pixel-a.asm.svn-base

H.264源码程序

    ; top
    movzx       eax, byte [edi+esi-FDEC_STRIDE+0]
    movzx       ebx, byte [edi+esi-FDEC_STRIDE+1]
    movzx       ecx, byte [edi+esi-FDEC_STRIDE+2]
    movzx       edx, byte [edi+esi-FDEC_STRIDE+3]
    SCALAR_SUMSUB eax, ebx, ecx, edx
    SCALAR_SUMSUB eax, ecx, ebx, edx
    add         ebp, eax
    mov         [top_1d+2*esi+0], ax
    mov         [top_1d+2*esi+2], bx
    mov         [top_1d+2*esi+4], cx
    mov         [top_1d+2*esi+6], dx
    sub         esi, 4
    jge .loop_edge

    ; dc
    shr         ebp, 1
    add         ebp, 8
    and         ebp, -16

    ; 2D hadamards
    mov         eax, [args+0] ; fenc
    xor         edi, edi
.loop_y:
    xor         esi, esi
.loop_x:
    LOAD_HADAMARD eax

    movq        mm4, mm1
    movq        mm5, mm2
    MMX_ABS_TWO mm4, mm5, mm6, mm7
    movq        mm7, mm3
    paddw       mm4, mm5
    MMX_ABS     mm7, mm6
    paddw       mm7, mm4 ; 3x4 sum

    movq        mm4, [left_1d+8*edi]
    movd        mm5, ebp
    psllw       mm4, 2
    psubw       mm4, mm0
    psubw       mm5, mm0
    punpcklwd   mm0, mm1
    punpcklwd   mm2, mm3
    punpckldq   mm0, mm2 ; transpose
    movq        mm1, [top_1d+8*esi]
    psllw       mm1, 2
    psubw       mm0, mm1
    MMX_ABS     mm4, mm3 ; 1x4 sum
    MMX_ABS     mm5, mm2 ; 1x4 sum
    MMX_ABS     mm0, mm1 ; 4x1 sum
    pavgw       mm4, mm7
    pavgw       mm5, mm7
    paddw       mm0, [sums+0]  ; i4x4_v satd
    paddw       mm4, [sums+8]  ; i4x4_h satd
    paddw       mm5, [sums+16] ; i4x4_dc satd
    movq        [sums+0], mm0
    movq        [sums+8], mm4
    movq        [sums+16], mm5

    add         eax, 4
    inc         esi
    cmp         esi, 4
    jl  .loop_x
    add         eax, 4*FENC_STRIDE-16
    inc         edi
    cmp         edi, 4
    jl  .loop_y

; horizontal sum
    movq        mm2, [sums+16]
    movq        mm0, [sums+0]
    movq        mm1, [sums+8]
    movq        mm7, mm2
    SUM_MM_X3   mm0, mm1, mm2, mm3, mm4, mm5, mm6, paddd
    psrld       mm0, 1
    pslld       mm7, 16
    psrld       mm7, 16
    paddd       mm0, mm2
    psubd       mm0, mm7
    mov         eax, [args+8] ; res
    movd        [eax+0], mm0 ; i16x16_v satd
    movd        [eax+4], mm1 ; i16x16_h satd
    movd        [eax+8], mm2 ; i16x16_dc satd

    add  esp, 88
    pop  esi
    pop  edi
    pop  ebp
    pop  ebx
    ret

ALIGN 16
;-----------------------------------------------------------------------------
;  void x264_intra_satd_x3_8x8c_mmxext( uint8_t *fenc, uint8_t *fdec, int *res )
;-----------------------------------------------------------------------------
x264_intra_satd_x3_8x8c_mmxext:
    push ebx
    push ebp
    push edi
    push esi
    sub  esp, 72
%define  args    esp+92
%define  sums    esp+48 ; +24
%define  dc_1d   esp+32 ; +16
%define  top_1d  esp+16 ; +16
%define  left_1d esp+0  ; +16

    pxor        mm0, mm0
    movq        [sums+0], mm0
    movq        [sums+8], mm0
    movq        [sums+16], mm0

