pixel-a.asm

DM642 H.264 codec DM642 H.264 codec DM642 H.264 codec DM642 H.264 codec

SAD  8, 16
SAD  8,  8
SAD  8,  4
SAD  4,  8
SAD  4,  4

;-----------------------------------------------------------------------------
;  void x264_pixel_sad_x3_16x16_mmxext( uint8_t *fenc, uint8_t *pix0, uint8_t *pix1,
;                                       uint8_t *pix2, int i_stride, int scores[3] )
;-----------------------------------------------------------------------------
%macro SAD_X 3
ALIGN 16
x264_pixel_sad_x%1_%2x%3_mmxext:
    SAD_X%1_2x%2P 1
%rep %3/2-1
    SAD_X%1_2x%2P 0
%endrep
    SAD_X%1_END
%endmacro

SAD_X 3, 16, 16
SAD_X 3, 16,  8
SAD_X 3,  8, 16
SAD_X 3,  8,  8
SAD_X 3,  8,  4
SAD_X 3,  4,  8
SAD_X 3,  4,  4
SAD_X 4, 16, 16
SAD_X 4, 16,  8
SAD_X 4,  8, 16
SAD_X 4,  8,  8
SAD_X 4,  8,  4
SAD_X 4,  4,  8
SAD_X 4,  4,  4


;-----------------------------------------------------------------------------
;   int __cdecl x264_pixel_sad_pde_16x16_mmxext (uint8_t *, int, uint8_t *, int, int )
;-----------------------------------------------------------------------------
%macro SAD_PDE 2
ALIGN 16
x264_pixel_sad_pde_%1x%2_mmxext:
    SAD_START
%rep %2/4
    SAD_INC_2x%1P
%endrep

    movd ebx, mm0
    cmp  ebx, [esp+24] ; prev_score
    jl   .continue
    pop  ebx
    mov  eax, 0xffff
    ret
ALIGN 4
.continue:
    mov  ebx, [esp+12]

%rep %2/4
    SAD_INC_2x%1P
%endrep
    SAD_END
%endmacro

SAD_PDE 16, 16
SAD_PDE 16 , 8
SAD_PDE  8, 16



%macro SSD_START 0
    push    ebx

    mov     eax,    [esp+ 8]    ; pix1
    mov     ebx,    [esp+12]    ; stride1
    mov     ecx,    [esp+16]    ; pix2
    mov     edx,    [esp+20]    ; stride2

    pxor    mm7,    mm7         ; zero
    pxor    mm0,    mm0         ; mm0 holds the sum
%endmacro

%macro SSD_END 0
    movq    mm1,    mm0
    psrlq   mm1,    32
    paddd   mm0,    mm1
    movd    eax,    mm0

    pop ebx
    ret
%endmacro

;-----------------------------------------------------------------------------
;   int __cdecl x264_pixel_ssd_16x16_mmx (uint8_t *, int, uint8_t *, int )
;-----------------------------------------------------------------------------
%macro SSD 2
ALIGN 16
x264_pixel_ssd_%1x%2_mmx:
    SSD_START
%rep %2
    SSD_INC_1x%1P
%endrep
    SSD_END
%endmacro

SSD 16, 16
SSD 16,  8
SSD  8, 16
SSD  8,  8
SSD  8,  4
SSD  4,  8
SSD  4,  4



%macro SATD_START 0
    push        ebx

    mov         eax, [esp+ 8]       ; pix1
    mov         ebx, [esp+12]       ; stride1
    mov         ecx, [esp+16]       ; pix2
    mov         edx, [esp+20]       ; stride2
%endmacro

%macro SATD_END 0
    pshufw      mm1, mm0, 01001110b
    paddw       mm0, mm1
    pshufw      mm1, mm0, 10110001b
    paddw       mm0, mm1
    movd        eax, mm0
    and         eax, 0xffff
    pop         ebx
    ret
%endmacro

ALIGN 16
;-----------------------------------------------------------------------------
;   int __cdecl x264_pixel_satd_4x4_mmxext (uint8_t *, int, uint8_t *, int )
;-----------------------------------------------------------------------------
x264_pixel_satd_4x4_mmxext:
    SATD_START
    LOAD_DIFF_HADAMARD_SUM mm0, 0, 0
    SATD_END

ALIGN 16
;-----------------------------------------------------------------------------
;   int __cdecl x264_pixel_satd_4x8_mmxext (uint8_t *, int, uint8_t *, int )
;-----------------------------------------------------------------------------
x264_pixel_satd_4x8_mmxext:
    SATD_START
    LOAD_DIFF_HADAMARD_SUM mm0, 0, 1
    LOAD_DIFF_HADAMARD_SUM mm1, 0, 0
    paddw       mm0, mm1
    SATD_END

