Parent Directory | Revision Log
Revision 1.4 - (view) (download)
1 : | chl | 1.2 | ;/***************************************************************************** |
2 : | suxen_drol | 1.1 | ; * |
3 : | chl | 1.2 | ; * XVID MPEG-4 VIDEO CODEC |
4 : | ; * 3dnow (*but without xmm*) sum of absolute difference | ||
5 : | suxen_drol | 1.1 | ; * |
6 : | chl | 1.2 | ; * Copyright(C) 2002 Peter Ross <pross@xvid.org> |
7 : | suxen_drol | 1.1 | ; * |
8 : | edgomez | 1.4 | ; * This file is part of XviD, a free MPEG-4 video encoder/decoder |
9 : | chl | 1.2 | ; * |
10 : | edgomez | 1.4 | ; * XviD is free software; you can redistribute it and/or modify it |
11 : | ; * under the terms of the GNU General Public License as published by | ||
12 : | chl | 1.2 | ; * the Free Software Foundation; either version 2 of the License, or |
13 : | ; * (at your option) any later version. | ||
14 : | ; * | ||
15 : | ; * This program is distributed in the hope that it will be useful, | ||
16 : | ; * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
17 : | ; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
18 : | ; * GNU General Public License for more details. | ||
19 : | ; * | ||
20 : | ; * You should have received a copy of the GNU General Public License | ||
21 : | ; * along with this program; if not, write to the Free Software | ||
22 : | ; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
23 : | edgomez | 1.4 | ; * |
24 : | ; * Under section 8 of the GNU General Public License, the copyright | ||
25 : | ; * holders of XVID explicitly forbid distribution in the following | ||
26 : | ; * countries: | ||
27 : | ; * | ||
28 : | ; * - Japan | ||
29 : | ; * - United States of America | ||
30 : | ; * | ||
31 : | ; * Linking XviD statically or dynamically with other modules is making a | ||
32 : | ; * combined work based on XviD. Thus, the terms and conditions of the | ||
33 : | ; * GNU General Public License cover the whole combination. | ||
34 : | ; * | ||
35 : | ; * As a special exception, the copyright holders of XviD give you | ||
36 : | ; * permission to link XviD with independent modules that communicate with | ||
37 : | ; * XviD solely through the VFW1.1 and DShow interfaces, regardless of the | ||
38 : | ; * license terms of these independent modules, and to copy and distribute | ||
39 : | ; * the resulting combined work under terms of your choice, provided that | ||
40 : | ; * every copy of the combined work is accompanied by a complete copy of | ||
41 : | ; * the source code of XviD (the version of XviD used to produce the | ||
42 : | ; * combined work), being distributed under the terms of the GNU General | ||
43 : | ; * Public License plus this exception. An independent module is a module | ||
44 : | ; * which is not derived from or based on XviD. | ||
45 : | ; * | ||
46 : | ; * Note that people who make modified versions of XviD are not obligated | ||
47 : | ; * to grant this special exception for their modified versions; it is | ||
48 : | ; * their choice whether to do so. The GNU General Public License gives | ||
49 : | ; * permission to release a modified version without this exception; this | ||
50 : | ; * exception also makes it possible to release a modified version which | ||
51 : | ; * carries forward this exception. | ||
52 : | ; * | ||
53 : | ; * $Id$ | ||
54 : | suxen_drol | 1.1 | ; * |
55 : | chl | 1.3 | ; ****************************************************************************/ |
56 : | suxen_drol | 1.1 | |
57 : | bits 32 | ||
58 : | |||
59 : | %macro cglobal 1 | ||
60 : | %ifdef PREFIX | ||
61 : | global _%1 | ||
62 : | %define %1 _%1 | ||
63 : | %else | ||
64 : | global %1 | ||
65 : | %endif | ||
66 : | %endmacro | ||
67 : | |||
68 : | section .data | ||
69 : | |||
70 : | align 16 | ||
71 : | mmx_one times 4 dw 1 | ||
72 : | |||
73 : | section .text | ||
74 : | |||
75 : | cglobal sad16bi_3dn | ||
76 : | cglobal sad8bi_3dn | ||
77 : | |||
78 : | ;=========================================================================== | ||
79 : | ; | ||
80 : | ; uint32_t sad16bi_3dn(const uint8_t * const cur, | ||
81 : | ; const uint8_t * const ref1, | ||
82 : | ; const uint8_t * const ref2, | ||
83 : | ; const uint32_t stride); | ||
84 : | ; | ||