    ; 1D hadamards
    mov         edi, [args+4] ; fdec
    xor         ebp, ebp
    mov         esi, 12
.loop_edge:
    ; left
    shl         esi, 5 ; log(FDEC_STRIDE)
    movzx       eax, byte [edi+esi-1+0*FDEC_STRIDE]
    movzx       ebx, byte [edi+esi-1+1*FDEC_STRIDE]
    movzx       ecx, byte [edi+esi-1+2*FDEC_STRIDE]
    movzx       edx, byte [edi+esi-1+3*FDEC_STRIDE]
    shr         esi, 5
    SCALAR_SUMSUB eax, ebx, ecx, edx
    SCALAR_SUMSUB eax, ecx, ebx, edx
    mov         [left_1d+2*esi+0], ax
    mov         [left_1d+2*esi+2], bx
    mov         [left_1d+2*esi+4], cx
    mov         [left_1d+2*esi+6], dx

    ; top
    movzx       eax, byte [edi+esi-FDEC_STRIDE+0]
    movzx       ebx, byte [edi+esi-FDEC_STRIDE+1]
    movzx       ecx, byte [edi+esi-FDEC_STRIDE+2]
    movzx       edx, byte [edi+esi-FDEC_STRIDE+3]
    SCALAR_SUMSUB eax, ebx, ecx, edx
    SCALAR_SUMSUB eax, ecx, ebx, edx
    mov         [top_1d+2*esi+0], ax
    mov         [top_1d+2*esi+2], bx
    mov         [top_1d+2*esi+4], cx
    mov         [top_1d+2*esi+6], dx
    sub         esi, 4
    jge .loop_edge

    ; dc
    movzx       eax, word [left_1d+0]
    movzx       ebx, word [top_1d+0]
    movzx       ecx, word [left_1d+8]
    movzx       edx, word [top_1d+8]
    add         eax, ebx
    lea         ebx, [ecx + edx]
    lea         eax, [2*eax + 8]
    lea         ebx, [2*ebx + 8]
    lea         ecx, [4*ecx + 8]
    lea         edx, [4*edx + 8]
    and         eax, -16
    and         ebx, -16
    and         ecx, -16
    and         edx, -16
    mov         [dc_1d+ 0], eax ; tl
    mov         [dc_1d+ 4], edx ; tr
    mov         [dc_1d+ 8], ecx ; bl
    mov         [dc_1d+12], ebx ; br
    lea         ebp, [dc_1d]

    ; 2D hadamards
    mov         eax, [args+0] ; fenc
    xor         edi, edi
.loop_y:
    xor         esi, esi
.loop_x:
    LOAD_HADAMARD eax

    movq        mm4, mm1
    movq        mm5, mm2
    MMX_ABS_TWO mm4, mm5, mm6, mm7
    movq        mm7, mm3
    paddw       mm4, mm5
    MMX_ABS     mm7, mm6
    paddw       mm7, mm4 ; 3x4 sum

    movq        mm4, [left_1d+8*edi]
    movd        mm5, [ebp]
    psllw       mm4, 2
    psubw       mm4, mm0
    psubw       mm5, mm0
    punpcklwd   mm0, mm1
    punpcklwd   mm2, mm3
    punpckldq   mm0, mm2 ; transpose
    movq        mm1, [top_1d+8*esi]
    psllw       mm1, 2
    psubw       mm0, mm1
    MMX_ABS     mm4, mm3 ; 1x4 sum
    MMX_ABS     mm5, mm2 ; 1x4 sum
    MMX_ABS     mm0, mm1 ; 4x1 sum
    pavgw       mm4, mm7
    pavgw       mm5, mm7
    paddw       mm0, [sums+16] ; i4x4_v satd
    paddw       mm4, [sums+8]  ; i4x4_h satd
    paddw       mm5, [sums+0]  ; i4x4_dc satd
    movq        [sums+16], mm0
    movq        [sums+8], mm4
    movq        [sums+0], mm5

    add         eax, 4
    add         ebp, 4
    inc         esi
    cmp         esi, 2
    jl  .loop_x
    add         eax, 4*FENC_STRIDE-8
    inc         edi
    cmp         edi, 2
    jl  .loop_y

