Mercurial > libavcodec.hg
annotate snow.c @ 3777:20545fbb6f7c libavcodec
add some #ifdef CONFIG_ENCODERS/DECODERS
author | mru |
---|---|
date | Wed, 27 Sep 2006 19:54:07 +0000 |
parents | acf9ca729bd2 |
children | 3a151ccc6ed7 |
rev | line source |
---|---|
2138 | 1 /* |
2 * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at> | |
3 * | |
4 * This library is free software; you can redistribute it and/or | |
5 * modify it under the terms of the GNU Lesser General Public | |
6 * License as published by the Free Software Foundation; either | |
7 * version 2 of the License, or (at your option) any later version. | |
8 * | |
9 * This library is distributed in the hope that it will be useful, | |
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
12 * Lesser General Public License for more details. | |
13 * | |
14 * You should have received a copy of the GNU Lesser General Public | |
15 * License along with this library; if not, write to the Free Software | |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
3035
diff
changeset
|
16 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
2138 | 17 */ |
18 | |
19 #include "avcodec.h" | |
20 #include "common.h" | |
21 #include "dsputil.h" | |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
22 #include "snow.h" |
2335 | 23 |
24 #include "rangecoder.h" | |
2138 | 25 |
26 #include "mpegvideo.h" | |
27 | |
28 #undef NDEBUG | |
29 #include <assert.h> | |
30 | |
31 static const int8_t quant3[256]={ | |
32 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
33 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
34 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
35 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
36 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
37 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
38 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
39 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
40 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, | |
41 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, | |
42 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, | |
43 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, | |
44 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, | |
45 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, | |
46 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, | |
47 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, 0, | |
48 }; | |
49 static const int8_t quant3b[256]={ | |
50 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
51 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
52 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
53 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
54 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
55 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
56 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
57 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
58 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, | |
59 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, | |
60 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, | |
61 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, | |
62 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, | |
63 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, | |
64 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, | |
65 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, | |
66 }; | |
2596 | 67 static const int8_t quant3bA[256]={ |
68 0, 0, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
69 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
70 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
71 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
72 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
73 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
74 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
75 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
76 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
77 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
78 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
79 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
80 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
81 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
82 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
83 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
84 }; | |
2138 | 85 static const int8_t quant5[256]={ |
86 0, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | |
87 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | |
88 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | |
89 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | |
90 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | |
91 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | |
92 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | |
93 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | |
94 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2, | |
95 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2, | |
96 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2, | |
97 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2, | |
98 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2, | |
99 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2, | |
100 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2, | |
101 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-1,-1,-1, | |
102 }; | |
103 static const int8_t quant7[256]={ | |
104 0, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | |
105 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | |
106 2, 2, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, | |
107 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, | |
108 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, | |
109 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, | |
110 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, | |
111 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, | |
112 -3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3, | |
113 -3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3, | |
114 -3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3, | |
115 -3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3, | |
116 -3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3, | |
117 -3,-3,-3,-3,-3,-3,-3,-3,-3,-2,-2,-2,-2,-2,-2,-2, | |
118 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2, | |
119 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-1,-1, | |
120 }; | |
121 static const int8_t quant9[256]={ | |
122 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3, | |
123 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, | |
124 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, | |
125 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, | |
126 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, | |
127 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, | |
128 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, | |
129 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, | |
130 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4, | |
131 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4, | |
132 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4, | |
133 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4, | |
134 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4, | |
135 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4, | |
136 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-3,-3,-3,-3, | |
137 -3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-2,-2,-2,-2,-1,-1, | |
138 }; | |
139 static const int8_t quant11[256]={ | |
140 0, 1, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, | |
141 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, | |
142 4, 4, 4, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, | |
143 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, | |
144 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, | |
145 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, | |
146 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, | |
147 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, | |
148 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5, | |
149 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5, | |
150 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5, | |
151 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5, | |
152 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5, | |
153 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-4,-4, | |
154 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4, | |
155 -4,-4,-4,-4,-4,-3,-3,-3,-3,-3,-3,-3,-2,-2,-2,-1, | |
156 }; | |
157 static const int8_t quant13[256]={ | |
158 0, 1, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, | |
159 4, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, | |
160 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, | |
161 5, 5, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, | |
162 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, | |
163 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, | |
164 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, | |
165 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, | |
166 -6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6, | |
167 -6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6, | |
168 -6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6, | |
169 -6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6, | |
170 -6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-5, | |
171 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5, | |
172 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5, | |
173 -4,-4,-4,-4,-4,-4,-4,-4,-4,-3,-3,-3,-3,-2,-2,-1, | |
174 }; | |
175 | |
176 #if 0 //64*cubic | |
177 static const uint8_t obmc32[1024]={ | |
178 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, | |
179 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, | |
180 0, 0, 0, 1, 1, 1, 1, 2, 2, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, 3, 3, 3, 2, 2, 1, 1, 1, 1, 0, 0, 0, | |
181 0, 0, 1, 1, 2, 2, 3, 4, 4, 5, 6, 6, 7, 7, 8, 8, 8, 8, 7, 7, 6, 6, 5, 4, 4, 3, 2, 2, 1, 1, 0, 0, | |
182 0, 0, 1, 2, 2, 3, 4, 6, 7, 8, 9,10,11,12,12,12,12,12,12,11,10, 9, 8, 7, 6, 4, 3, 2, 2, 1, 0, 0, | |
183 0, 1, 1, 2, 3, 5, 6, 8,10,11,13,14,15,16,17,18,18,17,16,15,14,13,11,10, 8, 6, 5, 3, 2, 1, 1, 0, | |
184 0, 1, 1, 3, 4, 6, 8,10,13,15,17,19,20,22,22,23,23,22,22,20,19,17,15,13,10, 8, 6, 4, 3, 1, 1, 0, | |
185 0, 1, 2, 4, 6, 8,10,13,16,19,21,23,25,27,28,29,29,28,27,25,23,21,19,16,13,10, 8, 6, 4, 2, 1, 0, | |
186 0, 1, 2, 4, 7,10,13,16,19,22,25,28,31,33,34,35,35,34,33,31,28,25,22,19,16,13,10, 7, 4, 2, 1, 0, | |
187 0, 1, 3, 5, 8,11,15,19,22,26,30,33,36,38,40,41,41,40,38,36,33,30,26,22,19,15,11, 8, 5, 3, 1, 0, | |
188 0, 1, 3, 6, 9,12,17,21,25,30,34,38,41,44,45,46,46,45,44,41,38,34,30,25,21,17,12, 9, 6, 3, 1, 0, | |
189 0, 1, 3, 6,10,14,19,23,28,33,38,42,45,48,51,52,52,51,48,45,42,38,33,28,23,19,14,10, 6, 3, 1, 0, | |
190 0, 1, 4, 7,11,15,20,25,31,36,41,45,49,52,55,56,56,55,52,49,45,41,36,31,25,20,15,11, 7, 4, 1, 0, | |
191 0, 2, 4, 7,12,16,22,27,33,38,44,48,52,56,58,60,60,58,56,52,48,44,38,33,27,22,16,12, 7, 4, 2, 0, | |
192 0, 1, 4, 8,12,17,22,28,34,40,45,51,55,58,61,62,62,61,58,55,51,45,40,34,28,22,17,12, 8, 4, 1, 0, | |
193 0, 2, 4, 8,12,18,23,29,35,41,46,52,56,60,62,64,64,62,60,56,52,46,41,35,29,23,18,12, 8, 4, 2, 0, | |
194 0, 2, 4, 8,12,18,23,29,35,41,46,52,56,60,62,64,64,62,60,56,52,46,41,35,29,23,18,12, 8, 4, 2, 0, | |
195 0, 1, 4, 8,12,17,22,28,34,40,45,51,55,58,61,62,62,61,58,55,51,45,40,34,28,22,17,12, 8, 4, 1, 0, | |
196 0, 2, 4, 7,12,16,22,27,33,38,44,48,52,56,58,60,60,58,56,52,48,44,38,33,27,22,16,12, 7, 4, 2, 0, | |
197 0, 1, 4, 7,11,15,20,25,31,36,41,45,49,52,55,56,56,55,52,49,45,41,36,31,25,20,15,11, 7, 4, 1, 0, | |
198 0, 1, 3, 6,10,14,19,23,28,33,38,42,45,48,51,52,52,51,48,45,42,38,33,28,23,19,14,10, 6, 3, 1, 0, | |
199 0, 1, 3, 6, 9,12,17,21,25,30,34,38,41,44,45,46,46,45,44,41,38,34,30,25,21,17,12, 9, 6, 3, 1, 0, | |
200 0, 1, 3, 5, 8,11,15,19,22,26,30,33,36,38,40,41,41,40,38,36,33,30,26,22,19,15,11, 8, 5, 3, 1, 0, | |
201 0, 1, 2, 4, 7,10,13,16,19,22,25,28,31,33,34,35,35,34,33,31,28,25,22,19,16,13,10, 7, 4, 2, 1, 0, | |
202 0, 1, 2, 4, 6, 8,10,13,16,19,21,23,25,27,28,29,29,28,27,25,23,21,19,16,13,10, 8, 6, 4, 2, 1, 0, | |
203 0, 1, 1, 3, 4, 6, 8,10,13,15,17,19,20,22,22,23,23,22,22,20,19,17,15,13,10, 8, 6, 4, 3, 1, 1, 0, | |
204 0, 1, 1, 2, 3, 5, 6, 8,10,11,13,14,15,16,17,18,18,17,16,15,14,13,11,10, 8, 6, 5, 3, 2, 1, 1, 0, | |
205 0, 0, 1, 2, 2, 3, 4, 6, 7, 8, 9,10,11,12,12,12,12,12,12,11,10, 9, 8, 7, 6, 4, 3, 2, 2, 1, 0, 0, | |
206 0, 0, 1, 1, 2, 2, 3, 4, 4, 5, 6, 6, 7, 7, 8, 8, 8, 8, 7, 7, 6, 6, 5, 4, 4, 3, 2, 2, 1, 1, 0, 0, | |
207 0, 0, 0, 1, 1, 1, 1, 2, 2, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, 3, 3, 3, 2, 2, 1, 1, 1, 1, 0, 0, 0, | |
208 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, | |
209 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, | |
210 //error:0.000022 | |
211 }; | |
212 static const uint8_t obmc16[256]={ | |
213 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, | |
214 0, 1, 1, 2, 4, 5, 5, 6, 6, 5, 5, 4, 2, 1, 1, 0, | |
215 0, 1, 4, 6, 9,11,13,15,15,13,11, 9, 6, 4, 1, 0, | |
216 0, 2, 6,11,15,20,24,26,26,24,20,15,11, 6, 2, 0, | |
217 0, 4, 9,15,23,29,34,38,38,34,29,23,15, 9, 4, 0, | |
218 0, 5,11,20,29,38,45,49,49,45,38,29,20,11, 5, 0, | |
219 1, 5,13,24,34,45,53,57,57,53,45,34,24,13, 5, 1, | |
220 1, 6,15,26,38,49,57,62,62,57,49,38,26,15, 6, 1, | |
221 1, 6,15,26,38,49,57,62,62,57,49,38,26,15, 6, 1, | |
222 1, 5,13,24,34,45,53,57,57,53,45,34,24,13, 5, 1, | |
223 0, 5,11,20,29,38,45,49,49,45,38,29,20,11, 5, 0, | |
224 0, 4, 9,15,23,29,34,38,38,34,29,23,15, 9, 4, 0, | |
225 0, 2, 6,11,15,20,24,26,26,24,20,15,11, 6, 2, 0, | |
226 0, 1, 4, 6, 9,11,13,15,15,13,11, 9, 6, 4, 1, 0, | |
227 0, 1, 1, 2, 4, 5, 5, 6, 6, 5, 5, 4, 2, 1, 1, 0, | |
228 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, | |
229 //error:0.000033 | |
230 }; | |
231 #elif 1 // 64*linear | |
232 static const uint8_t obmc32[1024]={ | |
3206 | 233 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8, 8, 4, 4, 4, 4, 4, 4, 4, 4, 0, 0, 0, 0, |
234 0, 4, 4, 4, 8, 8, 8, 12, 12, 16, 16, 16, 20, 20, 20, 24, 24, 20, 20, 20, 16, 16, 16, 12, 12, 8, 8, 8, 4, 4, 4, 0, | |
235 0, 4, 8, 8, 12, 12, 16, 20, 20, 24, 28, 28, 32, 32, 36, 40, 40, 36, 32, 32, 28, 28, 24, 20, 20, 16, 12, 12, 8, 8, 4, 0, | |
236 0, 4, 8, 12, 16, 20, 24, 28, 28, 32, 36, 40, 44, 48, 52, 56, 56, 52, 48, 44, 40, 36, 32, 28, 28, 24, 20, 16, 12, 8, 4, 0, | |
237 4, 8, 12, 16, 20, 24, 28, 32, 40, 44, 48, 52, 56, 60, 64, 68, 68, 64, 60, 56, 52, 48, 44, 40, 32, 28, 24, 20, 16, 12, 8, 4, | |
238 4, 8, 12, 20, 24, 32, 36, 40, 48, 52, 56, 64, 68, 76, 80, 84, 84, 80, 76, 68, 64, 56, 52, 48, 40, 36, 32, 24, 20, 12, 8, 4, | |
239 4, 8, 16, 24, 28, 36, 44, 48, 56, 60, 68, 76, 80, 88, 96,100,100, 96, 88, 80, 76, 68, 60, 56, 48, 44, 36, 28, 24, 16, 8, 4, | |
240 4, 12, 20, 28, 32, 40, 48, 56, 64, 72, 80, 88, 92,100,108,116,116,108,100, 92, 88, 80, 72, 64, 56, 48, 40, 32, 28, 20, 12, 4, | |
241 4, 12, 20, 28, 40, 48, 56, 64, 72, 80, 88, 96,108,116,124,132,132,124,116,108, 96, 88, 80, 72, 64, 56, 48, 40, 28, 20, 12, 4, | |
242 4, 16, 24, 32, 44, 52, 60, 72, 80, 92,100,108,120,128,136,148,148,136,128,120,108,100, 92, 80, 72, 60, 52, 44, 32, 24, 16, 4, | |
243 4, 16, 28, 36, 48, 56, 68, 80, 88,100,112,120,132,140,152,164,164,152,140,132,120,112,100, 88, 80, 68, 56, 48, 36, 28, 16, 4, | |
244 4, 16, 28, 40, 52, 64, 76, 88, 96,108,120,132,144,156,168,180,180,168,156,144,132,120,108, 96, 88, 76, 64, 52, 40, 28, 16, 4, | |
245 8, 20, 32, 44, 56, 68, 80, 92,108,120,132,144,156,168,180,192,192,180,168,156,144,132,120,108, 92, 80, 68, 56, 44, 32, 20, 8, | |
246 8, 20, 32, 48, 60, 76, 88,100,116,128,140,156,168,184,196,208,208,196,184,168,156,140,128,116,100, 88, 76, 60, 48, 32, 20, 8, | |
247 8, 20, 36, 52, 64, 80, 96,108,124,136,152,168,180,196,212,224,224,212,196,180,168,152,136,124,108, 96, 80, 64, 52, 36, 20, 8, | |
248 8, 24, 40, 56, 68, 84,100,116,132,148,164,180,192,208,224,240,240,224,208,192,180,164,148,132,116,100, 84, 68, 56, 40, 24, 8, | |
249 8, 24, 40, 56, 68, 84,100,116,132,148,164,180,192,208,224,240,240,224,208,192,180,164,148,132,116,100, 84, 68, 56, 40, 24, 8, | |
250 8, 20, 36, 52, 64, 80, 96,108,124,136,152,168,180,196,212,224,224,212,196,180,168,152,136,124,108, 96, 80, 64, 52, 36, 20, 8, | |
251 8, 20, 32, 48, 60, 76, 88,100,116,128,140,156,168,184,196,208,208,196,184,168,156,140,128,116,100, 88, 76, 60, 48, 32, 20, 8, | |
252 8, 20, 32, 44, 56, 68, 80, 92,108,120,132,144,156,168,180,192,192,180,168,156,144,132,120,108, 92, 80, 68, 56, 44, 32, 20, 8, | |
253 4, 16, 28, 40, 52, 64, 76, 88, 96,108,120,132,144,156,168,180,180,168,156,144,132,120,108, 96, 88, 76, 64, 52, 40, 28, 16, 4, | |
254 4, 16, 28, 36, 48, 56, 68, 80, 88,100,112,120,132,140,152,164,164,152,140,132,120,112,100, 88, 80, 68, 56, 48, 36, 28, 16, 4, | |
255 4, 16, 24, 32, 44, 52, 60, 72, 80, 92,100,108,120,128,136,148,148,136,128,120,108,100, 92, 80, 72, 60, 52, 44, 32, 24, 16, 4, | |
256 4, 12, 20, 28, 40, 48, 56, 64, 72, 80, 88, 96,108,116,124,132,132,124,116,108, 96, 88, 80, 72, 64, 56, 48, 40, 28, 20, 12, 4, | |
257 4, 12, 20, 28, 32, 40, 48, 56, 64, 72, 80, 88, 92,100,108,116,116,108,100, 92, 88, 80, 72, 64, 56, 48, 40, 32, 28, 20, 12, 4, | |
258 4, 8, 16, 24, 28, 36, 44, 48, 56, 60, 68, 76, 80, 88, 96,100,100, 96, 88, 80, 76, 68, 60, 56, 48, 44, 36, 28, 24, 16, 8, 4, | |
259 4, 8, 12, 20, 24, 32, 36, 40, 48, 52, 56, 64, 68, 76, 80, 84, 84, 80, 76, 68, 64, 56, 52, 48, 40, 36, 32, 24, 20, 12, 8, 4, | |
260 4, 8, 12, 16, 20, 24, 28, 32, 40, 44, 48, 52, 56, 60, 64, 68, 68, 64, 60, 56, 52, 48, 44, 40, 32, 28, 24, 20, 16, 12, 8, 4, | |
261 0, 4, 8, 12, 16, 20, 24, 28, 28, 32, 36, 40, 44, 48, 52, 56, 56, 52, 48, 44, 40, 36, 32, 28, 28, 24, 20, 16, 12, 8, 4, 0, | |
262 0, 4, 8, 8, 12, 12, 16, 20, 20, 24, 28, 28, 32, 32, 36, 40, 40, 36, 32, 32, 28, 28, 24, 20, 20, 16, 12, 12, 8, 8, 4, 0, | |
263 0, 4, 4, 4, 8, 8, 8, 12, 12, 16, 16, 16, 20, 20, 20, 24, 24, 20, 20, 20, 16, 16, 16, 12, 12, 8, 8, 8, 4, 4, 4, 0, | |
264 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8, 8, 4, 4, 4, 4, 4, 4, 4, 4, 0, 0, 0, 0, | |
2138 | 265 //error:0.000020 |
266 }; | |
267 static const uint8_t obmc16[256]={ | |
3206 | 268 0, 4, 4, 8, 8, 12, 12, 16, 16, 12, 12, 8, 8, 4, 4, 0, |
269 4, 8, 16, 20, 28, 32, 40, 44, 44, 40, 32, 28, 20, 16, 8, 4, | |
270 4, 16, 24, 36, 44, 56, 64, 76, 76, 64, 56, 44, 36, 24, 16, 4, | |
271 8, 20, 36, 48, 64, 76, 92,104,104, 92, 76, 64, 48, 36, 20, 8, | |
272 8, 28, 44, 64, 80,100,116,136,136,116,100, 80, 64, 44, 28, 8, | |
273 12, 32, 56, 76,100,120,144,164,164,144,120,100, 76, 56, 32, 12, | |
274 12, 40, 64, 92,116,144,168,196,196,168,144,116, 92, 64, 40, 12, | |
275 16, 44, 76,104,136,164,196,224,224,196,164,136,104, 76, 44, 16, | |
276 16, 44, 76,104,136,164,196,224,224,196,164,136,104, 76, 44, 16, | |
277 12, 40, 64, 92,116,144,168,196,196,168,144,116, 92, 64, 40, 12, | |
278 12, 32, 56, 76,100,120,144,164,164,144,120,100, 76, 56, 32, 12, | |
279 8, 28, 44, 64, 80,100,116,136,136,116,100, 80, 64, 44, 28, 8, | |
280 8, 20, 36, 48, 64, 76, 92,104,104, 92, 76, 64, 48, 36, 20, 8, | |
281 4, 16, 24, 36, 44, 56, 64, 76, 76, 64, 56, 44, 36, 24, 16, 4, | |
282 4, 8, 16, 20, 28, 32, 40, 44, 44, 40, 32, 28, 20, 16, 8, 4, | |
283 0, 4, 4, 8, 8, 12, 12, 16, 16, 12, 12, 8, 8, 4, 4, 0, | |
2138 | 284 //error:0.000015 |
285 }; | |
286 #else //64*cos | |
287 static const uint8_t obmc32[1024]={ | |
288 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, | |
289 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, | |
290 0, 0, 0, 1, 1, 1, 1, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 3, 3, 3, 3, 2, 2, 1, 1, 1, 1, 0, 0, 0, | |
291 0, 0, 1, 1, 1, 2, 2, 3, 4, 5, 5, 6, 7, 7, 7, 7, 7, 7, 7, 7, 6, 5, 5, 4, 3, 2, 2, 1, 1, 1, 0, 0, | |
292 0, 0, 1, 1, 2, 3, 4, 5, 6, 7, 9,10,11,11,12,12,12,12,11,11,10, 9, 7, 6, 5, 4, 3, 2, 1, 1, 0, 0, | |
293 0, 0, 1, 2, 3, 5, 6, 8, 9,11,12,14,15,16,17,17,17,17,16,15,14,12,11, 9, 8, 6, 5, 3, 2, 1, 0, 0, | |
294 0, 1, 1, 2, 4, 6, 8,10,12,15,17,19,20,21,22,23,23,22,21,20,19,17,15,12,10, 8, 6, 4, 2, 1, 1, 0, | |
295 0, 1, 2, 3, 5, 8,10,13,16,19,21,24,26,27,28,29,29,28,27,26,24,21,19,16,13,10, 8, 5, 3, 2, 1, 0, | |
296 0, 1, 2, 4, 6, 9,12,16,19,23,26,29,31,33,34,35,35,34,33,31,29,26,23,19,16,12, 9, 6, 4, 2, 1, 0, | |
297 0, 1, 3, 5, 7,11,15,19,23,26,30,34,37,39,40,41,41,40,39,37,34,30,26,23,19,15,11, 7, 5, 3, 1, 0, | |
298 0, 1, 3, 5, 9,12,17,21,26,30,35,38,42,44,46,47,47,46,44,42,38,35,30,26,21,17,12, 9, 5, 3, 1, 0, | |
299 0, 1, 3, 6, 9,14,19,24,29,34,38,43,46,49,51,52,52,51,49,46,43,38,34,29,24,19,14, 9, 6, 3, 1, 0, | |
300 0, 1, 3, 6,11,15,20,26,31,37,42,46,50,53,56,57,57,56,53,50,46,42,37,31,26,20,15,11, 6, 3, 1, 0, | |
301 0, 1, 3, 7,11,16,21,27,33,39,44,49,53,57,59,60,60,59,57,53,49,44,39,33,27,21,16,11, 7, 3, 1, 0, | |
302 0, 1, 4, 7,12,17,22,28,34,40,46,51,56,59,61,63,63,61,59,56,51,46,40,34,28,22,17,12, 7, 4, 1, 0, | |
303 0, 1, 4, 7,12,17,23,29,35,41,47,52,57,60,63,64,64,63,60,57,52,47,41,35,29,23,17,12, 7, 4, 1, 0, | |
304 0, 1, 4, 7,12,17,23,29,35,41,47,52,57,60,63,64,64,63,60,57,52,47,41,35,29,23,17,12, 7, 4, 1, 0, | |
305 0, 1, 4, 7,12,17,22,28,34,40,46,51,56,59,61,63,63,61,59,56,51,46,40,34,28,22,17,12, 7, 4, 1, 0, | |
306 0, 1, 3, 7,11,16,21,27,33,39,44,49,53,57,59,60,60,59,57,53,49,44,39,33,27,21,16,11, 7, 3, 1, 0, | |
307 0, 1, 3, 6,11,15,20,26,31,37,42,46,50,53,56,57,57,56,53,50,46,42,37,31,26,20,15,11, 6, 3, 1, 0, | |
308 0, 1, 3, 6, 9,14,19,24,29,34,38,43,46,49,51,52,52,51,49,46,43,38,34,29,24,19,14, 9, 6, 3, 1, 0, | |
309 0, 1, 3, 5, 9,12,17,21,26,30,35,38,42,44,46,47,47,46,44,42,38,35,30,26,21,17,12, 9, 5, 3, 1, 0, | |
310 0, 1, 3, 5, 7,11,15,19,23,26,30,34,37,39,40,41,41,40,39,37,34,30,26,23,19,15,11, 7, 5, 3, 1, 0, | |
311 0, 1, 2, 4, 6, 9,12,16,19,23,26,29,31,33,34,35,35,34,33,31,29,26,23,19,16,12, 9, 6, 4, 2, 1, 0, | |
312 0, 1, 2, 3, 5, 8,10,13,16,19,21,24,26,27,28,29,29,28,27,26,24,21,19,16,13,10, 8, 5, 3, 2, 1, 0, | |
313 0, 1, 1, 2, 4, 6, 8,10,12,15,17,19,20,21,22,23,23,22,21,20,19,17,15,12,10, 8, 6, 4, 2, 1, 1, 0, | |
314 0, 0, 1, 2, 3, 5, 6, 8, 9,11,12,14,15,16,17,17,17,17,16,15,14,12,11, 9, 8, 6, 5, 3, 2, 1, 0, 0, | |
315 0, 0, 1, 1, 2, 3, 4, 5, 6, 7, 9,10,11,11,12,12,12,12,11,11,10, 9, 7, 6, 5, 4, 3, 2, 1, 1, 0, 0, | |
316 0, 0, 1, 1, 1, 2, 2, 3, 4, 5, 5, 6, 7, 7, 7, 7, 7, 7, 7, 7, 6, 5, 5, 4, 3, 2, 2, 1, 1, 1, 0, 0, | |
317 0, 0, 0, 1, 1, 1, 1, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 3, 3, 3, 3, 2, 2, 1, 1, 1, 1, 0, 0, 0, | |
318 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, | |
319 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, | |
320 //error:0.000022 | |
321 }; | |
322 static const uint8_t obmc16[256]={ | |
323 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, | |
324 0, 0, 1, 2, 3, 4, 5, 5, 5, 5, 4, 3, 2, 1, 0, 0, | |
325 0, 1, 3, 6, 8,11,13,14,14,13,11, 8, 6, 3, 1, 0, | |
326 0, 2, 6,10,15,20,24,26,26,24,20,15,10, 6, 2, 0, | |
327 0, 3, 8,16,23,30,35,38,38,35,30,23,16, 8, 3, 0, | |
328 1, 4,11,20,30,39,46,49,49,46,39,30,20,11, 4, 1, | |
329 1, 5,13,24,35,46,54,58,58,54,46,35,24,13, 5, 1, | |
330 0, 5,14,26,38,49,58,63,63,58,49,38,26,14, 5, 0, | |
331 0, 5,14,26,38,49,58,63,63,58,49,38,26,14, 5, 0, | |
332 1, 5,13,24,35,46,54,58,58,54,46,35,24,13, 5, 1, | |
333 1, 4,11,20,30,39,46,49,49,46,39,30,20,11, 4, 1, | |
334 0, 3, 8,16,23,30,35,38,38,35,30,23,16, 8, 3, 0, | |
335 0, 2, 6,10,15,20,24,26,26,24,20,15,10, 6, 2, 0, | |
336 0, 1, 3, 6, 8,11,13,14,14,13,11, 8, 6, 3, 1, 0, | |
337 0, 0, 1, 2, 3, 4, 5, 5, 5, 5, 4, 3, 2, 1, 0, 0, | |
338 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, | |
339 //error:0.000022 | |
340 }; | |
341 #endif | |
342 | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
343 //linear *64 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
344 static const uint8_t obmc8[64]={ |
3206 | 345 4, 12, 20, 28, 28, 20, 12, 4, |
346 12, 36, 60, 84, 84, 60, 36, 12, | |
347 20, 60,100,140,140,100, 60, 20, | |
348 28, 84,140,196,196,140, 84, 28, | |
349 28, 84,140,196,196,140, 84, 28, | |
350 20, 60,100,140,140,100, 60, 20, | |
351 12, 36, 60, 84, 84, 60, 36, 12, | |
352 4, 12, 20, 28, 28, 20, 12, 4, | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
353 //error:0.000000 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
354 }; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
355 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
356 //linear *64 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
357 static const uint8_t obmc4[16]={ |
3206 | 358 16, 48, 48, 16, |
359 48,144,144, 48, | |
360 48,144,144, 48, | |
361 16, 48, 48, 16, | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
362 //error:0.000000 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
363 }; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
364 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
365 static const uint8_t *obmc_tab[4]={ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
366 obmc32, obmc16, obmc8, obmc4 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
367 }; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
368 |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
369 static int scale_mv_ref[MAX_REF_FRAMES][MAX_REF_FRAMES]; |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
370 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
371 typedef struct BlockNode{ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
372 int16_t mx; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
373 int16_t my; |
3314 | 374 uint8_t ref; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
375 uint8_t color[3]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
376 uint8_t type; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
377 //#define TYPE_SPLIT 1 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
378 #define BLOCK_INTRA 1 |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
379 #define BLOCK_OPT 2 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
380 //#define TYPE_NOCOLOR 4 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
381 uint8_t level; //FIXME merge into type? |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
382 }BlockNode; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
383 |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
384 static const BlockNode null_block= { //FIXME add border maybe |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
385 .color= {128,128,128}, |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
386 .mx= 0, |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
387 .my= 0, |
3314 | 388 .ref= 0, |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
389 .type= 0, |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
390 .level= 0, |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
391 }; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
392 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
393 #define LOG2_MB_SIZE 4 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
394 #define MB_SIZE (1<<LOG2_MB_SIZE) |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
395 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
396 typedef struct x_and_coeff{ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
397 int16_t x; |
2596 | 398 uint16_t coeff; |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
399 } x_and_coeff; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
400 |
2138 | 401 typedef struct SubBand{ |
402 int level; | |
403 int stride; | |
404 int width; | |
405 int height; | |
406 int qlog; ///< log(qscale)/log[2^(1/6)] | |
407 DWTELEM *buf; | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
408 int buf_x_offset; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
409 int buf_y_offset; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
410 int stride_line; ///< Stride measured in lines, not pixels. |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
411 x_and_coeff * x_coeff; |
2138 | 412 struct SubBand *parent; |
413 uint8_t state[/*7*2*/ 7 + 512][32]; | |
414 }SubBand; | |
415 | |
416 typedef struct Plane{ | |
417 int width; | |
418 int height; | |
419 SubBand band[MAX_DECOMPOSITIONS][4]; | |
420 }Plane; | |
421 | |
422 typedef struct SnowContext{ | |
423 // MpegEncContext m; // needed for motion estimation, should not be used for anything else, the idea is to make the motion estimation eventually independant of MpegEncContext, so this will be removed then (FIXME/XXX) | |
424 | |
425 AVCodecContext *avctx; | |
2335 | 426 RangeCoder c; |
2138 | 427 DSPContext dsp; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
428 AVFrame new_picture; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
429 AVFrame input_picture; ///< new_picture with the internal linesizes |
2138 | 430 AVFrame current_picture; |
3314 | 431 AVFrame last_picture[MAX_REF_FRAMES]; |
2138 | 432 AVFrame mconly_picture; |
433 // uint8_t q_context[16]; | |
434 uint8_t header_state[32]; | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
435 uint8_t block_state[128 + 32*128]; |
2138 | 436 int keyframe; |
2199 | 437 int always_reset; |
2138 | 438 int version; |
439 int spatial_decomposition_type; | |
440 int temporal_decomposition_type; | |
441 int spatial_decomposition_count; | |
442 int temporal_decomposition_count; | |
3314 | 443 int max_ref_frames; |
444 int ref_frames; | |
445 int16_t (*ref_mvs[MAX_REF_FRAMES])[2]; | |
446 uint32_t *ref_scores[MAX_REF_FRAMES]; | |
2138 | 447 DWTELEM *spatial_dwt_buffer; |
448 int colorspace_type; | |
449 int chroma_h_shift; | |
450 int chroma_v_shift; | |
451 int spatial_scalability; | |
452 int qlog; | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
453 int lambda; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
454 int lambda2; |
3313 | 455 int pass1_rc; |
2138 | 456 int mv_scale; |
457 int qbias; | |
458 #define QBIAS_SHIFT 3 | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
459 int b_width; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
460 int b_height; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
461 int block_max_depth; |
2138 | 462 Plane plane[MAX_PLANES]; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
463 BlockNode *block; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
464 #define ME_CACHE_SIZE 1024 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
465 int me_cache[ME_CACHE_SIZE]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
466 int me_cache_generation; |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
467 slice_buffer sb; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
468 |
2138 | 469 MpegEncContext m; // needed for motion estimation, should not be used for anything else, the idea is to make the motion estimation eventually independant of MpegEncContext, so this will be removed then (FIXME/XXX) |
470 }SnowContext; | |
471 | |
2562 | 472 typedef struct { |
473 DWTELEM *b0; | |
474 DWTELEM *b1; | |
475 DWTELEM *b2; | |
476 DWTELEM *b3; | |
477 int y; | |
478 } dwt_compose_t; | |
479 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
480 #define slice_buffer_get_line(slice_buf, line_num) ((slice_buf)->line[line_num] ? (slice_buf)->line[line_num] : slice_buffer_load_line((slice_buf), (line_num))) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
481 //#define slice_buffer_get_line(slice_buf, line_num) (slice_buffer_load_line((slice_buf), (line_num))) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
482 |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
483 static void iterative_me(SnowContext *s); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
484 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
485 static void slice_buffer_init(slice_buffer * buf, int line_count, int max_allocated_lines, int line_width, DWTELEM * base_buffer) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
486 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
487 int i; |
2967 | 488 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
489 buf->base_buffer = base_buffer; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
490 buf->line_count = line_count; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
491 buf->line_width = line_width; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
492 buf->data_count = max_allocated_lines; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
493 buf->line = (DWTELEM * *) av_mallocz (sizeof(DWTELEM *) * line_count); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
494 buf->data_stack = (DWTELEM * *) av_malloc (sizeof(DWTELEM *) * max_allocated_lines); |
2967 | 495 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
496 for (i = 0; i < max_allocated_lines; i++) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
497 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
498 buf->data_stack[i] = (DWTELEM *) av_malloc (sizeof(DWTELEM) * line_width); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
499 } |
2967 | 500 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
501 buf->data_stack_top = max_allocated_lines - 1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
502 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
503 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
504 static DWTELEM * slice_buffer_load_line(slice_buffer * buf, int line) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
505 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
506 int offset; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
507 DWTELEM * buffer; |
2967 | 508 |
509 // av_log(NULL, AV_LOG_DEBUG, "Cache hit: %d\n", line); | |
510 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
511 assert(buf->data_stack_top >= 0); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
512 // assert(!buf->line[line]); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
513 if (buf->line[line]) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
514 return buf->line[line]; |
2967 | 515 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
516 offset = buf->line_width * line; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
517 buffer = buf->data_stack[buf->data_stack_top]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
518 buf->data_stack_top--; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
519 buf->line[line] = buffer; |
2967 | 520 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
521 // av_log(NULL, AV_LOG_DEBUG, "slice_buffer_load_line: line: %d remaining: %d\n", line, buf->data_stack_top + 1); |
2967 | 522 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
523 return buffer; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
524 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
525 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
526 static void slice_buffer_release(slice_buffer * buf, int line) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
527 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
528 int offset; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
529 DWTELEM * buffer; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
530 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
531 assert(line >= 0 && line < buf->line_count); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
532 assert(buf->line[line]); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
533 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
534 offset = buf->line_width * line; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
535 buffer = buf->line[line]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
536 buf->data_stack_top++; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
