE    %assign disp disp+16    %endrep    retALIGN 16;-----------------------------------------------------------------------------;   void __cdecl x264_transpose_8x8_mmx( int16_t src[8][8] );;-----------------------------------------------------------------------------x264_transpose_8x8_mmx:    mov   eax, [esp+4]    movq  mm0, [eax    ]    movq  mm1, [eax+ 16]    movq  mm2, [eax+ 32]    movq  mm3, [eax+ 48]    MMX_TRANSPOSE  mm0, mm1, mm2, mm3, mm4    movq  [eax    ], mm0    movq  [eax+ 16], mm3    movq  [eax+ 32], mm4    movq  [eax+ 48], mm2    movq  mm0, [eax+ 72]    movq  mm1, [eax+ 88]    movq  mm2, [eax+104]    movq  mm3, [eax+120]    MMX_TRANSPOSE  mm0, mm1, mm2, mm3, mm4    movq  [eax+ 72], mm0    movq  [eax+ 88], mm3    movq  [eax+104], mm4    movq  [eax+120], mm2    movq  mm0, [eax+  8]    movq  mm1, [eax+ 24]    movq  mm2, [eax+ 40]    movq  mm3, [eax+ 56]    MMX_TRANSPOSE  mm0, mm1, mm2, mm3, mm4    movq  mm1, [eax+ 64]    movq  mm5, [eax+ 80]    movq  mm6, [eax+ 96]    movq  mm7, [eax+112]    movq  [eax+ 64], mm0    movq  [eax+ 80], mm3    movq  [eax+ 96], mm4    movq  [eax+112], mm2    MMX_TRANSPOSE  mm1, mm5, mm6, mm7, mm4    movq  [eax+  8], mm1    movq  [eax+ 24], mm7    movq  [eax+ 40], mm4    movq  [eax+ 56], mm6    ret