; horizontal sum
    movq        mm0, [sums+0]
    movq        mm1, [sums+8]
    movq        mm2, [sums+16]
    movq        mm6, mm0
    psrlq       mm6, 15
    paddw       mm2, mm6
    SUM_MM_X3   mm0, mm1, mm2, mm3, mm4, mm5, mm7, paddd
    psrld       mm2, 1
    mov         eax, [args+8] ; res
    movd        [eax+0], mm0 ; i8x8c_dc satd
    movd        [eax+4], mm1 ; i8x8c_h satd
    movd        [eax+8], mm2 ; i8x8c_v satd

    add  esp, 72
    pop  esi
    pop  edi
    pop  ebp
    pop  ebx
    ret

%macro LOAD_4x8P 1 ; dx
    pxor        mm7, mm7
    movd        mm6, [eax+%1+7*FENC_STRIDE]
    movd        mm0, [eax+%1+0*FENC_STRIDE]
    movd        mm1, [eax+%1+1*FENC_STRIDE]
    movd        mm2, [eax+%1+2*FENC_STRIDE]
    movd        mm3, [eax+%1+3*FENC_STRIDE]
    movd        mm4, [eax+%1+4*FENC_STRIDE]
    movd        mm5, [eax+%1+5*FENC_STRIDE]
    punpcklbw   mm6, mm7
    punpcklbw   mm0, mm7
    punpcklbw   mm1, mm7
    movq    [spill], mm6
    punpcklbw   mm2, mm7
    punpcklbw   mm3, mm7
    movd        mm6, [eax+%1+6*FENC_STRIDE]
    punpcklbw   mm4, mm7
    punpcklbw   mm5, mm7
    punpcklbw   mm6, mm7
    movq        mm7, [spill]
%endmacro

ALIGN 16
;-----------------------------------------------------------------------------
;  void x264_intra_sa8d_x3_8x8_core_mmxext( uint8_t *fenc, int16_t edges[2][8], int *res )
;-----------------------------------------------------------------------------
x264_intra_sa8d_x3_8x8_core_mmxext:
    mov    eax, [esp+4]
    mov    ecx, [esp+8]
    sub    esp, 0x70
%define args  esp+0x74
%define spill esp+0x60 ; +16
%define trans esp+0    ; +96
%define sum   esp+0    ; +32
    LOAD_4x8P 0
    HADAMARD1x8 mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7

    movq   [spill], mm0
    TRANSPOSE4x4 mm4, mm5, mm6, mm7, mm0
    movq   [trans+0x00], mm4
    movq   [trans+0x08], mm7
    movq   [trans+0x10], mm0
    movq   [trans+0x18], mm6
    movq   mm0, [spill]
    TRANSPOSE4x4 mm0, mm1, mm2, mm3, mm4
    movq   [trans+0x20], mm0
    movq   [trans+0x28], mm3
    movq   [trans+0x30], mm4
    movq   [trans+0x38], mm2

    LOAD_4x8P 4
    HADAMARD1x8 mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7

    movq   [spill], mm7
    TRANSPOSE4x4 mm0, mm1, mm2, mm3, mm7
    movq   [trans+0x40], mm0
    movq   [trans+0x48], mm3
    movq   [trans+0x50], mm7
    movq   [trans+0x58], mm2
    movq   mm7, [spill]
    TRANSPOSE4x4 mm4, mm5, mm6, mm7, mm0
    movq   mm5, [trans+0x00]
    movq   mm1, [trans+0x08]
    movq   mm2, [trans+0x10]
    movq   mm3, [trans+0x18]

    HADAMARD1x8 mm5, mm1, mm2, mm3, mm4, mm7, mm0, mm6

    movq [spill+0], mm5
    movq [spill+8], mm7
    MMX_ABS_TWO  mm0, mm1, mm5, mm7
    MMX_ABS_TWO  mm2, mm3, mm5, mm7
    paddw    mm0, mm2
    paddw    mm1, mm3
    paddw    mm0, mm1
    MMX_ABS_TWO  mm4, mm6, mm2, mm3
    movq     mm5, [spill+0]
    movq     mm7, [spill+8]
    paddw    mm0, mm4
    paddw    mm0, mm6
    MMX_ABS  mm7, mm1
    paddw    mm0, mm7 ; 7x4 sum
    movq     mm6, mm5
    movq     mm7, [ecx+8] ; left bottom
    psllw    mm7, 3
    psubw    mm6, mm7
    MMX_ABS_TWO  mm5, mm6, mm2, mm3
    paddw    mm5, mm0
    paddw    mm6, mm0
    movq [sum+0], mm5 ; dc
    movq [sum+8], mm6 ; left

    movq   mm0, [trans+0x20]
    movq   mm1, [trans+0x28]
    movq   mm2, [trans+0x30]
    movq   mm3, [trans+0x38]
    movq   mm4, [trans+0x40]
    movq   mm5, [trans+0x48]
    movq   mm6, [trans+0x50]
    movq   mm7, [trans+0x58]
    
