pixel-a.asm

从服务器上下的x264编码器C源码……希望对大家有帮助……这个是09年4月的

    paddw       m0, m1
    pshufw      m1, m0, 10110001b
    paddw       m0, m1
    movd       eax, m0
    and        eax, 0xffff
    RET
%endmacro

; FIXME avoid the spilling of regs to hold 3*stride.
; for small blocks on x86_32, modify pixel pointer instead.

;-----------------------------------------------------------------------------
; int x264_pixel_satd_16x16_mmxext (uint8_t *, int, uint8_t *, int )
;-----------------------------------------------------------------------------
INIT_MMX
cglobal x264_pixel_satd_16x4_internal_mmxext
    SATD_4x4_MMX m2,  0, 0
    SATD_4x4_MMX m1,  4, 0
    paddw        m0, m2
    SATD_4x4_MMX m2,  8, 0
    paddw        m0, m1
    SATD_4x4_MMX m1, 12, 0
    paddw        m0, m2
    paddw        m0, m1
    ret

cglobal x264_pixel_satd_8x8_internal_mmxext
    SATD_4x4_MMX m2,  0, 0
    SATD_4x4_MMX m1,  4, 1
    paddw        m0, m2
    paddw        m0, m1
x264_pixel_satd_8x4_internal_mmxext:
    SATD_4x4_MMX m2,  0, 0
    SATD_4x4_MMX m1,  4, 0
    paddw        m0, m2
    paddw        m0, m1
    ret

cglobal x264_pixel_satd_16x16_mmxext, 4,6
    SATD_START_MMX
    pxor   m0, m0
%rep 3
    call x264_pixel_satd_16x4_internal_mmxext
    lea  r0, [r0+4*r1]
    lea  r2, [r2+4*r3]
%endrep
    call x264_pixel_satd_16x4_internal_mmxext
    HADDUW m0, m1
    movd  eax, m0
    RET

cglobal x264_pixel_satd_16x8_mmxext, 4,6
    SATD_START_MMX
    pxor   m0, m0
    call x264_pixel_satd_16x4_internal_mmxext
    lea  r0, [r0+4*r1]
    lea  r2, [r2+4*r3]
    call x264_pixel_satd_16x4_internal_mmxext
    SATD_END_MMX

cglobal x264_pixel_satd_8x16_mmxext, 4,6
    SATD_START_MMX
    pxor   m0, m0
    call x264_pixel_satd_8x8_internal_mmxext
    lea  r0, [r0+4*r1]
    lea  r2, [r2+4*r3]
    call x264_pixel_satd_8x8_internal_mmxext
    SATD_END_MMX

cglobal x264_pixel_satd_8x8_mmxext, 4,6
    SATD_START_MMX
    pxor   m0, m0
    call x264_pixel_satd_8x8_internal_mmxext
    SATD_END_MMX

cglobal x264_pixel_satd_8x4_mmxext, 4,6
    SATD_START_MMX
    pxor   m0, m0
    call x264_pixel_satd_8x4_internal_mmxext
    SATD_END_MMX

cglobal x264_pixel_satd_4x8_mmxext, 4,6
    SATD_START_MMX
    SATD_4x4_MMX m0, 0, 1
    SATD_4x4_MMX m1, 0, 0
    paddw  m0, m1
    SATD_END_MMX

cglobal x264_pixel_satd_4x4_mmxext, 4,6
    SATD_START_MMX
    SATD_4x4_MMX m0, 0, 0
    SATD_END_MMX

%macro SATD_START_SSE2 3
%ifnidn %1, sse2
    mova    %3, [hmul_8p GLOBAL]
%endif
    lea     r4, [3*r1]
    lea     r5, [3*r3]
    pxor    %2, %2
%endmacro

%macro SATD_END_SSE2 2
    HADDW   %2, m7
    movd   eax, %2
    RET
%endmacro

%macro BACKUP_POINTERS 0
%ifdef ARCH_X86_64
    mov    r10, r0
    mov    r11, r2
%endif
%endmacro

