📄 reduced_mmx.asm

📁 从FFMPEG转换而来的H264解码程序,VC下编译..
💻 ASM
📖 第 1 页 / 共 2 页
字号:
12 下一页
;/*****************************************************************************
; *
; *  XVID MPEG-4 VIDEO CODEC
; *  - Reduced-Resolution utilities -
; *
; *  Copyright(C) 2002 Pascal Massimino <skal@planet-d.net>
; *
; *  XviD 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., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
; *
; * $Id: reduced_mmx.asm,v 1.6 2004/08/29 10:02:38 edgomez Exp $
; *
; *************************************************************************/

BITS 32

%macro cglobal 1
	%ifdef PREFIX
		%ifdef MARK_FUNCS
			global _%1:function %1.endfunc-%1
			%define %1 _%1:function %1.endfunc-%1
		%else
			global _%1
			%define %1 _%1
		%endif
	%else
		%ifdef MARK_FUNCS
			global %1:function %1.endfunc-%1
		%else
			global %1
		%endif
	%endif
%endmacro

;===========================================================================

%ifdef FORMAT_COFF
SECTION .rodata
%else
SECTION .rodata align=16
%endif

align 16
Up31 dw  3, 1, 3, 1
Up13 dw  1, 3, 1, 3
Up93 dw  9, 3, 9, 3
Up39 dw  3, 9, 3, 9
Cst0 dw  0, 0, 0, 0
Cst2 dw  2, 2, 2, 2
Cst3 dw  3, 3, 3, 3
Cst32 dw 32,32,32,32
Cst2000 dw  2, 0, 0, 0
Cst0002 dw  0, 0, 0, 2

Mask_ff dw 0xff,0xff,0xff,0xff

;===========================================================================

SECTION .text

cglobal xvid_Copy_Upsampled_8x8_16To8_mmx
cglobal xvid_Add_Upsampled_8x8_16To8_mmx
cglobal xvid_Copy_Upsampled_8x8_16To8_xmm
cglobal xvid_Add_Upsampled_8x8_16To8_xmm

cglobal xvid_HFilter_31_mmx
cglobal xvid_VFilter_31_x86
cglobal xvid_HFilter_31_x86

cglobal xvid_Filter_18x18_To_8x8_mmx
cglobal xvid_Filter_Diff_18x18_To_8x8_mmx


;//////////////////////////////////////////////////////////////////////
;// 8x8 -> 16x16 upsampling (16b)
;//////////////////////////////////////////////////////////////////////

%macro MUL_PACK 4     ; %1/%2: regs   %3/%4/%5: Up13/Up31
  pmullw %1,  %3 ; [Up13]
  pmullw mm4, %4 ; [Up31]
  pmullw %2,  %3 ; [Up13]
  pmullw mm5, %4 ; [Up31]
  paddsw %1, [Cst2]
  paddsw %2, [Cst2]
  paddsw %1, mm4
  paddsw %2, mm5
%endmacro

    ; MMX-way of reordering columns...

%macro COL03 3    ;%1/%2: regs, %3: row   -output: mm4/mm5
  movq %1, [edx+%3*16+0*2]   ; %1  = 0|1|2|3
  movq %2,[edx+%3*16+1*2]    ; %2  = 1|2|3|4
  movq mm5, %1               ; mm5 = 0|1|2|3
  movq mm4, %1               ; mm4 = 0|1|2|3
  punpckhwd mm5,%2           ; mm5 = 2|3|3|4
  punpcklwd mm4,%2           ; mm4 = 0|1|1|2
  punpcklwd %1,%1            ; %1  = 0|0|1|1
  punpcklwd %2, mm5          ; %2  = 1|2|2|3
  punpcklwd %1, mm4          ; %1  = 0|0|0|1
%endmacro

%macro COL47 3    ;%1-%2: regs, %3: row   -output: mm4/mm5
  movq mm5, [edx+%3*16+4*2]   ; mm5 = 4|5|6|7
  movq %1, [edx+%3*16+3*2]    ; %1  = 3|4|5|6
  movq %2,  mm5               ; %2  = 4|5|6|7
  movq mm4, mm5               ; mm4 = 4|5|6|7
  punpckhwd %2, %2            ; %2  = 6|6|7|7
  punpckhwd mm5, %2           ; mm5 = 6|7|7|7
  movq %2,  %1                ; %2  = 3|4|5|6
  punpcklwd %1, mm4           ; %1  = 3|4|4|5
  punpckhwd %2, mm4           ; %2  = 5|6|6|7
  punpcklwd mm4, %2           ; mm4 = 4|5|5|6
%endmacro