    HADAMARD1x8 mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7

    movd   [sum+0x10], mm0
    movd   [sum+0x12], mm1
    movd   [sum+0x14], mm2
    movd   [sum+0x16], mm3
    movd   [sum+0x18], mm4
    movd   [sum+0x1a], mm5
    movd   [sum+0x1c], mm6
    movd   [sum+0x1e], mm7

    movq [spill],   mm0
    movq [spill+8], mm1
    MMX_ABS_TWO  mm2, mm3, mm0, mm1
    MMX_ABS_TWO  mm4, mm5, mm0, mm1
    paddw    mm2, mm3
    paddw    mm4, mm5
    paddw    mm2, mm4
    movq     mm0, [spill]
    movq     mm1, [spill+8]
    MMX_ABS_TWO  mm6, mm7, mm4, mm5
    MMX_ABS  mm1, mm4
    paddw    mm2, mm7
    paddw    mm1, mm6
    paddw    mm2, mm1 ; 7x4 sum
    movq     mm1, mm0

    movq     mm7, [ecx+0]
    psllw    mm7, 3   ; left top

    movzx    edx, word [ecx+0]
    add      dx,  [ecx+16]
    lea      edx, [4*edx+32]
    and      edx, -64
    movd     mm6, edx ; dc

    psubw    mm1, mm7
    psubw    mm0, mm6
    MMX_ABS_TWO  mm0, mm1, mm5, mm6
    movq     mm3, [sum+0] ; dc
    paddw    mm0, mm2
    paddw    mm1, mm2
    movq     mm2, mm0
    paddw    mm0, mm3
    paddw    mm1, [sum+8] ; h
    psrlq    mm2, 16
    paddw    mm2, mm3

    movq     mm3, [ecx+16] ; top left
    movq     mm4, [ecx+24] ; top right
    psllw    mm3, 3
    psllw    mm4, 3
    psubw    mm3, [sum+16]
    psubw    mm4, [sum+24]
    MMX_ABS_TWO  mm3, mm4, mm5, mm6
    paddw    mm2, mm3
    paddw    mm2, mm4 ; v

    SUM_MM_X3   mm0, mm1, mm2, mm3, mm4, mm5, mm6, paddd
    mov      eax, [args+8]
    movd     ecx, mm2
    movd     edx, mm1
    add      ecx, 2
    add      edx, 2
    shr      ecx, 2
    shr      edx, 2
    mov      [eax+0], ecx ; i8x8_v satd
    mov      [eax+4], edx ; i8x8_h satd
    movd     ecx, mm0
    add      ecx, 2
    shr      ecx, 2
    mov      [eax+8], ecx ; i8x8_dc satd