537 buf->data_stack[buf->data_stack_top] = buffer; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
538 buf->line[line] = NULL; |
2967 | 539 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
540 // av_log(NULL, AV_LOG_DEBUG, "slice_buffer_release: line: %d remaining: %d\n", line, buf->data_stack_top + 1); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
541 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
542 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
543 static void slice_buffer_flush(slice_buffer * buf) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
544 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
545 int i; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
546 for (i = 0; i < buf->line_count; i++) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
547 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
548 if (buf->line[i]) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
549 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
550 // av_log(NULL, AV_LOG_DEBUG, "slice_buffer_flush: line: %d \n", i); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
551 slice_buffer_release(buf, i); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
552 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
553 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
554 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
555 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
556 static void slice_buffer_destroy(slice_buffer * buf) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
557 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
558 int i; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
559 slice_buffer_flush(buf); |
2967 | 560 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
561 for (i = buf->data_count - 1; i >= 0; i--) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
562 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
563 assert(buf->data_stack[i]); |
3190 | 564 av_freep(&buf->data_stack[i]); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
565 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
566 assert(buf->data_stack); |
3190 | 567 av_freep(&buf->data_stack); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
568 assert(buf->line); |
3190 | 569 av_freep(&buf->line); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
570 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
571 |
2979 | 572 #ifdef __sgi |
2368
a7ac68734a91
fix for build on IRIX by (Michel Bardiaux {mbardiaux peaktime be})
michael
parents:
2335
diff
changeset
|
573 // Avoid a name clash on SGI IRIX |
2979 | 574 #undef qexp |
2368
a7ac68734a91
fix for build on IRIX by (Michel Bardiaux {mbardiaux peaktime be})
michael
parents:
2335
diff
changeset
|
575 #endif |
2246 | 576 #define QEXPSHIFT (7-FRAC_BITS+8) //FIXME try to change this to 0 |
2600 | 577 static uint8_t qexp[QROOT]; |
2138 | 578 |
579 static inline int mirror(int v, int m){ | |
2998 | 580 while((unsigned)v > (unsigned)m){ |
581 v=-v; | |
582 if(v<0) v+= 2*m; | |
583 } | |
584 return v; | |
2138 | 585 } |
586 | |
2335 | 587 static inline void put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed){ |
2138 | 588 int i; |
589 | |
590 if(v){ | |
591 const int a= ABS(v); | |
592 const int e= av_log2(a); | |
593 #if 1 | |
2967 | 594 const int el= FFMIN(e, 10); |
2335 | 595 put_rac(c, state+0, 0); |
2138 | 596 |
597 for(i=0; i<el; i++){ | |
2335 | 598 put_rac(c, state+1+i, 1); //1..10 |
2138 | 599 } |
600 for(; i<e; i++){ | |
2335 | 601 put_rac(c, state+1+9, 1); //1..10 |
2138 | 602 } |
2335 | 603 put_rac(c, state+1+FFMIN(i,9), 0); |
2138 | 604 |
605 for(i=e-1; i>=el; i--){ | |
2335 | 606 put_rac(c, state+22+9, (a>>i)&1); //22..31 |
2138 | 607 } |
608 for(; i>=0; i--){ | |
2335 | 609 put_rac(c, state+22+i, (a>>i)&1); //22..31 |
2138 | 610 } |
611 | |
612 if(is_signed) | |
2335 | 613 put_rac(c, state+11 + el, v < 0); //11..21 |
2138 | 614 #else |
2967 | 615 |
2335 | 616 put_rac(c, state+0, 0); |
2138 | 617 if(e<=9){ |
618 for(i=0; i<e; i++){ | |
2335 | 619 put_rac(c, state+1+i, 1); //1..10 |
2138 | 620 } |
2335 | 621 put_rac(c, state+1+i, 0); |
2138 | 622 |
623 for(i=e-1; i>=0; i--){ | |
2335 | 624 put_rac(c, state+22+i, (a>>i)&1); //22..31 |
2138 | 625 } |
626 | |
627 if(is_signed) | |
2335 | 628 put_rac(c, state+11 + e, v < 0); //11..21 |
2138 | 629 }else{ |
630 for(i=0; i<e; i++){ | |
2335 | 631 put_rac(c, state+1+FFMIN(i,9), 1); //1..10 |
2138 | 632 } |
2335 | 633 put_rac(c, state+1+FFMIN(i,9), 0); |
2138 | 634 |
635 for(i=e-1; i>=0; i--){ | |
2335 | 636 put_rac(c, state+22+FFMIN(i,9), (a>>i)&1); //22..31 |
2138 | 637 } |
638 | |
639 if(is_signed) | |
2335 | 640 put_rac(c, state+11 + FFMIN(e,10), v < 0); //11..21 |
2138 | 641 } |
642 #endif | |
643 }else{ | |
2335 | 644 put_rac(c, state+0, 1); |
2138 | 645 } |
646 } | |
647 | |
2335 | 648 static inline int get_symbol(RangeCoder *c, uint8_t *state, int is_signed){ |
649 if(get_rac(c, state+0)) | |
2138 | 650 return 0; |
651 else{ | |
2240
c46fed9b7575
simplify getsymbol patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2232
diff
changeset
|
652 int i, e, a; |
c46fed9b7575
simplify getsymbol patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2232
diff
changeset
|
653 e= 0; |
2335 | 654 while(get_rac(c, state+1 + FFMIN(e,9))){ //1..10 |
2240
c46fed9b7575
simplify getsymbol patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2232
diff
changeset
|
655 e++; |
2138 | 656 } |
657 | |
2240
c46fed9b7575
simplify getsymbol patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2232
diff
changeset
|
658 a= 1; |
c46fed9b7575
simplify getsymbol patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2232
diff
changeset
|
659 for(i=e-1; i>=0; i--){ |
2335 | 660 a += a + get_rac(c, state+22 + FFMIN(i,9)); //22..31 |
2240
c46fed9b7575
simplify getsymbol patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2232
diff
changeset
|
661 } |
c46fed9b7575
simplify getsymbol patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2232
diff
changeset
|
662 |
2335 | 663 if(is_signed && get_rac(c, state+11 + FFMIN(e,10))) //11..21 |
2138 | 664 return -a; |
665 else | |
666 return a; | |
667 } | |
668 } | |
669 | |
2335 | 670 static inline void put_symbol2(RangeCoder *c, uint8_t *state, int v, int log2){ |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
671 int i; |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
672 int r= log2>=0 ? 1<<log2 : 1; |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
673 |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
674 assert(v>=0); |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
675 assert(log2>=-4); |
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
676 |
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
677 while(v >= r){ |
2335 | 678 put_rac(c, state+4+log2, 1); |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
679 v -= r; |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
680 log2++; |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
681 if(log2>0) r+=r; |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
682 } |
2335 | 683 put_rac(c, state+4+log2, 0); |
2967 | 684 |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
685 for(i=log2-1; i>=0; i--){ |
2335 | 686 put_rac(c, state+31-i, (v>>i)&1); |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
687 } |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
688 } |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
689 |
2335 | 690 static inline int get_symbol2(RangeCoder *c, uint8_t *state, int log2){ |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
691 int i; |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
692 int r= log2>=0 ? 1<<log2 : 1; |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
693 int v=0; |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
694 |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
695 assert(log2>=-4); |
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
696 |
2335 | 697 while(get_rac(c, state+4+log2)){ |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
698 v+= r; |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
699 log2++; |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
700 if(log2>0) r+=r; |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
701 } |
2967 | 702 |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
703 for(i=log2-1; i>=0; i--){ |
2335 | 704 v+= get_rac(c, state+31-i)<<i; |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
705 } |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
706 |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
707 return v; |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
708 } |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
709 |
2138 | 710 static always_inline void lift(DWTELEM *dst, DWTELEM *src, DWTELEM *ref, int dst_step, int src_step, int ref_step, int width, int mul, int add, int shift, int highpass, int inverse){ |
711 const int mirror_left= !highpass; | |
712 const int mirror_right= (width&1) ^ highpass; | |
713 const int w= (width>>1) - 1 + (highpass & width); | |
714 int i; | |
715 | |
716 #define LIFT(src, ref, inv) ((src) + ((inv) ? - (ref) : + (ref))) | |
717 if(mirror_left){ | |
718 dst[0] = LIFT(src[0], ((mul*2*ref[0]+add)>>shift), inverse); | |
719 dst += dst_step; | |
720 src += src_step; | |
721 } | |
2967 | 722 |
2138 | 723 for(i=0; i<w; i++){ |
724 dst[i*dst_step] = LIFT(src[i*src_step], ((mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add)>>shift), inverse); | |
725 } | |
2967 | 726 |
2138 | 727 if(mirror_right){ |
728 dst[w*dst_step] = LIFT(src[w*src_step], ((mul*2*ref[w*ref_step]+add)>>shift), inverse); | |
729 } | |
730 } | |
731 | |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
732 #ifndef lift5 |
2138 | 733 static always_inline void lift5(DWTELEM *dst, DWTELEM *src, DWTELEM *ref, int dst_step, int src_step, int ref_step, int width, int mul, int add, int shift, int highpass, int inverse){ |
734 const int mirror_left= !highpass; | |
735 const int mirror_right= (width&1) ^ highpass; | |
736 const int w= (width>>1) - 1 + (highpass & width); | |
737 int i; | |
738 | |
739 if(mirror_left){ | |
740 int r= 3*2*ref[0]; | |
741 r += r>>4; | |
742 r += r>>8; | |
743 dst[0] = LIFT(src[0], ((r+add)>>shift), inverse); | |
744 dst += dst_step; | |
745 src += src_step; | |
746 } | |
2967 | 747 |
2138 | 748 for(i=0; i<w; i++){ |
749 int r= 3*(ref[i*ref_step] + ref[(i+1)*ref_step]); | |
750 r += r>>4; | |
751 r += r>>8; | |
752 dst[i*dst_step] = LIFT(src[i*src_step], ((r+add)>>shift), inverse); | |
753 } | |
2967 | 754 |
2138 | 755 if(mirror_right){ |
756 int r= 3*2*ref[w*ref_step]; | |
757 r += r>>4; | |
758 r += r>>8; | |
759 dst[w*dst_step] = LIFT(src[w*src_step], ((r+add)>>shift), inverse); | |
760 } | |
761 } | |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
762 #endif |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
763 |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
764 #ifndef liftS |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
765 static always_inline void liftS(DWTELEM *dst, DWTELEM *src, DWTELEM *ref, int dst_step, int src_step, int ref_step, int width, int mul, int add, int shift, int highpass, int inverse){ |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
766 const int mirror_left= !highpass; |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
767 const int mirror_right= (width&1) ^ highpass; |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
768 const int w= (width>>1) - 1 + (highpass & width); |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
769 int i; |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
770 |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
771 assert(shift == 4); |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
772 #define LIFTS(src, ref, inv) ((inv) ? (src) - (((ref) - 4*(src))>>shift): (16*4*(src) + 4*(ref) + 8 + (5<<27))/(5*16) - (1<<23)) |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
773 if(mirror_left){ |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
774 dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse); |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
775 dst += dst_step; |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
776 src += src_step; |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
777 } |
2967 | 778 |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
779 for(i=0; i<w; i++){ |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
780 dst[i*dst_step] = LIFTS(src[i*src_step], mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add, inverse); |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
781 } |
2967 | 782 |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
783 if(mirror_right){ |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
784 dst[w*dst_step] = LIFTS(src[w*src_step], mul*2*ref[w*ref_step]+add, inverse); |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
785 } |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
786 } |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
787 #endif |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
788 |
2138 | 789 |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
790 static void inplace_lift(DWTELEM *dst, int width, int *coeffs, int n, int shift, int start, int inverse){ |
2138 | 791 int x, i; |
2967 | 792 |
2138 | 793 for(x=start; x<width; x+=2){ |
794 int64_t sum=0; | |
795 | |
796 for(i=0; i<n; i++){ | |
797 int x2= x + 2*i - n + 1; | |
798 if (x2< 0) x2= -x2; | |
799 else if(x2>=width) x2= 2*width-x2-2; | |
800 sum += coeffs[i]*(int64_t)dst[x2]; | |
801 } | |
802 if(inverse) dst[x] -= (sum + (1<<shift)/2)>>shift; | |
803 else dst[x] += (sum + (1<<shift)/2)>>shift; | |
804 } | |
805 } | |
806 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
807 static void inplace_liftV(DWTELEM *dst, int width, int height, int stride, int *coeffs, int n, int shift, int start, int inverse){ |
2138 | 808 int x, y, i; |
809 for(y=start; y<height; y+=2){ | |
810 for(x=0; x<width; x++){ | |
811 int64_t sum=0; | |
2967 | 812 |
2138 | 813 for(i=0; i<n; i++){ |
814 int y2= y + 2*i - n + 1; | |
815 if (y2< 0) y2= -y2; | |
816 else if(y2>=height) y2= 2*height-y2-2; | |
817 sum += coeffs[i]*(int64_t)dst[x + y2*stride]; | |
818 } | |
819 if(inverse) dst[x + y*stride] -= (sum + (1<<shift)/2)>>shift; | |
820 else dst[x + y*stride] += (sum + (1<<shift)/2)>>shift; | |
821 } | |
822 } | |
823 } | |
824 | |
825 #define SCALEX 1 | |
826 #define LX0 0 | |
827 #define LX1 1 | |
828 | |
2139 | 829 #if 0 // more accurate 9/7 |
2138 | 830 #define N1 2 |
831 #define SHIFT1 14 | |
832 #define COEFFS1 (int[]){-25987,-25987} | |
833 #define N2 2 | |
834 #define SHIFT2 19 | |
835 #define COEFFS2 (int[]){-27777,-27777} | |
836 #define N3 2 | |
837 #define SHIFT3 15 | |
838 #define COEFFS3 (int[]){28931,28931} | |
839 #define N4 2 | |
840 #define SHIFT4 15 | |
841 #define COEFFS4 (int[]){14533,14533} | |
842 #elif 1 // 13/7 CRF | |
843 #define N1 4 | |
844 #define SHIFT1 4 | |
845 #define COEFFS1 (int[]){1,-9,-9,1} | |
2139 | 846 #define N2 4 |
2138 | 847 #define SHIFT2 4 |
848 #define COEFFS2 (int[]){-1,5,5,-1} | |
849 #define N3 0 | |
850 #define SHIFT3 1 | |
851 #define COEFFS3 NULL | |
852 #define N4 0 | |
853 #define SHIFT4 1 | |
854 #define COEFFS4 NULL | |
855 #elif 1 // 3/5 | |
856 #define LX0 1 | |
857 #define LX1 0 | |
858 #define SCALEX 0.5 | |
859 #define N1 2 | |
860 #define SHIFT1 1 | |
861 #define COEFFS1 (int[]){1,1} | |
862 #define N2 2 | |
863 #define SHIFT2 2 | |
864 #define COEFFS2 (int[]){-1,-1} | |
865 #define N3 0 | |
866 #define SHIFT3 0 | |
867 #define COEFFS3 NULL | |
868 #define N4 0 | |
869 #define SHIFT4 0 | |
870 #define COEFFS4 NULL | |
2967 | 871 #elif 1 // 11/5 |
2138 | 872 #define N1 0 |
873 #define SHIFT1 1 | |
874 #define COEFFS1 NULL | |
875 #define N2 2 | |
876 #define SHIFT2 2 | |
877 #define COEFFS2 (int[]){-1,-1} | |
878 #define N3 2 | |
879 #define SHIFT3 0 | |
880 #define COEFFS3 (int[]){-1,-1} | |
881 #define N4 4 | |
882 #define SHIFT4 7 | |
883 #define COEFFS4 (int[]){-5,29,29,-5} | |
884 #define SCALEX 4 | |
885 #elif 1 // 9/7 CDF | |
886 #define N1 2 | |
887 #define SHIFT1 7 | |
888 #define COEFFS1 (int[]){-203,-203} | |
889 #define N2 2 | |
890 #define SHIFT2 12 | |
891 #define COEFFS2 (int[]){-217,-217} | |
892 #define N3 2 | |
893 #define SHIFT3 7 | |
894 #define COEFFS3 (int[]){113,113} | |
895 #define N4 2 | |
896 #define SHIFT4 9 | |
897 #define COEFFS4 (int[]){227,227} | |
898 #define SCALEX 1 | |
899 #elif 1 // 7/5 CDF | |
900 #define N1 0 | |
901 #define SHIFT1 1 | |
902 #define COEFFS1 NULL | |
903 #define N2 2 | |
904 #define SHIFT2 2 | |
905 #define COEFFS2 (int[]){-1,-1} | |
906 #define N3 2 | |
907 #define SHIFT3 0 | |
908 #define COEFFS3 (int[]){-1,-1} | |
909 #define N4 2 | |
910 #define SHIFT4 4 | |
911 #define COEFFS4 (int[]){3,3} | |
912 #elif 1 // 9/7 MN | |
913 #define N1 4 | |
914 #define SHIFT1 4 | |
915 #define COEFFS1 (int[]){1,-9,-9,1} | |
916 #define N2 2 | |
917 #define SHIFT2 2 | |
918 #define COEFFS2 (int[]){1,1} | |
919 #define N3 0 | |
920 #define SHIFT3 1 | |
921 #define COEFFS3 NULL | |
922 #define N4 0 | |
923 #define SHIFT4 1 | |
924 #define COEFFS4 NULL | |
925 #else // 13/7 CRF | |
926 #define N1 4 | |
927 #define SHIFT1 4 | |
928 #define COEFFS1 (int[]){1,-9,-9,1} | |
929 #define N2 4 | |
930 #define SHIFT2 4 | |
931 #define COEFFS2 (int[]){-1,5,5,-1} | |
932 #define N3 0 | |
933 #define SHIFT3 1 | |
934 #define COEFFS3 NULL | |
935 #define N4 0 | |
936 #define SHIFT4 1 | |
937 #define COEFFS4 NULL | |
938 #endif | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
939 static void horizontal_decomposeX(DWTELEM *b, int width){ |
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
940 DWTELEM temp[width]; |
2138 | 941 const int width2= width>>1; |
942 const int w2= (width+1)>>1; | |
2893 | 943 int x; |
2138 | 944 |
945 inplace_lift(b, width, COEFFS1, N1, SHIFT1, LX1, 0); | |
946 inplace_lift(b, width, COEFFS2, N2, SHIFT2, LX0, 0); | |
947 inplace_lift(b, width, COEFFS3, N3, SHIFT3, LX1, 0); | |
948 inplace_lift(b, width, COEFFS4, N4, SHIFT4, LX0, 0); | |
2967 | 949 |
2138 | 950 for(x=0; x<width2; x++){ |
951 temp[x ]= b[2*x ]; | |
952 temp[x+w2]= b[2*x + 1]; | |
953 } | |
954 if(width&1) | |
955 temp[x ]= b[2*x ]; | |
956 memcpy(b, temp, width*sizeof(int)); | |
957 } | |
958 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
959 static void horizontal_composeX(DWTELEM *b, int width){ |
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
960 DWTELEM temp[width]; |
2138 | 961 const int width2= width>>1; |
2893 | 962 int x; |
2138 | 963 const int w2= (width+1)>>1; |
964 | |
965 memcpy(temp, b, width*sizeof(int)); | |
966 for(x=0; x<width2; x++){ | |
967 b[2*x ]= temp[x ]; | |
968 b[2*x + 1]= temp[x+w2]; | |
969 } | |
970 if(width&1) | |
971 b[2*x ]= temp[x ]; | |
972 | |
973 inplace_lift(b, width, COEFFS4, N4, SHIFT4, LX0, 1); | |
974 inplace_lift(b, width, COEFFS3, N3, SHIFT3, LX1, 1); | |
975 inplace_lift(b, width, COEFFS2, N2, SHIFT2, LX0, 1); | |
976 inplace_lift(b, width, COEFFS1, N1, SHIFT1, LX1, 1); | |
977 } | |
978 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
979 static void spatial_decomposeX(DWTELEM *buffer, int width, int height, int stride){ |
2138 | 980 int x, y; |
2967 | 981 |
2138 | 982 for(y=0; y<height; y++){ |
983 for(x=0; x<width; x++){ | |
984 buffer[y*stride + x] *= SCALEX; | |
985 } | |
986 } | |
987 | |
988 for(y=0; y<height; y++){ | |
989 horizontal_decomposeX(buffer + y*stride, width); | |
990 } | |
2967 | 991 |
2138 | 992 inplace_liftV(buffer, width, height, stride, COEFFS1, N1, SHIFT1, LX1, 0); |
993 inplace_liftV(buffer, width, height, stride, COEFFS2, N2, SHIFT2, LX0, 0); | |
994 inplace_liftV(buffer, width, height, stride, COEFFS3, N3, SHIFT3, LX1, 0); | |
2967 | 995 inplace_liftV(buffer, width, height, stride, COEFFS4, N4, SHIFT4, LX0, 0); |
2138 | 996 } |
997 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
998 static void spatial_composeX(DWTELEM *buffer, int width, int height, int stride){ |
2138 | 999 int x, y; |
2967 | 1000 |
2138 | 1001 inplace_liftV(buffer, width, height, stride, COEFFS4, N4, SHIFT4, LX0, 1); |
1002 inplace_liftV(buffer, width, height, stride, COEFFS3, N3, SHIFT3, LX1, 1); | |
1003 inplace_liftV(buffer, width, height, stride, COEFFS2, N2, SHIFT2, LX0, 1); | |
1004 inplace_liftV(buffer, width, height, stride, COEFFS1, N1, SHIFT1, LX1, 1); | |
1005 | |
1006 for(y=0; y<height; y++){ | |
1007 horizontal_composeX(buffer + y*stride, width); | |
1008 } | |
1009 | |
1010 for(y=0; y<height; y++){ | |
1011 for(x=0; x<width; x++){ | |
1012 buffer[y*stride + x] /= SCALEX; | |
1013 } | |
1014 } | |
1015 } | |
1016 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1017 static void horizontal_decompose53i(DWTELEM *b, int width){ |
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1018 DWTELEM temp[width]; |
2138 | 1019 const int width2= width>>1; |
2893 | 1020 int x; |
2138 | 1021 const int w2= (width+1)>>1; |
1022 | |
1023 for(x=0; x<width2; x++){ | |
1024 temp[x ]= b[2*x ]; | |
1025 temp[x+w2]= b[2*x + 1]; | |
1026 } | |
1027 if(width&1) | |
1028 temp[x ]= b[2*x ]; | |
1029 #if 0 | |
2893 | 1030 { |
1031 int A1,A2,A3,A4; | |
2138 | 1032 A2= temp[1 ]; |
1033 A4= temp[0 ]; | |
1034 A1= temp[0+width2]; | |
1035 A1 -= (A2 + A4)>>1; | |
1036 A4 += (A1 + 1)>>1; | |
1037 b[0+width2] = A1; | |
1038 b[0 ] = A4; | |
1039 for(x=1; x+1<width2; x+=2){ | |
1040 A3= temp[x+width2]; | |
1041 A4= temp[x+1 ]; | |
1042 A3 -= (A2 + A4)>>1; | |
1043 A2 += (A1 + A3 + 2)>>2; | |
1044 b[x+width2] = A3; | |
1045 b[x ] = A2; | |
1046 | |
1047 A1= temp[x+1+width2]; | |
1048 A2= temp[x+2 ]; | |
1049 A1 -= (A2 + A4)>>1; | |
1050 A4 += (A1 + A3 + 2)>>2; | |
1051 b[x+1+width2] = A1; | |
1052 b[x+1 ] = A4; | |
1053 } | |
1054 A3= temp[width-1]; | |
1055 A3 -= A2; | |
1056 A2 += (A1 + A3 + 2)>>2; | |
1057 b[width -1] = A3; | |
1058 b[width2-1] = A2; | |
2893 | 1059 } |
2967 | 1060 #else |
2138 | 1061 lift(b+w2, temp+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 0); |
1062 lift(b , temp , b+w2, 1, 1, 1, width, 1, 2, 2, 0, 0); | |
1063 #endif | |
1064 } | |
1065 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1066 static void vertical_decompose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1067 int i; |
2967 | 1068 |
2138 | 1069 for(i=0; i<width; i++){ |
1070 b1[i] -= (b0[i] + b2[i])>>1; | |
1071 } | |
1072 } | |
1073 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1074 static void vertical_decompose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1075 int i; |
2967 | 1076 |
2138 | 1077 for(i=0; i<width; i++){ |
1078 b1[i] += (b0[i] + b2[i] + 2)>>2; | |
1079 } | |
1080 } | |
1081 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1082 static void spatial_decompose53i(DWTELEM *buffer, int width, int height, int stride){ |
2198 | 1083 int y; |
2138 | 1084 DWTELEM *b0= buffer + mirror(-2-1, height-1)*stride; |
1085 DWTELEM *b1= buffer + mirror(-2 , height-1)*stride; | |
2967 | 1086 |
2138 | 1087 for(y=-2; y<height; y+=2){ |
1088 DWTELEM *b2= buffer + mirror(y+1, height-1)*stride; | |
1089 DWTELEM *b3= buffer + mirror(y+2, height-1)*stride; | |
1090 | |
1091 {START_TIMER | |
2998 | 1092 if(y+1<(unsigned)height) horizontal_decompose53i(b2, width); |
1093 if(y+2<(unsigned)height) horizontal_decompose53i(b3, width); | |
2138 | 1094 STOP_TIMER("horizontal_decompose53i")} |
2967 | 1095 |
2138 | 1096 {START_TIMER |
2998 | 1097 if(y+1<(unsigned)height) vertical_decompose53iH0(b1, b2, b3, width); |
1098 if(y+0<(unsigned)height) vertical_decompose53iL0(b0, b1, b2, width); | |
2138 | 1099 STOP_TIMER("vertical_decompose53i*")} |
2967 | 1100 |
2138 | 1101 b0=b2; |
1102 b1=b3; | |
1103 } | |
1104 } | |
1105 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1106 static void horizontal_decompose97i(DWTELEM *b, int width){ |
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1107 DWTELEM temp[width]; |
2138 | 1108 const int w2= (width+1)>>1; |
1109 | |
1110 lift (temp+w2, b +1, b , 1, 2, 2, width, -W_AM, W_AO, W_AS, 1, 0); | |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1111 liftS(temp , b , temp+w2, 1, 2, 1, width, -W_BM, W_BO, W_BS, 0, 0); |
2138 | 1112 lift5(b +w2, temp+w2, temp , 1, 1, 1, width, W_CM, W_CO, W_CS, 1, 0); |
1113 lift (b , temp , b +w2, 1, 1, 1, width, W_DM, W_DO, W_DS, 0, 0); | |
1114 } | |
1115 | |
1116 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1117 static void vertical_decompose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1118 int i; |
2967 | 1119 |
2138 | 1120 for(i=0; i<width; i++){ |
1121 b1[i] -= (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS; | |
1122 } | |
1123 } | |
1124 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1125 static void vertical_decompose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1126 int i; |
2967 | 1127 |
2138 | 1128 for(i=0; i<width; i++){ |
1129 #ifdef lift5 | |
1130 b1[i] += (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS; | |
1131 #else | |
1132 int r= 3*(b0[i] + b2[i]); | |
1133 r+= r>>4; | |
1134 r+= r>>8; | |
1135 b1[i] += (r+W_CO)>>W_CS; | |
1136 #endif | |
1137 } | |
1138 } | |
1139 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1140 static void vertical_decompose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1141 int i; |
2967 | 1142 |
2138 | 1143 for(i=0; i<width; i++){ |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1144 #ifdef liftS |
2138 | 1145 b1[i] -= (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS; |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1146 #else |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1147 b1[i] = (16*4*b1[i] - 4*(b0[i] + b2[i]) + 8*5 + (5<<27)) / (5*16) - (1<<23); |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1148 #endif |
2138 | 1149 } |
1150 } | |
1151 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1152 static void vertical_decompose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1153 int i; |
2967 | 1154 |
2138 | 1155 for(i=0; i<width; i++){ |
1156 b1[i] += (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS; | |
1157 } | |
1158 } | |
1159 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1160 static void spatial_decompose97i(DWTELEM *buffer, int width, int height, int stride){ |
2198 | 1161 int y; |
2138 | 1162 DWTELEM *b0= buffer + mirror(-4-1, height-1)*stride; |
1163 DWTELEM *b1= buffer + mirror(-4 , height-1)*stride; | |
1164 DWTELEM *b2= buffer + mirror(-4+1, height-1)*stride; | |
1165 DWTELEM *b3= buffer + mirror(-4+2, height-1)*stride; | |
2967 | 1166 |
2138 | 1167 for(y=-4; y<height; y+=2){ |
1168 DWTELEM *b4= buffer + mirror(y+3, height-1)*stride; | |
1169 DWTELEM *b5= buffer + mirror(y+4, height-1)*stride; | |
1170 | |
1171 {START_TIMER | |
2998 | 1172 if(y+3<(unsigned)height) horizontal_decompose97i(b4, width); |
1173 if(y+4<(unsigned)height) horizontal_decompose97i(b5, width); | |
2138 | 1174 if(width>400){ |
1175 STOP_TIMER("horizontal_decompose97i") | |
1176 }} | |
2967 | 1177 |
2138 | 1178 {START_TIMER |
2998 | 1179 if(y+3<(unsigned)height) vertical_decompose97iH0(b3, b4, b5, width); |
1180 if(y+2<(unsigned)height) vertical_decompose97iL0(b2, b3, b4, width); | |
1181 if(y+1<(unsigned)height) vertical_decompose97iH1(b1, b2, b3, width); | |
1182 if(y+0<(unsigned)height) vertical_decompose97iL1(b0, b1, b2, width); | |
2138 | 1183 |
1184 if(width>400){ | |
1185 STOP_TIMER("vertical_decompose97i") | |
1186 }} | |
2967 | 1187 |
2138 | 1188 b0=b2; |
1189 b1=b3; | |
1190 b2=b4; | |
1191 b3=b5; | |
1192 } | |
1193 } | |
1194 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1195 void ff_spatial_dwt(DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){ |
2138 | 1196 int level; |
2967 | 1197 |
2164 | 1198 for(level=0; level<decomposition_count; level++){ |
1199 switch(type){ | |
3326 | 1200 case DWT_97: spatial_decompose97i(buffer, width>>level, height>>level, stride<<level); break; |
1201 case DWT_53: spatial_decompose53i(buffer, width>>level, height>>level, stride<<level); break; | |
1202 case DWT_X: spatial_decomposeX (buffer, width>>level, height>>level, stride<<level); break; | |
2138 | 1203 } |
1204 } | |
1205 } | |
1206 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1207 static void horizontal_compose53i(DWTELEM *b, int width){ |
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1208 DWTELEM temp[width]; |
2138 | 1209 const int width2= width>>1; |
1210 const int w2= (width+1)>>1; | |
2893 | 1211 int x; |
2138 | 1212 |
1213 #if 0 | |
2893 | 1214 int A1,A2,A3,A4; |
2138 | 1215 A2= temp[1 ]; |
1216 A4= temp[0 ]; | |
1217 A1= temp[0+width2]; | |
1218 A1 -= (A2 + A4)>>1; | |
1219 A4 += (A1 + 1)>>1; | |
1220 b[0+width2] = A1; | |
1221 b[0 ] = A4; | |
1222 for(x=1; x+1<width2; x+=2){ | |
1223 A3= temp[x+width2]; | |
1224 A4= temp[x+1 ]; | |
1225 A3 -= (A2 + A4)>>1; | |
1226 A2 += (A1 + A3 + 2)>>2; | |
1227 b[x+width2] = A3; | |
1228 b[x ] = A2; | |
1229 | |
1230 A1= temp[x+1+width2]; | |
1231 A2= temp[x+2 ]; | |
1232 A1 -= (A2 + A4)>>1; | |
1233 A4 += (A1 + A3 + 2)>>2; | |
1234 b[x+1+width2] = A1; | |
1235 b[x+1 ] = A4; | |
1236 } | |
1237 A3= temp[width-1]; | |
1238 A3 -= A2; | |
1239 A2 += (A1 + A3 + 2)>>2; | |
1240 b[width -1] = A3; | |
1241 b[width2-1] = A2; | |
2967 | 1242 #else |
2138 | 1243 lift(temp , b , b+w2, 1, 1, 1, width, 1, 2, 2, 0, 1); |
1244 lift(temp+w2, b+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 1); | |
1245 #endif | |
1246 for(x=0; x<width2; x++){ | |
1247 b[2*x ]= temp[x ]; | |
1248 b[2*x + 1]= temp[x+w2]; | |
1249 } | |
1250 if(width&1) | |
1251 b[2*x ]= temp[x ]; | |
1252 } | |
1253 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1254 static void vertical_compose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1255 int i; |
2967 | 1256 |
2138 | 1257 for(i=0; i<width; i++){ |
1258 b1[i] += (b0[i] + b2[i])>>1; | |
1259 } | |
1260 } | |
1261 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1262 static void vertical_compose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1263 int i; |
2967 | 1264 |
2138 | 1265 for(i=0; i<width; i++){ |
1266 b1[i] -= (b0[i] + b2[i] + 2)>>2; | |
1267 } | |
1268 } | |
1269 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1270 static void spatial_compose53i_buffered_init(dwt_compose_t *cs, slice_buffer * sb, int height, int stride_line){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1271 cs->b0 = slice_buffer_get_line(sb, mirror(-1-1, height-1) * stride_line); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1272 cs->b1 = slice_buffer_get_line(sb, mirror(-1 , height-1) * stride_line); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1273 cs->y = -1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1274 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1275 |
2562 | 1276 static void spatial_compose53i_init(dwt_compose_t *cs, DWTELEM *buffer, int height, int stride){ |
1277 cs->b0 = buffer + mirror(-1-1, height-1)*stride; | |
1278 cs->b1 = buffer + mirror(-1 , height-1)*stride; | |
1279 cs->y = -1; | |
1280 } | |
1281 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1282 static void spatial_compose53i_dy_buffered(dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1283 int y= cs->y; |
2967 | 1284 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1285 DWTELEM *b0= cs->b0; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1286 DWTELEM *b1= cs->b1; |
3012 | 1287 DWTELEM *b2= slice_buffer_get_line(sb, mirror(y+1, height-1) * stride_line); |
1288 DWTELEM *b3= slice_buffer_get_line(sb, mirror(y+2, height-1) * stride_line); | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1289 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1290 {START_TIMER |
2998 | 1291 if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width); |
1292 if(y+0<(unsigned)height) vertical_compose53iH0(b0, b1, b2, width); | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1293 STOP_TIMER("vertical_compose53i*")} |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1294 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1295 {START_TIMER |
2998 | 1296 if(y-1<(unsigned)height) horizontal_compose53i(b0, width); |
1297 if(y+0<(unsigned)height) horizontal_compose53i(b1, width); | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1298 STOP_TIMER("horizontal_compose53i")} |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1299 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1300 cs->b0 = b2; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1301 cs->b1 = b3; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1302 cs->y += 2; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1303 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1304 |
2562 | 1305 static void spatial_compose53i_dy(dwt_compose_t *cs, DWTELEM *buffer, int width, int height, int stride){ |
1306 int y= cs->y; | |
1307 DWTELEM *b0= cs->b0; | |
1308 DWTELEM *b1= cs->b1; | |
1309 DWTELEM *b2= buffer + mirror(y+1, height-1)*stride; | |
1310 DWTELEM *b3= buffer + mirror(y+2, height-1)*stride; | |
2138 | 1311 |
1312 {START_TIMER | |
2998 | 1313 if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width); |
1314 if(y+0<(unsigned)height) vertical_compose53iH0(b0, b1, b2, width); | |
2138 | 1315 STOP_TIMER("vertical_compose53i*")} |
1316 | |
1317 {START_TIMER | |
2998 | 1318 if(y-1<(unsigned)height) horizontal_compose53i(b0, width); |
1319 if(y+0<(unsigned)height) horizontal_compose53i(b1, width); | |
2138 | 1320 STOP_TIMER("horizontal_compose53i")} |
1321 | |
2562 | 1322 cs->b0 = b2; |
1323 cs->b1 = b3; | |
1324 cs->y += 2; | |
1325 } | |
1326 | |
1327 static void spatial_compose53i(DWTELEM *buffer, int width, int height, int stride){ | |
1328 dwt_compose_t cs; | |
1329 spatial_compose53i_init(&cs, buffer, height, stride); | |
1330 while(cs.y <= height) | |
1331 spatial_compose53i_dy(&cs, buffer, width, height, stride); | |
2967 | 1332 } |
1333 | |
1334 | |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
1335 void ff_snow_horizontal_compose97i(DWTELEM *b, int width){ |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1336 DWTELEM temp[width]; |
2138 | 1337 const int w2= (width+1)>>1; |
1338 | |
1339 lift (temp , b , b +w2, 1, 1, 1, width, W_DM, W_DO, W_DS, 0, 1); | |
1340 lift5(temp+w2, b +w2, temp , 1, 1, 1, width, W_CM, W_CO, W_CS, 1, 1); | |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1341 liftS(b , temp , temp+w2, 2, 1, 1, width, -W_BM, W_BO, W_BS, 0, 1); |
2138 | 1342 lift (b+1 , temp+w2, b , 2, 1, 2, width, -W_AM, W_AO, W_AS, 1, 1); |
1343 } | |
1344 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1345 static void vertical_compose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1346 int i; |
2967 | 1347 |
2138 | 1348 for(i=0; i<width; i++){ |
1349 b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS; | |
1350 } | |
1351 } | |
1352 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1353 static void vertical_compose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1354 int i; |
2967 | 1355 |
2138 | 1356 for(i=0; i<width; i++){ |
1357 #ifdef lift5 | |
1358 b1[i] -= (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS; | |
1359 #else | |
1360 int r= 3*(b0[i] + b2[i]); | |
1361 r+= r>>4; | |
1362 r+= r>>8; | |
1363 b1[i] -= (r+W_CO)>>W_CS; | |
1364 #endif | |
1365 } | |
1366 } | |
1367 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1368 static void vertical_compose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1369 int i; |
2967 | 1370 |
2138 | 1371 for(i=0; i<width; i++){ |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1372 #ifdef liftS |
2138 | 1373 b1[i] += (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS; |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1374 #else |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1375 b1[i] += (W_BM*(b0[i] + b2[i])+4*b1[i]+W_BO)>>W_BS; |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1376 #endif |
2138 | 1377 } |
1378 } | |
1379 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1380 static void vertical_compose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1381 int i; |
2967 | 1382 |
2138 | 1383 for(i=0; i<width; i++){ |
1384 b1[i] -= (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS; | |
1385 } | |
1386 } | |
1387 | |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
1388 void ff_snow_vertical_compose97i(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width){ |
2592 | 1389 int i; |
2967 | 1390 |
2592 | 1391 for(i=0; i<width; i++){ |
2893 | 1392 #ifndef lift5 |
2592 | 1393 int r; |
2893 | 1394 #endif |
2592 | 1395 b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS; |
1396 #ifdef lift5 | |
1397 b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS; | |
1398 #else | |
1399 r= 3*(b2[i] + b4[i]); | |
1400 r+= r>>4; | |
1401 r+= r>>8; | |
1402 b3[i] -= (r+W_CO)>>W_CS; | |
1403 #endif | |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1404 #ifdef liftS |
2592 | 1405 b2[i] += (W_BM*(b1[i] + b3[i])+W_BO)>>W_BS; |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1406 #else |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1407 b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS; |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1408 #endif |
2592 | 1409 b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS; |
1410 } | |
1411 } | |
1412 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1413 static void spatial_compose97i_buffered_init(dwt_compose_t *cs, slice_buffer * sb, int height, int stride_line){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1414 cs->b0 = slice_buffer_get_line(sb, mirror(-3-1, height-1) * stride_line); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1415 cs->b1 = slice_buffer_get_line(sb, mirror(-3 , height-1) * stride_line); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1416 cs->b2 = slice_buffer_get_line(sb, mirror(-3+1, height-1) * stride_line); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1417 cs->b3 = slice_buffer_get_line(sb, mirror(-3+2, height-1) * stride_line); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1418 cs->y = -3; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1419 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1420 |
2562 | 1421 static void spatial_compose97i_init(dwt_compose_t *cs, DWTELEM *buffer, int height, int stride){ |
1422 cs->b0 = buffer + mirror(-3-1, height-1)*stride; | |
1423 cs->b1 = buffer + mirror(-3 , height-1)*stride; | |
1424 cs->b2 = buffer + mirror(-3+1, height-1)*stride; | |
1425 cs->b3 = buffer + mirror(-3+2, height-1)*stride; | |
1426 cs->y = -3; | |
1427 } | |
2138 | 1428 |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
1429 static void spatial_compose97i_dy_buffered(DSPContext *dsp, dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1430 int y = cs->y; |
2967 | 1431 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1432 DWTELEM *b0= cs->b0; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1433 DWTELEM *b1= cs->b1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1434 DWTELEM *b2= cs->b2; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1435 DWTELEM *b3= cs->b3; |
3012 | 1436 DWTELEM *b4= slice_buffer_get_line(sb, mirror(y + 3, height - 1) * stride_line); |
1437 DWTELEM *b5= slice_buffer_get_line(sb, mirror(y + 4, height - 1) * stride_line); | |
2967 | 1438 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1439 {START_TIMER |
2592 | 1440 if(y>0 && y+4<height){ |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
1441 dsp->vertical_compose97i(b0, b1, b2, b3, b4, b5, width); |
2592 | 1442 }else{ |
2998 | 1443 if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width); |
1444 if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width); | |
1445 if(y+1<(unsigned)height) vertical_compose97iL0(b1, b2, b3, width); | |
1446 if(y+0<(unsigned)height) vertical_compose97iH0(b0, b1, b2, width); | |
2592 | 1447 } |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1448 if(width>400){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1449 STOP_TIMER("vertical_compose97i")}} |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1450 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1451 {START_TIMER |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
1452 if(y-1<(unsigned)height) dsp->horizontal_compose97i(b0, width); |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
1453 if(y+0<(unsigned)height) dsp->horizontal_compose97i(b1, width); |
3012 | 1454 if(width>400 && y+0<(unsigned)height){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1455 STOP_TIMER("horizontal_compose97i")}} |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1456 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1457 cs->b0=b2; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1458 cs->b1=b3; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1459 cs->b2=b4; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1460 cs->b3=b5; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1461 cs->y += 2; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1462 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1463 |
2562 | 1464 static void spatial_compose97i_dy(dwt_compose_t *cs, DWTELEM *buffer, int width, int height, int stride){ |
1465 int y = cs->y; | |
1466 DWTELEM *b0= cs->b0; | |
1467 DWTELEM *b1= cs->b1; | |
1468 DWTELEM *b2= cs->b2; | |
1469 DWTELEM *b3= cs->b3; | |
1470 DWTELEM *b4= buffer + mirror(y+3, height-1)*stride; | |
1471 DWTELEM *b5= buffer + mirror(y+4, height-1)*stride; | |
2138 | 1472 |
1473 {START_TIMER | |
2998 | 1474 if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width); |
1475 if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width); | |
1476 if(y+1<(unsigned)height) vertical_compose97iL0(b1, b2, b3, width); | |
1477 if(y+0<(unsigned)height) vertical_compose97iH0(b0, b1, b2, width); | |
2138 | 1478 if(width>400){ |
1479 STOP_TIMER("vertical_compose97i")}} | |
1480 | |
1481 {START_TIMER | |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
1482 if(y-1<(unsigned)height) ff_snow_horizontal_compose97i(b0, width); |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
1483 if(y+0<(unsigned)height) ff_snow_horizontal_compose97i(b1, width); |
2138 | 1484 if(width>400 && b0 <= b2){ |
1485 STOP_TIMER("horizontal_compose97i")}} | |
2562 | 1486 |
1487 cs->b0=b2; | |
1488 cs->b1=b3; | |
1489 cs->b2=b4; | |
1490 cs->b3=b5; | |
1491 cs->y += 2; | |
1492 } | |
1493 | |
1494 static void spatial_compose97i(DWTELEM *buffer, int width, int height, int stride){ | |
1495 dwt_compose_t cs; | |
1496 spatial_compose97i_init(&cs, buffer, height, stride); | |
1497 while(cs.y <= height) | |
1498 spatial_compose97i_dy(&cs, buffer, width, height, stride); | |
1499 } | |
1500 | |
3075 | 1501 static void ff_spatial_idwt_buffered_init(dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line, int type, int decomposition_count){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1502 int level; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1503 for(level=decomposition_count-1; level>=0; level--){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1504 switch(type){ |
3326 | 1505 case DWT_97: spatial_compose97i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break; |
1506 case DWT_53: spatial_compose53i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break; | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1507 /* not slicified yet */ |
3326 | 1508 case DWT_X: /*spatial_composeX(buffer, width>>level, height>>level, stride<<level); break;*/ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1509 av_log(NULL, AV_LOG_ERROR, "spatial_composeX neither buffered nor slicified yet.\n"); break; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1510 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1511 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1512 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1513 |
3075 | 1514 static void ff_spatial_idwt_init(dwt_compose_t *cs, DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){ |
2562 | 1515 int level; |
1516 for(level=decomposition_count-1; level>=0; level--){ | |
1517 switch(type){ | |
3326 | 1518 case DWT_97: spatial_compose97i_init(cs+level, buffer, height>>level, stride<<level); break; |
1519 case DWT_53: spatial_compose53i_init(cs+level, buffer, height>>level, stride<<level); break; | |
2562 | 1520 /* not slicified yet */ |
3326 | 1521 case DWT_X: spatial_composeX(buffer, width>>level, height>>level, stride<<level); break; |
2562 | 1522 } |
1523 } | |
1524 } | |
1525 | |
3075 | 1526 static void ff_spatial_idwt_slice(dwt_compose_t *cs, DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count, int y){ |
2562 | 1527 const int support = type==1 ? 3 : 5; |
1528 int level; | |
1529 if(type==2) return; | |
1530 | |
1531 for(level=decomposition_count-1; level>=0; level--){ | |
1532 while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){ | |
1533 switch(type){ | |
3326 | 1534 case DWT_97: spatial_compose97i_dy(cs+level, buffer, width>>level, height>>level, stride<<level); |
2562 | 1535 break; |
3326 | 1536 case DWT_53: spatial_compose53i_dy(cs+level, buffer, width>>level, height>>level, stride<<level); |
2562 | 1537 break; |
3326 | 1538 case DWT_X: break; |
2562 | 1539 } |
1540 } | |
2138 | 1541 } |
1542 } | |
1543 | |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
1544 static void ff_spatial_idwt_buffered_slice(DSPContext *dsp, dwt_compose_t *cs, slice_buffer * slice_buf, int width, int height, int stride_line, int type, int decomposition_count, int y){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1545 const int support = type==1 ? 3 : 5; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1546 int level; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1547 if(type==2) return; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1548 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1549 for(level=decomposition_count-1; level>=0; level--){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1550 while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1551 switch(type){ |
3326 | 1552 case DWT_97: spatial_compose97i_dy_buffered(dsp, cs+level, slice_buf, width>>level, height>>level, stride_line<<level); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1553 break; |
3326 | 1554 case DWT_53: spatial_compose53i_dy_buffered(cs+level, slice_buf, width>>level, height>>level, stride_line<<level); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1555 break; |
3326 | 1556 case DWT_X: break; |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1557 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1558 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1559 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1560 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1561 |
3075 | 1562 static void ff_spatial_idwt(DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){ |
2562 | 1563 if(type==2){ |
1564 int level; | |
1565 for(level=decomposition_count-1; level>=0; level--) | |
1566 spatial_composeX (buffer, width>>level, height>>level, stride<<level); | |
1567 }else{ | |
1568 dwt_compose_t cs[MAX_DECOMPOSITIONS]; | |
1569 int y; | |
1570 ff_spatial_idwt_init(cs, buffer, width, height, stride, type, decomposition_count); | |
1571 for(y=0; y<height; y+=4) | |
1572 ff_spatial_idwt_slice(cs, buffer, width, height, stride, type, decomposition_count, y); | |
2138 | 1573 } |
1574 } | |
1575 | |
2422 | 1576 static int encode_subband_c0run(SnowContext *s, SubBand *b, DWTELEM *src, DWTELEM *parent, int stride, int orientation){ |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1577 const int w= b->width; |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1578 const int h= b->height; |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1579 int x, y; |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1580 |
2138 | 1581 if(1){ |
1582 int run=0; | |
2149 | 1583 int runs[w*h]; |
2138 | 1584 int run_index=0; |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1585 int max_index; |
2967 | 1586 |
2138 | 1587 for(y=0; y<h; y++){ |
1588 for(x=0; x<w; x++){ | |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1589 int v, p=0; |
2144 | 1590 int /*ll=0, */l=0, lt=0, t=0, rt=0; |
2149 | 1591 v= src[x + y*stride]; |
2138 | 1592 |
1593 if(y){ | |
2149 | 1594 t= src[x + (y-1)*stride]; |
2138 | 1595 if(x){ |
2149 | 1596 lt= src[x - 1 + (y-1)*stride]; |
2138 | 1597 } |
1598 if(x + 1 < w){ | |
2149 | 1599 rt= src[x + 1 + (y-1)*stride]; |
2138 | 1600 } |
1601 } | |
1602 if(x){ | |
2149 | 1603 l= src[x - 1 + y*stride]; |
2144 | 1604 /*if(x > 1){ |
1605 if(orientation==1) ll= src[y + (x-2)*stride]; | |
1606 else ll= src[x - 2 + y*stride]; | |
2138 | 1607 }*/ |
1608 } | |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1609 if(parent){ |
2149 | 1610 int px= x>>1; |
1611 int py= y>>1; | |
2967 | 1612 if(px<b->parent->width && py<b->parent->height) |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1613 p= parent[px + py*2*stride]; |
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1614 } |
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1615 if(!(/*ll|*/l|lt|t|rt|p)){ |
2138 | 1616 if(v){ |
1617 runs[run_index++]= run; | |
1618 run=0; | |
1619 }else{ | |
1620 run++; | |
1621 } | |
1622 } | |
1623 } | |
1624 } | |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1625 max_index= run_index; |
2138 | 1626 runs[run_index++]= run; |
1627 run_index=0; | |
1628 run= runs[run_index++]; | |
1629 | |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1630 put_symbol2(&s->c, b->state[30], max_index, 0); |
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1631 if(run_index <= max_index) |
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1632 put_symbol2(&s->c, b->state[1], run, 3); |
2967 | 1633 |
2138 | 1634 for(y=0; y<h; y++){ |
2435
c89ac0e70c66
10l patch by (matthieu castet <castet.matthieu free fr>)
michael
parents:
2422
diff
changeset
|
1635 if(s->c.bytestream_end - s->c.bytestream < w*40){ |
2422 | 1636 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n"); |
1637 return -1; | |
1638 } | |
2138 | 1639 for(x=0; x<w; x++){ |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1640 int v, p=0; |
2144 | 1641 int /*ll=0, */l=0, lt=0, t=0, rt=0; |
2149 | 1642 v= src[x + y*stride]; |
2138 | 1643 |
1644 if(y){ | |
2149 | 1645 t= src[x + (y-1)*stride]; |
2138 | 1646 if(x){ |
2149 | 1647 lt= src[x - 1 + (y-1)*stride]; |
2138 | 1648 } |
1649 if(x + 1 < w){ | |
2149 | 1650 rt= src[x + 1 + (y-1)*stride]; |
2138 | 1651 } |
1652 } | |
1653 if(x){ | |
2149 | 1654 l= src[x - 1 + y*stride]; |
2144 | 1655 /*if(x > 1){ |
1656 if(orientation==1) ll= src[y + (x-2)*stride]; | |
1657 else ll= src[x - 2 + y*stride]; | |
2138 | 1658 }*/ |
1659 } | |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1660 if(parent){ |
2149 | 1661 int px= x>>1; |
1662 int py= y>>1; | |
2967 | 1663 if(px<b->parent->width && py<b->parent->height) |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1664 p= parent[px + py*2*stride]; |
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1665 } |
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1666 if(/*ll|*/l|lt|t|rt|p){ |
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1667 int context= av_log2(/*ABS(ll) + */3*ABS(l) + ABS(lt) + 2*ABS(t) + ABS(rt) + ABS(p)); |
2144 | 1668 |
2335 | 1669 put_rac(&s->c, &b->state[0][context], !!v); |
2138 | 1670 }else{ |
1671 if(!run){ | |
1672 run= runs[run_index++]; | |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1673 |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1674 if(run_index <= max_index) |
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1675 put_symbol2(&s->c, b->state[1], run, 3); |
2138 | 1676 assert(v); |
1677 }else{ | |
1678 run--; | |
1679 assert(!v); | |
1680 } | |
1681 } | |
1682 if(v){ | |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1683 int context= av_log2(/*ABS(ll) + */3*ABS(l) + ABS(lt) + 2*ABS(t) + ABS(rt) + ABS(p)); |
2596 | 1684 int l2= 2*ABS(l) + (l<0); |
1685 int t2= 2*ABS(t) + (t<0); | |
2144 | 1686 |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
1687 put_symbol2(&s->c, b->state[context + 2], ABS(v)-1, context-4); |
2596 | 1688 put_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l2&0xFF] + 3*quant3bA[t2&0xFF]], v<0); |
2138 | 1689 } |
1690 } | |
1691 } | |
1692 } | |
2422 | 1693 return 0; |
2138 | 1694 } |
1695 | |
2967 | 1696 static int encode_subband(SnowContext *s, SubBand *b, DWTELEM *src, DWTELEM *parent, int stride, int orientation){ |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1697 // encode_subband_qtree(s, b, src, parent, stride, orientation); |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1698 // encode_subband_z0run(s, b, src, parent, stride, orientation); |
2422 | 1699 return encode_subband_c0run(s, b, src, parent, stride, orientation); |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1700 // encode_subband_dzr(s, b, src, parent, stride, orientation); |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1701 } |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1702 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1703 static inline void unpack_coeffs(SnowContext *s, SubBand *b, SubBand * parent, int orientation){ |
2138 | 1704 const int w= b->width; |
1705 const int h= b->height; | |
1706 int x,y; | |
2967 | 1707 |
2138 | 1708 if(1){ |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1709 int run, runs; |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1710 x_and_coeff *xc= b->x_coeff; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1711 x_and_coeff *prev_xc= NULL; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1712 x_and_coeff *prev2_xc= xc; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1713 x_and_coeff *parent_xc= parent ? parent->x_coeff : NULL; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1714 x_and_coeff *prev_parent_xc= parent_xc; |
2138 | 1715 |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1716 runs= get_symbol2(&s->c, b->state[30], 0); |
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1717 if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3); |
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1718 else run= INT_MAX; |
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1719 |
2138 | 1720 for(y=0; y<h; y++){ |
2193 | 1721 int v=0; |
1722 int lt=0, t=0, rt=0; | |
1723 | |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1724 if(y && prev_xc->x == 0){ |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1725 rt= prev_xc->coeff; |
2193 | 1726 } |
2138 | 1727 for(x=0; x<w; x++){ |
2193 | 1728 int p=0; |
1729 const int l= v; | |
2967 | 1730 |
2193 | 1731 lt= t; t= rt; |
1732 | |
2194 | 1733 if(y){ |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1734 if(prev_xc->x <= x) |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1735 prev_xc++; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1736 if(prev_xc->x == x + 1) |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1737 rt= prev_xc->coeff; |
2194 | 1738 else |
1739 rt=0; | |
1740 } | |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1741 if(parent_xc){ |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1742 if(x>>1 > parent_xc->x){ |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1743 parent_xc++; |
2192 | 1744 } |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1745 if(x>>1 == parent_xc->x){ |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1746 p= parent_xc->coeff; |
2194 | 1747 } |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1748 } |
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1749 if(/*ll|*/l|lt|t|rt|p){ |
2596 | 1750 int context= av_log2(/*ABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1)); |
2144 | 1751 |
2335 | 1752 v=get_rac(&s->c, &b->state[0][context]); |
2605 | 1753 if(v){ |
1754 v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1); | |
1755 v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l&0xFF] + 3*quant3bA[t&0xFF]]); | |
2967 | 1756 |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1757 xc->x=x; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1758 (xc++)->coeff= v; |
2605 | 1759 } |
2138 | 1760 }else{ |
1761 if(!run){ | |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1762 if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3); |
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1763 else run= INT_MAX; |
2605 | 1764 v= 2*(get_symbol2(&s->c, b->state[0 + 2], 0-4) + 1); |
1765 v+=get_rac(&s->c, &b->state[0][16 + 1 + 3]); | |
2967 | 1766 |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1767 xc->x=x; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1768 (xc++)->coeff= v; |
2138 | 1769 }else{ |
2606 | 1770 int max_run; |
2138 | 1771 run--; |
1772 v=0; | |
2191 | 1773 |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1774 if(y) max_run= FFMIN(run, prev_xc->x - x - 2); |
2606 | 1775 else max_run= FFMIN(run, w-x-1); |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1776 if(parent_xc) |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1777 max_run= FFMIN(max_run, 2*parent_xc->x - x - 1); |
2606 | 1778 x+= max_run; |
1779 run-= max_run; | |
2138 | 1780 } |
1781 } | |
2192 | 1782 } |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1783 (xc++)->x= w+1; //end marker |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1784 prev_xc= prev2_xc; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1785 prev2_xc= xc; |
2967 | 1786 |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1787 if(parent_xc){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1788 if(y&1){ |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1789 while(parent_xc->x != parent->width+1) |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1790 parent_xc++; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1791 parent_xc++; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1792 prev_parent_xc= parent_xc; |
2192 | 1793 }else{ |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1794 parent_xc= prev_parent_xc; |
2138 | 1795 } |
1796 } | |
1797 } | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1798 |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1799 (xc++)->x= w+1; //end marker |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1800 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1801 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1802 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1803 static inline void decode_subband_slice_buffered(SnowContext *s, SubBand *b, slice_buffer * sb, int start_y, int h, int save_state[1]){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1804 const int w= b->width; |
2893 | 1805 int y; |
2600 | 1806 const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16); |
1807 int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1808 int qadd= (s->qbias*qmul)>>QBIAS_SHIFT; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1809 int new_index = 0; |
2967 | 1810 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1811 START_TIMER |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1812 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1813 if(b->buf == s->spatial_dwt_buffer || s->qlog == LOSSLESS_QLOG){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1814 qadd= 0; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1815 qmul= 1<<QEXPSHIFT; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1816 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1817 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1818 /* If we are on the second or later slice, restore our index. */ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1819 if (start_y != 0) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1820 new_index = save_state[0]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1821 |
2967 | 1822 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1823 for(y=start_y; y<h; y++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1824 int x = 0; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1825 int v; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1826 DWTELEM * line = slice_buffer_get_line(sb, y * b->stride_line + b->buf_y_offset) + b->buf_x_offset; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1827 memset(line, 0, b->width*sizeof(DWTELEM)); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1828 v = b->x_coeff[new_index].coeff; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1829 x = b->x_coeff[new_index++].x; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1830 while(x < w) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1831 { |
2596 | 1832 register int t= ( (v>>1)*qmul + qadd)>>QEXPSHIFT; |
1833 register int u= -(v&1); | |
1834 line[x] = (t^u) - u; | |
1835 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1836 v = b->x_coeff[new_index].coeff; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1837 x = b->x_coeff[new_index++].x; |
2138 | 1838 } |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1839 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1840 if(w > 200 && start_y != 0/*level+1 == s->spatial_decomposition_count*/){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1841 STOP_TIMER("decode_subband") |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1842 } |
2967 | 1843 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1844 /* Save our variables for the next slice. */ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1845 save_state[0] = new_index; |
2967 | 1846 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1847 return; |
2138 | 1848 } |
1849 | |
1850 static void reset_contexts(SnowContext *s){ | |
1851 int plane_index, level, orientation; | |
1852 | |
2199 | 1853 for(plane_index=0; plane_index<3; plane_index++){ |
2138 | 1854 for(level=0; level<s->spatial_decomposition_count; level++){ |
1855 for(orientation=level ? 1:0; orientation<4; orientation++){ | |
2335 | 1856 memset(s->plane[plane_index].band[level][orientation].state, MID_STATE, sizeof(s->plane[plane_index].band[level][orientation].state)); |
2138 | 1857 } |
1858 } | |
1859 } | |
2335 | 1860 memset(s->header_state, MID_STATE, sizeof(s->header_state)); |
1861 memset(s->block_state, MID_STATE, sizeof(s->block_state)); | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1862 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1863 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1864 static int alloc_blocks(SnowContext *s){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1865 int w= -((-s->avctx->width )>>LOG2_MB_SIZE); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1866 int h= -((-s->avctx->height)>>LOG2_MB_SIZE); |
2967 | 1867 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1868 s->b_width = w; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1869 s->b_height= h; |
2967 | 1870 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1871 s->block= av_mallocz(w * h * sizeof(BlockNode) << (s->block_max_depth*2)); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1872 return 0; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1873 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1874 |
2335 | 1875 static inline void copy_rac_state(RangeCoder *d, RangeCoder *s){ |
1876 uint8_t *bytestream= d->bytestream; | |
1877 uint8_t *bytestream_start= d->bytestream_start; | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1878 *d= *s; |
2335 | 1879 d->bytestream= bytestream; |
1880 d->bytestream_start= bytestream_start; | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1881 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1882 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1883 //near copy & paste from dsputil, FIXME |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1884 static int pix_sum(uint8_t * pix, int line_size, int w) |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1885 { |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1886 int s, i, j; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1887 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1888 s = 0; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1889 for (i = 0; i < w; i++) { |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1890 for (j = 0; j < w; j++) { |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1891 s += pix[0]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1892 pix ++; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1893 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1894 pix += line_size - w; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1895 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1896 return s; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1897 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1898 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1899 //near copy & paste from dsputil, FIXME |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1900 static int pix_norm1(uint8_t * pix, int line_size, int w) |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1901 { |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1902 int s, i, j; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1903 uint32_t *sq = squareTbl + 256; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1904 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1905 s = 0; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1906 for (i = 0; i < w; i++) { |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1907 for (j = 0; j < w; j ++) { |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1908 s += sq[pix[0]]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1909 pix ++; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1910 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1911 pix += line_size - w; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1912 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1913 return s; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1914 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1915 |
3314 | 1916 static inline void set_blocks(SnowContext *s, int level, int x, int y, int l, int cb, int cr, int mx, int my, int ref, int type){ |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1917 const int w= s->b_width << s->block_max_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1918 const int rem_depth= s->block_max_depth - level; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1919 const int index= (x + y*w) << rem_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1920 const int block_w= 1<<rem_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1921 BlockNode block; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1922 int i,j; |
2967 | 1923 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1924 block.color[0]= l; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1925 block.color[1]= cb; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1926 block.color[2]= cr; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1927 block.mx= mx; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1928 block.my= my; |
3314 | 1929 block.ref= ref; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1930 block.type= type; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1931 block.level= level; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1932 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1933 for(j=0; j<block_w; j++){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1934 for(i=0; i<block_w; i++){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1935 s->block[index + i + j*w]= block; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1936 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1937 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1938 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1939 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1940 static inline void init_ref(MotionEstContext *c, uint8_t *src[3], uint8_t *ref[3], uint8_t *ref2[3], int x, int y, int ref_index){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1941 const int offset[3]= { |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1942 y*c-> stride + x, |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1943 ((y*c->uvstride + x)>>1), |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1944 ((y*c->uvstride + x)>>1), |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1945 }; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1946 int i; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1947 for(i=0; i<3; i++){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1948 c->src[0][i]= src [i]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1949 c->ref[0][i]= ref [i] + offset[i]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1950 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1951 assert(!ref_index); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1952 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1953 |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1954 static inline void pred_mv(SnowContext *s, int *mx, int *my, int ref, |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1955 BlockNode *left, BlockNode *top, BlockNode *tr){ |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1956 if(s->ref_frames == 1){ |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1957 *mx = mid_pred(left->mx, top->mx, tr->mx); |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1958 *my = mid_pred(left->my, top->my, tr->my); |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1959 }else{ |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1960 const int *scale = scale_mv_ref[ref]; |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1961 *mx = mid_pred(left->mx * scale[left->ref] + 128 >>8, |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1962 top ->mx * scale[top ->ref] + 128 >>8, |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1963 tr ->mx * scale[tr ->ref] + 128 >>8); |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1964 *my = mid_pred(left->my * scale[left->ref] + 128 >>8, |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1965 top ->my * scale[top ->ref] + 128 >>8, |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1966 tr ->my * scale[tr ->ref] + 128 >>8); |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1967 } |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1968 } |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1969 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1970 //FIXME copy&paste |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1971 #define P_LEFT P[1] |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1972 #define P_TOP P[2] |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1973 #define P_TOPRIGHT P[3] |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1974 #define P_MEDIAN P[4] |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1975 #define P_MV1 P[9] |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1976 #define FLAG_QPEL 1 //must be 1 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1977 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1978 static int encode_q_branch(SnowContext *s, int level, int x, int y){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1979 uint8_t p_buffer[1024]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1980 uint8_t i_buffer[1024]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1981 uint8_t p_state[sizeof(s->block_state)]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1982 uint8_t i_state[sizeof(s->block_state)]; |
2335 | 1983 RangeCoder pc, ic; |
1984 uint8_t *pbbak= s->c.bytestream; | |