85 : | ;=========================================================================== | ||
86 : | |||
87 : | %macro SADBI_16x16_3DN 0 | ||
88 : | movq mm0, [eax] ; src | ||
89 : | movq mm2, [eax+8] | ||
90 : | |||
91 : | movq mm1, [edx] ; ref1 | ||
92 : | movq mm3, [edx+8] | ||
93 : | pavgusb mm1, [ebx] ; ref2 | ||
94 : | lea edx,[edx+ecx] | ||
95 : | pavgusb mm3, [ebx+8] | ||
96 : | lea ebx,[ebx+ecx] | ||
97 : | |||
98 : | movq mm4, mm0 | ||
99 : | lea eax,[eax+ecx] | ||
100 : | psubusb mm0, mm1 | ||
101 : | movq mm5, mm2 | ||
102 : | psubusb mm2, mm3 | ||
103 : | |||
104 : | psubusb mm1, mm4 | ||
105 : | por mm0, mm1 | ||
106 : | psubusb mm3, mm5 | ||
107 : | por mm2, mm3 | ||
108 : | |||
109 : | movq mm1,mm0 | ||
110 : | movq mm3,mm2 | ||
111 : | |||
112 : | punpcklbw mm0,mm7 | ||
113 : | punpckhbw mm1,mm7 | ||
114 : | punpcklbw mm2,mm7 | ||
115 : | punpckhbw mm3,mm7 | ||
116 : | |||
117 : | paddusw mm0,mm1 | ||
118 : | paddusw mm2,mm3 | ||
119 : | paddusw mm6,mm0 | ||
120 : | paddusw mm6,mm2 | ||
121 : | %endmacro | ||
122 : | |||
123 : | align 16 | ||
124 : | sad16bi_3dn: | ||
125 : | push ebx | ||
126 : | mov eax, [esp+4+ 4] ; Src | ||
127 : | mov edx, [esp+4+ 8] ; Ref1 | ||
128 : | mov ebx, [esp+4+12] ; Ref2 | ||
129 : | mov ecx, [esp+4+16] ; Stride | ||
130 : | |||
131 : | pxor mm6, mm6 ; accum2 | ||
132 : | pxor mm7, mm7 | ||
133 : | .Loop | ||
134 : | SADBI_16x16_3DN | ||
135 : | SADBI_16x16_3DN | ||
136 : | SADBI_16x16_3DN | ||
137 : | SADBI_16x16_3DN | ||
138 : | SADBI_16x16_3DN | ||
139 : | SADBI_16x16_3DN | ||
140 : | SADBI_16x16_3DN | ||
141 : | SADBI_16x16_3DN | ||
142 : | |||
143 : | SADBI_16x16_3DN | ||
144 : | SADBI_16x16_3DN | ||
145 : | SADBI_16x16_3DN | ||
146 : | SADBI_16x16_3DN | ||
147 : | SADBI_16x16_3DN | ||
148 : | SADBI_16x16_3DN | ||
149 : | SADBI_16x16_3DN | ||
150 : | SADBI_16x16_3DN | ||
151 : | |||
152 : | pmaddwd mm6, [mmx_one] ; collapse | ||
153 : | movq mm7, mm6 | ||
154 : | psrlq mm7, 32 | ||
155 : | paddd mm6, mm7 | ||
156 : | |||
157 : | movd eax, mm6 | ||
158 : | |||
159 : | pop ebx | ||
160 : | |||
161 : | ret | ||
162 : | |||
163 : | |||
164 : | |||
165 : | ;=========================================================================== | ||
166 : | ; | ||
167 : | ; uint32_t sad8bi_3dn(const uint8_t * const cur, | ||
168 : | ; const uint8_t * const ref1, | ||
169 : | ; const uint8_t * const ref2, | ||
170 : | ; const uint32_t stride); | ||
171 : | ; | ||
172 : | ;=========================================================================== | ||
173 : | |||
174 : | %macro SADBI_8x8_3DN 0 | ||
175 : | movq mm0, [eax] ; src | ||
176 : | movq mm2, [eax+ecx] | ||
177 : | |||
178 : | movq mm1, [edx] ; ref1 | ||
179 : | movq mm3, [edx+ecx] | ||
180 : | pavgusb mm1, [ebx] ; ref2 | ||
181 : | lea edx,[edx+2*ecx] | ||
182 : | pavgusb mm3, [ebx+ecx] | ||
183 : | lea ebx,[ebx+2*ecx] | ||
184 : | |||
185 : | movq mm4, mm0 | ||
186 : | lea eax,[eax+2*ecx] | ||
187 : | psubusb mm0, mm1 | ||
188 : | movq mm5, mm2 | ||
189 : | psubusb mm2, mm3 | ||
190 : | |||
191 : | psubusb mm1, mm4 | ||
192 : | por mm0, mm1 | ||
193 : | psubusb mm3, mm5 | ||
194 : | por mm2, mm3 | ||
195 : | |||
196 : | movq mm1,mm0 | ||
197 : | movq mm3,mm2 | ||
198 : | |||
199 : | punpcklbw mm0,mm7 | ||
200 : | punpckhbw mm1,mm7 | ||
201 : | punpcklbw mm2,mm7 | ||
202 : | punpckhbw mm3,mm7 | ||
203 : | |||
204 : | paddusw mm0,mm1 | ||
205 : | paddusw mm2,mm3 | ||
206 : | paddusw mm6,mm0 | ||
207 : | paddusw mm6,mm2 | ||
208 : | %endmacro | ||
209 : | |||
210 : | align 16 | ||
211 : | sad8bi_3dn: | ||
212 : | push ebx | ||
213 : | mov eax, [esp+4+ 4] ; Src | ||
214 : | mov edx, [esp+4+ 8] ; Ref1 | ||
215 : | mov ebx, [esp+4+12] ; Ref2 | ||
216 : | mov ecx, [esp+4+16] ; Stride | ||
217 : | |||
218 : | pxor mm6, mm6 ; accum2 | ||
219 : | pxor mm7, mm7 | ||
220 : | .Loop | ||
221 : | SADBI_8x8_3DN | ||
222 : | SADBI_8x8_3DN | ||
223 : | SADBI_8x8_3DN | ||
224 : | SADBI_8x8_3DN | ||
225 : | |||
226 : | pmaddwd mm6, [mmx_one] ; collapse | ||
227 : | movq mm7, mm6 | ||
228 : | psrlq mm7, 32 | ||
229 : | paddd mm6, mm7 | ||
230 : | |||
231 : | movd eax, mm6 | ||
232 : | |||
233 : | pop ebx | ||
234 : | |||
235 : | ret |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |