mc-a.asm

在dsp上实现h.264编解码

    BIWEIGHT_START_MMX
    mov     eax, [esp+32] ; i_height
    ALIGN 4
    .height_loop

    BIWEIGHT_4P_MMX  [edi   ], [edx   ]
    BIWEIGHT_4P_MMX  [edi+ 4], [edx+ 4]
    BIWEIGHT_4P_MMX  [edi+ 8], [edx+ 8]
    BIWEIGHT_4P_MMX  [edi+12], [edx+12]

    add  edi, esi
    add  edx, ecx
    dec  eax
    jnz  .height_loop
    BIWEIGHT_END_MMX

ALIGN 16
;-----------------------------------------------------------------------------
;   int __cdecl SHU264_pixel_avg_weight_w8_mmxext( uint8_t *, int, uint8_t *, int, int, int )
;-----------------------------------------------------------------------------
SHU264_pixel_avg_weight_w8_mmxext:
    BIWEIGHT_START_MMX
    mov     eax, [esp+32]
    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
    jnz  .height_loop
    BIWEIGHT_END_MMX

ALIGN 16
;-----------------------------------------------------------------------------
;   int __cdecl SHU264_pixel_avg_weight_4x4_mmxext( uint8_t *, int, uint8_t *, int, int )
;-----------------------------------------------------------------------------
SHU264_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
;=============================================================================

ALIGN 16
;-----------------------------------------------------------------------------
;  void SHU264_mc_copy_w4_mmxext( uint8_t *src, int i_src_stride,
;                               uint8_t *dst, int i_dst_stride, int i_height )
;-----------------------------------------------------------------------------
SHU264_mc_copy_w4_mmxext:
    push    ebx
    push    esi
    push    edi

    mov     esi, [esp+16]       ; src
    mov     edi, [esp+24]       ; dst
    mov     ebx, [esp+20]       ; i_src_stride
    mov     edx, [esp+28]       ; 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
    jne     .height_loop

    pop     edi
    pop     esi
    pop     ebx
    ret

ALIGN 16
;-----------------------------------------------------------------------------
;   void SHU264_mc_copy_w8_mmxext( uint8_t *src, int i_src_stride,
;                                uint8_t *dst, int i_dst_stride, int i_height )
;-----------------------------------------------------------------------------
SHU264_mc_copy_w8_mmxext:
    push    ebx
    push    esi
    push    edi

    mov     esi, [esp+16]       ; src
    mov     edi, [esp+24]       ; dst
    mov     ebx, [esp+20]       ; i_src_stride
    mov     edx, [esp+28]       ; 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
    jnz     .height_loop

    pop     edi
    pop     esi
    pop     ebx
    ret

ALIGN 16
;-----------------------------------------------------------------------------
;   void SHU264_mc_copy_w16_mmxext( uint8_t *src, int i_src_stride,
;                                 uint8_t *dst, int i_dst_stride, int i_height )
;-----------------------------------------------------------------------------
SHU264_mc_copy_w16_mmxext:
    push    ebx
    push    esi
    push    edi

    mov     esi, [esp+16]       ; src
    mov     edi, [esp+24]       ; dst
    mov     ebx, [esp+20]       ; i_src_stride
    mov     edx, [esp+28]       ; 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
    jnz     .height_loop
    
    pop     edi
    pop     esi
    pop     ebx
    ret


ALIGN 16
;-----------------------------------------------------------------------------
;   void SHU264_mc_copy_w16_sse2( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
;-----------------------------------------------------------------------------
SHU264_mc_copy_w16_sse2:
    push    ebx
    push    esi
    push    edi

    mov     esi, [esp+16]       ; src
    mov     edi, [esp+24]       ; dst
    mov     ebx, [esp+20]       ; i_src_stride
    mov     edx, [esp+28]       ; 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]
    jnz     .height_loop
    
    pop     edi
    pop     esi
    pop     ebx
    ret



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

ALIGN 16
;-----------------------------------------------------------------------------
;   void SHU264_mc_chroma_sse( uint8_t *src, int i_src_stride,
;                               uint8_t *dst, int i_dst_stride,
;                               int dx, int dy,
;                               int i_height, int i_width )
;-----------------------------------------------------------------------------

SHU264_mc_chroma_sse:

    pxor    mm3, mm3

    pshufw  mm5, [esp+20], 0    ; mm5 - dx
    pshufw  mm6, [esp+24], 0    ; mm6 - dy

    movq    mm4, [pw_8]
    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

    push    edi

    mov     eax, [esp+4+4]     ; src
    mov     edi, [esp+4+12]    ; dst
    mov     ecx, [esp+4+8]     ; i_src_stride
    mov     edx, [esp+4+28]    ; 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]

    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, [esp+4+16]

    dec     edx
    jnz     .height_loop

    mov     eax, [esp+4+32]
    sub     eax, 8
    jnz     .finish              ; width != 8 so assume 4

    mov     [esp+4+32], eax
    mov     edi, [esp+4+12]    ; dst
    mov     eax, [esp+4+4]     ; src
    mov     edx, [esp+4+28]    ; i_height
    add     edi, 4
    add     eax, 4
    jmp    .height_loop

.finish
    pop     edi
    ret