1985 uint8_t *pbbak_start= s->c.bytestream_start; | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1986 int score, score2, iscore, i_len, p_len, block_s, sum; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1987 const int w= s->b_width << s->block_max_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1988 const int h= s->b_height << s->block_max_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1989 const int rem_depth= s->block_max_depth - level; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1990 const int index= (x + y*w) << rem_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1991 const int block_w= 1<<(LOG2_MB_SIZE - level); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1992 int trx= (x+1)<<rem_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1993 int try= (y+1)<<rem_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1994 BlockNode *left = x ? &s->block[index-1] : &null_block; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1995 BlockNode *top = y ? &s->block[index-w] : &null_block; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1996 BlockNode *right = trx<w ? &s->block[index+1] : &null_block; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1997 BlockNode *bottom= try<h ? &s->block[index+w] : &null_block; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1998 BlockNode *tl = y && x ? &s->block[index-w-1] : left; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1999 BlockNode *tr = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2000 int pl = left->color[0]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2001 int pcb= left->color[1]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2002 int pcr= left->color[2]; |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2003 int pmx, pmy; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2004 int mx=0, my=0; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2005 int l,cr,cb; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2006 const int stride= s->current_picture.linesize[0]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2007 const int uvstride= s->current_picture.linesize[1]; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2008 uint8_t *current_data[3]= { s->input_picture.data[0] + (x + y* stride)*block_w, |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2009 s->input_picture.data[1] + (x + y*uvstride)*block_w/2, |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2010 s->input_picture.data[2] + (x + y*uvstride)*block_w/2}; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2011 int P[10][2]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2012 int16_t last_mv[3][2]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2013 int qpel= !!(s->avctx->flags & CODEC_FLAG_QPEL); //unused |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2014 const int shift= 1+qpel; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2015 MotionEstContext *c= &s->m.me; |
3314 | 2016 int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2017 int mx_context= av_log2(2*ABS(left->mx - top->mx)); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2018 int my_context= av_log2(2*ABS(left->my - top->my)); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2019 int s_context= 2*left->level + 2*top->level + tl->level + tr->level; |
3314 | 2020 int ref, best_ref, ref_score, ref_mx, ref_my; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2021 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2022 assert(sizeof(s->block_state) >= 256); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2023 if(s->keyframe){ |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2024 set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2025 return 0; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2026 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2027 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2028 // clip predictors / edge ? |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2029 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2030 P_LEFT[0]= left->mx; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2031 P_LEFT[1]= left->my; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2032 P_TOP [0]= top->mx; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2033 P_TOP [1]= top->my; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2034 P_TOPRIGHT[0]= tr->mx; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2035 P_TOPRIGHT[1]= tr->my; |
2967 | 2036 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2037 last_mv[0][0]= s->block[index].mx; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2038 last_mv[0][1]= s->block[index].my; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2039 last_mv[1][0]= right->mx; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2040 last_mv[1][1]= right->my; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2041 last_mv[2][0]= bottom->mx; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2042 last_mv[2][1]= bottom->my; |
2967 | 2043 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2044 s->m.mb_stride=2; |
2967 | 2045 s->m.mb_x= |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2046 s->m.mb_y= 0; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2047 s->m.me.skip= 0; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2048 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2049 assert(s->m.me. stride == stride); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2050 assert(s->m.me.uvstride == uvstride); |
2967 | 2051 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2052 c->penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2053 c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2054 c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2055 c->current_mv_penalty= c->mv_penalty[s->m.f_code=1] + MAX_MV; |
2967 | 2056 |
2206 | 2057 c->xmin = - x*block_w - 16+2; |
2058 c->ymin = - y*block_w - 16+2; | |
2059 c->xmax = - (x+1)*block_w + (w<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-2; | |
2060 c->ymax = - (y+1)*block_w + (h<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-2; | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2061 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2062 if(P_LEFT[0] > (c->xmax<<shift)) P_LEFT[0] = (c->xmax<<shift); |
2967 | 2063 if(P_LEFT[1] > (c->ymax<<shift)) P_LEFT[1] = (c->ymax<<shift); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2064 if(P_TOP[0] > (c->xmax<<shift)) P_TOP[0] = (c->xmax<<shift); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2065 if(P_TOP[1] > (c->ymax<<shift)) P_TOP[1] = (c->ymax<<shift); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2066 if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2067 if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift); //due to pmx no clip |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2068 if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2069 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2070 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2071 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2072 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2073 if (!y) { |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2074 c->pred_x= P_LEFT[0]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2075 c->pred_y= P_LEFT[1]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2076 } else { |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2077 c->pred_x = P_MEDIAN[0]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2078 c->pred_y = P_MEDIAN[1]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2079 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2080 |
3314 | 2081 score= INT_MAX; |
2082 best_ref= 0; | |
2083 for(ref=0; ref<s->ref_frames; ref++){ | |
2084 init_ref(c, current_data, s->last_picture[ref].data, NULL, block_w*x, block_w*y, 0); | |
2085 | |
2086 ref_score= ff_epzs_motion_search(&s->m, &ref_mx, &ref_my, P, 0, /*ref_index*/ 0, last_mv, | |
2087 (1<<16)>>shift, level-LOG2_MB_SIZE+4, block_w); | |
2088 | |
2089 assert(ref_mx >= c->xmin); | |
2090 assert(ref_mx <= c->xmax); | |
2091 assert(ref_my >= c->ymin); | |
2092 assert(ref_my <= c->ymax); | |
2093 | |
2094 ref_score= s->m.me.sub_motion_search(&s->m, &ref_mx, &ref_my, ref_score, 0, 0, level-LOG2_MB_SIZE+4, block_w); | |
2095 ref_score= ff_get_mb_score(&s->m, ref_mx, ref_my, 0, 0, level-LOG2_MB_SIZE+4, block_w, 0); | |
2096 ref_score+= 2*av_log2(2*ref)*c->penalty_factor; | |
2097 if(s->ref_mvs[ref]){ | |
2098 s->ref_mvs[ref][index][0]= ref_mx; | |
2099 s->ref_mvs[ref][index][1]= ref_my; | |
2100 s->ref_scores[ref][index]= ref_score; | |
2101 } | |
2102 if(score > ref_score){ | |
2103 score= ref_score; | |
2104 best_ref= ref; | |
2105 mx= ref_mx; | |
2106 my= ref_my; | |
2107 } | |
2108 } | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2109 //FIXME if mb_cmp != SSE then intra cant be compared currently and mb_penalty vs. lambda2 |
2967 | 2110 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2111 // subpel search |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2112 pc= s->c; |
2335 | 2113 pc.bytestream_start= |
2114 pc.bytestream= p_buffer; //FIXME end/start? and at the other stoo | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2115 memcpy(p_state, s->block_state, sizeof(s->block_state)); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2116 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2117 if(level!=s->block_max_depth) |
2335 | 2118 put_rac(&pc, &p_state[4 + s_context], 1); |
2119 put_rac(&pc, &p_state[1 + left->type + top->type], 0); | |
3314 | 2120 if(s->ref_frames > 1) |
2121 put_symbol(&pc, &p_state[128 + 1024 + 32*ref_context], best_ref, 0); | |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2122 pred_mv(s, &pmx, &pmy, best_ref, left, top, tr); |
3314 | 2123 put_symbol(&pc, &p_state[128 + 32*(mx_context + 16*!!best_ref)], mx - pmx, 1); |
2124 put_symbol(&pc, &p_state[128 + 32*(my_context + 16*!!best_ref)], my - pmy, 1); | |
2335 | 2125 p_len= pc.bytestream - pc.bytestream_start; |
2126 score += (s->lambda2*(p_len*8 | |
2127 + (pc.outstanding_count - s->c.outstanding_count)*8 | |
2128 + (-av_log2(pc.range) + av_log2(s->c.range)) | |
2129 ))>>FF_LAMBDA_SHIFT; | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2130 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2131 block_s= block_w*block_w; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2132 sum = pix_sum(current_data[0], stride, block_w); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2133 l= (sum + block_s/2)/block_s; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2134 iscore = pix_norm1(current_data[0], stride, block_w) - 2*l*sum + l*l*block_s; |
2967 | 2135 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2136 block_s= block_w*block_w>>2; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2137 sum = pix_sum(current_data[1], uvstride, block_w>>1); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2138 cb= (sum + block_s/2)/block_s; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2139 // iscore += pix_norm1(¤t_mb[1][0], uvstride, block_w>>1) - 2*cb*sum + cb*cb*block_s; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2140 sum = pix_sum(current_data[2], uvstride, block_w>>1); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2141 cr= (sum + block_s/2)/block_s; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2142 // iscore += pix_norm1(¤t_mb[2][0], uvstride, block_w>>1) - 2*cr*sum + cr*cr*block_s; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2143 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2144 ic= s->c; |
2335 | 2145 ic.bytestream_start= |
2146 ic.bytestream= i_buffer; //FIXME end/start? and at the other stoo | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2147 memcpy(i_state, s->block_state, sizeof(s->block_state)); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2148 if(level!=s->block_max_depth) |
2335 | 2149 put_rac(&ic, &i_state[4 + s_context], 1); |
2150 put_rac(&ic, &i_state[1 + left->type + top->type], 1); | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2151 put_symbol(&ic, &i_state[32], l-pl , 1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2152 put_symbol(&ic, &i_state[64], cb-pcb, 1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2153 put_symbol(&ic, &i_state[96], cr-pcr, 1); |
2335 | 2154 i_len= ic.bytestream - ic.bytestream_start; |
2155 iscore += (s->lambda2*(i_len*8 | |
2156 + (ic.outstanding_count - s->c.outstanding_count)*8 | |
2157 + (-av_log2(ic.range) + av_log2(s->c.range)) | |
2158 ))>>FF_LAMBDA_SHIFT; | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2159 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2160 // assert(score==256*256*256*64-1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2161 assert(iscore < 255*255*256 + s->lambda2*10); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2162 assert(iscore >= 0); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2163 assert(l>=0 && l<=255); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2164 assert(pl>=0 && pl<=255); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2165 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2166 if(level==0){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2167 int varc= iscore >> 8; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2168 int vard= score >> 8; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2169 if (vard <= 64 || vard < varc) |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2170 c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2171 else |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2172 c->scene_change_score+= s->m.qscale; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2173 } |
2967 | 2174 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2175 if(level!=s->block_max_depth){ |
2335 | 2176 put_rac(&s->c, &s->block_state[4 + s_context], 0); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2177 score2 = encode_q_branch(s, level+1, 2*x+0, 2*y+0); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2178 score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+0); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2179 score2+= encode_q_branch(s, level+1, 2*x+0, 2*y+1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2180 score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2181 score2+= s->lambda2>>FF_LAMBDA_SHIFT; //FIXME exact split overhead |
2967 | 2182 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2183 if(score2 < score && score2 < iscore) |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2184 return score2; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2185 } |
2967 | 2186 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2187 if(iscore < score){ |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2188 pred_mv(s, &pmx, &pmy, 0, left, top, tr); |
2335 | 2189 memcpy(pbbak, i_buffer, i_len); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2190 s->c= ic; |
2335 | 2191 s->c.bytestream_start= pbbak_start; |
2192 s->c.bytestream= pbbak + i_len; | |
3314 | 2193 set_blocks(s, level, x, y, l, cb, cr, pmx, pmy, 0, BLOCK_INTRA); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2194 memcpy(s->block_state, i_state, sizeof(s->block_state)); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2195 return iscore; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2196 }else{ |
2335 | 2197 memcpy(pbbak, p_buffer, p_len); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2198 s->c= pc; |
2335 | 2199 s->c.bytestream_start= pbbak_start; |
2200 s->c.bytestream= pbbak + p_len; | |
3314 | 2201 set_blocks(s, level, x, y, pl, pcb, pcr, mx, my, best_ref, 0); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2202 memcpy(s->block_state, p_state, sizeof(s->block_state)); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2203 return score; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2204 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2205 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2206 |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2207 static always_inline int same_block(BlockNode *a, BlockNode *b){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2208 if((a->type&BLOCK_INTRA) && (b->type&BLOCK_INTRA)){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2209 return !((a->color[0] - b->color[0]) | (a->color[1] - b->color[1]) | (a->color[2] - b->color[2])); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2210 }else{ |
3314 | 2211 return !((a->mx - b->mx) | (a->my - b->my) | (a->ref - b->ref) | ((a->type ^ b->type)&BLOCK_INTRA)); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2212 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2213 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2214 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2215 static void encode_q_branch2(SnowContext *s, int level, int x, int y){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2216 const int w= s->b_width << s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2217 const int rem_depth= s->block_max_depth - level; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2218 const int index= (x + y*w) << rem_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2219 int trx= (x+1)<<rem_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2220 BlockNode *b= &s->block[index]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2221 BlockNode *left = x ? &s->block[index-1] : &null_block; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2222 BlockNode *top = y ? &s->block[index-w] : &null_block; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2223 BlockNode *tl = y && x ? &s->block[index-w-1] : left; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2224 BlockNode *tr = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2225 int pl = left->color[0]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2226 int pcb= left->color[1]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2227 int pcr= left->color[2]; |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2228 int pmx, pmy; |
3314 | 2229 int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref); |
2230 int mx_context= av_log2(2*ABS(left->mx - top->mx)) + 16*!!b->ref; | |
2231 int my_context= av_log2(2*ABS(left->my - top->my)) + 16*!!b->ref; | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2232 int s_context= 2*left->level + 2*top->level + tl->level + tr->level; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2233 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2234 if(s->keyframe){ |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2235 set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2236 return; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2237 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2238 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2239 if(level!=s->block_max_depth){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2240 if(same_block(b,b+1) && same_block(b,b+w) && same_block(b,b+w+1)){ |
2995
dfc271b90fe6
4mv + iter ME 10l fix (still not bugfree but better)
michael
parents:
2994
diff
changeset
|
2241 put_rac(&s->c, &s->block_state[4 + s_context], 1); |
dfc271b90fe6
4mv + iter ME 10l fix (still not bugfree but better)
michael
parents:
2994
diff
changeset
|
2242 }else{ |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2243 put_rac(&s->c, &s->block_state[4 + s_context], 0); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2244 encode_q_branch2(s, level+1, 2*x+0, 2*y+0); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2245 encode_q_branch2(s, level+1, 2*x+1, 2*y+0); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2246 encode_q_branch2(s, level+1, 2*x+0, 2*y+1); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2247 encode_q_branch2(s, level+1, 2*x+1, 2*y+1); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2248 return; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2249 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2250 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2251 if(b->type & BLOCK_INTRA){ |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2252 pred_mv(s, &pmx, &pmy, 0, left, top, tr); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2253 put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 1); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2254 put_symbol(&s->c, &s->block_state[32], b->color[0]-pl , 1); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2255 put_symbol(&s->c, &s->block_state[64], b->color[1]-pcb, 1); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2256 put_symbol(&s->c, &s->block_state[96], b->color[2]-pcr, 1); |
3314 | 2257 set_blocks(s, level, x, y, b->color[0], b->color[1], b->color[2], pmx, pmy, 0, BLOCK_INTRA); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2258 }else{ |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2259 pred_mv(s, &pmx, &pmy, b->ref, left, top, tr); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2260 put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 0); |
3314 | 2261 if(s->ref_frames > 1) |
2262 put_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], b->ref, 0); | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2263 put_symbol(&s->c, &s->block_state[128 + 32*mx_context], b->mx - pmx, 1); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2264 put_symbol(&s->c, &s->block_state[128 + 32*my_context], b->my - pmy, 1); |
3314 | 2265 set_blocks(s, level, x, y, pl, pcb, pcr, b->mx, b->my, b->ref, 0); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2266 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2267 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2268 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2269 static void decode_q_branch(SnowContext *s, int level, int x, int y){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2270 const int w= s->b_width << s->block_max_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2271 const int rem_depth= s->block_max_depth - level; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2272 const int index= (x + y*w) << rem_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2273 int trx= (x+1)<<rem_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2274 BlockNode *left = x ? &s->block[index-1] : &null_block; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2275 BlockNode *top = y ? &s->block[index-w] : &null_block; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2276 BlockNode *tl = y && x ? &s->block[index-w-1] : left; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2277 BlockNode *tr = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2278 int s_context= 2*left->level + 2*top->level + tl->level + tr->level; |
2967 | 2279 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2280 if(s->keyframe){ |
3314 | 2281 set_blocks(s, level, x, y, null_block.color[0], null_block.color[1], null_block.color[2], null_block.mx, null_block.my, null_block.ref, BLOCK_INTRA); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2282 return; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2283 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2284 |
2335 | 2285 if(level==s->block_max_depth || get_rac(&s->c, &s->block_state[4 + s_context])){ |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2286 int type; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2287 int l = left->color[0]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2288 int cb= left->color[1]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2289 int cr= left->color[2]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2290 int mx= mid_pred(left->mx, top->mx, tr->mx); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2291 int my= mid_pred(left->my, top->my, tr->my); |
3314 | 2292 int ref = 0; |
2293 int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref); | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2294 int mx_context= av_log2(2*ABS(left->mx - top->mx)) + 0*av_log2(2*ABS(tr->mx - top->mx)); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2295 int my_context= av_log2(2*ABS(left->my - top->my)) + 0*av_log2(2*ABS(tr->my - top->my)); |
2967 | 2296 |
2335 | 2297 type= get_rac(&s->c, &s->block_state[1 + left->type + top->type]) ? BLOCK_INTRA : 0; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2298 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2299 if(type){ |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2300 pred_mv(s, &mx, &my, 0, left, top, tr); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2301 l += get_symbol(&s->c, &s->block_state[32], 1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2302 cb+= get_symbol(&s->c, &s->block_state[64], 1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2303 cr+= get_symbol(&s->c, &s->block_state[96], 1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2304 }else{ |
3314 | 2305 if(s->ref_frames > 1) |
2306 ref= get_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], 0); | |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2307 pred_mv(s, &mx, &my, ref, left, top, tr); |
3314 | 2308 mx+= get_symbol(&s->c, &s->block_state[128 + 32*(mx_context + 16*!!ref)], 1); |
2309 my+= get_symbol(&s->c, &s->block_state[128 + 32*(my_context + 16*!!ref)], 1); | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2310 } |
3314 | 2311 set_blocks(s, level, x, y, l, cb, cr, mx, my, ref, type); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2312 }else{ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2313 decode_q_branch(s, level+1, 2*x+0, 2*y+0); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2314 decode_q_branch(s, level+1, 2*x+1, 2*y+0); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2315 decode_q_branch(s, level+1, 2*x+0, 2*y+1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2316 decode_q_branch(s, level+1, 2*x+1, 2*y+1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2317 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2318 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2319 |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
2320 static void encode_blocks(SnowContext *s, int search){ |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2321 int x, y; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2322 int w= s->b_width; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2323 int h= s->b_height; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2324 |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
2325 if(s->avctx->me_method == ME_ITER && !s->keyframe && search) |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2326 iterative_me(s); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2327 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2328 for(y=0; y<h; y++){ |
2435
c89ac0e70c66
10l patch by (matthieu castet <castet.matthieu free fr>)
michael
parents:
2422
diff
changeset
|
2329 if(s->c.bytestream_end - s->c.bytestream < w*MB_SIZE*MB_SIZE*3){ //FIXME nicer limit |
2422 | 2330 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n"); |
2331 return; | |
2332 } | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2333 for(x=0; x<w; x++){ |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
2334 if(s->avctx->me_method == ME_ITER || !search) |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2335 encode_q_branch2(s, 0, x, y); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2336 else |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2337 encode_q_branch (s, 0, x, y); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2338 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2339 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2340 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2341 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2342 static void decode_blocks(SnowContext *s){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2343 int x, y; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2344 int w= s->b_width; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2345 int h= s->b_height; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2346 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2347 for(y=0; y<h; y++){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2348 for(x=0; x<w; x++){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2349 decode_q_branch(s, 0, x, y); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2350 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2351 } |
2138 | 2352 } |
2353 | |
2354 static void mc_block(uint8_t *dst, uint8_t *src, uint8_t *tmp, int stride, int b_w, int b_h, int dx, int dy){ | |
2355 int x, y; | |
2221 | 2356 START_TIMER |
2138 | 2357 for(y=0; y < b_h+5; y++){ |
2358 for(x=0; x < b_w; x++){ | |
2221 | 2359 int a0= src[x ]; |
2360 int a1= src[x + 1]; | |
2361 int a2= src[x + 2]; | |
2362 int a3= src[x + 3]; | |
2363 int a4= src[x + 4]; | |
2364 int a5= src[x + 5]; | |
2138 | 2365 // int am= 9*(a1+a2) - (a0+a3); |
2366 int am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5); | |
2367 // int am= 18*(a2+a3) - 2*(a1+a4); | |
2368 // int aL= (-7*a0 + 105*a1 + 35*a2 - 5*a3)>>3; | |
2369 // int aR= (-7*a3 + 105*a2 + 35*a1 - 5*a0)>>3; | |
2370 | |
2371 // if(b_w==16) am= 8*(a1+a2); | |
2372 | |
2610
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2373 if(dx<8) am = (32*a2*( 8-dx) + am* dx + 128)>>8; |
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2374 else am = ( am*(16-dx) + 32*a3*(dx-8) + 128)>>8; |
2967 | 2375 |
2610
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2376 /* FIXME Try increasing tmp buffer to 16 bits and not clipping here. Should give marginally better results. - Robert*/ |
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2377 if(am&(~255)) am= ~(am>>31); |
2967 | 2378 |
2610
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2379 tmp[x] = am; |
2138 | 2380 |
2381 /* if (dx< 4) tmp[x + y*stride]= (16*a1*( 4-dx) + aL* dx + 32)>>6; | |
2382 else if(dx< 8) tmp[x + y*stride]= ( aL*( 8-dx) + am*(dx- 4) + 32)>>6; | |
2383 else if(dx<12) tmp[x + y*stride]= ( am*(12-dx) + aR*(dx- 8) + 32)>>6; | |
2384 else tmp[x + y*stride]= ( aR*(16-dx) + 16*a2*(dx-12) + 32)>>6;*/ | |
2385 } | |
2221 | 2386 tmp += stride; |
2387 src += stride; | |
2138 | 2388 } |
2221 | 2389 tmp -= (b_h+5)*stride; |
2967 | 2390 |
2138 | 2391 for(y=0; y < b_h; y++){ |
2392 for(x=0; x < b_w; x++){ | |
2221 | 2393 int a0= tmp[x + 0*stride]; |
2394 int a1= tmp[x + 1*stride]; | |
2395 int a2= tmp[x + 2*stride]; | |
2396 int a3= tmp[x + 3*stride]; | |
2397 int a4= tmp[x + 4*stride]; | |
2398 int a5= tmp[x + 5*stride]; | |
2138 | 2399 int am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5); |
2400 // int am= 18*(a2+a3) - 2*(a1+a4); | |
2401 /* int aL= (-7*a0 + 105*a1 + 35*a2 - 5*a3)>>3; | |
2402 int aR= (-7*a3 + 105*a2 + 35*a1 - 5*a0)>>3;*/ | |
2967 | 2403 |
2138 | 2404 // if(b_w==16) am= 8*(a1+a2); |
2405 | |
2610
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2406 if(dy<8) am = (32*a2*( 8-dy) + am* dy + 128)>>8; |
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2407 else am = ( am*(16-dy) + 32*a3*(dy-8) + 128)>>8; |
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2408 |
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2409 if(am&(~255)) am= ~(am>>31); |
2967 | 2410 |
2610
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2411 dst[x] = am; |
2138 | 2412 /* if (dy< 4) tmp[x + y*stride]= (16*a1*( 4-dy) + aL* dy + 32)>>6; |
2413 else if(dy< 8) tmp[x + y*stride]= ( aL*( 8-dy) + am*(dy- 4) + 32)>>6; | |
2414 else if(dy<12) tmp[x + y*stride]= ( am*(12-dy) + aR*(dy- 8) + 32)>>6; | |
2415 else tmp[x + y*stride]= ( aR*(16-dy) + 16*a2*(dy-12) + 32)>>6;*/ | |
2416 } | |
2221 | 2417 dst += stride; |
2418 tmp += stride; | |
2138 | 2419 } |
2221 | 2420 STOP_TIMER("mc_block") |
2138 | 2421 } |
2422 | |
2423 #define mca(dx,dy,b_w)\ | |
2224
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
2424 static void mc_block_hpel ## dx ## dy ## b_w(uint8_t *dst, uint8_t *src, int stride, int h){\ |
2138 | 2425 uint8_t tmp[stride*(b_w+5)];\ |
2426 assert(h==b_w);\ | |
2427 mc_block(dst, src-2-2*stride, tmp, stride, b_w, b_w, dx, dy);\ | |
2428 } | |
2429 | |
2430 mca( 0, 0,16) | |
2431 mca( 8, 0,16) | |
2432 mca( 0, 8,16) | |
2433 mca( 8, 8,16) | |
2224
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
2434 mca( 0, 0,8) |
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
2435 mca( 8, 0,8) |
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
2436 mca( 0, 8,8) |
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
2437 mca( 8, 8,8) |
2138 | 2438 |
3314 | 2439 static void pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, int stride, int sx, int sy, int b_w, int b_h, BlockNode *block, int plane_index, int w, int h){ |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2440 if(block->type & BLOCK_INTRA){ |
2206 | 2441 int x, y; |
3018 | 2442 const int color = block->color[plane_index]; |
2443 const int color4= color*0x01010101; | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2444 if(b_w==32){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2445 for(y=0; y < b_h; y++){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2446 *(uint32_t*)&dst[0 + y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2447 *(uint32_t*)&dst[4 + y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2448 *(uint32_t*)&dst[8 + y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2449 *(uint32_t*)&dst[12+ y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2450 *(uint32_t*)&dst[16+ y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2451 *(uint32_t*)&dst[20+ y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2452 *(uint32_t*)&dst[24+ y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2453 *(uint32_t*)&dst[28+ y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2454 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2455 }else if(b_w==16){ |
3018 | 2456 for(y=0; y < b_h; y++){ |
2457 *(uint32_t*)&dst[0 + y*stride]= color4; | |
2458 *(uint32_t*)&dst[4 + y*stride]= color4; | |
2459 *(uint32_t*)&dst[8 + y*stride]= color4; | |
2460 *(uint32_t*)&dst[12+ y*stride]= color4; | |
2461 } | |
2462 }else if(b_w==8){ | |
2463 for(y=0; y < b_h; y++){ | |
2464 *(uint32_t*)&dst[0 + y*stride]= color4; | |
2465 *(uint32_t*)&dst[4 + y*stride]= color4; | |
2466 } | |
2467 }else if(b_w==4){ | |
2468 for(y=0; y < b_h; y++){ | |
2469 *(uint32_t*)&dst[0 + y*stride]= color4; | |
2470 } | |
2471 }else{ | |
2472 for(y=0; y < b_h; y++){ | |
2473 for(x=0; x < b_w; x++){ | |
2474 dst[x + y*stride]= color; | |
2475 } | |
2138 | 2476 } |
2477 } | |
2478 }else{ | |
3314 | 2479 uint8_t *src= s->last_picture[block->ref].data[plane_index]; |
2206 | 2480 const int scale= plane_index ? s->mv_scale : 2*s->mv_scale; |
2481 int mx= block->mx*scale; | |
2482 int my= block->my*scale; | |
2223 | 2483 const int dx= mx&15; |
2484 const int dy= my&15; | |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3018
diff
changeset
|
2485 const int tab_index= 3 - (b_w>>2) + (b_w>>4); |
2206 | 2486 sx += (mx>>4) - 2; |
2487 sy += (my>>4) - 2; | |
2488 src += sx + sy*stride; | |
2489 if( (unsigned)sx >= w - b_w - 4 | |
2490 || (unsigned)sy >= h - b_h - 4){ | |
2491 ff_emulated_edge_mc(tmp + MB_SIZE, src, stride, b_w+5, b_h+5, sx, sy, w, h); | |
2492 src= tmp + MB_SIZE; | |
2138 | 2493 } |
3189 | 2494 // assert(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h); |
2495 // assert(!(b_w&(b_w-1))); | |
3018 | 2496 assert(b_w>1 && b_h>1); |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2497 assert(tab_index>=0 && tab_index<4 || b_w==32); |
3189 | 2498 if((dx&3) || (dy&3) || !(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h) || (b_w&(b_w-1))) |
2223 | 2499 mc_block(dst, src, tmp, stride, b_w, b_h, dx, dy); |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2500 else if(b_w==32){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2501 int y; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2502 for(y=0; y<b_h; y+=16){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2503 s->dsp.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + y*stride, src + 2 + (y+2)*stride,stride); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2504 s->dsp.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + 16 + y*stride, src + 18 + (y+2)*stride,stride); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2505 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2506 }else if(b_w==b_h) |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3018