%macro RESTORE_AND_INC_POINTERS 0
%ifdef ARCH_X86_64
    lea     r0, [r10+8]
    lea     r2, [r11+8]
%else
    mov     r0, r0mp
    mov     r2, r2mp
    add     r0, 8
    add     r2, 8
%endif
%endmacro

;-----------------------------------------------------------------------------
; int x264_pixel_satd_8x4_sse2 (uint8_t *, int, uint8_t *, int )
;-----------------------------------------------------------------------------
%macro SATDS_SSE2 1
INIT_XMM
%ifnidn %1, sse2
cglobal x264_pixel_satd_4x4_%1, 4, 6, 6
    SATD_START_MMX
    mova m4, [hmul_4p GLOBAL]
    LOAD_DUP_2x4P m2, m5, [r2], [r2+r3]
    LOAD_DUP_2x4P m3, m5, [r2+2*r3], [r2+r5]
    LOAD_DUP_2x4P m0, m5, [r0], [r0+r1]
    LOAD_DUP_2x4P m1, m5, [r0+2*r1], [r0+r4]
    DIFF_SUMSUB_SSSE3 0, 2, 1, 3, 4
    HADAMARD 0, sumsub, 0, 1, 2, 3
    HADAMARD 4, sumsub, 0, 1, 2, 3
    HADAMARD 1, amax, 0, 1, 2, 3
    HADDW m0, m1
    movd eax, m0
    RET
%endif

cglobal x264_pixel_satd_4x8_%1, 4, 6, 8
    SATD_START_MMX
%ifnidn %1, sse2
    mova m7, [hmul_4p GLOBAL]
%endif
    movd m4, [r2]
    movd m5, [r2+r3]
    movd m6, [r2+2*r3]
    add r2, r5
    movd m0, [r0]
    movd m1, [r0+r1]
    movd m2, [r0+2*r1]
    add r0, r4
    movd m3, [r2+r3]
    JDUP m4, m3
    movd m3, [r0+r1]
    JDUP m0, m3
    movd m3, [r2+2*r3]
    JDUP m5, m3
    movd m3, [r0+2*r1]
    JDUP m1, m3
    DIFFOP 0, 4, 1, 5, 7
    movd m5, [r2]
    add r2, r5
    movd m3, [r0]
    add r0, r4
    movd m4, [r2]
    JDUP m6, m4
    movd m4, [r0]
    JDUP m2, m4
    movd m4, [r2+r3]
    JDUP m5, m4
    movd m4, [r0+r1]
    JDUP m3, m4
    DIFFOP 2, 6, 3, 5, 7
    SATD_8x4_SSE %1, 0, 1, 2, 3, 4, 5, 6, swap
    HADDW m6, m1
    movd eax, m6
    RET

cglobal x264_pixel_satd_8x8_internal_%1
    LOAD_SUMSUB_8x4P 0, 1, 2, 3, 4, 5, 7, r0, r2, 1
    SATD_8x4_SSE %1, 0, 1, 2, 3, 4, 5, 6
x264_pixel_satd_8x4_internal_%1:
    LOAD_SUMSUB_8x4P 0, 1, 2, 3, 4, 5, 7, r0, r2, 1
    SATD_8x4_SSE %1, 0, 1, 2, 3, 4, 5, 6
    ret

%ifdef UNIX64 ; 16x8 regresses on phenom win64, 16x16 is almost the same
cglobal x264_pixel_satd_16x4_internal_%1
    LOAD_SUMSUB_16x4P 0, 1, 2, 3, 4, 8, 5, 9, 6, 7, r0, r2, 11
    lea  r2, [r2+4*r3]
    lea  r0, [r0+4*r1]
    SATD_8x4_SSE ssse3, 0, 1, 2, 3, 6, 11, 10
    SATD_8x4_SSE ssse3, 4, 8, 5, 9, 6, 3, 10
    ret

cglobal x264_pixel_satd_16x8_%1, 4,6,12
    SATD_START_SSE2 %1, m10, m7
%ifidn %1, sse2
    mova m7, [pw_00ff GLOBAL]
%endif
    jmp x264_pixel_satd_16x8_internal_%1

cglobal x264_pixel_satd_16x16_%1, 4,6,12
    SATD_START_SSE2 %1, m10, m7
%ifidn %1, sse2
    mova m7, [pw_00ff GLOBAL]
%endif
    call x264_pixel_satd_16x4_internal_%1
    call x264_pixel_satd_16x4_internal_%1
x264_pixel_satd_16x8_internal_%1:
    call x264_pixel_satd_16x4_internal_%1
    call x264_pixel_satd_16x4_internal_%1
    SATD_END_SSE2 %1, m10
%else
cglobal x264_pixel_satd_16x8_%1, 4,6,8
    SATD_START_SSE2 %1, m6, m7
    BACKUP_POINTERS
    call x264_pixel_satd_8x8_internal_%1
    RESTORE_AND_INC_POINTERS
    call x264_pixel_satd_8x8_internal_%1
    SATD_END_SSE2 %1, m6

cglobal x264_pixel_satd_16x16_%1, 4,6,8
    SATD_START_SSE2 %1, m6, m7
    BACKUP_POINTERS
    call x264_pixel_satd_8x8_internal_%1
    call x264_pixel_satd_8x8_internal_%1
    RESTORE_AND_INC_POINTERS
    call x264_pixel_satd_8x8_internal_%1
    call x264_pixel_satd_8x8_internal_%1
    SATD_END_SSE2 %1, m6
%endif

cglobal x264_pixel_satd_8x16_%1, 4,6,8
    SATD_START_SSE2 %1, m6, m7
    call x264_pixel_satd_8x8_internal_%1
    call x264_pixel_satd_8x8_internal_%1
    SATD_END_SSE2 %1, m6

cglobal x264_pixel_satd_8x8_%1, 4,6,8
    SATD_START_SSE2 %1, m6, m7
    call x264_pixel_satd_8x8_internal_%1
    SATD_END_SSE2 %1, m6

cglobal x264_pixel_satd_8x4_%1, 4,6,8
    SATD_START_SSE2 %1, m6, m7
    call x264_pixel_satd_8x4_internal_%1
    SATD_END_SSE2 %1, m6
%endmacro ; SATDS_SSE2

%macro SA8D 1
%ifdef ARCH_X86_64
;-----------------------------------------------------------------------------
; int x264_pixel_sa8d_8x8_sse2( uint8_t *, int, uint8_t *, int )
;-----------------------------------------------------------------------------
cglobal x264_pixel_sa8d_8x8_internal_%1
    lea  r10, [r0+4*r1]
    lea  r11, [r2+4*r3]
    LOAD_SUMSUB_8x4P 0, 1, 2, 8, 5, 6, 7, r0, r2
    LOAD_SUMSUB_8x4P 4, 5, 3, 9, 11, 6, 7, r10, r11
%ifidn %1, sse2 ; sse2 doesn't seem to like the horizontal way of doing things
    HADAMARD8_2D 0, 1, 2, 8, 4, 5, 3, 9, 6, amax
%else ; non-sse2
    HADAMARD4_V m0, m1, m2, m8, m6
    HADAMARD4_V m4, m5, m3, m9, m6
    SUMSUB_BADC m0, m4, m1, m5, m6
    HADAMARD 2, sumsub, 0, 4, 6, 11
    HADAMARD 2, sumsub, 1, 5, 6, 11
    SUMSUB_BADC m2, m3, m8, m9, m6
    HADAMARD 2, sumsub, 2, 3, 6, 11
    HADAMARD 2, sumsub, 8, 9, 6, 11
    HADAMARD 1, amax, 0, 4, 6, 11
    HADAMARD 1, amax, 1, 5, 6, 4
    HADAMARD 1, amax, 2, 3, 6, 4
    HADAMARD 1, amax, 8, 9, 6, 4
%endif
    paddw m0, m1
    paddw m0, m2
    paddw m0, m8
    SAVE_MM_PERMUTATION x264_pixel_sa8d_8x8_internal_%1
    ret