ALIGN 16
;-----------------------------------------------------------------------------
;   int __cdecl x264_pixel_satd_8x4_mmxext (uint8_t *, int, uint8_t *, int )
;-----------------------------------------------------------------------------
x264_pixel_satd_8x4_mmxext:
    SATD_START
    LOAD_DIFF_HADAMARD_SUM mm0, 0, 0
    sub         eax, ebx
    sub         ecx, edx
    LOAD_DIFF_HADAMARD_SUM mm1, 4, 0
    paddw       mm0, mm1
    SATD_END

ALIGN 16
;-----------------------------------------------------------------------------
;   int __cdecl x264_pixel_satd_8x8_mmxext (uint8_t *, int, uint8_t *, int )
;-----------------------------------------------------------------------------
x264_pixel_satd_8x8_mmxext:
    SATD_START
    LOAD_DIFF_HADAMARD_SUM mm0, 0, 1
    LOAD_DIFF_HADAMARD_SUM mm1, 0, 0

    mov         eax, [esp+ 8]       ; pix1
    mov         ecx, [esp+16]       ; pix2
    LOAD_DIFF_HADAMARD_SUM mm2, 4, 1
    paddw       mm0, mm1
    LOAD_DIFF_HADAMARD_SUM mm1, 4, 0
    paddw       mm0, mm2
    paddw       mm0, mm1
    SATD_END

ALIGN 16
;-----------------------------------------------------------------------------
;   int __cdecl x264_pixel_satd_16x8_mmxext (uint8_t *, int, uint8_t *, int )
;-----------------------------------------------------------------------------
x264_pixel_satd_16x8_mmxext:
    SATD_START
    LOAD_DIFF_HADAMARD_SUM mm0, 0, 1
    LOAD_DIFF_HADAMARD_SUM mm1, 0, 0

    mov         eax, [esp+ 8]       ; pix1
    mov         ecx, [esp+16]       ; pix2
    LOAD_DIFF_HADAMARD_SUM mm2, 4, 1
    paddw       mm0, mm1
    LOAD_DIFF_HADAMARD_SUM mm1, 4, 0
    paddw       mm0, mm2

    mov         eax, [esp+ 8]       ; pix1
    mov         ecx, [esp+16]       ; pix2
    LOAD_DIFF_HADAMARD_SUM mm2, 8, 1
    paddw       mm0, mm1
    LOAD_DIFF_HADAMARD_SUM mm1, 8, 0
    paddw       mm0, mm2

    mov         eax, [esp+ 8]       ; pix1
    mov         ecx, [esp+16]       ; pix2
    LOAD_DIFF_HADAMARD_SUM mm2, 12, 1
    paddw       mm0, mm1
    LOAD_DIFF_HADAMARD_SUM mm1, 12, 0
    paddw       mm0, mm2
    paddw       mm0, mm1
    SATD_END

ALIGN 16
;-----------------------------------------------------------------------------
;   int __cdecl x264_pixel_satd_8x16_mmxext (uint8_t *, int, uint8_t *, int )
;-----------------------------------------------------------------------------
x264_pixel_satd_8x16_mmxext:
    SATD_START
    LOAD_DIFF_HADAMARD_SUM mm0, 0, 1
    LOAD_DIFF_HADAMARD_SUM mm1, 0, 1
    LOAD_DIFF_HADAMARD_SUM mm2, 0, 1
    paddw       mm0, mm1
    LOAD_DIFF_HADAMARD_SUM mm1, 0, 0
    paddw       mm0, mm2

    mov         eax, [esp+ 8]       ; pix1
    mov         ecx, [esp+16]       ; pix2
    LOAD_DIFF_HADAMARD_SUM mm2, 4, 1
    paddw       mm0, mm1
    LOAD_DIFF_HADAMARD_SUM mm1, 4, 1
    paddw       mm0, mm2
    LOAD_DIFF_HADAMARD_SUM mm2, 4, 1
    paddw       mm0, mm1
    LOAD_DIFF_HADAMARD_SUM mm1, 4, 0
    paddw       mm0, mm2
    paddw       mm0, mm1
    SATD_END