diff
changeset
|
2507 s->dsp.put_h264_qpel_pixels_tab[tab_index ][dy+(dx>>2)](dst,src + 2 + 2*stride,stride); |
3018 | 2508 else if(b_w==2*b_h){ |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3018
diff
changeset
|
2509 s->dsp.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst ,src + 2 + 2*stride,stride); |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3018
diff
changeset
|
2510 s->dsp.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst+b_h,src + 2 + b_h + 2*stride,stride); |
3018 | 2511 }else{ |
2512 assert(2*b_w==b_h); | |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3018
diff
changeset
|
2513 s->dsp.put_h264_qpel_pixels_tab[tab_index ][dy+(dx>>2)](dst ,src + 2 + 2*stride ,stride); |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3018
diff
changeset
|
2514 s->dsp.put_h264_qpel_pixels_tab[tab_index ][dy+(dx>>2)](dst+b_w*stride,src + 2 + 2*stride+b_w*stride,stride); |
3018 | 2515 } |
2138 | 2516 } |
2517 } | |
2518 | |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2519 void ff_snow_inner_add_yblock(uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2520 int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){ |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2521 int y, x; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2522 DWTELEM * dst; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2523 for(y=0; y<b_h; y++){ |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2524 //FIXME ugly missue of obmc_stride |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2525 uint8_t *obmc1= obmc + y*obmc_stride; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2526 uint8_t *obmc2= obmc1+ (obmc_stride>>1); |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2527 uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1); |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2528 uint8_t *obmc4= obmc3+ (obmc_stride>>1); |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2529 dst = slice_buffer_get_line(sb, src_y + y); |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2530 for(x=0; x<b_w; x++){ |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2531 int v= obmc1[x] * block[3][x + y*src_stride] |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2532 +obmc2[x] * block[2][x + y*src_stride] |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2533 +obmc3[x] * block[1][x + y*src_stride] |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2534 +obmc4[x] * block[0][x + y*src_stride]; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2535 |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2536 v <<= 8 - LOG2_OBMC_MAX; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2537 if(FRAC_BITS != 8){ |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2538 v += 1<<(7 - FRAC_BITS); |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2539 v >>= 8 - FRAC_BITS; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2540 } |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2541 if(add){ |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2542 v += dst[x + src_x]; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2543 v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2544 if(v&(~255)) v= ~(v>>31); |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2545 dst8[x + y*src_stride] = v; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2546 }else{ |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2547 dst[x + src_x] -= v; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2548 } |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2549 } |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2550 } |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2551 } |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2552 |
2206 | 2553 //FIXME name clenup (b_w, block_w, b_width stuff) |
3662
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2554 static always_inline void add_yblock(SnowContext *s, int sliced, slice_buffer *sb, DWTELEM *dst, uint8_t *dst8, const uint8_t *obmc, int src_x, int src_y, int b_w, int b_h, int w, int h, int dst_stride, int src_stride, int obmc_stride, int b_x, int b_y, int add, int offset_dst, int plane_index){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2555 const int b_width = s->b_width << s->block_max_depth; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2556 const int b_height= s->b_height << s->block_max_depth; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2557 const int b_stride= b_width; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2558 BlockNode *lt= &s->block[b_x + b_y*b_stride]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2559 BlockNode *rt= lt+1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2560 BlockNode *lb= lt+b_stride; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2561 BlockNode *rb= lb+1; |
2967 | 2562 uint8_t *block[4]; |
2842 | 2563 int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride; |
2564 uint8_t tmp[src_stride*7*MB_SIZE]; //FIXME align | |
2565 uint8_t *ptmp; | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2566 int x,y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2567 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2568 if(b_x<0){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2569 lt= rt; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2570 lb= rb; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2571 }else if(b_x + 1 >= b_width){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2572 rt= lt; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2573 rb= lb; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2574 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2575 if(b_y<0){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2576 lt= lb; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2577 rt= rb; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2578 }else if(b_y + 1 >= b_height){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2579 lb= lt; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2580 rb= rt; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2581 } |
2967 | 2582 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2583 if(src_x<0){ //FIXME merge with prev & always round internal width upto *16 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2584 obmc -= src_x; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2585 b_w += src_x; |
3662
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2586 if(!sliced && !offset_dst) |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2587 dst -= src_x; |
2206 | 2588 src_x=0; |
2589 }else if(src_x + b_w > w){ | |
2590 b_w = w - src_x; | |
2591 } | |
2592 if(src_y<0){ | |
2593 obmc -= src_y*obmc_stride; | |
2594 b_h += src_y; | |
3662
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2595 if(!sliced && !offset_dst) |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2596 dst -= src_y*dst_stride; |
2206 | 2597 src_y=0; |
2598 }else if(src_y + b_h> h){ | |
2599 b_h = h - src_y; | |
2600 } | |
2967 | 2601 |
2206 | 2602 if(b_w<=0 || b_h<=0) return; |
2603 | |
2842 | 2604 assert(src_stride > 2*MB_SIZE + 5); |
3662
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2605 if(!sliced && offset_dst) |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2606 dst += src_x + src_y*dst_stride; |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2607 dst8+= src_x + src_y*src_stride; |
2206 | 2608 // src += src_x + src_y*src_stride; |
2609 | |
2842 | 2610 ptmp= tmp + 3*tmp_step; |
2611 block[0]= ptmp; | |
2612 ptmp+=tmp_step; | |
3314 | 2613 pred_block(s, block[0], tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h); |
2206 | 2614 |
2615 if(same_block(lt, rt)){ | |
2616 block[1]= block[0]; | |
2617 }else{ | |
2842 | 2618 block[1]= ptmp; |
2619 ptmp+=tmp_step; | |
3314 | 2620 pred_block(s, block[1], tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h); |
2206 | 2621 } |
2967 | 2622 |
2206 | 2623 if(same_block(lt, lb)){ |
2624 block[2]= block[0]; | |
2625 }else if(same_block(rt, lb)){ | |
2626 block[2]= block[1]; | |
2627 }else{ | |
2842 | 2628 block[2]= ptmp; |
2629 ptmp+=tmp_step; | |
3314 | 2630 pred_block(s, block[2], tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h); |
2206 | 2631 } |
2632 | |
2633 if(same_block(lt, rb) ){ | |
2634 block[3]= block[0]; | |
2635 }else if(same_block(rt, rb)){ | |
2636 block[3]= block[1]; | |
2637 }else if(same_block(lb, rb)){ | |
2638 block[3]= block[2]; | |
2639 }else{ | |
2842 | 2640 block[3]= ptmp; |
3314 | 2641 pred_block(s, block[3], tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h); |
2206 | 2642 } |
2643 #if 0 | |
2644 for(y=0; y<b_h; y++){ | |
2645 for(x=0; x<b_w; x++){ | |
2646 int v= obmc [x + y*obmc_stride] * block[3][x + y*src_stride] * (256/OBMC_MAX); | |
2647 if(add) dst[x + y*dst_stride] += v; | |
2648 else dst[x + y*dst_stride] -= v; | |
2649 } | |
2650 } | |
2651 for(y=0; y<b_h; y++){ | |
2652 uint8_t *obmc2= obmc + (obmc_stride>>1); | |
2653 for(x=0; x<b_w; x++){ | |
2654 int v= obmc2[x + y*obmc_stride] * block[2][x + y*src_stride] * (256/OBMC_MAX); | |
2655 if(add) dst[x + y*dst_stride] += v; | |
2656 else dst[x + y*dst_stride] -= v; | |
2657 } | |
2658 } | |
2659 for(y=0; y<b_h; y++){ | |
2660 uint8_t *obmc3= obmc + obmc_stride*(obmc_stride>>1); | |
2661 for(x=0; x<b_w; x++){ | |
2662 int v= obmc3[x + y*obmc_stride] * block[1][x + y*src_stride] * (256/OBMC_MAX); | |
2663 if(add) dst[x + y*dst_stride] += v; | |
2664 else dst[x + y*dst_stride] -= v; | |
2665 } | |
2666 } | |
2667 for(y=0; y<b_h; y++){ | |
2668 uint8_t *obmc3= obmc + obmc_stride*(obmc_stride>>1); | |
2669 uint8_t *obmc4= obmc3+ (obmc_stride>>1); | |
2670 for(x=0; x<b_w; x++){ | |
2671 int v= obmc4[x + y*obmc_stride] * block[0][x + y*src_stride] * (256/OBMC_MAX); | |
2672 if(add) dst[x + y*dst_stride] += v; | |
2673 else dst[x + y*dst_stride] -= v; | |
2674 } | |
2675 } | |
2676 #else | |
3662
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2677 if(sliced){ |
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2678 START_TIMER |
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2679 |
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2680 s->dsp.inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8); |
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2681 STOP_TIMER("inner_add_yblock") |
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2682 }else |
2206 | 2683 for(y=0; y<b_h; y++){ |
2684 //FIXME ugly missue of obmc_stride | |
2685 uint8_t *obmc1= obmc + y*obmc_stride; | |
2686 uint8_t *obmc2= obmc1+ (obmc_stride>>1); | |
2687 uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1); | |
2688 uint8_t *obmc4= obmc3+ (obmc_stride>>1); | |
2689 for(x=0; x<b_w; x++){ | |
2690 int v= obmc1[x] * block[3][x + y*src_stride] | |
2691 +obmc2[x] * block[2][x + y*src_stride] | |
2692 +obmc3[x] * block[1][x + y*src_stride] | |
2693 +obmc4[x] * block[0][x + y*src_stride]; | |
2967 | 2694 |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2695 v <<= 8 - LOG2_OBMC_MAX; |
2246 | 2696 if(FRAC_BITS != 8){ |
2697 v += 1<<(7 - FRAC_BITS); | |
2698 v >>= 8 - FRAC_BITS; | |
2699 } | |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2700 if(add){ |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2701 v += dst[x + y*dst_stride]; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2702 v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2703 if(v&(~255)) v= ~(v>>31); |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2704 dst8[x + y*src_stride] = v; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2705 }else{ |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2706 dst[x + y*dst_stride] -= v; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2707 } |
2206 | 2708 } |
2709 } | |
2710 #endif | |
2711 } | |
2712 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2713 static always_inline void predict_slice_buffered(SnowContext *s, slice_buffer * sb, DWTELEM * old_buffer, int plane_index, int add, int mb_y){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2714 Plane *p= &s->plane[plane_index]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2715 const int mb_w= s->b_width << s->block_max_depth; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2716 const int mb_h= s->b_height << s->block_max_depth; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2717 int x, y, mb_x; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2718 int block_size = MB_SIZE >> s->block_max_depth; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2719 int block_w = plane_index ? block_size/2 : block_size; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2720 const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2721 int obmc_stride= plane_index ? block_size : 2*block_size; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2722 int ref_stride= s->current_picture.linesize[plane_index]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2723 uint8_t *dst8= s->current_picture.data[plane_index]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2724 int w= p->width; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2725 int h= p->height; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2726 START_TIMER |
2967 | 2727 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2728 if(s->keyframe || (s->avctx->debug&512)){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2729 if(mb_y==mb_h) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2730 return; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2731 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2732 if(add){ |
2604
b7e6c3d31c65
Snow segfault bug in revision 1.43 patch by (Yartrebo /yartrebo earthlink net/)
michael
parents:
2602
diff
changeset
|
2733 for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++) |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2734 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2735 // DWTELEM * line = slice_buffer_get_line(sb, y); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2736 DWTELEM * line = sb->line[y]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2737 for(x=0; x<w; x++) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2738 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2739 // int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1)); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2740 int v= line[x] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1)); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2741 v >>= FRAC_BITS; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2742 if(v&(~255)) v= ~(v>>31); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2743 dst8[x + y*ref_stride]= v; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2744 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2745 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2746 }else{ |
2604
b7e6c3d31c65
Snow segfault bug in revision 1.43 patch by (Yartrebo /yartrebo earthlink net/)
michael
parents:
2602
diff
changeset
|
2747 for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++) |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2748 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2749 // DWTELEM * line = slice_buffer_get_line(sb, y); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2750 DWTELEM * line = sb->line[y]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2751 for(x=0; x<w; x++) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2752 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2753 line[x] -= 128 << FRAC_BITS; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2754 // buf[x + y*w]-= 128<<FRAC_BITS; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2755 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2756 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2757 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2758 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2759 return; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2760 } |
2967 | 2761 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2762 for(mb_x=0; mb_x<=mb_w; mb_x++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2763 START_TIMER |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2764 |
3662
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2765 add_yblock(s, 1, sb, old_buffer, dst8, obmc, |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2766 block_w*mb_x - block_w/2, |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2767 block_w*mb_y - block_w/2, |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2768 block_w, block_w, |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2769 w, h, |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2770 w, ref_stride, obmc_stride, |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2771 mb_x - 1, mb_y - 1, |
3662
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2772 add, 0, plane_index); |
2967 | 2773 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2774 STOP_TIMER("add_yblock") |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2775 } |
2967 | 2776 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2777 STOP_TIMER("predict_slice") |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2778 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2779 |
2562 | 2780 static always_inline void predict_slice(SnowContext *s, DWTELEM *buf, int plane_index, int add, int mb_y){ |
2138 | 2781 Plane *p= &s->plane[plane_index]; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2782 const int mb_w= s->b_width << s->block_max_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2783 const int mb_h= s->b_height << s->block_max_depth; |
2562 | 2784 int x, y, mb_x; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2785 int block_size = MB_SIZE >> s->block_max_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2786 int block_w = plane_index ? block_size/2 : block_size; |
2206 | 2787 const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth]; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2788 const int obmc_stride= plane_index ? block_size : 2*block_size; |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2789 int ref_stride= s->current_picture.linesize[plane_index]; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2790 uint8_t *dst8= s->current_picture.data[plane_index]; |
2138 | 2791 int w= p->width; |
2792 int h= p->height; | |
2197 | 2793 START_TIMER |
2967 | 2794 |
2206 | 2795 if(s->keyframe || (s->avctx->debug&512)){ |
2562 | 2796 if(mb_y==mb_h) |
2797 return; | |
2798 | |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2799 if(add){ |
2604
b7e6c3d31c65
Snow segfault bug in revision 1.43 patch by (Yartrebo /yartrebo earthlink net/)
michael
parents:
2602
diff
changeset
|
2800 for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){ |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2801 for(x=0; x<w; x++){ |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2802 int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1)); |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2803 v >>= FRAC_BITS; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2804 if(v&(~255)) v= ~(v>>31); |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2805 dst8[x + y*ref_stride]= v; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2806 } |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2807 } |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2808 }else{ |
2604
b7e6c3d31c65
Snow segfault bug in revision 1.43 patch by (Yartrebo /yartrebo earthlink net/)
michael
parents:
2602
diff
changeset
|
2809 for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){ |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2810 for(x=0; x<w; x++){ |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2811 buf[x + y*w]-= 128<<FRAC_BITS; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2812 } |
2206 | 2813 } |
2138 | 2814 } |
2206 | 2815 |
2816 return; | |
2138 | 2817 } |
2967 | 2818 |
2206 | 2819 for(mb_x=0; mb_x<=mb_w; mb_x++){ |
2197 | 2820 START_TIMER |
2206 | 2821 |
3662
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2822 add_yblock(s, 0, NULL, buf, dst8, obmc, |
2206 | 2823 block_w*mb_x - block_w/2, |
2138 | 2824 block_w*mb_y - block_w/2, |
2206 | 2825 block_w, block_w, |
2138 | 2826 w, h, |
2206 | 2827 w, ref_stride, obmc_stride, |
2828 mb_x - 1, mb_y - 1, | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2829 add, 1, plane_index); |
2967 | 2830 |
2206 | 2831 STOP_TIMER("add_yblock") |
2138 | 2832 } |
2967 | 2833 |
2562 | 2834 STOP_TIMER("predict_slice") |
2835 } | |
2836 | |
2837 static always_inline void predict_plane(SnowContext *s, DWTELEM *buf, int plane_index, int add){ | |
2838 const int mb_h= s->b_height << s->block_max_depth; | |
2839 int mb_y; | |
2840 for(mb_y=0; mb_y<=mb_h; mb_y++) | |
2841 predict_slice(s, buf, plane_index, add, mb_y); | |
2138 | 2842 } |
2843 | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2844 static int get_dc(SnowContext *s, int mb_x, int mb_y, int plane_index){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2845 int i, x2, y2; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2846 Plane *p= &s->plane[plane_index]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2847 const int block_size = MB_SIZE >> s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2848 const int block_w = plane_index ? block_size/2 : block_size; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2849 const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2850 const int obmc_stride= plane_index ? block_size : 2*block_size; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2851 const int ref_stride= s->current_picture.linesize[plane_index]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2852 uint8_t *src= s-> input_picture.data[plane_index]; |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2853 DWTELEM *dst= (DWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2854 const int b_stride = s->b_width << s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2855 const int w= p->width; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2856 const int h= p->height; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2857 int index= mb_x + mb_y*b_stride; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2858 BlockNode *b= &s->block[index]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2859 BlockNode backup= *b; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2860 int ab=0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2861 int aa=0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2862 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2863 b->type|= BLOCK_INTRA; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2864 b->color[plane_index]= 0; |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2865 memset(dst, 0, obmc_stride*obmc_stride*sizeof(DWTELEM)); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2866 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2867 for(i=0; i<4; i++){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2868 int mb_x2= mb_x + (i &1) - 1; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2869 int mb_y2= mb_y + (i>>1) - 1; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2870 int x= block_w*mb_x2 + block_w/2; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2871 int y= block_w*mb_y2 + block_w/2; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2872 |
3662
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2873 add_yblock(s, 0, NULL, dst + ((i&1)+(i>>1)*obmc_stride)*block_w, NULL, obmc, |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2874 x, y, block_w, block_w, w, h, obmc_stride, ref_stride, obmc_stride, mb_x2, mb_y2, 0, 0, plane_index); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2875 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2876 for(y2= FFMAX(y, 0); y2<FFMIN(h, y+block_w); y2++){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2877 for(x2= FFMAX(x, 0); x2<FFMIN(w, x+block_w); x2++){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2878 int index= x2-(block_w*mb_x - block_w/2) + (y2-(block_w*mb_y - block_w/2))*obmc_stride; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2879 int obmc_v= obmc[index]; |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2880 int d; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2881 if(y<0) obmc_v += obmc[index + block_w*obmc_stride]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2882 if(x<0) obmc_v += obmc[index + block_w]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2883 if(y+block_w>h) obmc_v += obmc[index - block_w*obmc_stride]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2884 if(x+block_w>w) obmc_v += obmc[index - block_w]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2885 //FIXME precalc this or simplify it somehow else |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2886 |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2887 d = -dst[index] + (1<<(FRAC_BITS-1)); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2888 dst[index] = d; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2889 ab += (src[x2 + y2*ref_stride] - (d>>FRAC_BITS)) * obmc_v; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2890 aa += obmc_v * obmc_v; //FIXME precalclate this |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2891 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2892 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2893 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2894 *b= backup; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2895 |
3206 | 2896 return clip(((ab<<LOG2_OBMC_MAX) + aa/2)/aa, 0, 255); //FIXME we shouldnt need cliping |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2897 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2898 |
3051 | 2899 static inline int get_block_bits(SnowContext *s, int x, int y, int w){ |
2900 const int b_stride = s->b_width << s->block_max_depth; | |
2901 const int b_height = s->b_height<< s->block_max_depth; | |
2902 int index= x + y*b_stride; | |
2903 BlockNode *b = &s->block[index]; | |
2904 BlockNode *left = x ? &s->block[index-1] : &null_block; | |
2905 BlockNode *top = y ? &s->block[index-b_stride] : &null_block; | |
2906 BlockNode *tl = y && x ? &s->block[index-b_stride-1] : left; | |
2907 BlockNode *tr = y && x+w<b_stride ? &s->block[index-b_stride+w] : tl; | |
2908 int dmx, dmy; | |
2909 // int mx_context= av_log2(2*ABS(left->mx - top->mx)); | |
2910 // int my_context= av_log2(2*ABS(left->my - top->my)); | |
2911 | |
2912 if(x<0 || x>=b_stride || y>=b_height) | |
2913 return 0; | |
2914 /* | |
2915 1 0 0 | |
2916 01X 1-2 1 | |
2917 001XX 3-6 2-3 | |
2918 0001XXX 7-14 4-7 | |
2919 00001XXXX 15-30 8-15 | |
2920 */ | |
2921 //FIXME try accurate rate | |
2922 //FIXME intra and inter predictors if surrounding blocks arent the same type | |
2923 if(b->type & BLOCK_INTRA){ | |
2924 return 3+2*( av_log2(2*ABS(left->color[0] - b->color[0])) | |
2925 + av_log2(2*ABS(left->color[1] - b->color[1])) | |
2926 + av_log2(2*ABS(left->color[2] - b->color[2]))); | |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2927 }else{ |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2928 pred_mv(s, &dmx, &dmy, b->ref, left, top, tr); |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2929 dmx-= b->mx; |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2930 dmy-= b->my; |
3314 | 2931 return 2*(1 + av_log2(2*ABS(dmx)) //FIXME kill the 2* can be merged in lambda |
2932 + av_log2(2*ABS(dmy)) | |
2933 + av_log2(2*b->ref)); | |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2934 } |
3051 | 2935 } |
2936 | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2937 static int get_block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index, const uint8_t *obmc_edged){ |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2938 Plane *p= &s->plane[plane_index]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2939 const int block_size = MB_SIZE >> s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2940 const int block_w = plane_index ? block_size/2 : block_size; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2941 const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2942 const int obmc_stride= plane_index ? block_size : 2*block_size; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2943 const int ref_stride= s->current_picture.linesize[plane_index]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2944 uint8_t *dst= s->current_picture.data[plane_index]; |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2945 uint8_t *src= s-> input_picture.data[plane_index]; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2946 DWTELEM *pred= (DWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2947 uint8_t cur[ref_stride*2*MB_SIZE]; //FIXME alignment |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2948 uint8_t tmp[ref_stride*(2*MB_SIZE+5)]; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2949 const int b_stride = s->b_width << s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2950 const int b_height = s->b_height<< s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2951 const int w= p->width; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2952 const int h= p->height; |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2953 int distortion; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2954 int rate= 0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2955 const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp); |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2956 int sx= block_w*mb_x - block_w/2; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2957 int sy= block_w*mb_y - block_w/2; |
3206 | 2958 int x0= FFMAX(0,-sx); |
2959 int y0= FFMAX(0,-sy); | |
2960 int x1= FFMIN(block_w*2, w-sx); | |
2961 int y1= FFMIN(block_w*2, h-sy); | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2962 int i,x,y; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2963 |
3314 | 2964 pred_block(s, cur, tmp, ref_stride, sx, sy, block_w*2, block_w*2, &s->block[mb_x + mb_y*b_stride], plane_index, w, h); |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2965 |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2966 for(y=y0; y<y1; y++){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2967 const uint8_t *obmc1= obmc_edged + y*obmc_stride; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2968 const DWTELEM *pred1 = pred + y*obmc_stride; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2969 uint8_t *cur1 = cur + y*ref_stride; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2970 uint8_t *dst1 = dst + sx + (sy+y)*ref_stride; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2971 for(x=x0; x<x1; x++){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2972 int v = (cur1[x] * obmc1[x]) << (FRAC_BITS - LOG2_OBMC_MAX); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2973 v = (v + pred1[x]) >> FRAC_BITS; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2974 if(v&(~255)) v= ~(v>>31); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2975 dst1[x] = v; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2976 } |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2977 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2978 |
3206 | 2979 /* copy the regions where obmc[] = (uint8_t)256 */ |
2980 if(LOG2_OBMC_MAX == 8 | |
2981 && (mb_x == 0 || mb_x == b_stride-1) | |
2982 && (mb_y == 0 || mb_y == b_height-1)){ | |
2983 if(mb_x == 0) | |
2984 x1 = block_w; | |
2985 else | |
2986 x0 = block_w; | |
2987 if(mb_y == 0) | |
2988 y1 = block_w; | |
2989 else | |
2990 y0 = block_w; | |
2991 for(y=y0; y<y1; y++) | |
2992 memcpy(dst + sx+x0 + (sy+y)*ref_stride, cur + x0 + y*ref_stride, x1-x0); | |
2993 } | |
2994 | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2995 if(block_w==16){ |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
2996 /* FIXME rearrange dsputil to fit 32x32 cmp functions */ |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
2997 /* FIXME check alignment of the cmp wavelet vs the encoding wavelet */ |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
2998 /* FIXME cmps overlap but don't cover the wavelet's whole support, |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
2999 * so improving the score of one block is not strictly guaranteed to |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
3000 * improve the score of the whole frame, so iterative motion est |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
3001 * doesn't always converge. */ |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
3002 if(s->avctx->me_cmp == FF_CMP_W97) |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
3003 distortion = w97_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32); |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
3004 else if(s->avctx->me_cmp == FF_CMP_W53) |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
3005 distortion = w53_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32); |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
3006 else{ |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
3007 distortion = 0; |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
3008 for(i=0; i<4; i++){ |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
3009 int off = sx+16*(i&1) + (sy+16*(i>>1))*ref_stride; |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
3010 distortion += s->dsp.me_cmp[0](&s->m, src + off, dst + off, ref_stride, 16); |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
3011 } |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3012 } |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3013 }else{ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3014 assert(block_w==8); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3015 distortion = s->dsp.me_cmp[0](&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, block_w*2); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3016 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3017 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3018 if(plane_index==0){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3019 for(i=0; i<4; i++){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3020 /* ..RRr |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3021 * .RXx. |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3022 * rxx.. |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3023 */ |
3051 | 3024 rate += get_block_bits(s, mb_x + (i&1) - (i>>1), mb_y + (i>>1), 1); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3025 } |
3057 | 3026 if(mb_x == b_stride-2) |
3027 rate += get_block_bits(s, mb_x + 1, mb_y + 1, 1); | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3028 } |
3051 | 3029 return distortion + rate*penalty_factor; |
3030 } | |
3031 | |
3032 static int get_4block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index){ | |
3033 int i, y2; | |
3034 Plane *p= &s->plane[plane_index]; | |
3035 const int block_size = MB_SIZE >> s->block_max_depth; | |
3036 const int block_w = plane_index ? block_size/2 : block_size; | |
3037 const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth]; | |
3038 const int obmc_stride= plane_index ? block_size : 2*block_size; | |
3039 const int ref_stride= s->current_picture.linesize[plane_index]; | |
3040 uint8_t *dst= s->current_picture.data[plane_index]; | |
3041 uint8_t *src= s-> input_picture.data[plane_index]; | |
3556
6805c1e5507e
Fix some "'static' is not at beginning of declaration" warnings.