cglobal x264_pixel_sa8d_8x8_%1, 4,6,12
    lea  r4, [3*r1]
    lea  r5, [3*r3]
%ifnidn %1, sse2
    mova m7, [hmul_8p GLOBAL]
%endif
    call x264_pixel_sa8d_8x8_internal_%1
    HADDW m0, m1
    movd eax, m0
    add eax, 1
    shr eax, 1
    RET

cglobal x264_pixel_sa8d_16x16_%1, 4,6,12
    lea  r4, [3*r1]
    lea  r5, [3*r3]
%ifnidn %1, sse2
    mova m7, [hmul_8p GLOBAL]
%endif
    call x264_pixel_sa8d_8x8_internal_%1 ; pix[0]
    add  r2, 8
    add  r0, 8
    mova m10, m0
    call x264_pixel_sa8d_8x8_internal_%1 ; pix[8]
    lea  r2, [r2+8*r3]
    lea  r0, [r0+8*r1]
    paddusw m10, m0
    call x264_pixel_sa8d_8x8_internal_%1 ; pix[8*stride+8]
    sub  r2, 8
    sub  r0, 8
    paddusw m10, m0
    call x264_pixel_sa8d_8x8_internal_%1 ; pix[8*stride]
    paddusw m0, m10
    HADDUW m0, m1
    movd eax, m0
    add  eax, 1
    shr  eax, 1
    RET

%else ; ARCH_X86_32
%ifnidn %1, mmxext
cglobal x264_pixel_sa8d_8x8_internal_%1
    %define spill0 [esp+4]
    %define spill1 [esp+20]
    %define spill2 [esp+36]
%ifidn %1, sse2
    LOAD_DIFF_8x4P 0, 1, 2, 3, 4, 5, 6, r0, r2, 1
    HADAMARD4_2D 0, 1, 2, 3, 4
    movdqa spill0, m3
    LOAD_DIFF_8x4P 4, 5, 6, 7, 3, 3, 2, r0, r2, 1
    HADAMARD4_2D 4, 5, 6, 7, 3
    HADAMARD2_2D 0, 4, 1, 5, 3, qdq, amax
    movdqa m3, spill0
    paddw m0, m1
    HADAMARD2_2D 2, 6, 3, 7, 5, qdq, amax
%else ; non-sse2
    mova m7, [hmul_8p GLOBAL]
    LOAD_SUMSUB_8x4P 0, 1, 2, 3, 5, 6, 7, r0, r2, 1
    ; could do first HADAMARD4_V here to save spilling later
    ; surprisingly, not a win on conroe or even p4
    mova spill0, m2
    mova spill1, m3
    mova spill2, m1
    SWAP 1, 7
    LOAD_SUMSUB_8x4P 4, 5, 6, 7, 2, 3, 1, r0, r2, 1
    HADAMARD4_V m4, m5, m6, m7, m3
    mova m1, spill2
    mova m2, spill0
    mova m3, spill1
    mova spill0, m6
    mova spill1, m7
    HADAMARD4_V m0, m1, m2, m3, m7
    SUMSUB_BADC m0, m4, m1, m5, m7
    HADAMARD 2, sumsub, 0, 4, 7, 6
    HADAMARD 2, sumsub, 1, 5, 7, 6
    HADAMARD 1, amax, 0, 4, 7, 6
    HADAMARD 1, amax, 1, 5, 7, 6
    mova m6, spill0
    mova m7, spill1
    paddw m0, m1
    SUMSUB_BADC m2, m6, m3, m7, m4
    HADAMARD 2, sumsub, 2, 6, 4, 5
    HADAMARD 2, sumsub, 3, 7, 4, 5
    HADAMARD 1, amax, 2, 6, 4, 5
    HADAMARD 1, amax, 3, 7, 4, 5
%endif ; sse2/non-sse2
    paddw m0, m2
    paddw m0, m3
    ret
%endif ; ifndef mmxext