    add      esp, 0x70
    ret
%undef args
%undef spill
%undef trans
%undef sum



;-----------------------------------------------------------------------------
; void x264_pixel_ssim_4x4x2_core_mmxext( const uint8_t *pix1, int stride1,
;                                         const uint8_t *pix2, int stride2, int sums[2][4] )
;-----------------------------------------------------------------------------
ALIGN 16
x264_pixel_ssim_4x4x2_core_mmxext:
    push      ebx
    push      edi
    mov       ebx, [esp+16]
    mov       edx, [esp+24]
    mov       edi, 4
    pxor      mm0, mm0
.loop
    mov       eax, [esp+12]
    mov       ecx, [esp+20]
    add       eax, edi
    add       ecx, edi
    pxor      mm1, mm1
    pxor      mm2, mm2
    pxor      mm3, mm3
    pxor      mm4, mm4
%rep 4
    movd      mm5, [eax]
    movd      mm6, [ecx]
    punpcklbw mm5, mm0
    punpcklbw mm6, mm0
    paddw     mm1, mm5
    paddw     mm2, mm6
    movq      mm7, mm5
    pmaddwd   mm5, mm5
    pmaddwd   mm7, mm6
    pmaddwd   mm6, mm6
    paddd     mm3, mm5
    paddd     mm4, mm7
    paddd     mm3, mm6
    add       eax, ebx
    add       ecx, edx
%endrep
    mov       eax, [esp+28]
    lea       eax, [eax+edi*4]
    pshufw    mm5, mm1, 0xE
    pshufw    mm6, mm2, 0xE
    paddusw   mm1, mm5
    paddusw   mm2, mm6
    punpcklwd mm1, mm2
    pshufw    mm2, mm1, 0xE
    pshufw    mm5, mm3, 0xE
    pshufw    mm6, mm4, 0xE
    paddusw   mm1, mm2
    paddd     mm3, mm5
    paddd     mm4, mm6
    punpcklwd mm1, mm0
    punpckldq mm3, mm4
    movq  [eax+0], mm1
    movq  [eax+8], mm3
    sub       edi, 4
    jge       .loop
    pop       edi
    pop       ebx
    emms
    ret



;-----------------------------------------------------------------------------
;  void x264_pixel_ads4_mmxext( int enc_dc[4], uint16_t *sums, int delta,
;                               uint16_t *res, int width )
;-----------------------------------------------------------------------------
ALIGN 16
x264_pixel_ads4_mmxext:
    push    ebx
    mov     eax, [esp+8]
    movq    mm6, [eax]
    movq    mm4, [eax+8]
    pshufw  mm7, mm6, 0
    pshufw  mm6, mm6, 0xAA
    pshufw  mm5, mm4, 0
    pshufw  mm4, mm4, 0xAA
    mov     eax, [esp+12]
    mov     ebx, [esp+16]
    mov     ecx, [esp+20]
    mov     edx, [esp+24]
    shl     ebx, 1
.loop:
    movq    mm0, [eax]
    movq    mm1, [eax+16]
    psubw   mm0, mm7
    psubw   mm1, mm6
    MMX_ABS mm0, mm2
    MMX_ABS mm1, mm3
    movq    mm2, [eax+ebx]
    movq    mm3, [eax+ebx+16]
    psubw   mm2, mm5
    psubw   mm3, mm4
    paddw   mm0, mm1
    MMX_ABS mm2, mm1
    MMX_ABS mm3, mm1
    paddw   mm0, mm2
    paddw   mm0, mm3
    movq    [ecx], mm0
    add     eax, 8
    add     ecx, 8
    sub     edx, 4
    jg      .loop
    pop     ebx
    ret

ALIGN 16
x264_pixel_ads2_mmxext:
    push    ebx
    mov     eax, [esp+8]
    movq    mm6, [eax]
    pshufw  mm7, mm6, 0
    pshufw  mm6, mm6, 0xAA
    mov     eax, [esp+12]
    mov     ebx, [esp+16]
    mov     ecx, [esp+20]
    mov     edx, [esp+24]
    shl     ebx, 1
.loop:
    movq    mm0, [eax]
    movq    mm1, [eax+ebx]
    psubw   mm0, mm7
    psubw   mm1, mm6
    MMX_ABS mm0, mm2
    MMX_ABS mm1, mm3
    paddw   mm0, mm1
    movq    [ecx], mm0
    add     eax, 8
    add     ecx, 8
    sub     edx, 4
    jg      .loop
    pop     ebx
    ret

ALIGN 16
x264_pixel_ads1_mmxext:
    mov     eax, [esp+4]
    pshufw  mm7, [eax], 0
    mov     eax, [esp+8]
    mov     ecx, [esp+16]
    mov     edx, [esp+20]
.loop:
    movq    mm0, [eax]
    movq    mm1, [eax+8]
    psubw   mm0, mm7
    psubw   mm1, mm7
    MMX_ABS mm0, mm2
    MMX_ABS mm1, mm3
    movq    [ecx], mm0
    movq    [ecx+8], mm1
    add     eax, 16
    add     ecx, 16
    sub     edx, 8
    jg      .loop
    nop
    ret