diego
parents:
3338
diff
changeset
|
3042 static const DWTELEM zero_dst[4096]; //FIXME |
3051 | 3043 const int b_stride = s->b_width << s->block_max_depth; |
3044 const int b_height = s->b_height<< s->block_max_depth; | |
3045 const int w= p->width; | |
3046 const int h= p->height; | |
3047 int distortion= 0; | |
3048 int rate= 0; | |
3049 const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp); | |
3050 | |
3051 for(i=0; i<9; i++){ | |
3052 int mb_x2= mb_x + (i%3) - 1; | |
3053 int mb_y2= mb_y + (i/3) - 1; | |
3054 int x= block_w*mb_x2 + block_w/2; | |
3055 int y= block_w*mb_y2 + block_w/2; | |
3056 | |
3662
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
3057 add_yblock(s, 0, NULL, zero_dst, dst, obmc, |
3051 | 3058 x, y, block_w, block_w, w, h, /*dst_stride*/0, ref_stride, obmc_stride, mb_x2, mb_y2, 1, 1, plane_index); |
3059 | |
3060 //FIXME find a cleaner/simpler way to skip the outside stuff | |
3061 for(y2= y; y2<0; y2++) | |
3062 memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w); | |
3063 for(y2= h; y2<y+block_w; y2++) | |
3064 memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w); | |
3065 if(x<0){ | |
3066 for(y2= y; y2<y+block_w; y2++) | |
3067 memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, -x); | |
3068 } | |
3069 if(x+block_w > w){ | |
3070 for(y2= y; y2<y+block_w; y2++) | |
3071 memcpy(dst + w + y2*ref_stride, src + w + y2*ref_stride, x+block_w - w); | |
3072 } | |
3073 | |
3074 assert(block_w== 8 || block_w==16); | |
3075 distortion += s->dsp.me_cmp[block_w==8](&s->m, src + x + y*ref_stride, dst + x + y*ref_stride, ref_stride, block_w); | |
3076 } | |
3077 | |
3078 if(plane_index==0){ | |
3079 BlockNode *b= &s->block[mb_x+mb_y*b_stride]; | |
3080 int merged= same_block(b,b+1) && same_block(b,b+b_stride) && same_block(b,b+b_stride+1); | |
3081 | |
3082 /* ..RRRr | |
3083 * .RXXx. | |
3084 * .RXXx. | |
3085 * rxxx. | |
3086 */ | |
3087 if(merged) | |
3088 rate = get_block_bits(s, mb_x, mb_y, 2); | |
3089 for(i=merged?4:0; i<9; i++){ | |
3090 static const int dxy[9][2] = {{0,0},{1,0},{0,1},{1,1},{2,0},{2,1},{-1,2},{0,2},{1,2}}; | |
3091 rate += get_block_bits(s, mb_x + dxy[i][0], mb_y + dxy[i][1], 1); | |
3092 } | |
3093 } | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3094 return distortion + rate*penalty_factor; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3095 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3096 |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3097 static always_inline int check_block(SnowContext *s, int mb_x, int mb_y, int p[3], int intra, const uint8_t *obmc_edged, int *best_rd){ |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3098 const int b_stride= s->b_width << s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3099 BlockNode *block= &s->block[mb_x + mb_y * b_stride]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3100 BlockNode backup= *block; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3101 int rd, index, value; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3102 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3103 assert(mb_x>=0 && mb_y>=0); |
2994 | 3104 assert(mb_x<b_stride); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3105 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3106 if(intra){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3107 block->color[0] = p[0]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3108 block->color[1] = p[1]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3109 block->color[2] = p[2]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3110 block->type |= BLOCK_INTRA; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3111 }else{ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3112 index= (p[0] + 31*p[1]) & (ME_CACHE_SIZE-1); |
3314 | 3113 value= s->me_cache_generation + (p[0]>>10) + (p[1]<<6) + (block->ref<<12); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3114 if(s->me_cache[index] == value) |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3115 return 0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3116 s->me_cache[index]= value; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3117 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3118 block->mx= p[0]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3119 block->my= p[1]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3120 block->type &= ~BLOCK_INTRA; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3121 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3122 |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3123 rd= get_block_rd(s, mb_x, mb_y, 0, obmc_edged); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3124 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3125 //FIXME chroma |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3126 if(rd < *best_rd){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3127 *best_rd= rd; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3128 return 1; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3129 }else{ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3130 *block= backup; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3131 return 0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3132 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3133 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3134 |
2999
bc83e7a080a5
this fixes compilation with gcc 2.95 which seems to not like array casting... special case inline of check_block which makes the array and passes it
mmu_man
parents:
2998
diff
changeset
|
3135 /* special case for int[2] args we discard afterward, fixes compilation prob with gcc 2.95 */ |
3197 | 3136 static always_inline int check_block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, const uint8_t *obmc_edged, int *best_rd){ |
2999
bc83e7a080a5
this fixes compilation with gcc 2.95 which seems to not like array casting... special case inline of check_block which makes the array and passes it
mmu_man
parents:
2998
diff
changeset
|
3137 int p[2] = {p0, p1}; |
3197 | 3138 return check_block(s, mb_x, mb_y, p, 0, obmc_edged, best_rd); |
2999
bc83e7a080a5
this fixes compilation with gcc 2.95 which seems to not like array casting... special case inline of check_block which makes the array and passes it
mmu_man
parents:
2998
diff
changeset
|
3139 } |
bc83e7a080a5
this fixes compilation with gcc 2.95 which seems to not like array casting... special case inline of check_block which makes the array and passes it
mmu_man
parents:
2998
diff
changeset
|
3140 |
3314 | 3141 static always_inline int check_4block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, int ref, int *best_rd){ |
3051 | 3142 const int b_stride= s->b_width << s->block_max_depth; |
3143 BlockNode *block= &s->block[mb_x + mb_y * b_stride]; | |
3144 BlockNode backup[4]= {block[0], block[1], block[b_stride], block[b_stride+1]}; | |
3145 int rd, index, value; | |
3146 | |
3147 assert(mb_x>=0 && mb_y>=0); | |
3148 assert(mb_x<b_stride); | |
3149 assert(((mb_x|mb_y)&1) == 0); | |
3150 | |
3151 index= (p0 + 31*p1) & (ME_CACHE_SIZE-1); | |
3314 | 3152 value= s->me_cache_generation + (p0>>10) + (p1<<6) + (block->ref<<12); |
3051 | 3153 if(s->me_cache[index] == value) |
3154 return 0; | |
3155 s->me_cache[index]= value; | |
3156 | |
3157 block->mx= p0; | |
3158 block->my= p1; | |
3314 | 3159 block->ref= ref; |
3051 | 3160 block->type &= ~BLOCK_INTRA; |
3161 block[1]= block[b_stride]= block[b_stride+1]= *block; | |
3162 | |
3163 rd= get_4block_rd(s, mb_x, mb_y, 0); | |
3164 | |
3165 //FIXME chroma | |
3166 if(rd < *best_rd){ | |
3167 *best_rd= rd; | |
3168 return 1; | |
3169 }else{ | |
3170 block[0]= backup[0]; | |
3171 block[1]= backup[1]; | |
3172 block[b_stride]= backup[2]; | |
3173 block[b_stride+1]= backup[3]; | |
3174 return 0; | |
3175 } | |
3176 } | |
3177 | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3178 static void iterative_me(SnowContext *s){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3179 int pass, mb_x, mb_y; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3180 const int b_width = s->b_width << s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3181 const int b_height= s->b_height << s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3182 const int b_stride= b_width; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3183 int color[3]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3184 |
3194
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
3185 { |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
3186 RangeCoder r = s->c; |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
3187 uint8_t state[sizeof(s->block_state)]; |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
3188 memcpy(state, s->block_state, sizeof(s->block_state)); |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
3189 for(mb_y= 0; mb_y<s->b_height; mb_y++) |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
3190 for(mb_x= 0; mb_x<s->b_width; mb_x++) |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
3191 encode_q_branch(s, 0, mb_x, mb_y); |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
3192 s->c = r; |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
3193 memcpy(s->block_state, state, sizeof(s->block_state)); |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
3194 } |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
3195 |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
3196 for(pass=0; pass<25; pass++){ |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3197 int change= 0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3198 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3199 for(mb_y= 0; mb_y<b_height; mb_y++){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3200 for(mb_x= 0; mb_x<b_width; mb_x++){ |
3314 | 3201 int dia_change, i, j, ref; |
3202 int best_rd= INT_MAX, ref_rd; | |
3203 BlockNode backup, ref_b; | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3204 const int index= mb_x + mb_y * b_stride; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3205 BlockNode *block= &s->block[index]; |
3324 | 3206 BlockNode *tb = mb_y ? &s->block[index-b_stride ] : NULL; |
3207 BlockNode *lb = mb_x ? &s->block[index -1] : NULL; | |
3208 BlockNode *rb = mb_x+1<b_width ? &s->block[index +1] : NULL; | |
3209 BlockNode *bb = mb_y+1<b_height ? &s->block[index+b_stride ] : NULL; | |
3210 BlockNode *tlb= mb_x && mb_y ? &s->block[index-b_stride-1] : NULL; | |
3211 BlockNode *trb= mb_x+1<b_width && mb_y ? &s->block[index-b_stride+1] : NULL; | |
3212 BlockNode *blb= mb_x && mb_y+1<b_height ? &s->block[index+b_stride-1] : NULL; | |
3213 BlockNode *brb= mb_x+1<b_width && mb_y+1<b_height ? &s->block[index+b_stride+1] : NULL; | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3214 const int b_w= (MB_SIZE >> s->block_max_depth); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3215 uint8_t obmc_edged[b_w*2][b_w*2]; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3216 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3217 if(pass && (block->type & BLOCK_OPT)) |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3218 continue; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3219 block->type |= BLOCK_OPT; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3220 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3221 backup= *block; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3222 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3223 if(!s->me_cache_generation) |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3224 memset(s->me_cache, 0, sizeof(s->me_cache)); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3225 s->me_cache_generation += 1<<22; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3226 |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3227 //FIXME precalc |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3228 { |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3229 int x, y; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3230 memcpy(obmc_edged, obmc_tab[s->block_max_depth], b_w*b_w*4); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3231 if(mb_x==0) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3232 for(y=0; y<b_w*2; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3233 memset(obmc_edged[y], obmc_edged[y][0] + obmc_edged[y][b_w-1], b_w); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3234 if(mb_x==b_stride-1) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3235 for(y=0; y<b_w*2; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3236 memset(obmc_edged[y]+b_w, obmc_edged[y][b_w] + obmc_edged[y][b_w*2-1], b_w); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3237 if(mb_y==0){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3238 for(x=0; x<b_w*2; x++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3239 obmc_edged[0][x] += obmc_edged[b_w-1][x]; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3240 for(y=1; y<b_w; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3241 memcpy(obmc_edged[y], obmc_edged[0], b_w*2); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3242 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3243 if(mb_y==b_height-1){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3244 for(x=0; x<b_w*2; x++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3245 obmc_edged[b_w*2-1][x] += obmc_edged[b_w][x]; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3246 for(y=b_w; y<b_w*2-1; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3247 memcpy(obmc_edged[y], obmc_edged[b_w*2-1], b_w*2); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3248 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3249 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3250 |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3251 //skip stuff outside the picture |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3252 if(mb_x==0 || mb_y==0 || mb_x==b_width-1 || mb_y==b_height-1) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3253 { |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3254 uint8_t *src= s-> input_picture.data[0]; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3255 uint8_t *dst= s->current_picture.data[0]; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3256 const int stride= s->current_picture.linesize[0]; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3257 const int block_w= MB_SIZE >> s->block_max_depth; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3258 const int sx= block_w*mb_x - block_w/2; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3259 const int sy= block_w*mb_y - block_w/2; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3260 const int w= s->plane[0].width; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3261 const int h= s->plane[0].height; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3262 int y; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3263 |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3264 for(y=sy; y<0; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3265 memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3266 for(y=h; y<sy+block_w*2; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3267 memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3268 if(sx<0){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3269 for(y=sy; y<sy+block_w*2; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3270 memcpy(dst + sx + y*stride, src + sx + y*stride, -sx); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3271 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3272 if(sx+block_w*2 > w){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3273 for(y=sy; y<sy+block_w*2; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3274 memcpy(dst + w + y*stride, src + w + y*stride, sx+block_w*2 - w); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3275 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3276 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3277 |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3278 // intra(black) = neighbors' contribution to the current block |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3279 for(i=0; i<3; i++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3280 color[i]= get_dc(s, mb_x, mb_y, i); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3281 |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3282 // get previous score (cant be cached due to OBMC) |
3057 | 3283 if(pass > 0 && (block->type&BLOCK_INTRA)){ |
3284 int color0[3]= {block->color[0], block->color[1], block->color[2]}; | |
3285 check_block(s, mb_x, mb_y, color0, 1, *obmc_edged, &best_rd); | |
3286 }else | |
3197 | 3287 check_block_inter(s, mb_x, mb_y, block->mx, block->my, *obmc_edged, &best_rd); |
3288 | |
3314 | 3289 ref_b= *block; |
3290 ref_rd= best_rd; | |
3291 for(ref=0; ref < s->ref_frames; ref++){ | |
3292 int16_t (*mvr)[2]= &s->ref_mvs[ref][index]; | |
3293 if(s->ref_scores[ref][index] > s->ref_scores[ref_b.ref][index]*3/2) //FIXME tune threshold | |
3294 continue; | |
3295 block->ref= ref; | |
3296 best_rd= INT_MAX; | |
3297 | |
3298 check_block_inter(s, mb_x, mb_y, mvr[0][0], mvr[0][1], *obmc_edged, &best_rd); | |
3299 check_block_inter(s, mb_x, mb_y, 0, 0, *obmc_edged, &best_rd); | |
3324 | 3300 if(tb) |
3314 | 3301 check_block_inter(s, mb_x, mb_y, mvr[-b_stride][0], mvr[-b_stride][1], *obmc_edged, &best_rd); |
3324 | 3302 if(lb) |
3314 | 3303 check_block_inter(s, mb_x, mb_y, mvr[-1][0], mvr[-1][1], *obmc_edged, &best_rd); |
3324 | 3304 if(rb) |
3314 | 3305 check_block_inter(s, mb_x, mb_y, mvr[1][0], mvr[1][1], *obmc_edged, &best_rd); |
3324 | 3306 if(bb) |
3314 | 3307 check_block_inter(s, mb_x, mb_y, mvr[b_stride][0], mvr[b_stride][1], *obmc_edged, &best_rd); |
3308 | |
3309 /* fullpel ME */ | |
3310 //FIXME avoid subpel interpol / round to nearest integer | |
3311 do{ | |
3312 dia_change=0; | |
3313 for(i=0; i<FFMAX(s->avctx->dia_size, 1); i++){ | |
3314 for(j=0; j<i; j++){ | |
3315 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my+(4*j), *obmc_edged, &best_rd); | |
3316 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my-(4*j), *obmc_edged, &best_rd); | |
3317 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my-(4*j), *obmc_edged, &best_rd); | |
3318 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my+(4*j), *obmc_edged, &best_rd); | |
3319 } | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3320 } |
3314 | 3321 }while(dia_change); |
3322 /* subpel ME */ | |
3323 do{ | |
3324 static const int square[8][2]= {{+1, 0},{-1, 0},{ 0,+1},{ 0,-1},{+1,+1},{-1,-1},{+1,-1},{-1,+1},}; | |
3325 dia_change=0; | |
3326 for(i=0; i<8; i++) | |
3327 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+square[i][0], block->my+square[i][1], *obmc_edged, &best_rd); | |
3328 }while(dia_change); | |
3329 //FIXME or try the standard 2 pass qpel or similar | |
3330 | |
3331 mvr[0][0]= block->mx; | |
3332 mvr[0][1]= block->my; | |
3333 if(ref_rd > best_rd){ | |
3334 ref_rd= best_rd; | |
3335 ref_b= *block; | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3336 } |
3314 | 3337 } |
3338 best_rd= ref_rd; | |
3339 *block= ref_b; | |
2998 | 3340 #if 1 |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3341 check_block(s, mb_x, mb_y, color, 1, *obmc_edged, &best_rd); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3342 //FIXME RD style color selection |
2998 | 3343 #endif |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3344 if(!same_block(block, &backup)){ |
3324 | 3345 if(tb ) tb ->type &= ~BLOCK_OPT; |
3346 if(lb ) lb ->type &= ~BLOCK_OPT; | |
3347 if(rb ) rb ->type &= ~BLOCK_OPT; | |
3348 if(bb ) bb ->type &= ~BLOCK_OPT; | |
3349 if(tlb) tlb->type &= ~BLOCK_OPT; | |
3350 if(trb) trb->type &= ~BLOCK_OPT; | |
3351 if(blb) blb->type &= ~BLOCK_OPT; | |
3352 if(brb) brb->type &= ~BLOCK_OPT; | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3353 change ++; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3354 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3355 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3356 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3357 av_log(NULL, AV_LOG_ERROR, "pass:%d changed:%d\n", pass, change); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3358 if(!change) |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3359 break; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3360 } |
3051 | 3361 |
3362 if(s->block_max_depth == 1){ | |
3363 int change= 0; | |
3364 for(mb_y= 0; mb_y<b_height; mb_y+=2){ | |
3365 for(mb_x= 0; mb_x<b_width; mb_x+=2){ | |
3324 | 3366 int i; |
3051 | 3367 int best_rd, init_rd; |
3368 const int index= mb_x + mb_y * b_stride; | |
3369 BlockNode *b[4]; | |
3370 | |
3371 b[0]= &s->block[index]; | |
3372 b[1]= b[0]+1; | |
3373 b[2]= b[0]+b_stride; | |
3374 b[3]= b[2]+1; | |
3375 if(same_block(b[0], b[1]) && | |
3376 same_block(b[0], b[2]) && | |
3377 same_block(b[0], b[3])) | |
3378 continue; | |
3379 | |
3380 if(!s->me_cache_generation) | |
3381 memset(s->me_cache, 0, sizeof(s->me_cache)); | |
3382 s->me_cache_generation += 1<<22; | |
3383 | |
3384 init_rd= best_rd= get_4block_rd(s, mb_x, mb_y, 0); | |
3385 | |
3314 | 3386 //FIXME more multiref search? |
3051 | 3387 check_4block_inter(s, mb_x, mb_y, |
3388 (b[0]->mx + b[1]->mx + b[2]->mx + b[3]->mx + 2) >> 2, | |
3314 | 3389 (b[0]->my + b[1]->my + b[2]->my + b[3]->my + 2) >> 2, 0, &best_rd); |
3051 | 3390 |
3391 for(i=0; i<4; i++) | |
3392 if(!(b[i]->type&BLOCK_INTRA)) | |
3314 | 3393 check_4block_inter(s, mb_x, mb_y, b[i]->mx, b[i]->my, b[i]->ref, &best_rd); |
3051 | 3394 |
3395 if(init_rd != best_rd) | |
3396 change++; | |
3397 } | |
3398 } | |
3399 av_log(NULL, AV_LOG_ERROR, "pass:4mv changed:%d\n", change*4); | |
3400 } | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3401 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3402 |
2138 | 3403 static void quantize(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int bias){ |
3404 const int level= b->level; | |
3405 const int w= b->width; | |
3406 const int h= b->height; | |
2600 | 3407 const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16); |
3408 const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); | |
2150 | 3409 int x,y, thres1, thres2; |
2893 | 3410 // START_TIMER |
2138 | 3411 |
2161 | 3412 if(s->qlog == LOSSLESS_QLOG) return; |
2967 | 3413 |
2138 | 3414 bias= bias ? 0 : (3*qmul)>>3; |
2150 | 3415 thres1= ((qmul - bias)>>QEXPSHIFT) - 1; |
3416 thres2= 2*thres1; | |
2967 | 3417 |
2138 | 3418 if(!bias){ |
3419 for(y=0; y<h; y++){ | |
3420 for(x=0; x<w; x++){ | |
2150 | 3421 int i= src[x + y*stride]; |
2967 | 3422 |
2150 | 3423 if((unsigned)(i+thres1) > thres2){ |
3424 if(i>=0){ | |
3425 i<<= QEXPSHIFT; | |
3426 i/= qmul; //FIXME optimize | |
3427 src[x + y*stride]= i; | |
3428 }else{ | |
3429 i= -i; | |
3430 i<<= QEXPSHIFT; | |
3431 i/= qmul; //FIXME optimize | |
3432 src[x + y*stride]= -i; | |
3433 } | |
3434 }else | |
3435 src[x + y*stride]= 0; | |
2138 | 3436 } |
3437 } | |
3438 }else{ | |
3439 for(y=0; y<h; y++){ | |
3440 for(x=0; x<w; x++){ | |
2967 | 3441 int i= src[x + y*stride]; |
3442 | |
2150 | 3443 if((unsigned)(i+thres1) > thres2){ |
3444 if(i>=0){ | |
3445 i<<= QEXPSHIFT; | |
3446 i= (i + bias) / qmul; //FIXME optimize | |
3447 src[x + y*stride]= i; | |
3448 }else{ | |
3449 i= -i; | |
3450 i<<= QEXPSHIFT; | |
3451 i= (i + bias) / qmul; //FIXME optimize | |
3452 src[x + y*stride]= -i; | |
3453 } | |
3454 }else | |
3455 src[x + y*stride]= 0; | |
2138 | 3456 } |
3457 } | |
3458 } | |
2150 | 3459 if(level+1 == s->spatial_decomposition_count){ |
3460 // STOP_TIMER("quantize") | |
3461 } | |
2138 | 3462 } |
3463 | |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
3464 static void dequantize_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, DWTELEM *src, int stride, int start_y, int end_y){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3465 const int w= b->width; |
2600 | 3466 const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16); |
3467 const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3468 const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3469 int x,y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3470 START_TIMER |
2967 | 3471 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3472 if(s->qlog == LOSSLESS_QLOG) return; |
2967 | 3473 |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
3474 for(y=start_y; y<end_y; y++){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3475 // DWTELEM * line = slice_buffer_get_line_from_address(sb, src + (y * stride)); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3476 DWTELEM * line = slice_buffer_get_line(sb, (y * b->stride_line) + b->buf_y_offset) + b->buf_x_offset; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3477 for(x=0; x<w; x++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3478 int i= line[x]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3479 if(i<0){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3480 line[x]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3481 }else if(i>0){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3482 line[x]= (( i*qmul + qadd)>>(QEXPSHIFT)); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3483 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3484 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3485 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3486 if(w > 200 /*level+1 == s->spatial_decomposition_count*/){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3487 STOP_TIMER("dquant") |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3488 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3489 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3490 |
2138 | 3491 static void dequantize(SnowContext *s, SubBand *b, DWTELEM *src, int stride){ |
3492 const int w= b->width; | |
3493 const int h= b->height; | |
2600 | 3494 const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16); |
3495 const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); | |
2138 | 3496 const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT; |
3497 int x,y; | |
2195 | 3498 START_TIMER |
2967 | 3499 |
2161 | 3500 if(s->qlog == LOSSLESS_QLOG) return; |
2967 | 3501 |
2138 | 3502 for(y=0; y<h; y++){ |
3503 for(x=0; x<w; x++){ | |
3504 int i= src[x + y*stride]; | |
3505 if(i<0){ | |
3506 src[x + y*stride]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias | |
3507 }else if(i>0){ | |
3508 src[x + y*stride]= (( i*qmul + qadd)>>(QEXPSHIFT)); | |
3509 } | |
3510 } | |
3511 } | |
2195 | 3512 if(w > 200 /*level+1 == s->spatial_decomposition_count*/){ |
3513 STOP_TIMER("dquant") | |
3514 } | |
2138 | 3515 } |
3516 | |
3517 static void decorrelate(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int inverse, int use_median){ | |
3518 const int w= b->width; | |
3519 const int h= b->height; | |
3520 int x,y; | |
2967 | 3521 |
2138 | 3522 for(y=h-1; y>=0; y--){ |
3523 for(x=w-1; x>=0; x--){ | |
3524 int i= x + y*stride; | |
2967 | 3525 |
2138 | 3526 if(x){ |
3527 if(use_median){ | |
3528 if(y && x+1<w) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]); | |
3529 else src[i] -= src[i - 1]; | |
3530 }else{ | |
3531 if(y) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]); | |
3532 else src[i] -= src[i - 1]; | |
3533 } | |
3534 }else{ | |
3535 if(y) src[i] -= src[i - stride]; | |
3536 } | |
3537 } | |
3538 } | |
3539 } | |
3540 | |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
3541 static void correlate_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, DWTELEM *src, int stride, int inverse, int use_median, int start_y, int end_y){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3542 const int w= b->width; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3543 int x,y; |
2967 | 3544 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3545 // START_TIMER |
2967 | 3546 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3547 DWTELEM * line; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3548 DWTELEM * prev; |
2967 | 3549 |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
3550 if (start_y != 0) |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
3551 line = slice_buffer_get_line(sb, ((start_y - 1) * b->stride_line) + b->buf_y_offset) + b->buf_x_offset; |
2967 | 3552 |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
3553 for(y=start_y; y<end_y; y++){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3554 prev = line; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3555 // line = slice_buffer_get_line_from_address(sb, src + (y * stride)); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3556 line = slice_buffer_get_line(sb, (y * b->stride_line) + b->buf_y_offset) + b->buf_x_offset; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3557 for(x=0; x<w; x++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3558 if(x){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3559 if(use_median){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3560 if(y && x+1<w) line[x] += mid_pred(line[x - 1], prev[x], prev[x + 1]); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3561 else line[x] += line[x - 1]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3562 }else{ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3563 if(y) line[x] += mid_pred(line[x - 1], prev[x], line[x - 1] + prev[x] - prev[x - 1]); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3564 else line[x] += line[x - 1]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3565 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3566 }else{ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3567 if(y) line[x] += prev[x]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3568 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3569 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3570 } |
2967 | 3571 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3572 // STOP_TIMER("correlate") |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3573 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3574 |
2138 | 3575 static void correlate(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int inverse, int use_median){ |
3576 const int w= b->width; | |
3577 const int h= b->height; | |
3578 int x,y; | |
2967 | 3579 |
2138 | 3580 for(y=0; y<h; y++){ |
3581 for(x=0; x<w; x++){ | |
3582 int i= x + y*stride; | |
2967 | 3583 |
2138 | 3584 if(x){ |
3585 if(use_median){ | |
3586 if(y && x+1<w) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]); | |
3587 else src[i] += src[i - 1]; | |
3588 }else{ | |
3589 if(y) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]); | |
3590 else src[i] += src[i - 1]; | |
3591 } | |
3592 }else{ | |
3593 if(y) src[i] += src[i - stride]; | |
3594 } | |
3595 } | |
3596 } | |
3597 } | |
3598 | |
3599 static void encode_header(SnowContext *s){ | |
3600 int plane_index, level, orientation; | |
2967 | 3601 uint8_t kstate[32]; |
3602 | |
3603 memset(kstate, MID_STATE, sizeof(kstate)); | |
2138 | 3604 |
2335 | 3605 put_rac(&s->c, kstate, s->keyframe); |
2199 | 3606 if(s->keyframe || s->always_reset) |
3607 reset_contexts(s); | |
2138 | 3608 if(s->keyframe){ |
3609 put_symbol(&s->c, s->header_state, s->version, 0); | |
2335 | 3610 put_rac(&s->c, s->header_state, s->always_reset); |
2138 | 3611 put_symbol(&s->c, s->header_state, s->temporal_decomposition_type, 0); |
3612 put_symbol(&s->c, s->header_state, s->temporal_decomposition_count, 0); | |
3613 put_symbol(&s->c, s->header_state, s->spatial_decomposition_count, 0); | |
3614 put_symbol(&s->c, s->header_state, s->colorspace_type, 0); | |
3615 put_symbol(&s->c, s->header_state, s->chroma_h_shift, 0); | |
3616 put_symbol(&s->c, s->header_state, s->chroma_v_shift, 0); | |
2335 | 3617 put_rac(&s->c, s->header_state, s->spatial_scalability); |
3618 // put_rac(&s->c, s->header_state, s->rate_scalability); | |
3314 | 3619 put_symbol(&s->c, s->header_state, s->max_ref_frames-1, 0); |
2138 | 3620 |
3621 for(plane_index=0; plane_index<2; plane_index++){ | |
3622 for(level=0; level<s->spatial_decomposition_count; level++){ | |
3623 for(orientation=level ? 1:0; orientation<4; orientation++){ | |
3624 if(orientation==2) continue; | |
3625 put_symbol(&s->c, s->header_state, s->plane[plane_index].band[level][orientation].qlog, 1); | |
3626 } | |
3627 } | |
3628 } | |
3629 } | |
3630 put_symbol(&s->c, s->header_state, s->spatial_decomposition_type, 0); | |
2967 | 3631 put_symbol(&s->c, s->header_state, s->qlog, 1); |
3632 put_symbol(&s->c, s->header_state, s->mv_scale, 0); | |
2138 | 3633 put_symbol(&s->c, s->header_state, s->qbias, 1); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3634 put_symbol(&s->c, s->header_state, s->block_max_depth, 0); |
2138 | 3635 } |
3636 | |
3637 static int decode_header(SnowContext *s){ | |
3638 int plane_index, level, orientation; | |
2335 | 3639 uint8_t kstate[32]; |
2138 | 3640 |
2967 | 3641 memset(kstate, MID_STATE, sizeof(kstate)); |
2335 | 3642 |
3643 s->keyframe= get_rac(&s->c, kstate); | |
2199 | 3644 if(s->keyframe || s->always_reset) |
3645 reset_contexts(s); | |
2138 | 3646 if(s->keyframe){ |
3647 s->version= get_symbol(&s->c, s->header_state, 0); | |
3648 if(s->version>0){ | |
3649 av_log(s->avctx, AV_LOG_ERROR, "version %d not supported", s->version); | |
3650 return -1; | |
3651 } | |
2335 | 3652 s->always_reset= get_rac(&s->c, s->header_state); |
2138 | 3653 s->temporal_decomposition_type= get_symbol(&s->c, s->header_state, 0); |
3654 s->temporal_decomposition_count= get_symbol(&s->c, s->header_state, 0); | |
3655 s->spatial_decomposition_count= get_symbol(&s->c, s->header_state, 0); | |
3656 s->colorspace_type= get_symbol(&s->c, s->header_state, 0); | |
3657 s->chroma_h_shift= get_symbol(&s->c, s->header_state, 0); | |
3658 s->chroma_v_shift= get_symbol(&s->c, s->header_state, 0); | |
2335 | 3659 s->spatial_scalability= get_rac(&s->c, s->header_state); |
3660 // s->rate_scalability= get_rac(&s->c, s->header_state); | |
3314 | 3661 s->max_ref_frames= get_symbol(&s->c, s->header_state, 0)+1; |
2138 | 3662 |
3663 for(plane_index=0; plane_index<3; plane_index++){ | |
3664 for(level=0; level<s->spatial_decomposition_count; level++){ | |
3665 for(orientation=level ? 1:0; orientation<4; orientation++){ | |
3666 int q; | |
3667 if (plane_index==2) q= s->plane[1].band[level][orientation].qlog; | |
3668 else if(orientation==2) q= s->plane[plane_index].band[level][1].qlog; | |
3669 else q= get_symbol(&s->c, s->header_state, 1); | |
3670 s->plane[plane_index].band[level][orientation].qlog= q; | |
3671 } | |
3672 } | |
3673 } | |
3674 } | |
2967 | 3675 |
2138 | 3676 s->spatial_decomposition_type= get_symbol(&s->c, s->header_state, 0); |
3677 if(s->spatial_decomposition_type > 2){ | |
3678 av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_type %d not supported", s->spatial_decomposition_type); | |
3679 return -1; | |
3680 } | |
2967 | 3681 |
2138 | 3682 s->qlog= get_symbol(&s->c, s->header_state, 1); |
3683 s->mv_scale= get_symbol(&s->c, s->header_state, 0); | |
3684 s->qbias= get_symbol(&s->c, s->header_state, 1); | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3685 s->block_max_depth= get_symbol(&s->c, s->header_state, 0); |
3303
68721b62a528
sanity checks, some might have been exploitable ...