cglobal x264_pixel_sa8d_8x8_%1, 4,7
    mov  r6, esp
    and  esp, ~15
    sub  esp, 48
    lea  r4, [3*r1]
    lea  r5, [3*r3]
    call x264_pixel_sa8d_8x8_internal_%1
    HADDW m0, m1
    movd eax, m0
    add  eax, 1
    shr  eax, 1
    mov  esp, r6
    RET

cglobal x264_pixel_sa8d_16x16_%1, 4,7
    mov  r6, esp
    and  esp, ~15
    sub  esp, 64
    lea  r4, [3*r1]
    lea  r5, [3*r3]
    call x264_pixel_sa8d_8x8_internal_%1
%ifidn %1, mmxext
    lea  r0, [r0+4*r1]
    lea  r2, [r2+4*r3]
%endif
    mova [esp+48], m0
    call x264_pixel_sa8d_8x8_internal_%1
    mov  r0, [r6+20]
    mov  r2, [r6+28]
    add  r0, 8
    add  r2, 8
    paddusw m0, [esp+48]
    mova [esp+48], m0
    call x264_pixel_sa8d_8x8_internal_%1
%ifidn %1, mmxext
    lea  r0, [r0+4*r1]
    lea  r2, [r2+4*r3]
%endif
%if mmsize == 16
    paddusw m0, [esp+48]
%endif
    mova [esp+64-mmsize], m0
    call x264_pixel_sa8d_8x8_internal_%1
    paddusw m0, [esp+64-mmsize]
%if mmsize == 16
    HADDUW m0, m1
%else
    mova m2, [esp+48]
    pxor m7, m7
    mova m1, m0
    mova m3, m2
    punpcklwd m0, m7
    punpckhwd m1, m7
    punpcklwd m2, m7
    punpckhwd m3, m7
    paddd m0, m1
    paddd m2, m3
    paddd m0, m2
    HADDD m0, m1
%endif
    movd eax, m0
    add  eax, 1
    shr  eax, 1
    mov  esp, r6
    RET
%endif ; !ARCH_X86_64
%endmacro ; SA8D

;=============================================================================
; INTRA SATD
;=============================================================================

%macro INTRA_SA8D_SSE2 1
%ifdef ARCH_X86_64
INIT_XMM
;-----------------------------------------------------------------------------
; void x264_intra_sa8d_x3_8x8_core_sse2( uint8_t *fenc, int16_t edges[2][8], int *res )
;-----------------------------------------------------------------------------
cglobal x264_intra_sa8d_x3_8x8_core_%1, 3,3,16
    ; 8x8 hadamard
    pxor        m8, m8
    movq        m0, [r0+0*FENC_STRIDE]
    movq        m1, [r0+1*FENC_STRIDE]
    movq        m2, [r0+2*FENC_STRIDE]
    movq        m3, [r0+3*FENC_STRIDE]
    movq        m4, [r0+4*FENC_STRIDE]
    movq        m5, [r0+5*FENC_STRIDE]
    movq        m6, [r0+6*FENC_STRIDE]
    movq        m7, [r0+7*FENC_STRIDE]
    punpcklbw   m0, m8
    punpcklbw   m1, m8
    punpcklbw   m2, m8
    punpcklbw   m3, m8
    punpcklbw   m4, m8
    punpcklbw   m5, m8
    punpcklbw   m6, m8
    punpcklbw   m7, m8

