1 |
;/************************************************************************** |
;/**************************************************************************** |
2 |
; * |
; * |
3 |
; * XVID MPEG-4 VIDEO CODEC |
; * XVID MPEG-4 VIDEO CODEC |
4 |
; * 3dnow (*but without xmm*) sum of absolute difference |
; * - 3DNow sad operators w/o XMM instructions - |
5 |
; * |
; * |
6 |
; * This program is free software; you can redistribute it and/or modify |
; * Copyright(C) 2002 Peter ross <pross@xvid.org> |
7 |
; * it under the terms of the GNU General Public License as published by |
; * |
8 |
|
; * This program is free software; you can redistribute it and/or modify it |
9 |
|
; * under the terms of the GNU General Public License as published by |
10 |
; * the Free Software Foundation; either version 2 of the License, or |
; * the Free Software Foundation; either version 2 of the License, or |
11 |
; * (at your option) any later version. |
; * (at your option) any later version. |
12 |
; * |
; * |
17 |
; * |
; * |
18 |
; * You should have received a copy of the GNU General Public License |
; * You should have received a copy of the GNU General Public License |
19 |
; * along with this program; if not, write to the Free Software |
; * along with this program; if not, write to the Free Software |
20 |
; * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
|
; * |
|
|
; *************************************************************************/ |
|
|
|
|
|
;/************************************************************************** |
|
|
; * |
|
|
; * History: |
|
21 |
; * |
; * |
22 |
; * 23.07.2002 sad[16,8]bi_3dn; <pross@xvid.org> |
; * $Id: sad_3dn.asm,v 1.6 2004-03-22 22:36:24 edgomez Exp $ |
23 |
; * |
; * |
24 |
; *************************************************************************/ |
; ***************************************************************************/ |
25 |
|
|
26 |
bits 32 |
BITS 32 |
27 |
|
|
28 |
%macro cglobal 1 |
%macro cglobal 1 |
29 |
%ifdef PREFIX |
%ifdef PREFIX |
34 |
%endif |
%endif |
35 |
%endmacro |
%endmacro |
36 |
|
|
37 |
section .data |
;============================================================================= |
38 |
|
; Read only data |
39 |
|
;============================================================================= |
40 |
|
|
41 |
align 16 |
%ifdef FORMAT_COFF |
42 |
mmx_one times 4 dw 1 |
SECTION .rodata data |
43 |
|
%else |
44 |
section .text |
SECTION .rodata data align=16 |
45 |
|
%endif |
|
cglobal sad16bi_3dn |
|
|
cglobal sad8bi_3dn |
|
|
|
|
|
;=========================================================================== |
|
|
; |
|
|
; uint32_t sad16bi_3dn(const uint8_t * const cur, |
|
|
; const uint8_t * const ref1, |
|
|
; const uint8_t * const ref2, |
|
|
; const uint32_t stride); |
|
|
; |
|
|
;=========================================================================== |
|
46 |
|
|
47 |
|
ALIGN 16 |
48 |
|
mmx_one: |
49 |
|
times 4 dw 1 |
50 |
|
|
51 |
|
;============================================================================= |
52 |
|
; Helper macros |
53 |
|
;============================================================================= |
54 |
%macro SADBI_16x16_3DN 0 |
%macro SADBI_16x16_3DN 0 |
55 |
movq mm0, [eax] ; src |
movq mm0, [eax] ; src |
56 |
movq mm2, [eax+8] |
