mc-a.asm.svn-base

关于x.264视频格式源码

    ALIGN 4
    .height_loop

    BIWEIGHT_4P_MMX  [edi      ], [edx      ]
    BIWEIGHT_4P_MMX  [edi+4    ], [edx+4    ]
    BIWEIGHT_4P_MMX  [edi+esi  ], [edx+ecx  ]
    BIWEIGHT_4P_MMX  [edi+esi+4], [edx+ecx+4]

    lea  edi, [edi+esi*2]
    lea  edx, [edx+ecx*2]
    sub  eax, byte 2
    jg   .height_loop
    BIWEIGHT_END_MMX

;-----------------------------------------------------------------------------
;   int __cdecl x264_pixel_avg_weight_4x4_mmxext( uint8_t *, int, uint8_t *, int, int )
;-----------------------------------------------------------------------------
cglobal x264_pixel_avg_weight_4x4_mmxext
    BIWEIGHT_START_MMX
    BIWEIGHT_4P_MMX  [edi      ], [edx      ]
    BIWEIGHT_4P_MMX  [edi+esi  ], [edx+ecx  ]
    BIWEIGHT_4P_MMX  [edi+esi*2], [edx+ecx*2]
    add  edi, esi
    add  edx, ecx
    BIWEIGHT_4P_MMX  [edi+esi*2], [edx+ecx*2]
    BIWEIGHT_END_MMX



;=============================================================================
; pixel copy
;=============================================================================

;-----------------------------------------------------------------------------
;  void x264_mc_copy_w4_mmx( uint8_t *src, int i_src_stride,
;                            uint8_t *dst, int i_dst_stride, int i_height )
;-----------------------------------------------------------------------------
cglobal x264_mc_copy_w4_mmx
    push    ebx
    push    esi
    push    edi

    mov     esi, [esp+24]       ; src
    mov     edi, [esp+16]       ; dst
    mov     ebx, [esp+28]       ; i_src_stride
    mov     edx, [esp+20]       ; i_dst_stride
    mov     ecx, [esp+32]       ; i_height
ALIGN 4
.height_loop
    mov     eax, [esi]
    mov     [edi], eax
    mov     eax, [esi+ebx]
    mov     [edi+edx], eax
    lea     esi, [esi+ebx*2]
    lea     edi, [edi+edx*2]
    dec     ecx
    dec     ecx
    jg      .height_loop

    pop     edi
    pop     esi
    pop     ebx
    ret

;-----------------------------------------------------------------------------
;   void x264_mc_copy_w8_mmx( uint8_t *src, int i_src_stride,
;                             uint8_t *dst, int i_dst_stride, int i_height )
;-----------------------------------------------------------------------------
cglobal x264_mc_copy_w8_mmx
    push    ebx
    push    esi
    push    edi

    mov     esi, [esp+24]       ; src
    mov     edi, [esp+16]       ; dst
    mov     ebx, [esp+28]       ; i_src_stride
    mov     edx, [esp+20]       ; i_dst_stride
    mov     ecx, [esp+32]       ; i_height
ALIGN 4
.height_loop
    movq    mm0, [esi]
    movq    [edi], mm0
    movq    mm1, [esi+ebx]
    movq    [edi+edx], mm1
    movq    mm2, [esi+ebx*2]
    movq    [edi+edx*2], mm2
    lea     esi, [esi+ebx*2]
    lea     edi, [edi+edx*2]
    movq    mm3, [esi+ebx]
    movq    [edi+edx], mm3
    lea     esi, [esi+ebx*2]
    lea     edi, [edi+edx*2]
    
    sub     ecx, byte 4
    jg      .height_loop

    pop     edi
    pop     esi
    pop     ebx
    ret

;-----------------------------------------------------------------------------
;   void x264_mc_copy_w16_mmx( uint8_t *src, int i_src_stride,
;                              uint8_t *dst, int i_dst_stride, int i_height )
;-----------------------------------------------------------------------------
cglobal x264_mc_copy_w16_mmx
    push    ebx
    push    esi
    push    edi

    mov     esi, [esp+24]       ; src
    mov     edi, [esp+16]       ; dst
    mov     ebx, [esp+28]       ; i_src_stride
    mov     edx, [esp+20]       ; i_dst_stride
    mov     ecx, [esp+32]       ; i_height