ALIGN 16
;-----------------------------------------------------------------------------
;   int __cdecl x264_pixel_satd_16x16_mmxext (uint8_t *, int, uint8_t *, int )
;-----------------------------------------------------------------------------
x264_pixel_satd_16x16_mmxext:
    SATD_START
    LOAD_DIFF_HADAMARD_SUM mm0, 0, 1
    LOAD_DIFF_HADAMARD_SUM mm1, 0, 1
    LOAD_DIFF_HADAMARD_SUM mm2, 0, 1
    paddw       mm0, mm1
    LOAD_DIFF_HADAMARD_SUM mm1, 0, 0
    paddw       mm0, mm2

    mov         eax, [esp+ 8]       ; pix1
    mov         ecx, [esp+16]       ; pix2
    LOAD_DIFF_HADAMARD_SUM mm2, 4, 1
    paddw       mm0, mm1
    LOAD_DIFF_HADAMARD_SUM mm1, 4, 1
    paddw       mm0, mm2
    LOAD_DIFF_HADAMARD_SUM mm2, 4, 1
    paddw       mm0, mm1
    LOAD_DIFF_HADAMARD_SUM mm1, 4, 0
    paddw       mm0, mm2

    mov         eax, [esp+ 8]       ; pix1
    mov         ecx, [esp+16]       ; pix2
    LOAD_DIFF_HADAMARD_SUM mm2, 8, 1
    paddw       mm0, mm1
    LOAD_DIFF_HADAMARD_SUM mm1, 8, 1
    paddw       mm0, mm2
    LOAD_DIFF_HADAMARD_SUM mm2, 8, 1
    paddw       mm0, mm1
    LOAD_DIFF_HADAMARD_SUM mm1, 8, 0
    paddw       mm0, mm2

    mov         eax, [esp+ 8]       ; pix1
    mov         ecx, [esp+16]       ; pix2
    LOAD_DIFF_HADAMARD_SUM mm2, 12, 1
    paddw       mm0, mm1
    LOAD_DIFF_HADAMARD_SUM mm1, 12, 1
    paddw       mm0, mm2
    LOAD_DIFF_HADAMARD_SUM mm2, 12, 1
    paddw       mm0, mm1
    LOAD_DIFF_HADAMARD_SUM mm1, 12, 0
    paddw       mm0, mm2
    paddw       mm0, mm1

    pxor        mm3, mm3
    pshufw      mm1, mm0, 01001110b
    paddw       mm0, mm1
    punpcklwd   mm0, mm3
    pshufw      mm1, mm0, 01001110b
    paddd       mm0, mm1
    movd        eax, mm0
    pop         ebx
    ret


%macro LOAD_DIFF_4x8P 1 ; dx
    LOAD_DIFF_4P  mm0, mm7, %1, 0
    LOAD_DIFF_4P  mm1, mm7, %1, 1
    lea  eax, [eax+2*ebx]
    lea  ecx, [ecx+2*edx]
    LOAD_DIFF_4P  mm2, mm7, %1, 0
    LOAD_DIFF_4P  mm3, mm7, %1, 1
    lea  eax, [eax+2*ebx]
    lea  ecx, [ecx+2*edx]
    LOAD_DIFF_4P  mm4, mm7, %1, 0
    LOAD_DIFF_4P  mm5, mm7, %1, 1
    lea  eax, [eax+2*ebx]
    lea  ecx, [ecx+2*edx]
    LOAD_DIFF_4P  mm6, mm7, %1, 0
    movq [spill], mm6
    LOAD_DIFF_4P  mm7, mm6, %1, 1
    movq mm6, [spill]
%endmacro

%macro HADAMARD1x8 8
    SUMSUB_BADC %1, %5, %2, %6
    SUMSUB_BADC %3, %7, %4, %8
    SUMSUB_BADC %1, %3, %2, %4
    SUMSUB_BADC %5, %7, %6, %8
    SUMSUB_BADC %1, %2, %3, %4
    SUMSUB_BADC %5, %6, %7, %8
%endmacro

%macro SUM4x8_MM 0
    movq [spill],   mm6
    movq [spill+8], mm7
    MMX_ABS_TWO  mm0, mm1, mm6, mm7
    MMX_ABS_TWO  mm2, mm3, mm6, mm7
    paddw    mm0, mm2
    paddw    mm1, mm3
    movq     mm6, [spill]
    movq     mm7, [spill+8]
    MMX_ABS_TWO  mm4, mm5, mm2, mm3
    MMX_ABS_TWO  mm6, mm7, mm2, mm3
    paddw    mm4, mm6
    paddw    mm5, mm7
    paddw    mm0, mm4
    paddw    mm1, mm5
    paddw    mm0, mm1
%endmacro