movq mm2, [eax+8] |
87 |
paddusw mm6,mm2 |
paddusw mm6,mm2 |
88 |
%endmacro |
%endmacro |
89 |
|
|
90 |
align 16 |
%macro SADBI_8x8_3DN 0 |
91 |
|
movq mm0, [eax] ; src |
92 |
|
movq mm2, [eax+ecx] |
93 |
|
|
94 |
|
movq mm1, [edx] ; ref1 |
95 |
|
movq mm3, [edx+ecx] |
96 |
|
pavgusb mm1, [ebx] ; ref2 |
97 |
|
lea edx, [edx+2*ecx] |
98 |
|
pavgusb mm3, [ebx+ecx] |
99 |
|
lea ebx, [ebx+2*ecx] |
100 |
|
|
101 |
|
movq mm4, mm0 |
102 |
|
lea eax, [eax+2*ecx] |
103 |
|
psubusb mm0, mm1 |
104 |
|
movq mm5, mm2 |
105 |
|
psubusb mm2, mm3 |
106 |
|
|
107 |
|
psubusb mm1, mm4 |
108 |
|
por mm0, mm1 |
109 |
|
psubusb mm3, mm5 |
110 |
|
por mm2, mm3 |
111 |
|
|
112 |
|
movq mm1, mm0 |
113 |
|
movq mm3, mm2 |
114 |
|
|
115 |
|
punpcklbw mm0,mm7 |
116 |
|
punpckhbw mm1,mm7 |
117 |
|
punpcklbw mm2,mm7 |
118 |
|
punpckhbw mm3,mm7 |
119 |
|
|
120 |
|
paddusw mm0,mm1 |
121 |
|
paddusw mm2,mm3 |
122 |
|
paddusw mm6,mm0 |
123 |
|
paddusw mm6,mm2 |
124 |
|
%endmacro |
125 |
|
|
126 |
|
;============================================================================= |
127 |
|
; Code |
128 |
|
;============================================================================= |
129 |
|
|
130 |
|
SECTION .text |
131 |
|
|
132 |
|
cglobal sad16bi_3dn |
133 |
|
cglobal sad8bi_3dn |
134 |
|
|
135 |
|
;----------------------------------------------------------------------------- |
136 |
|
; |
137 |
|
; uint32_t sad16bi_3dn(const uint8_t * const cur, |
138 |
|
; const uint8_t * const ref1, |
139 |
|
; const uint8_t * const ref2, |
140 |
|
; const uint32_t stride); |
141 |
|
; |
142 |
|
;----------------------------------------------------------------------------- |
143 |
|
|
144 |
|
ALIGN 16 |
145 |
sad16bi_3dn: |
sad16bi_3dn: |
146 |
push ebx |
push ebx |
147 |
mov eax, [esp+4+ 4] ; Src |
mov eax, [esp+4+ 4] ; Src |
181 |
|
|
182 |
ret |
ret |
183 |
|
|
184 |
|
;----------------------------------------------------------------------------- |
|
|
|
|
;=========================================================================== |
|
185 |
; |
; |
186 |
; uint32_t sad8bi_3dn(const uint8_t * const cur, |
; uint32_t sad8bi_3dn(const uint8_t * const cur, |
187 |
; const uint8_t * const ref1, |
; const uint8_t * const ref1, |
188 |
; const uint8_t * const ref2, |
; const uint8_t * const ref2, |
189 |
; const uint32_t stride); |
; const uint32_t stride); |
190 |
; |
; |
191 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
|
|
|
%macro SADBI_8x8_3DN 0 |
|
|
movq mm0, [eax] ; src |
|
|
movq mm2, [eax+ecx] |
|
|
|
|
|
movq mm1, [edx] ; ref1 |
|
|
movq mm3, [edx+ecx] |
|
|
pavgusb mm1, [ebx] ; ref2 |
|
|
lea edx,[edx+2*ecx] |
|
|
pavgusb mm3, [ebx+ecx] |
|
|
lea ebx,[ebx+2*ecx] |
|
|
|
|
|
movq mm4, mm0 |
|
|
lea eax,[eax+2*ecx] |
|
|
psubusb mm0, mm1 |
|
|
movq mm5, mm2 |
|
|
psubusb mm2, mm3 |
|
|
|
|
|
psubusb mm1, mm4 |
|
|
por mm0, mm1 |
|
|
psubusb mm3, mm5 |
|
|
por mm2, mm3 |
|
|
|
|
|
movq mm1,mm0 |
|
|
movq mm3,mm2 |
|
|
|
|
|
punpcklbw mm0,mm7 |
|
|
punpckhbw mm1,mm7 |
|
|
punpcklbw mm2,mm7 |
|
|
punpckhbw mm3,mm7 |
|
|
|
|
|
paddusw mm0,mm1 |
|
|
paddusw mm2,mm3 |
|
|
paddusw mm6,mm0 |
|
|
paddusw mm6,mm2 |
|
|
%endmacro |
|
192 |
|
|
193 |
align 16 |
ALIGN 16 |
194 |
sad8bi_3dn: |
sad8bi_3dn: |
195 |
push ebx |
push ebx |
196 |
mov eax, [esp+4+ 4] ; Src |
mov eax, [esp+4+ 4] ; Src |