    HADAMARD8_2D 0,  1,  2,  3,  4,  5,  6,  7,  8

    ; dc
    movzx       r0d, word [r1+0]
    add         r0w, word [r1+16]
    add         r0d, 8
    and         r0d, -16
    shl         r0d, 2

    pxor        m15, m15
    movdqa      m8,  m2
    movdqa      m9,  m3
    movdqa      m10, m4
    movdqa      m11, m5
    ABS4        m8, m9, m10, m11, m12, m13
    paddusw     m8,  m10
    paddusw     m9,  m11
%ifidn %1, ssse3
    pabsw       m10, m6
    pabsw       m11, m7
    pabsw       m15, m1
%else
    movdqa      m10, m6
    movdqa      m11, m7
    movdqa      m15, m1
    ABS2        m10, m11, m13, m14
    ABS1        m15, m13
%endif
    paddusw     m10, m11
    paddusw     m8,  m9
    paddusw     m15, m10
    paddusw     m15, m8
    movdqa      m14, m15 ; 7x8 sum

    movdqa      m8,  [r1+0] ; left edge
    movd        m9,  r0d
    psllw       m8,  3
    psubw       m8,  m0
    psubw       m9,  m0
    ABS1        m8,  m10
    ABS1        m9,  m11 ; 1x8 sum
    paddusw     m14, m8
    paddusw     m15, m9
    punpcklwd   m0,  m1
    punpcklwd   m2,  m3
    punpcklwd   m4,  m5
    punpcklwd   m6,  m7
    punpckldq   m0,  m2
    punpckldq   m4,  m6
    punpcklqdq  m0,  m4 ; transpose
    movdqa      m1,  [r1+16] ; top edge
    movdqa      m2,  m15
    psllw       m1,  3
    psrldq      m2,  2     ; 8x7 sum
    psubw       m0,  m1  ; 8x1 sum
    ABS1        m0,  m1
    paddusw     m2,  m0

    ; 3x HADDW
    movdqa      m7,  [pw_1 GLOBAL]
    pmaddwd     m2,  m7
    pmaddwd     m14, m7
    pmaddwd     m15, m7
    movdqa      m3,  m2
    punpckldq   m2,  m14
    punpckhdq   m3,  m14
    pshufd      m5,  m15, 0xf5
    paddd       m2,  m3
    paddd       m5,  m15
    movdqa      m3,  m2
    punpcklqdq  m2,  m5
    punpckhqdq  m3,  m5
    pavgw       m3,  m2
    pxor        m0,  m0
    pavgw       m3,  m0
    movq      [r2],  m3 ; i8x8_v, i8x8_h
    psrldq      m3,  8
    movd    [r2+8],  m3 ; i8x8_dc
    RET
%endif ; ARCH_X86_64
%endmacro ; INTRA_SA8D_SSE2

; in: r0 = fenc
; out: m0..m3 = hadamard coefs
INIT_MMX
ALIGN 16
load_hadamard:
    pxor        m7, m7
    movd        m0, [r0+0*FENC_STRIDE]
    movd        m1, [r0+1*FENC_STRIDE]
    movd        m2, [r0+2*FENC_STRIDE]
    movd        m3, [r0+3*FENC_STRIDE]
    punpcklbw   m0, m7
    punpcklbw   m1, m7
    punpcklbw   m2, m7
    punpcklbw   m3, m7
    HADAMARD4_2D 0, 1, 2, 3, 4
    SAVE_MM_PERMUTATION load_hadamard
    ret

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

%macro SCALAR_HADAMARD_LEFT 5 ; y, 4x tmp
%ifnidn %1, 0
    shl         %1d, 5 ; log(FDEC_STRIDE)
%endif
    movzx       %2d, byte [r1+%1-1+0*FDEC_STRIDE]
    movzx       %3d, byte [r1+%1-1+1*FDEC_STRIDE]
    movzx       %4d, byte [r1+%1-1+2*FDEC_STRIDE]
    movzx       %5d, byte [r1+%1-1+3*FDEC_STRIDE]
%ifnidn %1, 0
    shr         %1d, 5
%endif
    SCALAR_SUMSUB %2d, %3d, %4d, %5d