%macro HPEL 1;-----------------------------------------------------------------------------; void x264_hpel_filter_sse2( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc,;                             uint8_t *src, int stride, int width, int height);-----------------------------------------------------------------------------cglobal x264_hpel_filter_%1, 7,7,16%ifdef WIN64    movsxd   r4, r4d    movsxd   r5, r5d%endif    mov      r10, r3    sub       r5, 16    mov      r11, r1    and      r10, 15    sub       r3, r10    add       r0, r5    add      r11, r5    add      r10, r5    add       r5, r2    mov       r2, r4    neg      r10    lea       r1, [r3+r2]    sub       r3, r2    sub       r3, r2    mov       r4, r10%ifidn %1, sse2    pxor      m0, m0%endif    pcmpeqw  m15, m15    psrlw    m15, 15 ; pw_1    psllw    m15, 4;ALIGN 16.loopy:; first filter_v; prefetching does not help here! lots of variants tested, all slower    DO_FILT_V m8, m7, m13, m12, 0, %1;ALIGN 16.loopx:    DO_FILT_V m6, m5, m11, m10, 16, %1.lastx:    paddw    m15, m15    DO_FILT_CC m9, m8, m7, m6    movdqa   m7, m12        ; not really necessary, but seems free and    movdqa   m6, m11	     ; gives far shorter code    psrlw    m15, 5    DO_FILT_HH m14, m13, m7, m6    psllw    m15, 4 ; pw_16    movdqa   m7, m5    movdqa  m12, m10    add      r4, 16    jl .loopx    cmp      r4, 16    jl .lastx; setup regs for next y    sub      r4, r10    sub      r4, r2    sub      r1, r4    sub      r3, r4    add      r0, r2    add     r11, r2    add      r5, r2    mov      r4, r10    sub     r6d, 1    jg .loopy    sfence    RET%endmacro%define PALIGNR PALIGNR_MMXHPEL sse2%define PALIGNR PALIGNR_SSSE3HPEL ssse3%endifcglobal x264_sfence    sfence    ret;-----------------------------------------------------------------------------; void x264_plane_copy_mmxext( uint8_t *dst, int i_dst,;                              uint8_t *src, int i_src, int w, int h);-----------------------------------------------------------------------------cglobal x264_plane_copy_mmxext, 6,7    movsxdifnidn r1, r1d    movsxdifnidn r3, r3d    add    r4d, 3    and    r4d, ~3    mov    r6d, r4d    and    r6d, ~15    sub    r1,  r6    sub    r3,  r6.loopy:    mov    r6d, r4d    sub    r6d, 64    jl     .endx.loopx:    prefetchnta [r2+256]    movq   mm0, [r2   ]    movq   mm1, [r2+ 8]    movq   mm2, [r2+16]    movq   mm3, [r2+24]    movq   mm4, [r2+32]    movq   mm5, [r2+40]    movq   mm6, [r2+48]    movq   mm7, [r2+56]    movntq [r0   ], mm0    movntq [r0+ 8], mm1    movntq [r0+16], mm2    movntq [r0+24], mm3    movntq [r0+32], mm4    movntq [r0+40], mm5    movntq [r0+48], mm6    movntq [r0+56], mm7    add    r2,  64    add    r0,  64    sub    r6d, 64    jge    .loopx.endx:    prefetchnta [r2+256]    add    r6d, 48    jl .end16.loop16:    movq   mm0, [r2  ]    movq   mm1, [r2+8]    movntq [r0  ], mm0    movntq [r0+8], mm1    add    r2,  16    add    r0,  16    sub    r6d, 16    jge    .loop16.end16:    add    r6d, 12    jl .end4.loop4:    movd   mm2, [r2+r6]    movd   [r0+r6], mm2    sub    r6d, 4    jge .loop4.end4:    add    r2, r3    add    r0, r1    dec    r5d    jg     .loopy    sfence    emms    RET; These functions are not general-use; not only do the SSE ones require aligned input,; but they also will fail if given a non-mod16 size or a size less than 64.; memzero SSE will fail for non-mod128.;-----------------------------------------------------------------------------; void *x264_memcpy_aligned_mmx( void *dst, const void *src, size_t n );;-----------------------------------------------------------------------------cglobal x264_memcpy_aligned_mmx, 3,3    test r2d, 16    jz .copy32    sub r2d, 16    movq mm0, [r1 + r2 + 0]    movq mm1, [r1 + r2 + 8]    movq [r0 + r2 + 0], mm0    movq [r0 + r2 + 8], mm1.copy32:    sub r2d, 32    movq mm0, [r1 + r2 +  0]    movq mm1, [r1 + r2 +  8]    movq mm2, [r1 + r2 + 16]    