Mercurial > libavcodec.hg
annotate snow.c @ 3051:c0fde3eb7784 libavcodec
obmc-aware 4mv
author | lorenm |
---|---|
date | Tue, 17 Jan 2006 08:27:39 +0000 |
parents | 0b546eab515d |
children | 2991dc513726 |
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" | |
2335 | 22 |
23 #include "rangecoder.h" | |
24 #define MID_STATE 128 | |
2138 | 25 |
26 #include "mpegvideo.h" | |
27 | |
28 #undef NDEBUG | |
29 #include <assert.h> | |
30 | |
31 #define MAX_DECOMPOSITIONS 8 | |
32 #define MAX_PLANES 4 | |
33 #define DWTELEM int | |
2601
c31a28f27d9a
increasing precission of the quantization parameter
michael
parents:
2600
diff
changeset
|
34 #define QSHIFT 5 |
2600 | 35 #define QROOT (1<<QSHIFT) |
2161 | 36 #define LOSSLESS_QLOG -128 |
2246 | 37 #define FRAC_BITS 8 |
2138 | 38 |
39 static const int8_t quant3[256]={ | |
40 0, 0, 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, 1, | |
48 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, | |
49 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, | |
50 -1,-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, 0, | |
56 }; | |
57 static const int8_t quant3b[256]={ | |
58 0, 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 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, | |
67 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, | |
68 -1,-1,-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 }; | |
2596 | 75 static const int8_t quant3bA[256]={ |
76 0, 0, 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 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
85 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
86 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
87 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
88 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
89 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
90 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
91 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, | |
92 }; | |
2138 | 93 static const int8_t quant5[256]={ |
94 0, 1, 1, 1, 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, 2, 2, 2, | |
102 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2, | |
103 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2, | |
104 -2,-2,-2,-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,-2,-2,-2,-2,-2,-2,-2,-2, | |
107 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2, | |
108 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2, | |
109 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-1,-1,-1, | |
110 }; | |
111 static const int8_t quant7[256]={ | |
112 0, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | |
113 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | |
114 2, 2, 2, 2, 2, 2, 2, 2, 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, 3, 3, 3, 3, 3, 3, 3, | |
118 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, | |
119 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, | |
120 -3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3, | |
121 -3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3, | |
122 -3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3, | |
123 -3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3, | |
124 -3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3, | |
125 -3,-3,-3,-3,-3,-3,-3,-3,-3,-2,-2,-2,-2,-2,-2,-2, | |
126 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2, | |
127 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-1,-1, | |
128 }; | |
129 static const int8_t quant9[256]={ | |
130 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3, | |
131 3, 3, 3, 3, 3, 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, 4, 4, 4, 4, | |
137 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, | |
138 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4, | |
139 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4, | |
140 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-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,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4, | |
143 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4, | |
144 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-3,-3,-3,-3, | |
145 -3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-2,-2,-2,-2,-1,-1, | |
146 }; | |
147 static const int8_t quant11[256]={ | |
148 0, 1, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, | |
149 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, | |
150 4, 4, 4, 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, 5, 5, | |
154 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, | |
155 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, | |
156 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5, | |
157 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5, | |
158 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5, | |
159 -5,-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,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-4,-4, | |
162 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4, | |
163 -4,-4,-4,-4,-4,-3,-3,-3,-3,-3,-3,-3,-2,-2,-2,-1, | |
164 }; | |
165 static const int8_t quant13[256]={ | |
166 0, 1, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, | |
167 4, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, | |
168 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, | |
169 5, 5, 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, 6, | |
171 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, | |
172 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, | |
173 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, | |
174 -6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6, | |
175 -6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6, | |
176 -6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6, | |
177 -6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6, | |
178 -6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-5, | |
179 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5, | |