ALIGN 4
.height_loop
    movq    mm0, [esi]
    movq    mm1, [esi+8]
    movq    [edi], mm0
    movq    [edi+8], mm1
    movq    mm2, [esi+ebx]
    movq    mm3, [esi+ebx+8]
    movq    [edi+edx], mm2
    movq    [edi+edx+8], mm3
    movq    mm4, [esi+ebx*2]
    movq    mm5, [esi+ebx*2+8]
    movq    [edi+edx*2], mm4
    movq    [edi+edx*2+8], mm5
    lea     esi, [esi+ebx*2]
    lea     edi, [edi+edx*2]
    movq    mm6, [esi+ebx]
    movq    mm7, [esi+ebx+8]
    movq    [edi+edx], mm6
    movq    [edi+edx+8], mm7
    lea     esi, [esi+ebx*2]
    lea     edi, [edi+edx*2]
    sub     ecx, byte 4
    jg      .height_loop
    
    pop     edi
    pop     esi
    pop     ebx
    ret


;-----------------------------------------------------------------------------
;   void x264_mc_copy_w16_sse2( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
;-----------------------------------------------------------------------------
cglobal x264_mc_copy_w16_sse2
    push    ebx
    push    esi
    push    edi

    mov     esi, [esp+24]       ; src
    mov     edi, [esp+16]       ; dst
    mov     ebx, [esp+28]       ; i_src_stride
    mov     edx, [esp+20]       ; i_dst_stride
    mov     ecx, [esp+32]       ; i_height

ALIGN 4
.height_loop
    movdqu  xmm0, [esi]
    movdqu  xmm1, [esi+ebx]
    movdqu  [edi], xmm0
    movdqu  [edi+edx], xmm1
    dec     ecx
    dec     ecx
    lea     esi, [esi+ebx*2]
    lea     edi, [edi+edx*2]
    jg      .height_loop
    
    pop     edi
    pop     esi
    pop     ebx
    ret



;=============================================================================
; chroma MC
;=============================================================================

;-----------------------------------------------------------------------------
;   void x264_mc_chroma_mmxext( uint8_t *src, int i_src_stride,
;                               uint8_t *dst, int i_dst_stride,
;                               int dx, int dy,
;                               int i_width, int i_height )
;-----------------------------------------------------------------------------

cglobal x264_mc_chroma_mmxext
    picpush ebx
    picgetgot ebx
    push    edi

    mov     ecx, [picesp+4+24]
    mov     edx, [picesp+4+20]
    mov     eax, ecx
    mov     edi, edx
    sar     ecx, 3
    sar     edx, 3
    imul    ecx, [picesp+4+8]
    add     ecx, edx
    add     [picesp+4+4], ecx     ; src += (dx>>3) + (dy>>3) * src_stride

    pxor    mm3, mm3

    and     edi, 7
    and     eax, 7
    movd    mm5, edi
    movd    mm6, eax
    pshufw  mm5, mm5, 0         ; mm5 = dx&7
    pshufw  mm6, mm6, 0         ; mm6 = dy&7

    movq    mm4, [pw_8 GOT_ebx]
    movq    mm0, mm4

    psubw   mm4, mm5            ; mm4 = 8-dx
    psubw   mm0, mm6            ; mm0 = 8-dy

    movq    mm7, mm5
    pmullw  mm5, mm0            ; mm5 = dx*(8-dy) =     cB
    pmullw  mm7, mm6            ; mm7 = dx*dy =         cD
    pmullw  mm6, mm4            ; mm6 = (8-dx)*dy =     cC
    pmullw  mm4, mm0            ; mm4 = (8-dx)*(8-dy) = cA

    mov     eax, [picesp+4+4]   ; src
    mov     edi, [picesp+4+12]  ; dst
    mov     ecx, [picesp+4+8]   ; i_src_stride
    mov     edx, [picesp+4+32]  ; i_height

