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

;*****************************************************************************
;* pixel.asm: h264 encoder library
;*****************************************************************************
;* Copyright (C) 2003-2008 x264 project
;*
;* Authors: Loren Merritt <[e-mail]>
;*          Holger Lubitz <[e-mail]>
;*          Laurent Aimar <[e-mail]>
;*          Alex Izvorski <[e-mail]>
;*          Jason Garrett-Glaser <[e-mail]>
;*
;* This program is free software; you can redistribute it and/or modify
;* it under the terms of the GNU General Public License as published by
;* the Free Software Foundation; either version 2 of the License, or
;* (at your option) any later version.
;*
;* This program is distributed in the hope that it will be useful,
;* but WITHOUT ANY WARRANTY; without even the implied warranty of
;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
;* GNU General Public License for more details.
;*
;* You should have received a copy of the GNU General Public License
;* along with this program; if not, write to the Free Software
;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
;*****************************************************************************

%include "x86inc.asm"
%include "x86util.asm"

SECTION_RODATA
pw_1:      times 8 dw 1
pw_00ff:   times 8 dw 0xff
ssim_c1:   times 4 dd 416    ; .01*.01*255*255*64
ssim_c2:   times 4 dd 235963 ; .03*.03*255*255*64*63
mask_ff:   times 16 db 0xff
           times 16 db 0
mask_ac4:  dw 0, -1, -1, -1, 0, -1, -1, -1
mask_ac4b: dw 0, -1, 0, -1, -1, -1, -1, -1
mask_ac8:  dw 0, -1, -1, -1, -1, -1, -1, -1
hsub_mul:  times 8 db 1, -1
hmul_4p:   times 2 db 1, 1, 1, 1, 1, -1, 1, -1
hmul_8p:   times 8 db 1
           times 4 db 1, -1
mask_10:   times 4 dw 0, -1
mask_1100: times 2 dd 0, -1

SECTION .text

%macro HADDD 2 ; sum junk
%if mmsize == 16
    movhlps %2, %1
    paddd   %1, %2
    pshuflw %2, %1, 0xE
    paddd   %1, %2
%else
    pshufw  %2, %1, 0xE
    paddd   %1, %2
%endif
%endmacro

%macro HADDW 2
    pmaddwd %1, [pw_1 GLOBAL]
    HADDD   %1, %2
%endmacro

%macro HADDUW 2
    mova  %2, %1
    pslld %1, 16
    psrld %2, 16
    psrld %1, 16
    paddd %1, %2
    HADDD %1, %2
%endmacro

;=============================================================================
; SSD
;=============================================================================

%macro SSD_LOAD_FULL 5
    mova      m1, [r0+%1]
    mova      m2, [r2+%2]
    mova      m3, [r0+%3]
    mova      m4, [r2+%4]
%if %5
    lea       r0, [r0+2*r1]
    lea       r2, [r2+2*r3]
%endif
%endmacro

%macro LOAD 5
    movh      m%1, %3
    movh      m%2, %4
%if %5
    lea       r0, [r0+2*r1]
%endif
%endmacro

%macro JOIN 7
    movh      m%3, %5
    movh      m%4, %6
%if %7
    lea       r2, [r2+2*r3]
%endif
    punpcklbw m%1, m7
    punpcklbw m%3, m7
    psubw     m%1, m%3
    punpcklbw m%2, m7
    punpcklbw m%4, m7
    psubw     m%2, m%4
%endmacro

%macro JOIN_SSE2 7
    movh      m%3, %5
    movh      m%4, %6
%if %7
    lea       r2, [r2+2*r3]
%endif
    punpcklqdq m%1, m%2
    punpcklqdq m%3, m%4
    DEINTB %2, %1, %4, %3, 7
    psubw m%2, m%4
    psubw m%1, m%3
%endmacro

%macro JOIN_SSSE3 7
    movh      m%3, %5
    movh      m%4, %6
%if %7
    lea       r2, [r2+2*r3]
%endif
    punpcklbw m%1, m%3
    punpcklbw m%2, m%4
%endmacro