ALIGN 16
;-----------------------------------------------------------------------------
;   int __cdecl x264_pixel_sa8d_8x8_mmxext( uint8_t *, int, uint8_t *, int )
;-----------------------------------------------------------------------------
x264_pixel_sa8d_8x8_mmxext:
    SATD_START
    sub    esp, 0x70
%define args  esp+0x74
%define spill esp+0x60 ; +16
%define trans esp+0    ; +96
    LOAD_DIFF_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

    mov    eax, [args+4]
    mov    ecx, [args+12]
    LOAD_DIFF_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
    SUM4x8_MM
    movq   [trans], mm0

    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
    SUM4x8_MM

    pavgw  mm0, [esp]
    pshufw mm1, mm0, 01001110b
    paddw  mm0, mm1
    pshufw mm1, mm0, 10110001b
    paddw  mm0, mm1
    movd   eax, mm0
    and    eax, 0xffff
    mov    ecx, eax ; preserve rounding for 16x16
    add    eax, 1
    shr    eax, 1
    add    esp, 0x70
    pop    ebx
    ret
%undef args
%undef spill
%undef trans

ALIGN 16
;-----------------------------------------------------------------------------
;   int __cdecl x264_pixel_sa8d_16x16_mmxext( uint8_t *, int, uint8_t *, int )
;-----------------------------------------------------------------------------
;; violates calling convention
x264_pixel_sa8d_16x16_mmxext:
    push   esi
    push   edi
    push   ebp
    mov    esi, [esp+28]    ; stride2
    mov    edi, [esp+20]    ; stride1
    push   esi
    push   dword [esp+28]   ; pix2
    push   edi
    push   dword [esp+28]   ; pix1
    call x264_pixel_sa8d_8x8_mmxext
    mov    ebp, ecx
    shl    edi, 3
    shl    esi, 3
    add    [esp+0], edi     ; pix1+8*stride1
    add    [esp+8], esi     ; pix2+8*stride2
    call x264_pixel_sa8d_8x8_mmxext
    add    ebp, ecx
    add    dword [esp+0], 8 ; pix1+8*stride1+8
    add    dword [esp+8], 8 ; pix2+8*stride2+8
    call x264_pixel_sa8d_8x8_mmxext
    add    ebp, ecx
    sub    [esp+0], edi     ; pix1+8
    sub    [esp+8], esi     ; pix2+8
    call x264_pixel_sa8d_8x8_mmxext
    lea    eax, [ebp+ecx+1]
    shr    eax, 1
    add    esp, 16
    pop    ebp
    pop    edi
    pop    esi
    ret


; in: fenc
; out: mm0..mm3 = hadamard coefs
%macro LOAD_HADAMARD 1
    pxor        mm7, mm7
    movd        mm0, [%1+0*FENC_STRIDE]
    movd        mm4, [%1+1*FENC_STRIDE]
    movd        mm3, [%1+2*FENC_STRIDE]
    movd        mm1, [%1+3*FENC_STRIDE]
    punpcklbw   mm0, mm7
    punpcklbw   mm4, mm7
    punpcklbw   mm3, mm7
    punpcklbw   mm1, mm7
    HADAMARD4x4 mm0, mm4, mm3, mm1
    TRANSPOSE4x4 mm0, mm4, mm3, mm1, mm2
    HADAMARD4x4 mm0, mm1, mm2, mm3
%endmacro

%macro SCALAR_SUMSUB 4
    add %1, %2
    add %3, %4
    add %2, %2
    add %4, %4
    sub %2, %1
    sub %4, %3
%endmacro

%macro SUM_MM_X3 8 ; 3x sum, 4x tmp, op
    pxor        %7, %7
    pshufw      %4, %1, 01001110b
    pshufw      %5, %2, 01001110b
    pshufw      %6, %3, 01001110b
    paddusw     %1, %4
    paddusw     %2, %5
    paddusw     %3, %6
    punpcklwd   %1, %7
    punpcklwd   %2, %7
    punpcklwd   %3, %7
    pshufw      %4, %1, 01001110b
    pshufw      %5, %2, 01001110b
    pshufw      %6, %3, 01001110b
    %8          %1, %4
    %8          %2, %5
    %8          %3, %6
%endmacro

ALIGN 16
;-----------------------------------------------------------------------------
;  void x264_intra_satd_x3_4x4_mmxext( uint8_t *fenc, uint8_t *fdec, int *res )
;-----------------------------------------------------------------------------
x264_intra_satd_x3_4x4_mmxext:
    push ebx
    push edi
    push esi
    sub  esp, 16
%define  args esp+32
%define  top_1d  esp+8 ; +8
%define  left_1d esp+0 ; +8

    mov  eax, [args+0] ; fenc
    LOAD_HADAMARD eax

    mov  edi, [args+4] ; fdec