ALIGN 4
.height_loop

    movd    mm1, [eax+ecx]
    movd    mm0, [eax]
    punpcklbw mm1, mm3          ; 00 px1 | 00 px2 | 00 px3 | 00 px4
    punpcklbw mm0, mm3
    pmullw  mm1, mm6            ; 2nd line * cC
    pmullw  mm0, mm4            ; 1st line * cA

    paddw   mm0, mm1            ; mm0 <- result

    movd    mm2, [eax+1]
    movd    mm1, [eax+ecx+1]
    punpcklbw mm2, mm3
    punpcklbw mm1, mm3

    paddw   mm0, [pw_32 GOT_ebx]

    pmullw  mm2, mm5            ; line * cB
    pmullw  mm1, mm7            ; line * cD
    paddw   mm0, mm2
    paddw   mm0, mm1

    psrlw   mm0, 6
    packuswb mm0, mm3           ; 00 00 00 00 px1 px2 px3 px4
    movd    [edi], mm0

    add     eax, ecx
    add     edi, [picesp+4+16]

    dec     edx
    jnz     .height_loop

    sub     [picesp+4+28], dword 8
    jnz     .finish            ; width != 8 so assume 4

    mov     edi, [picesp+4+12] ; dst
    mov     eax, [picesp+4+4]  ; src
    mov     edx, [picesp+4+32] ; i_height
    add     edi, 4
    add     eax, 4
    jmp    .height_loop

.finish
    pop     edi
    picpop  ebx
    ret



; prefetches tuned for 64 byte cachelines (K7/K8/Core2)
; TODO add 32 and 128 byte versions for P3/P4

;-----------------------------------------------------------------------------
; void x264_prefetch_fenc_mmxext( uint8_t *pix_y, int stride_y, 
;                                 uint8_t *pix_uv, int stride_uv, int mb_x )
;-----------------------------------------------------------------------------
cglobal x264_prefetch_fenc_mmxext
    mov   eax, [esp+20]
    mov   ecx, [esp+8]
    mov   edx, [esp+4]
    and   eax, 3
    imul  eax, ecx
    lea   edx, [edx+eax*4+64]
    prefetcht0 [edx]
    prefetcht0 [edx+ecx]
    lea   edx, [edx+ecx*2]
    prefetcht0 [edx]
    prefetcht0 [edx+ecx]

    mov   eax, [esp+20]
    mov   ecx, [esp+16]
    mov   edx, [esp+12]
    and   eax, 6
    imul  eax, ecx
    lea   edx, [edx+eax+64]
    prefetcht0 [edx]
    prefetcht0 [edx+ecx]
    ret

;-----------------------------------------------------------------------------
; void x264_prefetch_ref_mmxext( uint8_t *pix, int stride, int parity )
;-----------------------------------------------------------------------------
cglobal x264_prefetch_ref_mmxext
    mov   eax, [esp+12]
    mov   ecx, [esp+8]
    mov   edx, [esp+4]
    sub   eax, 1
    and   eax, ecx
    lea   edx, [edx+eax*8+64]
    lea   eax, [ecx*3]
    prefetcht0 [edx]
    prefetcht0 [edx+ecx]
    prefetcht0 [edx+ecx*2]
    prefetcht0 [edx+eax]
    lea   edx, [edx+ecx*4]
    prefetcht0 [edx]
    prefetcht0 [edx+ecx]
    prefetcht0 [edx+ecx*2]
    prefetcht0 [edx+eax]
    ret