%macro SSD_LOAD_HALF 5
    LOAD      1, 2, [r0+%1], [r0+%3], 1
    JOIN      1, 2, 3, 4, [r2+%2], [r2+%4], 1
    LOAD      3, 4, [r0+%1], [r0+%3], %5
    JOIN      3, 4, 5, 6, [r2+%2], [r2+%4], %5
%endmacro

%macro SSD_CORE 7-8
%ifidn %8, FULL
    mova      m%6, m%2
    mova      m%7, m%4
    psubusb   m%2, m%1
    psubusb   m%4, m%3
    psubusb   m%1, m%6
    psubusb   m%3, m%7
    por       m%1, m%2
    por       m%3, m%4
    mova      m%2, m%1
    mova      m%4, m%3
    punpckhbw m%1, m%5
    punpckhbw m%3, m%5
    punpcklbw m%2, m%5
    punpcklbw m%4, m%5
%endif
    pmaddwd   m%1, m%1
    pmaddwd   m%2, m%2
    pmaddwd   m%3, m%3
    pmaddwd   m%4, m%4
%endmacro

%macro SSD_CORE_SSE2 7-8
%ifidn %8, FULL
    DEINTB %6, %1, %7, %2, %5
    psubw m%6, m%7
    psubw m%1, m%2
    SWAP %2, %6
    DEINTB %6, %3, %7, %4, %5
    psubw m%6, m%7
    psubw m%3, m%4
    SWAP %4, %6
%endif
    pmaddwd   m%1, m%1
    pmaddwd   m%2, m%2
    pmaddwd   m%3, m%3
    pmaddwd   m%4, m%4
%endmacro

%macro SSD_CORE_SSSE3 7-8
%ifidn %8, FULL
    mova      m%6, m%1
    mova      m%7, m%3
    punpcklbw m%1, m%2
    punpcklbw m%3, m%4
    punpckhbw m%6, m%2
    punpckhbw m%7, m%4
    SWAP %6, %2
    SWAP %7, %4
%endif
    pmaddubsw m%1, m%5
    pmaddubsw m%2, m%5
    pmaddubsw m%3, m%5
    pmaddubsw m%4, m%5
    pmaddwd   m%1, m%1
    pmaddwd   m%2, m%2
    pmaddwd   m%3, m%3
    pmaddwd   m%4, m%4
%endmacro

%macro SSD_END 1
    paddd     m1, m2
    paddd     m3, m4
%if %1
    paddd     m0, m1
%else
    SWAP      0, 1
%endif
    paddd     m0, m3
%endmacro

%macro SSD_ITER 7
    SSD_LOAD_%1 %2,%3,%4,%5,%7
    SSD_CORE  1, 2, 3, 4, 7, 5, 6, %1
    SSD_END  %6
%endmacro

;-----------------------------------------------------------------------------
; int x264_pixel_ssd_16x16_mmx( uint8_t *, int, uint8_t *, int )
;-----------------------------------------------------------------------------
%macro SSD 3-4 0
cglobal x264_pixel_ssd_%1x%2_%3, 4,4,%4
%ifidn %3, ssse3
    mova    m7, [hsub_mul GLOBAL]
%elifidn %3, sse2
    mova    m7, [pw_00ff GLOBAL]
%elif %1 >= mmsize
    pxor    m7, m7
%endif
%assign i 0
%rep %2/4
%if %1 > mmsize
    SSD_ITER FULL, 0,  0,     mmsize,    mmsize, i, 0
    SSD_ITER FULL, r1, r3, r1+mmsize, r3+mmsize, 1, 1
    SSD_ITER FULL, 0,  0,     mmsize,    mmsize, 1, 0
    SSD_ITER FULL, r1, r3, r1+mmsize, r3+mmsize, 1, i<%2/4-1
%elif %1 == mmsize
    SSD_ITER FULL, 0, 0, r1, r3, i, 1
    SSD_ITER FULL, 0, 0, r1, r3, 1, i<%2/4-1
%else
    SSD_ITER HALF, 0, 0, r1, r3, i, i<%2/4-1
%endif
%assign i i+1
%endrep
    HADDD   m0, m1
    movd   eax, m0
    RET
%endmacro