movq mm3, [r1 + r2 + 24]    movq [r0 + r2 +  0], mm0    movq [r0 + r2 +  8], mm1    movq [r0 + r2 + 16], mm2    movq [r0 + r2 + 24], mm3    jg .copy32    REP_RET;-----------------------------------------------------------------------------; void *x264_memcpy_aligned_sse2( void *dst, const void *src, size_t n );;-----------------------------------------------------------------------------cglobal x264_memcpy_aligned_sse2, 3,3    test r2d, 16    jz .copy32    sub r2d, 16    movdqa xmm0, [r1 + r2]    movdqa [r0 + r2], xmm0.copy32:    test r2d, 32    jz .copy64    sub r2d, 32    movdqa xmm0, [r1 + r2 +  0]    movdqa [r0 + r2 +  0], xmm0    movdqa xmm1, [r1 + r2 + 16]    movdqa [r0 + r2 + 16], xmm1.copy64:    sub r2d, 64    movdqa xmm0, [r1 + r2 +  0]    movdqa [r0 + r2 +  0], xmm0    movdqa xmm1, [r1 + r2 + 16]    movdqa [r0 + r2 + 16], xmm1    movdqa xmm2, [r1 + r2 + 32]    movdqa [r0 + r2 + 32], xmm2    movdqa xmm3, [r1 + r2 + 48]    movdqa [r0 + r2 + 48], xmm3    jg .copy64    REP_RET;-----------------------------------------------------------------------------; void *x264_memzero_aligned( void *dst, size_t n );;-----------------------------------------------------------------------------%macro MEMZERO 1cglobal x264_memzero_aligned_%1, 2,2    pxor m0, m0.loop:    sub r1d, mmsize*8%assign i 0%rep 8    mova [r0 + r1 + i], m0%assign i i+mmsize%endrep    jg .loop    REP_RET%endmacroINIT_MMXMEMZERO mmxINIT_XMMMEMZERO sse2;-----------------------------------------------------------------------------; void x264_integral_init4h_sse4( uint16_t *sum, uint8_t *pix, int stride );-----------------------------------------------------------------------------cglobal x264_integral_init4h_sse4, 3,4    lea     r3, [r0+r2*2]    add     r1, r2    neg     r2    pxor    m4, m4.loop:    movdqa  m0, [r1+r2]    movdqa  m1, [r1+r2+16]    palignr m1, m0, 8    mpsadbw m0, m4, 0    mpsadbw m1, m4, 0    paddw   m0, [r0+r2*2]    paddw   m1, [r0+r2*2+16]    movdqa  [r3+r2*2   ], m0    movdqa  [r3+r2*2+16], m1    add     r2, 16    jl .loop    REP_RETcglobal x264_integral_init8h_sse4, 3,4    lea     r3, [r0+r2*2]    add     r1, r2    neg     r2    pxor    m4, m4.loop:    movdqa  m0, [r1+r2]    movdqa  m1, [r1+r2+16]    palignr m1, m0, 8    movdqa  m2, m0    movdqa  m3, m1    mpsadbw m0, m4, 0    mpsadbw m1, m4, 0    mpsadbw m2, m4, 4    mpsadbw m3, m4, 4    paddw   m0, [r0+r2*2]    paddw   m1, [r0+r2*2+16]    paddw   m0, m2    paddw   m1, m3    movdqa  [r3+r2*2   ], m0    movdqa  [r3+r2*2+16], m1    add     r2, 16    jl .loop    REP_RET%macro INTEGRAL_INIT 1;-----------------------------------------------------------------------------; void x264_integral_init4v_mmx( uint16_t *sum8, uint16_t *sum4, int stride );-----------------------------------------------------------------------------cglobal x264_integral_init4v_%1, 3,5    shl   r2, 1    add   r0, r2    add   r1, r2    lea   r3, [r0+r2*4]    lea   r4, [r0+r2*8]    neg   r2.loop:    movu  m0, [r0+r2+8]    mova  m2, [r0+r2]    movu  m1, [r4+r2+8]    paddw m0, m2    paddw m1, [r4+r2]    mova  m3, [r3+r2]    psubw m1, m0    psubw m3, m2    mova  [r0+r2], m1    mova  [r1+r2], m3    add   r2, mmsize    jl .loop    REP_RET;-----------------------------------------------------------------------------; void x264_integral_init8v_mmx( uint16_t *sum8, int stride );-----------------------------------------------------------------------------cglobal x264_integral_init8v_%1, 3,3    shl   r1, 1    add   r0, r1    lea   r2, [r0+r1*8]    neg   r1.loop:    mova  m0, [r2+r1]    mova  m1, [r2+r1+mmsize]    psubw m0, [r0+r1]    psubw m1, [r0+r1+mmsize]    mova  [r0+r1], m0    mova  [r0+r1+mmsize], m1    add   r1, 2*mmsize    jl .loop    REP_RET%endmacroINIT_MMXINTEGRAL_INIT mmxINIT_XMMINTEGRAL_INIT sse2%macro FILT8x4 7    mova      %3, [r0+%7]    mova      %4, [r0+r5+%7]    pavgb     %3, %4    pavgb     %4, [r0+r5*2+%7]    PALIGNR   %1, %3, 