%macro MIX_ROWS 4   ; %1/%2:prev %3/4:cur (preserved)  mm4/mm5: output
  ; we need to perform: (%1,%3) -> (%1 = 3*%1+%3, mm4 = 3*%3+%1), %3 preserved.
  movq mm4, [Cst3]
  movq mm5, [Cst3]
  pmullw mm4, %3
  pmullw mm5, %4
  paddsw mm4, %1
  paddsw mm5, %2
  pmullw %1, [Cst3]
  pmullw %2, [Cst3]
  paddsw %1, %3
  paddsw %2, %4
%endmacro

;===========================================================================
;
; void xvid_Copy_Upsampled_8x8_16To8_mmx(uint8_t *Dst,
;                                        const int16_t *Src, const int BpS);
;
;===========================================================================

  ; Note: we can use ">>2" instead of "/4" here, since we
  ; are (supposed to be) averaging positive values

%macro STORE_1 2
  psraw %1, 2
  psraw %2, 2
  packuswb %1,%2
  movq [ecx], %1
%endmacro

%macro STORE_2 2    ; pack and store (%1,%2) + (mm4,mm5)
  psraw %1, 4
  psraw %2, 4
  psraw mm4, 4
  psraw mm5, 4
  packuswb %1,%2
  packuswb mm4, mm5
  movq [ecx], %1
  movq [ecx+eax], mm4
  lea ecx, [ecx+2*eax]
%endmacro

;//////////////////////////////////////////////////////////////////////

align 16
xvid_Copy_Upsampled_8x8_16To8_mmx:  ; 344c

  mov ecx, [esp+4]  ; Dst
  mov edx, [esp+8]  ; Src
  mov eax, [esp+12] ; BpS

  movq mm6, [Up13]
  movq mm7, [Up31]

  COL03 mm0, mm1, 0
  MUL_PACK mm0,mm1, mm6, mm7
  movq mm4, mm0
  movq mm5, mm1
  STORE_1 mm4, mm5
  add ecx, eax

  COL03 mm2, mm3, 1
  MUL_PACK mm2,mm3, mm6, mm7
  MIX_ROWS mm0, mm1, mm2, mm3
  STORE_2 mm0, mm1

  COL03 mm0, mm1, 2
  MUL_PACK mm0,mm1, mm6, mm7
  MIX_ROWS mm2, mm3, mm0, mm1
  STORE_2 mm2, mm3

  COL03 mm2, mm3, 3
  MUL_PACK mm2,mm3, mm6, mm7
  MIX_ROWS mm0, mm1, mm2, mm3
  STORE_2 mm0, mm1

  COL03 mm0, mm1, 4
  MUL_PACK mm0,mm1, mm6, mm7
  MIX_ROWS mm2, mm3, mm0, mm1
  STORE_2 mm2, mm3

  COL03 mm2, mm3, 5
  MUL_PACK mm2,mm3, mm6, mm7
  MIX_ROWS mm0, mm1, mm2, mm3
  STORE_2 mm0, mm1

  COL03 mm0, mm1, 6
  MUL_PACK mm0,mm1, mm6, mm7
  MIX_ROWS mm2, mm3, mm0, mm1
  STORE_2 mm2, mm3

  COL03 mm2, mm3, 7
  MUL_PACK mm2,mm3, mm6, mm7
  MIX_ROWS mm0, mm1, mm2, mm3
  STORE_2 mm0, mm1

  STORE_1 mm2, mm3

  mov ecx, [esp+4]
  add ecx, 8

  COL47 mm0, mm1, 0
  MUL_PACK mm0,mm1, mm6, mm7
  movq mm4, mm0
  movq mm5, mm1
  STORE_1 mm4, mm5
  add ecx, eax

  COL47 mm2, mm3, 1
  MUL_PACK mm2,mm3, mm6, mm7
  MIX_ROWS mm0, mm1, mm2, mm3
  STORE_2 mm0, mm1

  COL47 mm0, mm1, 2
  MUL_PACK mm0,mm1, mm6, mm7
  MIX_ROWS mm2, mm3, mm0, mm1
  STORE_2 mm2, mm3

  COL47 mm2, mm3, 3
  MUL_PACK mm2,mm3, mm6, mm7
  MIX_ROWS mm0, mm1, mm2, mm3
  STORE_2 mm0, mm1

  COL47 mm0, mm1, 4
  MUL_PACK mm0,mm1, mm6, mm7
  MIX_ROWS mm2, mm3, mm0, mm1
  STORE_2 mm2, mm3

  COL47 mm2, mm3, 5
  MUL_PACK mm2,mm3, mm6, mm7
  MIX_ROWS mm0, mm1, mm2, mm3
  STORE_2 mm0, mm1