INIT_MMX
SSD 16, 16, mmx
SSD 16,  8, mmx
SSD  8, 16, mmx
SSD  8,  8, mmx
SSD  8,  4, mmx
SSD  4,  8, mmx
SSD  4,  4, mmx
INIT_XMM
SSD 16, 16, sse2slow, 8
SSD 16,  8, sse2slow, 8
SSD  8, 16, sse2slow, 8
SSD  8,  8, sse2slow, 8
SSD  8,  4, sse2slow, 8
%define SSD_CORE SSD_CORE_SSE2
%define JOIN JOIN_SSE2
SSD 16, 16, sse2, 8
SSD 16,  8, sse2, 8
SSD  8, 16, sse2, 8
SSD  8,  8, sse2, 8
SSD  8,  4, sse2, 8
%define SSD_CORE SSD_CORE_SSSE3
%define JOIN JOIN_SSSE3
SSD 16, 16, ssse3, 8
SSD 16,  8, ssse3, 8
SSD  8, 16, ssse3, 8
SSD  8,  8, ssse3, 8
SSD  8,  4, ssse3, 8
INIT_MMX
SSD  4,  8, ssse3
SSD  4,  4, ssse3

;=============================================================================
; variance
;=============================================================================

%macro VAR_START 1
    pxor  m5, m5    ; sum
    pxor  m6, m6    ; sum squared
%if %1
    mova  m7, [pw_00ff GLOBAL]
%else
    pxor  m7, m7    ; zero
%endif
%endmacro

%macro VAR_END 1
    HADDW   m5, m7
    movd   r1d, m5
    imul   r1d, r1d
    HADDD   m6, m1
    shr    r1d, %1
    movd   eax, m6
    sub    eax, r1d  ; sqr - (sum * sum >> shift)
    RET
%endmacro

%macro VAR_CORE 0
    paddw     m5, m0
    paddw     m5, m3
    paddw     m5, m1
    paddw     m5, m4
    pmaddwd   m0, m0
    pmaddwd   m3, m3
    pmaddwd   m1, m1
    pmaddwd   m4, m4
    paddd     m6, m0
    paddd     m6, m3
    paddd     m6, m1
    paddd     m6, m4
%endmacro

%macro VAR_2ROW 2
    mov      r2d, %2
.loop:
    mova      m0, [r0]
    mova      m1, m0
    mova      m3, [r0+%1]
    mova      m4, m3
    punpcklbw m0, m7
    punpckhbw m1, m7
%ifidn %1, r1
    lea       r0, [r0+%1*2]
%else
    add       r0, r1
%endif
    punpcklbw m3, m7
    punpckhbw m4, m7
    dec r2d
    VAR_CORE
    jg .loop
%endmacro

;-----------------------------------------------------------------------------
; int x264_pixel_var_wxh_mmxext( uint8_t *, int )
;-----------------------------------------------------------------------------
INIT_MMX
cglobal x264_pixel_var_16x16_mmxext, 2,3
    VAR_START 0
    VAR_2ROW 8, 16
    VAR_END 8

cglobal x264_pixel_var_8x8_mmxext, 2,3
    VAR_START 0
    VAR_2ROW r1, 4
    VAR_END 6

INIT_XMM
cglobal x264_pixel_var_16x16_sse2, 2,3,8
    VAR_START 1
    mov      r2d, 8
.loop:
    mova      m0, [r0]
    mova      m3, [r0+r1]
    DEINTB    1, 0, 4, 3, 7
    lea       r0, [r0+r1*2]
    VAR_CORE
    dec r2d
    jg .loop
    VAR_END 8

cglobal x264_pixel_var_8x8_sse2, 2,4,8
    VAR_START 1
    mov      r2d, 2
    lea       r3, [r1*3]
.loop:
    movh      m0, [r0]
    movh      m3, [r0+r1]
    movhps    m0, [r0+r1*2]
    movhps    m3, [r0+r3]
    DEINTB    1, 0, 4, 3, 7
    lea       r0, [r0+r1*4]
    VAR_CORE
    dec r2d
    jg .loop
    VAR_END 6