michael
parents:
3206
diff
changeset
|
3686 if(s->block_max_depth > 1 || s->block_max_depth < 0){ |
2952 | 3687 av_log(s->avctx, AV_LOG_ERROR, "block_max_depth= %d is too large", s->block_max_depth); |
3688 s->block_max_depth= 0; | |
3689 return -1; | |
3690 } | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3691 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3692 return 0; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3693 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3694 |
3075 | 3695 static void init_qexp(void){ |
2600 | 3696 int i; |
3697 double v=128; | |
3698 | |
3699 for(i=0; i<QROOT; i++){ | |
3700 qexp[i]= lrintf(v); | |
2967 | 3701 v *= pow(2, 1.0 / QROOT); |
2600 | 3702 } |
3703 } | |
3704 | |
2138 | 3705 static int common_init(AVCodecContext *avctx){ |
3706 SnowContext *s = avctx->priv_data; | |
3707 int width, height; | |
3708 int level, orientation, plane_index, dec; | |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
3709 int i, j; |
2138 | 3710 |
3711 s->avctx= avctx; | |
2967 | 3712 |
2138 | 3713 dsputil_init(&s->dsp, avctx); |
3714 | |
3715 #define mcf(dx,dy)\ | |
3716 s->dsp.put_qpel_pixels_tab [0][dy+dx/4]=\ | |
3717 s->dsp.put_no_rnd_qpel_pixels_tab[0][dy+dx/4]=\ | |
2224
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
3718 s->dsp.put_h264_qpel_pixels_tab[0][dy+dx/4];\ |
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
3719 s->dsp.put_qpel_pixels_tab [1][dy+dx/4]=\ |
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
3720 s->dsp.put_no_rnd_qpel_pixels_tab[1][dy+dx/4]=\ |
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
3721 s->dsp.put_h264_qpel_pixels_tab[1][dy+dx/4]; |
2138 | 3722 |
3723 mcf( 0, 0) | |
3724 mcf( 4, 0) | |
3725 mcf( 8, 0) | |
3726 mcf(12, 0) | |
3727 mcf( 0, 4) | |
3728 mcf( 4, 4) | |
3729 mcf( 8, 4) | |
3730 mcf(12, 4) | |
3731 mcf( 0, 8) | |
3732 mcf( 4, 8) | |
3733 mcf( 8, 8) | |
3734 mcf(12, 8) | |
3735 mcf( 0,12) | |
3736 mcf( 4,12) | |
3737 mcf( 8,12) | |
3738 mcf(12,12) | |
3739 | |
3740 #define mcfh(dx,dy)\ | |
3741 s->dsp.put_pixels_tab [0][dy/4+dx/8]=\ | |
3742 s->dsp.put_no_rnd_pixels_tab[0][dy/4+dx/8]=\ | |
2224
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
3743 mc_block_hpel ## dx ## dy ## 16;\ |
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
3744 s->dsp.put_pixels_tab [1][dy/4+dx/8]=\ |
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
3745 s->dsp.put_no_rnd_pixels_tab[1][dy/4+dx/8]=\ |
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
3746 mc_block_hpel ## dx ## dy ## 8; |
2138 | 3747 |
3748 mcfh(0, 0) | |
3749 mcfh(8, 0) | |
3750 mcfh(0, 8) | |
3751 mcfh(8, 8) | |
2600 | 3752 |
3753 if(!qexp[0]) | |
3754 init_qexp(); | |
3755 | |
2138 | 3756 dec= s->spatial_decomposition_count= 5; |
3757 s->spatial_decomposition_type= avctx->prediction_method; //FIXME add decorrelator type r transform_type | |
2967 | 3758 |
2138 | 3759 s->chroma_h_shift= 1; //FIXME XXX |
3760 s->chroma_v_shift= 1; | |
2967 | 3761 |
2138 | 3762 // dec += FFMAX(s->chroma_h_shift, s->chroma_v_shift); |
2967 | 3763 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3764 width= s->avctx->width; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3765 height= s->avctx->height; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3766 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3767 s->spatial_dwt_buffer= av_mallocz(width*height*sizeof(DWTELEM)); |
2967 | 3768 |
2138 | 3769 s->mv_scale= (s->avctx->flags & CODEC_FLAG_QPEL) ? 2 : 4; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3770 s->block_max_depth= (s->avctx->flags & CODEC_FLAG_4MV) ? 1 : 0; |
2967 | 3771 |
3772 for(plane_index=0; plane_index<3; plane_index++){ | |
2138 | 3773 int w= s->avctx->width; |
3774 int h= s->avctx->height; | |
3775 | |
3776 if(plane_index){ | |
3777 w>>= s->chroma_h_shift; | |
3778 h>>= s->chroma_v_shift; | |
3779 } | |
3780 s->plane[plane_index].width = w; | |
3781 s->plane[plane_index].height= h; | |
2160 | 3782 //av_log(NULL, AV_LOG_DEBUG, "%d %d\n", w, h); |
2138 | 3783 for(level=s->spatial_decomposition_count-1; level>=0; level--){ |
3784 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
3785 SubBand *b= &s->plane[plane_index].band[level][orientation]; | |
2967 | 3786 |
2138 | 3787 b->buf= s->spatial_dwt_buffer; |
3788 b->level= level; | |
3789 b->stride= s->plane[plane_index].width << (s->spatial_decomposition_count - level); | |
3790 b->width = (w + !(orientation&1))>>1; | |
3791 b->height= (h + !(orientation>1))>>1; | |
2967 | 3792 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3793 b->stride_line = 1 << (s->spatial_decomposition_count - level); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3794 b->buf_x_offset = 0; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3795 b->buf_y_offset = 0; |
2967 | 3796 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3797 if(orientation&1){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3798 b->buf += (w+1)>>1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3799 b->buf_x_offset = (w+1)>>1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3800 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3801 if(orientation>1){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3802 b->buf += b->stride>>1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3803 b->buf_y_offset = b->stride_line >> 1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3804 } |
2967 | 3805 |
2138 | 3806 if(level) |
3807 b->parent= &s->plane[plane_index].band[level-1][orientation]; | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3808 b->x_coeff=av_mallocz(((b->width+1) * b->height+1)*sizeof(x_and_coeff)); |
2138 | 3809 } |
3810 w= (w+1)>>1; | |
3811 h= (h+1)>>1; | |
3812 } | |
3813 } | |
2967 | 3814 |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
3815 for(i=0; i<MAX_REF_FRAMES; i++) |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
3816 for(j=0; j<MAX_REF_FRAMES; j++) |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
3817 scale_mv_ref[i][j] = 256*(i+1)/(j+1); |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
3818 |
2138 | 3819 reset_contexts(s); |
2967 | 3820 /* |
2138 | 3821 width= s->width= avctx->width; |
3822 height= s->height= avctx->height; | |
2967 | 3823 |
2138 | 3824 assert(width && height); |
3825 */ | |
3826 s->avctx->get_buffer(s->avctx, &s->mconly_picture); | |
2967 | 3827 |
2138 | 3828 return 0; |
3829 } | |
3830 | |
3322
0b4f548dfb44
Snow: 10l. 1pass ratecontrol failed to set snow's internal quant.
lorenm
parents:
3314
diff
changeset
|
3831 static int qscale2qlog(int qscale){ |
0b4f548dfb44
Snow: 10l. 1pass ratecontrol failed to set snow's internal quant.
lorenm
parents:
3314
diff
changeset
|
3832 return rint(QROOT*log(qscale / (float)FF_QP2LAMBDA)/log(2)) |
0b4f548dfb44
Snow: 10l. 1pass ratecontrol failed to set snow's internal quant.
lorenm
parents:
3314
diff
changeset
|
3833 + 61*QROOT/8; //<64 >60 |
0b4f548dfb44
Snow: 10l. 1pass ratecontrol failed to set snow's internal quant.
lorenm
parents:
3314
diff
changeset
|
3834 } |
0b4f548dfb44
Snow: 10l. 1pass ratecontrol failed to set snow's internal quant.
lorenm
parents:
3314
diff
changeset
|
3835 |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
3836 static int ratecontrol_1pass(SnowContext *s, AVFrame *pict) |
3313 | 3837 { |
3838 /* estimate the frame's complexity as a sum of weighted dwt coefs. | |
3839 * FIXME we know exact mv bits at this point, | |
3840 * but ratecontrol isn't set up to include them. */ | |
3841 uint32_t coef_sum= 0; | |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
3842 int level, orientation, delta_qlog; |
3313 | 3843 |
3844 for(level=0; level<s->spatial_decomposition_count; level++){ | |
3845 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
3846 SubBand *b= &s->plane[0].band[level][orientation]; | |
3847 DWTELEM *buf= b->buf; | |
3848 const int w= b->width; | |
3849 const int h= b->height; | |
3850 const int stride= b->stride; | |
3851 const int qlog= clip(2*QROOT + b->qlog, 0, QROOT*16); | |
3852 const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); | |
3853 const int qdiv= (1<<16)/qmul; | |
3854 int x, y; | |
3855 if(orientation==0) | |
3856 decorrelate(s, b, buf, stride, 1, 0); | |
3857 for(y=0; y<h; y++) | |
3858 for(x=0; x<w; x++) | |
3859 coef_sum+= abs(buf[x+y*stride]) * qdiv >> 16; | |
3860 if(orientation==0) | |
3861 correlate(s, b, buf, stride, 1, 0); | |
3862 } | |
3863 } | |
3864 | |
3865 /* ugly, ratecontrol just takes a sqrt again */ | |
3866 coef_sum = (uint64_t)coef_sum * coef_sum >> 16; | |
3867 assert(coef_sum < INT_MAX); | |
3868 | |
3869 if(pict->pict_type == I_TYPE){ | |
3870 s->m.current_picture.mb_var_sum= coef_sum; | |
3871 s->m.current_picture.mc_mb_var_sum= 0; | |
3872 }else{ | |
3873 s->m.current_picture.mc_mb_var_sum= coef_sum; | |
3874 s->m.current_picture.mb_var_sum= 0; | |
3875 } | |
3876 | |
3877 pict->quality= ff_rate_estimate_qscale(&s->m, 1); | |
3766 | 3878 if (pict->quality < 0) |
3879 return -1; | |
3313 | 3880 s->lambda= pict->quality * 3/2; |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
3881 delta_qlog= qscale2qlog(pict->quality) - s->qlog; |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
3882 s->qlog+= delta_qlog; |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
3883 return delta_qlog; |
3313 | 3884 } |
2138 | 3885 |
3886 static void calculate_vissual_weight(SnowContext *s, Plane *p){ | |
3887 int width = p->width; | |
3888 int height= p->height; | |
2198 | 3889 int level, orientation, x, y; |
2138 | 3890 |
3891 for(level=0; level<s->spatial_decomposition_count; level++){ | |
3892 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
3893 SubBand *b= &p->band[level][orientation]; | |
3894 DWTELEM *buf= b->buf; | |
3895 int64_t error=0; | |
2967 | 3896 |
2138 | 3897 memset(s->spatial_dwt_buffer, 0, sizeof(int)*width*height); |
3898 buf[b->width/2 + b->height/2*b->stride]= 256*256; | |
2164 | 3899 ff_spatial_idwt(s->spatial_dwt_buffer, width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count); |
2138 | 3900 for(y=0; y<height; y++){ |
3901 for(x=0; x<width; x++){ | |
3902 int64_t d= s->spatial_dwt_buffer[x + y*width]; | |
3903 error += d*d; | |
3904 } | |
3905 } | |
3906 | |
3907 b->qlog= (int)(log(352256.0/sqrt(error)) / log(pow(2.0, 1.0/QROOT))+0.5); | |
2164 | 3908 // av_log(NULL, AV_LOG_DEBUG, "%d %d %d\n", level, orientation, b->qlog/*, sqrt(error)*/); |
2138 | 3909 } |
3910 } | |
3911 } | |
3912 | |
3913 static int encode_init(AVCodecContext *avctx) | |
3914 { | |
3915 SnowContext *s = avctx->priv_data; | |
2198 | 3916 int plane_index; |
2138 | 3917 |
2658
d1609cfeb1d0
#defines for strict_std_compliance and split between inofficial extensions and non standarized things
michael
parents:
2635
diff
changeset
|
3918 if(avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL){ |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2610
diff
changeset
|
3919 av_log(avctx, AV_LOG_ERROR, "this codec is under development, files encoded with it may not be decodable with future versions!!!\n" |
2658
d1609cfeb1d0
#defines for strict_std_compliance and split between inofficial extensions and non standarized things
michael
parents:
2635
diff
changeset
|
3920 "use vstrict=-2 / -strict -2 to use it anyway\n"); |
2151 | 3921 return -1; |
3922 } | |
2967 | 3923 |
3327
955096780e7c
Snow: don't try to encode lossless with 9/7 wavelet, because it isn't lossless.
lorenm
parents:
3326
diff
changeset
|
3924 if(avctx->prediction_method == DWT_97 |
955096780e7c
Snow: don't try to encode lossless with 9/7 wavelet, because it isn't lossless.
lorenm
parents:
3326
diff
changeset
|
3925 && (avctx->flags & CODEC_FLAG_QSCALE) |
955096780e7c
Snow: don't try to encode lossless with 9/7 wavelet, because it isn't lossless.
lorenm
parents:
3326
diff
changeset
|
3926 && avctx->global_quality == 0){ |
955096780e7c
Snow: don't try to encode lossless with 9/7 wavelet, because it isn't lossless.
lorenm
parents:
3326
diff
changeset
|
3927 av_log(avctx, AV_LOG_ERROR, "the 9/7 wavelet is incompatible with lossless mode\n"); |
955096780e7c
Snow: don't try to encode lossless with 9/7 wavelet, because it isn't lossless.
lorenm
parents:
3326
diff
changeset
|
3928 return -1; |
955096780e7c
Snow: don't try to encode lossless with 9/7 wavelet, because it isn't lossless.
lorenm
parents:
3326
diff
changeset
|
3929 } |
955096780e7c
Snow: don't try to encode lossless with 9/7 wavelet, because it isn't lossless.
lorenm
parents:
3326
diff
changeset
|
3930 |
2138 | 3931 common_init(avctx); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3932 alloc_blocks(s); |
2967 | 3933 |
2138 | 3934 s->version=0; |
2967 | 3935 |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3936 s->m.avctx = avctx; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3937 s->m.flags = avctx->flags; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3938 s->m.bit_rate= avctx->bit_rate; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3939 |
2138 | 3940 s->m.me.scratchpad= av_mallocz((avctx->width+64)*2*16*2*sizeof(uint8_t)); |
3941 s->m.me.map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t)); | |
3942 s->m.me.score_map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t)); | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3943 s->m.obmc_scratchpad= av_mallocz(MB_SIZE*MB_SIZE*12*sizeof(uint32_t)); |
2138 | 3944 h263_encode_init(&s->m); //mv_penalty |
3945 | |
3314 | 3946 s->max_ref_frames = FFMAX(FFMIN(avctx->refs, MAX_REF_FRAMES), 1); |
3947 | |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3948 if(avctx->flags&CODEC_FLAG_PASS1){ |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3949 if(!avctx->stats_out) |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3950 avctx->stats_out = av_mallocz(256); |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3951 } |
3322
0b4f548dfb44
Snow: 10l. 1pass ratecontrol failed to set snow's internal quant.
lorenm
parents:
3314
diff
changeset
|
3952 if((avctx->flags&CODEC_FLAG_PASS2) || !(avctx->flags&CODEC_FLAG_QSCALE)){ |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3953 if(ff_rate_control_init(&s->m) < 0) |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3954 return -1; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3955 } |
3313 | 3956 s->pass1_rc= !(avctx->flags & (CODEC_FLAG_QSCALE|CODEC_FLAG_PASS2)); |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3957 |
2138 | 3958 for(plane_index=0; plane_index<3; plane_index++){ |
3959 calculate_vissual_weight(s, &s->plane[plane_index]); | |
3960 } | |
2967 | 3961 |
3962 | |
2138 | 3963 avctx->coded_frame= &s->current_picture; |
3964 switch(avctx->pix_fmt){ | |
3965 // case PIX_FMT_YUV444P: | |
3966 // case PIX_FMT_YUV422P: | |
3967 case PIX_FMT_YUV420P: | |
3968 case PIX_FMT_GRAY8: | |
3969 // case PIX_FMT_YUV411P: | |
3970 // case PIX_FMT_YUV410P: | |
3971 s->colorspace_type= 0; | |
3972 break; | |
3973 /* case PIX_FMT_RGBA32: | |
3974 s->colorspace= 1; | |
3975 break;*/ | |
3976 default: | |
3977 av_log(avctx, AV_LOG_ERROR, "format not supported\n"); | |
3978 return -1; | |
3979 } | |
3980 // avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_h_shift, &s->chroma_v_shift); | |
3981 s->chroma_h_shift= 1; | |
3982 s->chroma_v_shift= 1; | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3983 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3984 ff_set_cmp(&s->dsp, s->dsp.me_cmp, s->avctx->me_cmp); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3985 ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, s->avctx->me_sub_cmp); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3986 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3987 s->avctx->get_buffer(s->avctx, &s->input_picture); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3988 |
3314 | 3989 if(s->avctx->me_method == ME_ITER){ |
3990 int i; | |
3991 int size= s->b_width * s->b_height << 2*s->block_max_depth; | |
3992 for(i=0; i<s->max_ref_frames; i++){ | |
3993 s->ref_mvs[i]= av_mallocz(size*sizeof(int16_t[2])); | |
3994 s->ref_scores[i]= av_mallocz(size*sizeof(uint32_t)); | |
3995 } | |
3996 } | |
3997 | |
2138 | 3998 return 0; |
3999 } | |
4000 | |
4001 static int frame_start(SnowContext *s){ | |
4002 AVFrame tmp; | |
2187 | 4003 int w= s->avctx->width; //FIXME round up to x16 ? |
4004 int h= s->avctx->height; | |
2138 | 4005 |
2187 | 4006 if(s->current_picture.data[0]){ |
4007 draw_edges(s->current_picture.data[0], s->current_picture.linesize[0], w , h , EDGE_WIDTH ); | |
4008 draw_edges(s->current_picture.data[1], s->current_picture.linesize[1], w>>1, h>>1, EDGE_WIDTH/2); | |
4009 draw_edges(s->current_picture.data[2], s->current_picture.linesize[2], w>>1, h>>1, EDGE_WIDTH/2); | |
4010 } | |
4011 | |
3314 | 4012 tmp= s->last_picture[s->max_ref_frames-1]; |
4013 memmove(s->last_picture+1, s->last_picture, (s->max_ref_frames-1)*sizeof(AVFrame)); | |
4014 s->last_picture[0]= s->current_picture; | |
2138 | 4015 s->current_picture= tmp; |
2967 | 4016 |
3314 | 4017 if(s->keyframe){ |
4018 s->ref_frames= 0; | |
4019 }else{ | |
4020 int i; | |
4021 for(i=0; i<s->max_ref_frames && s->last_picture[i].data[0]; i++) | |
4022 if(i && s->last_picture[i-1].key_frame) | |
4023 break; | |
4024 s->ref_frames= i; | |
4025 } | |
4026 | |
2138 | 4027 s->current_picture.reference= 1; |
4028 if(s->avctx->get_buffer(s->avctx, &s->current_picture) < 0){ | |
4029 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n"); | |
4030 return -1; | |
4031 } | |
2967 | 4032 |
3314 | 4033 s->current_picture.key_frame= s->keyframe; |
4034 | |
2138 | 4035 return 0; |
4036 } | |
4037 | |
4038 static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size, void *data){ | |
4039 SnowContext *s = avctx->priv_data; | |
2335 | 4040 RangeCoder * const c= &s->c; |
2138 | 4041 AVFrame *pict = data; |
4042 const int width= s->avctx->width; | |
4043 const int height= s->avctx->height; | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4044 int level, orientation, plane_index, i, y; |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4045 uint8_t rc_header_bak[sizeof(s->header_state)]; |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4046 uint8_t rc_block_bak[sizeof(s->block_state)]; |
2138 | 4047 |
2335 | 4048 ff_init_range_encoder(c, buf, buf_size); |
4049 ff_build_rac_states(c, 0.05*(1LL<<32), 256-8); | |
2967 | 4050 |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4051 for(i=0; i<3; i++){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4052 int shift= !!i; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4053 for(y=0; y<(height>>shift); y++) |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4054 memcpy(&s->input_picture.data[i][y * s->input_picture.linesize[i]], |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4055 &pict->data[i][y * pict->linesize[i]], |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4056 width>>shift); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4057 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4058 s->new_picture = *pict; |
2138 | 4059 |
3313 | 4060 s->m.picture_number= avctx->frame_number; |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4061 if(avctx->flags&CODEC_FLAG_PASS2){ |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4062 s->m.pict_type = |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4063 pict->pict_type= s->m.rc_context.entry[avctx->frame_number].new_pict_type; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4064 s->keyframe= pict->pict_type==FF_I_TYPE; |
3766 | 4065 if(!(avctx->flags&CODEC_FLAG_QSCALE)) { |
3193 | 4066 pict->quality= ff_rate_estimate_qscale(&s->m, 0); |
3766 | 4067 if (pict->quality < 0) |
4068 return -1; | |
4069 } | |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4070 }else{ |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4071 s->keyframe= avctx->gop_size==0 || avctx->frame_number % avctx->gop_size == 0; |
3313 | 4072 s->m.pict_type= |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4073 pict->pict_type= s->keyframe ? FF_I_TYPE : FF_P_TYPE; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4074 } |
2967 | 4075 |
3313 | 4076 if(s->pass1_rc && avctx->frame_number == 0) |
4077 pict->quality= 2*FF_QP2LAMBDA; | |
2161 | 4078 if(pict->quality){ |
3322
0b4f548dfb44
Snow: 10l. 1pass ratecontrol failed to set snow's internal quant.