  COL47 mm0, mm1, 6
  MUL_PACK mm0,mm1, mm6, mm7
  MIX_ROWS mm2, mm3, mm0, mm1
  STORE_2 mm2, mm3

  COL47 mm2, mm3, 7
  MUL_PACK mm2,mm3, mm6, mm7
  MIX_ROWS mm0, mm1, mm2, mm3
  STORE_2 mm0, mm1

  STORE_1 mm2, mm3

  ret
.endfunc

;===========================================================================
;
; void xvid_Add_Upsampled_8x8_16To8_mmx(uint8_t *Dst,
;                                       const int16_t *Src, const int BpS);
;
;===========================================================================

    ; Note: grrr... the 'pcmpgtw' stuff are the "/4" and "/16" operators
    ; implemented with ">>2" and ">>4" using:
    ;       x/4  = ( (x-(x<0))>>2 ) + (x<0)
    ;       x/16 = ( (x-(x<0))>>4 ) + (x<0)

%macro STORE_ADD_1 2
    ; We substract the rounder '2' for corner pixels,
    ; since when 'x' is negative, (x*4 + 2)/4 is *not*
    ; equal to 'x'. In fact, the correct relation is:
    ;         (x*4 + 2)/4 = x - (x<0)
    ; So, better revert to (x*4)/4 = x.

  psubsw %1, [Cst2000]
  psubsw %2, [Cst0002]
  pxor mm6, mm6
  pxor mm7, mm7
  pcmpgtw mm6, %1
  pcmpgtw mm7, %2
  paddsw %1, mm6
  paddsw %2, mm7
  psraw %1, 2
  psraw %2, 2
  psubsw %1, mm6
  psubsw %2, mm7

    ; mix with destination [ecx]
  movq mm6, [ecx]
  movq mm7, [ecx]
  punpcklbw mm6, [Cst0]
  punpckhbw mm7, [Cst0]
  paddsw %1, mm6
  paddsw %2, mm7
  packuswb %1,%2
  movq [ecx], %1
%endmacro

%macro STORE_ADD_2 2
  pxor mm6, mm6
  pxor mm7, mm7
  pcmpgtw mm6, %1
  pcmpgtw mm7, %2
  paddsw %1, mm6
  paddsw %2, mm7
  psraw %1, 4
  psraw %2, 4
  psubsw %1, mm6
  psubsw %2, mm7

  pxor mm6, mm6
  pxor mm7, mm7
  pcmpgtw mm6, mm4
  pcmpgtw mm7, mm5
  paddsw mm4, mm6
  paddsw mm5, mm7
  psraw mm4, 4
  psraw mm5, 4
  psubsw mm4, mm6
  psubsw mm5, mm7

    ; mix with destination
  movq mm6, [ecx]
  movq mm7, [ecx]
  punpcklbw mm6, [Cst0]
  punpckhbw mm7, [Cst0]
  paddsw %1, mm6
  paddsw %2, mm7

  movq mm6, [ecx+eax]
  movq mm7, [ecx+eax]

  punpcklbw mm6, [Cst0]
  punpckhbw mm7, [Cst0]
  paddsw mm4, mm6
  paddsw mm5, mm7

  packuswb %1,%2
  packuswb mm4, mm5

  movq [ecx], %1
  movq [ecx+eax], mm4

  lea ecx, [ecx+2*eax]
%endmacro

;//////////////////////////////////////////////////////////////////////

align 16
xvid_Add_Upsampled_8x8_16To8_mmx:  ; 579c

  mov ecx, [esp+4]  ; Dst
  mov edx, [esp+8]  ; Src
  mov eax, [esp+12] ; BpS

  COL03 mm0, mm1, 0
  MUL_PACK mm0,mm1, [Up13], [Up31]
  movq mm4, mm0
  movq mm5, mm1
  STORE_ADD_1 mm4, mm5
  add ecx, eax

  COL03 mm2, mm3, 1
  MUL_PACK mm2,mm3, [Up13], [Up31]
  MIX_ROWS mm0, mm1, mm2, mm3
  STORE_ADD_2 mm0, mm1

  COL03 mm0, mm1, 2
  MUL_PACK mm0,mm1, [Up13], [Up31]
  MIX_ROWS mm2, mm3, mm0, mm1
  STORE_ADD_2 mm2, mm3

  COL03 mm2, mm3, 3
  MUL_PACK mm2,mm3, [Up13], [Up31]
  MIX_ROWS mm0, mm1, mm2, mm3
  STORE_ADD_2 mm0, mm1

  COL03 mm0, mm1, 4
  MUL_PACK mm0,mm1, [Up13], [Up31]
  MIX_ROWS mm2, mm3, mm0, mm1
  STORE_ADD_2 mm2, mm3