;=============================================================================
; SATD
;=============================================================================

%macro TRANS_SSE2 5-6
; TRANSPOSE2x2
; %1: transpose width (d/q) - use SBUTTERFLY qdq for dq
; %2: ord/unord (for compat with sse4, unused)
; %3/%4: source regs
; %5/%6: tmp regs
%ifidn %1, d
%define mask [mask_10 GLOBAL]
%define shift 16
%elifidn %1, q
%define mask [mask_1100 GLOBAL]
%define shift 32
%endif
%if %0==6 ; less dependency if we have two tmp
    mova   m%5, mask   ; ff00
    mova   m%6, m%4    ; x5x4
    psll%1 m%4, shift  ; x4..
    pand   m%6, m%5    ; x5..
    pandn  m%5, m%3    ; ..x0
    psrl%1 m%3, shift  ; ..x1
    por    m%4, m%5    ; x4x0
    por    m%3, m%6    ; x5x1
%else ; more dependency, one insn less. sometimes faster, sometimes not
    mova   m%5, m%4    ; x5x4
    psll%1 m%4, shift  ; x4..
    pxor   m%4, m%3    ; (x4^x1)x0
    pand   m%4, mask   ; (x4^x1)..
    pxor   m%3, m%4    ; x4x0
    psrl%1 m%4, shift  ; ..(x1^x4)
    pxor   m%5, m%4    ; x5x1
    SWAP   %4, %3, %5
%endif
%endmacro

%define TRANS TRANS_SSE2

%macro TRANS_SSE4 5-6 ; see above
%ifidn %1, d
    mova   m%5, m%3
%ifidn %2, ord
    psrl%1 m%3, 16
%endif
    pblendw m%3, m%4, 10101010b
    psll%1 m%4, 16
%ifidn %2, ord
    pblendw m%4, m%5, 01010101b
%else
    psrl%1 m%5, 16
    por    m%4, m%5
%endif
%elifidn %1, q
    mova   m%5, m%3
    shufps m%3, m%4, 10001000b
    shufps m%5, m%4, 11011101b
    SWAP   %4, %5
%endif
%endmacro

%macro JDUP_SSE2 2
    punpckldq %1, %2
    ; doesn't need to dup. sse2 does things by zero extending to words and full h_2d
%endmacro

%macro JDUP_CONROE 2
    ; join 2x 32 bit and duplicate them
    ; emulating shufps is faster on conroe
    punpcklqdq %1, %2
    movsldup %1, %1
%endmacro

%macro JDUP_PENRYN 2
    ; just use shufps on anything post conroe
    shufps %1, %2, 0
%endmacro

%macro HSUMSUB 5
    pmaddubsw m%2, m%5
    pmaddubsw m%1, m%5
    pmaddubsw m%4, m%5
    pmaddubsw m%3, m%5
%endmacro

%macro DIFF_UNPACK_SSE2 5
    punpcklbw m%1, m%5
    punpcklbw m%2, m%5
    punpcklbw m%3, m%5
    punpcklbw m%4, m%5
    psubw m%1, m%2
    psubw m%3, m%4
%endmacro

%macro DIFF_SUMSUB_SSSE3 5
    HSUMSUB %1, %2, %3, %4, %5
    psubw m%1, m%2
    psubw m%3, m%4
%endmacro

%macro LOAD_DUP_2x4P 4 ; dst, tmp, 2* pointer
    movd %1, %3
    movd %2, %4
    JDUP %1, %2
%endmacro

%macro LOAD_DUP_4x8P_CONROE 8 ; 4*dst, 4*pointer
    movddup m%3, %6
    movddup m%4, %8
    movddup m%1, %5
    movddup m%2, %7
%endmacro

%macro LOAD_DUP_4x8P_PENRYN 8
    ; penryn and nehalem run punpcklqdq and movddup in different units
    movh m%3, %6
    movh m%4, %8
    punpcklqdq m%3, m%3
    movddup m%1, %5
    punpcklqdq m%4, m%4
    movddup m%2, %7
%endmacro

%macro LOAD_SUMSUB_8x2P 9
    LOAD_DUP_4x8P %1, %2, %3, %4, %6, %7, %8, %9
    DIFF_SUMSUB_SSSE3 %1, %3, %2, %4, %5
%endmacro

%macro LOAD_SUMSUB_8x4P_SSSE3 7-10 r0, r2, 0
; 4x dest, 2x tmp, 1x mul, [2* ptr], [increment?]
    LOAD_SUMSUB_8x2P %1, %2, %5, %6, %7, [%8], [%9], [%8+r1], [%9+r3]
    LOAD_SUMSUB_8x2P %3, %4, %5, %6, %7, [%8+2*r1], [%9+2*r3], [%8+r4], [%9+r5]
%if %10
    lea %8, [%8+4*r1]
    lea %9, [%9+4*r3]
%endif
%endmacro

%macro LOAD_SUMSUB_16P_SSSE3 7 ; 2*dst, 2*tmp, mul, 2*ptr
    movddup m%1, [%7]
    movddup m%2, [%7+8]
    mova m%4, [%6]
    movddup m%3, m%4
    punpckhqdq m%4, m%4
    DIFF_SUMSUB_SSSE3 %1, %3, %2, %4, %5
%endmacro

%macro LOAD_SUMSUB_16P_SSE2 7 ; 2*dst, 2*tmp, mask, 2*ptr
    movu  m%4, [%7]
    mova  m%2, [%6]
    DEINTB %1, %2, %3, %4, %5
    psubw m%1, m%3
    psubw m%2, m%4
    SUMSUB_BA m%1, m%2, m%3
%endmacro

%macro LOAD_SUMSUB_16x4P 10-13 r0, r2, none
; 8x dest, 1x tmp, 1x mul, [2* ptr] [2nd tmp]
    LOAD_SUMSUB_16P %1, %5, %2, %3, %10, %11, %12
    LOAD_SUMSUB_16P %2, %6, %3, %4, %10, %11+r1, %12+r3
    LOAD_SUMSUB_16P %3, %7, %4, %9, %10, %11+2*r1, %12+2*r3
    LOAD_SUMSUB_16P %4, %8, %13, %9, %10, %11+r4, %12+r5
%endmacro

; in: r4=3*stride1, r5=3*stride2
; in: %2 = horizontal offset
; in: %3 = whether we need to increment pix1 and pix2
; clobber: m3..m7
; out: %1 = satd
%macro SATD_4x4_MMX 3
    %xdefine %%n n%1
    LOAD_DIFF m4, m3, none, [r0+%2],      [r2+%2]
    LOAD_DIFF m5, m3, none, [r0+r1+%2],   [r2+r3+%2]
    LOAD_DIFF m6, m3, none, [r0+2*r1+%2], [r2+2*r3+%2]
    LOAD_DIFF m7, m3, none, [r0+r4+%2],   [r2+r5+%2]
%if %3
    lea  r0, [r0+4*r1]
    lea  r2, [r2+4*r3]
%endif
    HADAMARD4_2D 4, 5, 6, 7, 3, %%n
    paddw m4, m6
    SWAP %%n, 4
%endmacro

%macro SATD_8x4_SSE 8-9
%ifidn %1, sse2
    HADAMARD4_2D_SSE %2, %3, %4, %5, %6, amax
%else
    HADAMARD4_V m%2, m%3, m%4, m%5, m%6
    ; doing the abs first is a slight advantage
    ABS4 m%2, m%4, m%3, m%5, m%6, m%7
    HADAMARD 1, max, %2, %4, %6, %7
%endif
%ifnidn %9, swap
    paddw m%8, m%2
%else
    SWAP %8, %2
%endif
%ifidn %1, sse2
    paddw m%8, m%4
%else
    HADAMARD 1, max, %3, %5, %6, %7
    paddw m%8, m%3
%endif
%endmacro

%macro SATD_START_MMX 0
    lea  r4, [3*r1] ; 3*stride1
    lea  r5, [3*r3] ; 3*stride2
%endmacro

%macro SATD_END_MMX 0
    pshufw      m1, m0, 01001110b