1, m6    PALIGNR   %2, %4, 1, m6    pavgb     %1, %3    pavgb     %2, %4    mova      %5, %1    mova      %6, %2    pand      %1, m7    pand      %2, m7    psrlw     %5, 8    psrlw     %6, 8%endmacro%macro FILT16x2 4    mova      m3, [r0+%4+mmsize]    mova      m2, [r0+%4]    pavgb     m3, [r0+%4+r5+mmsize]    pavgb     m2, [r0+%4+r5]    PALIGNR   %1, m3, 1, m6    pavgb     %1, m3    PALIGNR   m3, m2, 1, m6    pavgb     m3, m2    mova      m5, m3    mova      m4, %1    pand      m3, m7    pand      %1, m7    psrlw     m5, 8    psrlw     m4, 8    packuswb  m3, %1    packuswb  m5, m4    mova    [%2], m3    mova    [%3], m5    mova      %1, m2%endmacro%macro FILT8x2U 3    mova      m3, [r0+%3+8]    mova      m2, [r0+%3]    pavgb     m3, [r0+%3+r5+8]    pavgb     m2, [r0+%3+r5]    mova      m1, [r0+%3+9]    mova      m0, [r0+%3+1]    pavgb     m1, [r0+%3+r5+9]    pavgb     m0, [r0+%3+r5+1]    pavgb     m1, m3    pavgb     m0, m2    mova      m3, m1    mova      m2, m0    pand      m1, m7    pand      m0, m7    psrlw     m3, 8    psrlw     m2, 8    packuswb  m0, m1    packuswb  m2, m3    mova    [%1], m0    mova    [%2], m2%endmacro;-----------------------------------------------------------------------------; void frame_init_lowres_core( uint8_t *src0, uint8_t *dst0, uint8_t *dsth, uint8_t *dstv, uint8_t *dstc,;                              int src_stride, int dst_stride, int width, int height );-----------------------------------------------------------------------------%macro FRAME_INIT_LOWRES 1-2 0 ; FIXMEcglobal x264_frame_init_lowres_core_%1, 6,7,%2%ifdef WIN64    movsxd   r5, r5d%endif    ; src += 2*(height-1)*stride + 2*width    mov      r6d, r8m    dec      r6d    imul     r6d, r5d    add      r6d, r7m    lea       r0, [r0+r6*2]    ; dst += (height-1)*stride + width    mov      r6d, r8m    dec      r6d    imul     r6d, r6m    add      r6d, r7m    add       r1, r6    add       r2, r6    add       r3, r6    add       r4, r6    ; gap = stride - width    mov      r6d, r6m    sub      r6d, r7m    PUSH      r6    %define dst_gap [rsp+gprsize]    mov      r6d, r5d    sub      r6d, r7m    shl      r6d, 1    PUSH      r6    %define src_gap [rsp]%if mmsize == 16    ; adjust for the odd end case    mov      r6d, r7m    and      r6d, 8    sub       r1, r6    sub       r2, r6    sub       r3, r6    sub       r4, r6    add  dst_gap, r6d%endif ; mmsize    pcmpeqb   m7, m7    psrlw     m7, 8.vloop:    mov      r6d, r7m%ifnidn %1, mmxext    mova      m0, [r0]    mova      m1, [r0+r5]    pavgb     m0, m1    pavgb     m1, [r0+r5*2]%endif%if mmsize == 16    test     r6d, 8    jz .hloop    sub       r0, 16    FILT8x4   m0, m1, m2, m3, m4, m5, 0    packuswb  m0, m4    packuswb  m1, m5    movq    [r1], m0    movhps  [r2], m0    movq    [r3], m1    movhps  [r4], m1    mova      m0, m2    mova      m1, m3    sub      r6d, 8%endif ; mmsize.hloop:    sub       r0, mmsize*2    sub       r1, mmsize    sub       r2, mmsize    sub       r3, mmsize    sub       r4, mmsize%ifdef m8    FILT8x4   m0, m1, m2, m3, m10, m11, mmsize    mova      m8, m0    mova      m9, m1    FILT8x4   m2, m3, m0, m1, m4, m5, 0    packuswb  m2, m8    packuswb  m3, m9    packuswb  m4, m10    packuswb  m5, m11    mova    [r1], m2    mova    [r2], m4    mova    [r3], m3    mova    [r4], m5%elifidn %1, mmxext    FILT8x2U  r1, r2, 0    FILT8x2U  r3, r4, r5%else    FILT16x2  m0, r1, r2, 0    FILT16x2  m1, r3, r4, r5%endif    sub      r6d, mmsize    jg .hloop.skip:    mov       r6, dst_gap    sub       r0, src_gap    sub       r1, r6    sub       r2, r6    sub       r3, r6    sub       r4, r6    dec    dword r8m    jg .vloop    ADD      rsp, 2*gprsize    emms    RET%endmacro ; FRAME_INIT_LOWRESINIT_MMX%define PALIGNR PALIGNR_MMXFRAME_INIT_LOWRES mmxext%ifndef ARCH_X86_64FRAME_INIT_LOWRES cache32_mmxext%endifINIT_XMMFRAME_INIT_LOWRES sse2, 12%define PALIGNR PALIGNR_SSSE3FRAME_INIT_LOWRES ssse3, 12