lorenm
parents:
3314
diff
changeset
|
4079 s->qlog= qscale2qlog(pict->quality); |
3313 | 4080 s->lambda = pict->quality * 3/2; |
4081 } | |
4082 if(s->qlog < 0 || (!pict->quality && (avctx->flags & CODEC_FLAG_QSCALE))){ | |
2161 | 4083 s->qlog= LOSSLESS_QLOG; |
3313 | 4084 s->lambda = 0; |
4085 }//else keep previous frame's qlog until after motion est | |
2138 | 4086 |
4087 frame_start(s); | |
4088 | |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4089 s->m.current_picture_ptr= &s->m.current_picture; |
2138 | 4090 if(pict->pict_type == P_TYPE){ |
4091 int block_width = (width +15)>>4; | |
4092 int block_height= (height+15)>>4; | |
4093 int stride= s->current_picture.linesize[0]; | |
2967 | 4094 |
2138 | 4095 assert(s->current_picture.data[0]); |
3314 | 4096 assert(s->last_picture[0].data[0]); |
2967 | 4097 |
2138 | 4098 s->m.avctx= s->avctx; |
4099 s->m.current_picture.data[0]= s->current_picture.data[0]; | |
3314 | 4100 s->m. last_picture.data[0]= s->last_picture[0].data[0]; |
2138 | 4101 s->m. new_picture.data[0]= s-> input_picture.data[0]; |
4102 s->m. last_picture_ptr= &s->m. last_picture; | |
4103 s->m.linesize= | |
4104 s->m. last_picture.linesize[0]= | |
4105 s->m. new_picture.linesize[0]= | |
4106 s->m.current_picture.linesize[0]= stride; | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4107 s->m.uvlinesize= s->current_picture.linesize[1]; |
2138 | 4108 s->m.width = width; |
4109 s->m.height= height; | |
4110 s->m.mb_width = block_width; | |
4111 s->m.mb_height= block_height; | |
4112 s->m.mb_stride= s->m.mb_width+1; | |
4113 s->m.b8_stride= 2*s->m.mb_width+1; | |
4114 s->m.f_code=1; | |
4115 s->m.pict_type= pict->pict_type; | |
4116 s->m.me_method= s->avctx->me_method; | |
4117 s->m.me.scene_change_score=0; | |
4118 s->m.flags= s->avctx->flags; | |
4119 s->m.quarter_sample= (s->avctx->flags & CODEC_FLAG_QPEL)!=0; | |
4120 s->m.out_format= FMT_H263; | |
4121 s->m.unrestricted_mv= 1; | |
4122 | |
3313 | 4123 s->m.lambda = s->lambda; |
2138 | 4124 s->m.qscale= (s->m.lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4125 s->lambda2= s->m.lambda2= (s->m.lambda*s->m.lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4126 |
2138 | 4127 s->m.dsp= s->dsp; //move |
4128 ff_init_me(&s->m); | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4129 s->dsp= s->m.dsp; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4130 } |
2967 | 4131 |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4132 if(s->pass1_rc){ |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4133 memcpy(rc_header_bak, s->header_state, sizeof(s->header_state)); |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4134 memcpy(rc_block_bak, s->block_state, sizeof(s->block_state)); |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4135 } |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4136 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4137 redo_frame: |
2967 | 4138 |
3313 | 4139 s->m.pict_type = pict->pict_type; |
2138 | 4140 s->qbias= pict->pict_type == P_TYPE ? 2 : 0; |
4141 | |
4142 encode_header(s); | |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4143 s->m.misc_bits = 8*(s->c.bytestream - s->c.bytestream_start); |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4144 encode_blocks(s, 1); |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4145 s->m.mv_bits = 8*(s->c.bytestream - s->c.bytestream_start) - s->m.misc_bits; |
2967 | 4146 |
2138 | 4147 for(plane_index=0; plane_index<3; plane_index++){ |
4148 Plane *p= &s->plane[plane_index]; | |
4149 int w= p->width; | |
4150 int h= p->height; | |
4151 int x, y; | |
2198 | 4152 // int bits= put_bits_count(&s->c.pb); |
2138 | 4153 |
3338
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4154 if(!(avctx->flags2 & CODEC_FLAG2_MEMC_ONLY)){ |
2138 | 4155 //FIXME optimize |
4156 if(pict->data[plane_index]) //FIXME gray hack | |
4157 for(y=0; y<h; y++){ | |
4158 for(x=0; x<w; x++){ | |
2246 | 4159 s->spatial_dwt_buffer[y*w + x]= pict->data[plane_index][y*pict->linesize[plane_index] + x]<<FRAC_BITS; |
2138 | 4160 } |
4161 } | |
4162 predict_plane(s, s->spatial_dwt_buffer, plane_index, 0); | |
2967 | 4163 |
4164 if( plane_index==0 | |
4165 && pict->pict_type == P_TYPE | |
3313 | 4166 && !(avctx->flags&CODEC_FLAG_PASS2) |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4167 && s->m.me.scene_change_score > s->avctx->scenechange_threshold){ |
2335 | 4168 ff_init_range_encoder(c, buf, buf_size); |
4169 ff_build_rac_states(c, 0.05*(1LL<<32), 256-8); | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4170 pict->pict_type= FF_I_TYPE; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4171 s->keyframe=1; |
3314 | 4172 s->current_picture.key_frame=1; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4173 reset_contexts(s); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4174 goto redo_frame; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4175 } |
2967 | 4176 |
2161 | 4177 if(s->qlog == LOSSLESS_QLOG){ |
4178 for(y=0; y<h; y++){ | |
4179 for(x=0; x<w; x++){ | |
2595 | 4180 s->spatial_dwt_buffer[y*w + x]= (s->spatial_dwt_buffer[y*w + x] + (1<<(FRAC_BITS-1))-1)>>FRAC_BITS; |
2161 | 4181 } |
4182 } | |
4183 } | |
2967 | 4184 |
2164 | 4185 ff_spatial_dwt(s->spatial_dwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count); |
2161 | 4186 |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4187 if(s->pass1_rc && plane_index==0){ |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4188 int delta_qlog = ratecontrol_1pass(s, pict); |
3766 | 4189 if (delta_qlog < 0) |
4190 return -1; | |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4191 if(delta_qlog){ |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4192 //reordering qlog in the bitstream would eliminate this reset |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4193 ff_init_range_encoder(c, buf, buf_size); |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4194 memcpy(s->header_state, rc_header_bak, sizeof(s->header_state)); |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4195 memcpy(s->block_state, rc_block_bak, sizeof(s->block_state)); |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4196 encode_header(s); |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4197 encode_blocks(s, 0); |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4198 } |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4199 } |
3313 | 4200 |
2138 | 4201 for(level=0; level<s->spatial_decomposition_count; level++){ |
4202 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
4203 SubBand *b= &p->band[level][orientation]; | |
2967 | 4204 |
2138 | 4205 quantize(s, b, b->buf, b->stride, s->qbias); |
4206 if(orientation==0) | |
4207 decorrelate(s, b, b->buf, b->stride, pict->pict_type == P_TYPE, 0); | |
4208 encode_subband(s, b, b->buf, b->parent ? b->parent->buf : NULL, b->stride, orientation); | |
4209 assert(b->parent==NULL || b->parent->stride == b->stride*2); | |
4210 if(orientation==0) | |
4211 correlate(s, b, b->buf, b->stride, 1, 0); | |
4212 } | |
4213 } | |
4214 // av_log(NULL, AV_LOG_DEBUG, "plane:%d bits:%d\n", plane_index, put_bits_count(&s->c.pb) - bits); | |
4215 | |
4216 for(level=0; level<s->spatial_decomposition_count; level++){ | |
4217 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
4218 SubBand *b= &p->band[level][orientation]; | |
4219 | |
4220 dequantize(s, b, b->buf, b->stride); | |
4221 } | |
4222 } | |
2161 | 4223 |
2164 | 4224 ff_spatial_idwt(s->spatial_dwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count); |
2161 | 4225 if(s->qlog == LOSSLESS_QLOG){ |
4226 for(y=0; y<h; y++){ | |
4227 for(x=0; x<w; x++){ | |
2246 | 4228 s->spatial_dwt_buffer[y*w + x]<<=FRAC_BITS; |
2161 | 4229 } |
4230 } | |
4231 } | |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
4232 {START_TIMER |
2138 | 4233 predict_plane(s, s->spatial_dwt_buffer, plane_index, 1); |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
4234 STOP_TIMER("pred-conv")} |
3338
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4235 }else{ |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4236 //ME/MC only |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4237 if(pict->pict_type == I_TYPE){ |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4238 for(y=0; y<h; y++){ |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4239 for(x=0; x<w; x++){ |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4240 s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x]= |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4241 pict->data[plane_index][y*pict->linesize[plane_index] + x]; |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4242 } |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4243 } |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4244 }else{ |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4245 memset(s->spatial_dwt_buffer, 0, sizeof(DWTELEM)*w*h); |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4246 predict_plane(s, s->spatial_dwt_buffer, plane_index, 1); |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4247 } |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4248 } |
2138 | 4249 if(s->avctx->flags&CODEC_FLAG_PSNR){ |
4250 int64_t error= 0; | |
2967 | 4251 |
2138 | 4252 if(pict->data[plane_index]) //FIXME gray hack |
4253 for(y=0; y<h; y++){ | |
4254 for(x=0; x<w; x++){ | |
2161 | 4255 int d= s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x] - pict->data[plane_index][y*pict->linesize[plane_index] + x]; |
2138 | 4256 error += d*d; |
4257 } | |
4258 } | |
4259 s->avctx->error[plane_index] += error; | |
2232 | 4260 s->current_picture.error[plane_index] = error; |
2138 | 4261 } |
4262 } | |
4263 | |
3314 | 4264 if(s->last_picture[s->max_ref_frames-1].data[0]) |
4265 avctx->release_buffer(avctx, &s->last_picture[s->max_ref_frames-1]); | |
2138 | 4266 |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4267 s->current_picture.coded_picture_number = avctx->frame_number; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4268 s->current_picture.pict_type = pict->pict_type; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4269 s->current_picture.quality = pict->quality; |
3313 | 4270 s->m.frame_bits = 8*(s->c.bytestream - s->c.bytestream_start); |
4271 s->m.p_tex_bits = s->m.frame_bits - s->m.misc_bits - s->m.mv_bits; | |
4272 s->m.current_picture.display_picture_number = | |
4273 s->m.current_picture.coded_picture_number = avctx->frame_number; | |
4274 s->m.current_picture.quality = pict->quality; | |
4275 s->m.total_bits += 8*(s->c.bytestream - s->c.bytestream_start); | |
4276 if(s->pass1_rc) | |
3766 | 4277 if (ff_rate_estimate_qscale(&s->m, 0) < 0) |
4278 return -1; | |
3313 | 4279 if(avctx->flags&CODEC_FLAG_PASS1) |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4280 ff_write_pass1_stats(&s->m); |
3313 | 4281 s->m.last_pict_type = s->m.pict_type; |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4282 |
2138 | 4283 emms_c(); |
2967 | 4284 |
2335 | 4285 return ff_rac_terminate(c); |
2138 | 4286 } |
4287 | |
4288 static void common_end(SnowContext *s){ | |
3314 | 4289 int plane_index, level, orientation, i; |
2192 | 4290 |
2138 | 4291 av_freep(&s->spatial_dwt_buffer); |
4292 | |
2967 | 4293 av_freep(&s->m.me.scratchpad); |
2138 | 4294 av_freep(&s->m.me.map); |
4295 av_freep(&s->m.me.score_map); | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
4296 av_freep(&s->m.obmc_scratchpad); |
2967 | 4297 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4298 av_freep(&s->block); |
2192 | 4299 |
3314 | 4300 for(i=0; i<MAX_REF_FRAMES; i++){ |
4301 av_freep(&s->ref_mvs[i]); | |
4302 av_freep(&s->ref_scores[i]); | |
4303 if(s->last_picture[i].data[0]) | |
4304 s->avctx->release_buffer(s->avctx, &s->last_picture[i]); | |
4305 } | |
4306 | |
2967 | 4307 for(plane_index=0; plane_index<3; plane_index++){ |
2192 | 4308 for(level=s->spatial_decomposition_count-1; level>=0; level--){ |
4309 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
4310 SubBand *b= &s->plane[plane_index].band[level][orientation]; | |
2967 | 4311 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4312 av_freep(&b->x_coeff); |
2192 | 4313 } |
4314 } | |
4315 } | |
2138 | 4316 } |
4317 | |
4318 static int encode_end(AVCodecContext *avctx) | |
4319 { | |
4320 SnowContext *s = avctx->priv_data; | |
4321 | |
4322 common_end(s); | |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4323 av_free(avctx->stats_out); |
2138 | 4324 |
4325 return 0; | |
4326 } | |
4327 | |
4328 static int decode_init(AVCodecContext *avctx) | |
4329 { | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4330 SnowContext *s = avctx->priv_data; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4331 int block_size; |
2967 | 4332 |
2635 | 4333 avctx->pix_fmt= PIX_FMT_YUV420P; |
2138 | 4334 |
4335 common_init(avctx); | |
2967 | 4336 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4337 block_size = MB_SIZE >> s->block_max_depth; |
3017 | 4338 slice_buffer_init(&s->sb, s->plane[0].height, (block_size) + (s->spatial_decomposition_count * (s->spatial_decomposition_count + 3)) + 1, s->plane[0].width, s->spatial_dwt_buffer); |
2967 | 4339 |
2138 | 4340 return 0; |
4341 } | |
4342 | |
4343 static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, uint8_t *buf, int buf_size){ | |
4344 SnowContext *s = avctx->priv_data; | |
2335 | 4345 RangeCoder * const c= &s->c; |
2138 | 4346 int bytes_read; |
4347 AVFrame *picture = data; | |
2198 | 4348 int level, orientation, plane_index; |
2138 | 4349 |
2335 | 4350 ff_init_range_decoder(c, buf, buf_size); |
4351 ff_build_rac_states(c, 0.05*(1LL<<32), 256-8); | |
2138 | 4352 |
4353 s->current_picture.pict_type= FF_I_TYPE; //FIXME I vs. P | |
4354 decode_header(s); | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4355 if(!s->block) alloc_blocks(s); |
2138 | 4356 |
4357 frame_start(s); | |
4358 //keyframe flag dupliaction mess FIXME | |
4359 if(avctx->debug&FF_DEBUG_PICT_INFO) | |
4360 av_log(avctx, AV_LOG_ERROR, "keyframe:%d qlog:%d\n", s->keyframe, s->qlog); | |
2967 | 4361 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4362 decode_blocks(s); |
2138 | 4363 |
4364 for(plane_index=0; plane_index<3; plane_index++){ | |
4365 Plane *p= &s->plane[plane_index]; | |
4366 int w= p->width; | |
4367 int h= p->height; | |
4368 int x, y; | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4369 int decode_state[MAX_DECOMPOSITIONS][4][1]; /* Stored state info for unpack_coeffs. 1 variable per instance. */ |
2967 | 4370 |
2138 | 4371 if(s->avctx->debug&2048){ |
4372 memset(s->spatial_dwt_buffer, 0, sizeof(DWTELEM)*w*h); | |
4373 predict_plane(s, s->spatial_dwt_buffer, plane_index, 1); | |
4374 | |
4375 for(y=0; y<h; y++){ | |
4376 for(x=0; x<w; x++){ | |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
4377 int v= s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x]; |
2138 | 4378 s->mconly_picture.data[plane_index][y*s->mconly_picture.linesize[plane_index] + x]= v; |
4379 } | |
4380 } | |
4381 } | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4382 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4383 { START_TIMER |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4384 for(level=0; level<s->spatial_decomposition_count; level++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4385 for(orientation=level ? 1 : 0; orientation<4; orientation++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4386 SubBand *b= &p->band[level][orientation]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4387 unpack_coeffs(s, b, b->parent, orientation); |
2138 | 4388 } |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4389 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4390 STOP_TIMER("unpack coeffs"); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4391 } |
2138 | 4392 |
2562 | 4393 {START_TIMER |
4394 const int mb_h= s->b_height << s->block_max_depth; | |
4395 const int block_size = MB_SIZE >> s->block_max_depth; | |
4396 const int block_w = plane_index ? block_size/2 : block_size; | |
4397 int mb_y; | |
4398 dwt_compose_t cs[MAX_DECOMPOSITIONS]; | |
4399 int yd=0, yq=0; | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4400 int y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4401 int end_y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4402 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4403 ff_spatial_idwt_buffered_init(cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count); |
2562 | 4404 for(mb_y=0; mb_y<=mb_h; mb_y++){ |
2967 | 4405 |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4406 int slice_starty = block_w*mb_y; |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4407 int slice_h = block_w*(mb_y+1); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4408 if (!(s->keyframe || s->avctx->debug&512)){ |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4409 slice_starty = FFMAX(0, slice_starty - (block_w >> 1)); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4410 slice_h -= (block_w >> 1); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4411 } |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4412 |
2967 | 4413 { |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4414 START_TIMER |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4415 for(level=0; level<s->spatial_decomposition_count; level++){ |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4416 for(orientation=level ? 1 : 0; orientation<4; orientation++){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4417 SubBand *b= &p->band[level][orientation]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4418 int start_y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4419 int end_y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4420 int our_mb_start = mb_y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4421 int our_mb_end = (mb_y + 1); |
3012 | 4422 const int extra= 3; |
4423 start_y = (mb_y ? ((block_w * our_mb_start) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra: 0); | |
4424 end_y = (((block_w * our_mb_end) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra); | |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4425 if (!(s->keyframe || s->avctx->debug&512)){ |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4426 start_y = FFMAX(0, start_y - (block_w >> (1+s->spatial_decomposition_count - level))); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4427 end_y = FFMAX(0, end_y - (block_w >> (1+s->spatial_decomposition_count - level))); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4428 } |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4429 start_y = FFMIN(b->height, start_y); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4430 end_y = FFMIN(b->height, end_y); |
2967 | 4431 |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4432 if (start_y != end_y){ |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4433 if (orientation == 0){ |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4434 SubBand * correlate_band = &p->band[0][0]; |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4435 int correlate_end_y = FFMIN(b->height, end_y + 1); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4436 int correlate_start_y = FFMIN(b->height, (start_y ? start_y + 1 : 0)); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4437 decode_subband_slice_buffered(s, correlate_band, &s->sb, correlate_start_y, correlate_end_y, decode_state[0][0]); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4438 correlate_slice_buffered(s, &s->sb, correlate_band, correlate_band->buf, correlate_band->stride, 1, 0, correlate_start_y, correlate_end_y); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4439 dequantize_slice_buffered(s, &s->sb, correlate_band, correlate_band->buf, correlate_band->stride, start_y, end_y); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4440 } |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4441 else |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4442 decode_subband_slice_buffered(s, b, &s->sb, start_y, end_y, decode_state[level][orientation]); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4443 } |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4444 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4445 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4446 STOP_TIMER("decode_subband_slice"); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4447 } |
2967 | 4448 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4449 { START_TIMER |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4450 for(; yd<slice_h; yd+=4){ |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
4451 ff_spatial_idwt_buffered_slice(&s->dsp, cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count, yd); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4452 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4453 STOP_TIMER("idwt slice");} |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4454 |
2967 | 4455 |
2161 | 4456 if(s->qlog == LOSSLESS_QLOG){ |
2562 | 4457 for(; yq<slice_h && yq<h; yq++){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4458 DWTELEM * line = slice_buffer_get_line(&s->sb, yq); |
2161 | 4459 for(x=0; x<w; x++){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4460 line[x] <<= FRAC_BITS; |
2161 | 4461 } |
4462 } | |
4463 } | |
2562 | 4464 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4465 predict_slice_buffered(s, &s->sb, s->spatial_dwt_buffer, plane_index, 1, mb_y); |
2967 | 4466 |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4467 y = FFMIN(p->height, slice_starty); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4468 end_y = FFMIN(p->height, slice_h); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4469 while(y < end_y) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4470 slice_buffer_release(&s->sb, y++); |
2562 | 4471 } |
2967 | 4472 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4473 slice_buffer_flush(&s->sb); |
2967 | 4474 |
2562 | 4475 STOP_TIMER("idwt + predict_slices")} |
2138 | 4476 } |
2967 | 4477 |
2138 | 4478 emms_c(); |
4479 | |
3314 | 4480 if(s->last_picture[s->max_ref_frames-1].data[0]) |
4481 avctx->release_buffer(avctx, &s->last_picture[s->max_ref_frames-1]); | |
2138 | 4482 |
2967 | 4483 if(!(s->avctx->debug&2048)) |
2138 | 4484 *picture= s->current_picture; |
4485 else | |
4486 *picture= s->mconly_picture; | |
2967 | 4487 |
2138 | 4488 *data_size = sizeof(AVFrame); |
2967 | 4489 |
2335 | 4490 bytes_read= c->bytestream - c->bytestream_start; |
4491 if(bytes_read ==0) av_log(s->avctx, AV_LOG_ERROR, "error at end of frame\n"); //FIXME | |
2138 | 4492 |
4493 return bytes_read; | |
4494 } | |
4495 | |
4496 static int decode_end(AVCodecContext *avctx) | |
4497 { | |
4498 SnowContext *s = avctx->priv_data; | |
4499 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4500 slice_buffer_destroy(&s->sb); |
2967 | 4501 |
2138 | 4502 common_end(s); |
4503 | |
4504 return 0; | |
4505 } | |
4506 | |
4507 AVCodec snow_decoder = { | |
4508 "snow", | |
4509 CODEC_TYPE_VIDEO, | |
4510 CODEC_ID_SNOW, | |
4511 sizeof(SnowContext), | |
4512 decode_init, | |
4513 NULL, | |
4514 decode_end, | |
4515 decode_frame, | |
4516 0 /*CODEC_CAP_DR1*/ /*| CODEC_CAP_DRAW_HORIZ_BAND*/, | |
4517 NULL | |
4518 }; | |
4519 | |
2408
a6e4da1c28ee
Disable encoders patch by (Gianluigi Tiesi <mplayer netfarm it>)
michael
parents:
2368
diff
changeset
|
4520 #ifdef CONFIG_ENCODERS |
2138 | 4521 AVCodec snow_encoder = { |
4522 "snow", | |
4523 CODEC_TYPE_VIDEO, | |
4524 CODEC_ID_SNOW, | |
4525 sizeof(SnowContext), | |
4526 encode_init, | |
4527 encode_frame, | |
4528 encode_end, | |
4529 }; | |
2408
a6e4da1c28ee
Disable encoders patch by (Gianluigi Tiesi <mplayer netfarm it>)
michael
parents:
2368
diff
changeset
|
4530 #endif |
2138 | 4531 |
4532 | |
4533 #if 0 | |
4534 #undef malloc | |
4535 #undef free | |
4536 #undef printf | |
4537 | |
4538 int main(){ | |
4539 int width=256; | |
4540 int height=256; | |
4541 int buffer[2][width*height]; | |
4542 SnowContext s; | |
4543 int i; | |
4544 s.spatial_decomposition_count=6; | |
4545 s.spatial_decomposition_type=1; | |
2967 | 4546 |
2138 | 4547 printf("testing 5/3 DWT\n"); |
4548 for(i=0; i<width*height; i++) | |
4549 buffer[0][i]= buffer[1][i]= random()%54321 - 12345; | |
2967 | 4550 |
2951 | 4551 ff_spatial_dwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); |
4552 ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); | |
2967 | 4553 |
2138 | 4554 for(i=0; i<width*height; i++) |
4555 if(buffer[0][i]!= buffer[1][i]) printf("fsck: %d %d %d\n",i, buffer[0][i], buffer[1][i]); | |
4556 | |
4557 printf("testing 9/7 DWT\n"); | |
4558 s.spatial_decomposition_type=0; | |
4559 for(i=0; i<width*height; i++) | |
4560 buffer[0][i]= buffer[1][i]= random()%54321 - 12345; | |
2967 | 4561 |
2951 | 4562 ff_spatial_dwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); |
4563 ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); | |
2967 | 4564 |
2138 | 4565 for(i=0; i<width*height; i++) |
2951 | 4566 if(ABS(buffer[0][i] - buffer[1][i])>20) printf("fsck: %d %d %d\n",i, buffer[0][i], buffer[1][i]); |
2967 | 4567 |
2951 | 4568 #if 0 |
2138 | 4569 printf("testing AC coder\n"); |
4570 memset(s.header_state, 0, sizeof(s.header_state)); | |
2335 | 4571 ff_init_range_encoder(&s.c, buffer[0], 256*256); |
2138 | 4572 ff_init_cabac_states(&s.c, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64); |
2967 | 4573 |
2138 | 4574 for(i=-256; i<256; i++){ |
4575 START_TIMER | |
4576 put_symbol(&s.c, s.header_state, i*i*i/3*ABS(i), 1); | |
4577 STOP_TIMER("put_symbol") | |
4578 } | |
2335 | 4579 ff_rac_terminate(&s.c); |
2138 | 4580 |
4581 memset(s.header_state, 0, sizeof(s.header_state)); | |
2335 | 4582 ff_init_range_decoder(&s.c, buffer[0], 256*256); |
2138 | 4583 ff_init_cabac_states(&s.c, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64); |
2967 | 4584 |
2138 | 4585 for(i=-256; i<256; i++){ |
4586 int j; | |
4587 START_TIMER | |
4588 j= get_symbol(&s.c, s.header_state, 1); | |
4589 STOP_TIMER("get_symbol") | |
4590 if(j!=i*i*i/3*ABS(i)) printf("fsck: %d != %d\n", i, j); | |
4591 } | |
2951 | 4592 #endif |
2138 | 4593 { |
4594 int level, orientation, x, y; | |
4595 int64_t errors[8][4]; | |
4596 int64_t g=0; | |
4597 | |
4598 memset(errors, 0, sizeof(errors)); | |
4599 s.spatial_decomposition_count=3; | |
4600 s.spatial_decomposition_type=0; | |
4601 for(level=0; level<s.spatial_decomposition_count; level++){ | |
4602 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
4603 int w= width >> (s.spatial_decomposition_count-level); | |
4604 int h= height >> (s.spatial_decomposition_count-level); | |
4605 int stride= width << (s.spatial_decomposition_count-level); | |
4606 DWTELEM *buf= buffer[0]; | |
4607 int64_t error=0; | |
4608 | |
4609 if(orientation&1) buf+=w; | |
4610 if(orientation>1) buf+=stride>>1; | |
2967 | 4611 |
2138 | 4612 memset(buffer[0], 0, sizeof(int)*width*height); |
4613 buf[w/2 + h/2*stride]= 256*256; | |
2951 | 4614 ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); |
2138 | 4615 for(y=0; y<height; y++){ |
4616 for(x=0; x<width; x++){ | |
4617 int64_t d= buffer[0][x + y*width]; | |
4618 error += d*d; | |
4619 if(ABS(width/2-x)<9 && ABS(height/2-y)<9 && level==2) printf("%8lld ", d); | |
4620 } | |
4621 if(ABS(height/2-y)<9 && level==2) printf("\n"); | |
4622 } | |
4623 error= (int)(sqrt(error)+0.5); | |
4624 errors[level][orientation]= error; | |
4625 if(g) g=ff_gcd(g, error); | |
4626 else g= error; | |
4627 } | |
4628 } | |
4629 printf("static int const visual_weight[][4]={\n"); | |
4630 for(level=0; level<s.spatial_decomposition_count; level++){ | |
4631 printf(" {"); | |
4632 for(orientation=0; orientation<4; orientation++){ | |
4633 printf("%8lld,", errors[level][orientation]/g); | |
4634 } | |
4635 printf("},\n"); | |
4636 } | |
4637 printf("};\n"); | |
4638 { | |
4639 int level=2; | |
4640 int orientation=3; | |
4641 int w= width >> (s.spatial_decomposition_count-level); | |
4642 int h= height >> (s.spatial_decomposition_count-level); | |
4643 int stride= width << (s.spatial_decomposition_count-level); | |
4644 DWTELEM *buf= buffer[0]; | |
4645 int64_t error=0; | |
4646 | |
4647 buf+=w; | |
4648 buf+=stride>>1; | |
2967 | 4649 |
2138 | 4650 memset(buffer[0], 0, sizeof(int)*width*height); |
4651 #if 1 | |
4652 for(y=0; y<height; y++){ | |
4653 for(x=0; x<width; x++){ | |
4654 int tab[4]={0,2,3,1}; | |
4655 buffer[0][x+width*y]= 256*256*tab[(x&1) + 2*(y&1)]; | |
4656 } | |
4657 } | |
2951 | 4658 ff_spatial_dwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); |
2138 | 4659 #else |
4660 for(y=0; y<h; y++){ | |
4661 for(x=0; x<w; x++){ | |
4662 buf[x + y*stride ]=169; | |
4663 buf[x + y*stride-w]=64; | |
4664 } | |
4665 } | |
2951 | 4666 ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); |
2138 | 4667 #endif |
4668 for(y=0; y<height; y++){ | |
4669 for(x=0; x<width; x++){ | |
4670 int64_t d= buffer[0][x + y*width]; | |
4671 error += d*d; | |
4672 if(ABS(width/2-x)<9 && ABS(height/2-y)<9) printf("%8lld ", d); | |
4673 } | |
4674 if(ABS(height/2-y)<9) printf("\n"); | |
4675 } | |
4676 } | |
4677 | |
4678 } | |
4679 return 0; | |
4680 } | |
4681 #endif | |
4682 |