  COL03 mm2, mm3, 5
  MUL_PACK mm2,mm3, [Up13], [Up31]
  MIX_ROWS mm0, mm1, mm2, mm3
  STORE_ADD_2 mm0, mm1

  COL03 mm0, mm1, 6
  MUL_PACK mm0,mm1, [Up13], [Up31]
  MIX_ROWS mm2, mm3, mm0, mm1
  STORE_ADD_2 mm2, mm3

  COL03 mm2, mm3, 7
  MUL_PACK mm2,mm3, [Up13], [Up31]
  MIX_ROWS mm0, mm1, mm2, mm3
  STORE_ADD_2 mm0, mm1

  STORE_ADD_1 mm2, mm3


  mov ecx, [esp+4]
  add ecx, 8

  COL47 mm0, mm1, 0
  MUL_PACK mm0,mm1, [Up13], [Up31]
  movq mm4, mm0
  movq mm5, mm1
  STORE_ADD_1 mm4, mm5
  add ecx, eax

  COL47 mm2, mm3, 1
  MUL_PACK mm2,mm3, [Up13], [Up31]
  MIX_ROWS mm0, mm1, mm2, mm3
  STORE_ADD_2 mm0, mm1

  COL47 mm0, mm1, 2
  MUL_PACK mm0,mm1, [Up13], [Up31]
  MIX_ROWS mm2, mm3, mm0, mm1
  STORE_ADD_2 mm2, mm3

  COL47 mm2, mm3, 3
  MUL_PACK mm2,mm3, [Up13], [Up31]
  MIX_ROWS mm0, mm1, mm2, mm3
  STORE_ADD_2 mm0, mm1

  COL47 mm0, mm1, 4
  MUL_PACK mm0,mm1, [Up13], [Up31]
  MIX_ROWS mm2, mm3, mm0, mm1
  STORE_ADD_2 mm2, mm3

  COL47 mm2, mm3, 5
  MUL_PACK mm2,mm3, [Up13], [Up31]
  MIX_ROWS mm0, mm1, mm2, mm3
  STORE_ADD_2 mm0, mm1

  COL47 mm0, mm1, 6
  MUL_PACK mm0,mm1, [Up13], [Up31]
  MIX_ROWS mm2, mm3, mm0, mm1
  STORE_ADD_2 mm2, mm3

  COL47 mm2, mm3, 7
  MUL_PACK mm2,mm3, [Up13], [Up31]
  MIX_ROWS mm0, mm1, mm2, mm3
  STORE_ADD_2 mm0, mm1

  STORE_ADD_1 mm2, mm3

  ret
.endfunc

;===========================================================================
;
; void xvid_Copy_Upsampled_8x8_16To8_xmm(uint8_t *Dst,
;                                        const int16_t *Src, const int BpS);
;
;===========================================================================

  ; xmm version can take (little) advantage of 'pshufw'

%macro COL03_SSE 3    ;%1/%2: regs, %3: row   -trashes mm4/mm5
  movq %2, [edx+%3*16+0*2]               ; <- 0|1|2|3
  pshufw %1,  %2,  (0+0*4+0*16+1*64)     ; %1 = 0|0|0|1
  pshufw mm4, %2,  (0+1*4+1*16+2*64)     ; mm4= 0|1|1|2
  pshufw %2,  %2,  (1+2*4+2*16+3*64)     ; %2 = 1|2|2|3
  pshufw mm5, [edx+%3*16+2*2],  (0+1*4+1*16+2*64) ; mm5 = 2|3|3|4
%endmacro

%macro COL47_SSE 3    ;%1-%2: regs, %3: row   -trashes mm4/mm5
  pshufw %1, [edx+%3*16+2*2],  (1+2*4+2*16+3*64) ; 3|4|4|5
  movq mm5, [edx+%3*16+2*4]                      ; <- 4|5|6|7
  pshufw mm4, mm5,  (0+1*4+1*16+2*64)            ; 4|5|5|6
  pshufw %2,  mm5,  (1+2*4+2*16+3*64)            ; 5|6|6|7
  pshufw mm5, mm5,  (2+3*4+3*16+3*64)            ; 6|7|7|7
%endmacro


;//////////////////////////////////////////////////////////////////////

align 16
xvid_Copy_Upsampled_8x8_16To8_xmm:  ; 315c

  mov ecx, [esp+4]  ; Dst
  mov edx, [esp+8]  ; Src
12 下一页
⌨️ 快捷键说明

复制代码 Ctrl + C
搜索代码 Ctrl + F
全屏模式 F11
切换主题 Ctrl + Shift + D
显示快捷键 ?
增大字号 Ctrl + =
减小字号 Ctrl + -