180 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5, | |
181 -4,-4,-4,-4,-4,-4,-4,-4,-4,-3,-3,-3,-3,-2,-2,-1, | |
182 }; | |
183 | |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
184 #define LOG2_OBMC_MAX 6 |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
185 #define OBMC_MAX (1<<(LOG2_OBMC_MAX)) |
2138 | 186 #if 0 //64*cubic |
187 static const uint8_t obmc32[1024]={ | |
188 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, | |
189 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, | |
190 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, | |
191 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, | |
192 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, | |
193 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, | |
194 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, | |
195 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, | |
196 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, | |
197 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, | |
198 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, | |
199 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, | |
200 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, | |
201 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, | |
202 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, | |
203 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, | |
204 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, | |
205 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, | |
206 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, | |
207 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, | |
208 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, | |
209 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, | |
210 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, | |
211 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, | |
212 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, | |
213 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, | |
214 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, | |
215 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, | |
216 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, | |
217 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, | |
218 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, | |
219 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, | |
220 //error:0.000022 | |
221 }; | |
222 static const uint8_t obmc16[256]={ | |
223 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, | |
224 0, 1, 1, 2, 4, 5, 5, 6, 6, 5, 5, 4, 2, 1, 1, 0, | |
225 0, 1, 4, 6, 9,11,13,15,15,13,11, 9, 6, 4, 1, 0, | |
226 0, 2, 6,11,15,20,24,26,26,24,20,15,11, 6, 2, 0, | |
227 0, 4, 9,15,23,29,34,38,38,34,29,23,15, 9, 4, 0, | |
228 0, 5,11,20,29,38,45,49,49,45,38,29,20,11, 5, 0, | |
229 1, 5,13,24,34,45,53,57,57,53,45,34,24,13, 5, 1, | |
230 1, 6,15,26,38,49,57,62,62,57,49,38,26,15, 6, 1, | |
231 1, 6,15,26,38,49,57,62,62,57,49,38,26,15, 6, 1, | |
232 1, 5,13,24,34,45,53,57,57,53,45,34,24,13, 5, 1, | |
233 0, 5,11,20,29,38,45,49,49,45,38,29,20,11, 5, 0, | |
234 0, 4, 9,15,23,29,34,38,38,34,29,23,15, 9, 4, 0, | |
235 0, 2, 6,11,15,20,24,26,26,24,20,15,11, 6, 2, 0, | |
236 0, 1, 4, 6, 9,11,13,15,15,13,11, 9, 6, 4, 1, 0, | |
237 0, 1, 1, 2, 4, 5, 5, 6, 6, 5, 5, 4, 2, 1, 1, 0, | |
238 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, | |
239 //error:0.000033 | |
240 }; | |
241 #elif 1 // 64*linear | |
242 static const uint8_t obmc32[1024]={ | |
243 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, | |
244 0, 1, 1, 1, 2, 2, 2, 3, 3, 4, 4, 4, 5, 5, 5, 6, 6, 5, 5, 5, 4, 4, 4, 3, 3, 2, 2, 2, 1, 1, 1, 0, | |
245 0, 1, 2, 2, 3, 3, 4, 5, 5, 6, 7, 7, 8, 8, 9,10,10, 9, 8, 8, 7, 7, 6, 5, 5, 4, 3, 3, 2, 2, 1, 0, | |
246 0, 1, 2, 3, 4, 5, 6, 7, 7, 8, 9,10,11,12,13,14,14,13,12,11,10, 9, 8, 7, 7, 6, 5, 4, 3, 2, 1, 0, | |
247 1, 2, 3, 4, 5, 6, 7, 8,10,11,12,13,14,15,16,17,17,16,15,14,13,12,11,10, 8, 7, 6, 5, 4, 3, 2, 1, | |
248 1, 2, 3, 5, 6, 8, 9,10,12,13,14,16,17,19,20,21,21,20,19,17,16,14,13,12,10, 9, 8, 6, 5, 3, 2, 1, | |
249 1, 2, 4, 6, 7, 9,11,12,14,15,17,19,20,22,24,25,25,24,22,20,19,17,15,14,12,11, 9, 7, 6, 4, 2, 1, | |
250 1, 3, 5, 7, 8,10,12,14,16,18,20,22,23,25,27,29,29,27,25,23,22,20,18,16,14,12,10, 8, 7, 5, 3, 1, | |
251 1, 3, 5, 7,10,12,14,16,18,20,22,24,27,29,31,33,33,31,29,27,24,22,20,18,16,14,12,10, 7, 5, 3, 1, | |
252 1, 4, 6, 8,11,13,15,18,20,23,25,27,30,32,34,37,37,34,32,30,27,25,23,20,18,15,13,11, 8, 6, 4, 1, | |
253 1, 4, 7, 9,12,14,17,20,22,25,28,30,33,35,38,41,41,38,35,33,30,28,25,22,20,17,14,12, 9, 7, 4, 1, | |
254 1, 4, 7,10,13,16,19,22,24,27,30,33,36,39,42,45,45,42,39,36,33,30,27,24,22,19,16,13,10, 7, 4, 1, | |
255 2, 5, 8,11,14,17,20,23,27,30,33,36,39,42,45,48,48,45,42,39,36,33,30,27,23,20,17,14,11, 8, 5, 2, | |
256 2, 5, 8,12,15,19,22,25,29,32,35,39,42,46,49,52,52,49,46,42,39,35,32,29,25,22,19,15,12, 8, 5, 2, | |
257 2, 5, 9,13,16,20,24,27,31,34,38,42,45,49,53,56,56,53,49,45,42,38,34,31,27,24,20,16,13, 9, 5, 2, | |
258 2, 6,10,14,17,21,25,29,33,37,41,45,48,52,56,60,60,56,52,48,45,41,37,33,29,25,21,17,14,10, 6, 2, | |
259 2, 6,10,14,17,21,25,29,33,37,41,45,48,52,56,60,60,56,52,48,45,41,37,33,29,25,21,17,14,10, 6, 2, | |
260 2, 5, 9,13,16,20,24,27,31,34,38,42,45,49,53,56,56,53,49,45,42,38,34,31,27,24,20,16,13, 9, 5, 2, | |
261 2, 5, 8,12,15,19,22,25,29,32,35,39,42,46,49,52,52,49,46,42,39,35,32,29,25,22,19,15,12, 8, 5, 2, | |
262 2, 5, 8,11,14,17,20,23,27,30,33,36,39,42,45,48,48,45,42,39,36,33,30,27,23,20,17,14,11, 8, 5, 2, | |
263 1, 4, 7,10,13,16,19,22,24,27,30,33,36,39,42,45,45,42,39,36,33,30,27,24,22,19,16,13,10, 7, 4, 1, | |
264 1, 4, 7, 9,12,14,17,20,22,25,28,30,33,35,38,41,41,38,35,33,30,28,25,22,20,17,14,12, 9, 7, 4, 1, | |
265 1, 4, 6, 8,11,13,15,18,20,23,25,27,30,32,34,37,37,34,32,30,27,25,23,20,18,15,13,11, 8, 6, 4, 1, | |
266 1, 3, 5, 7,10,12,14,16,18,20,22,24,27,29,31,33,33,31,29,27,24,22,20,18,16,14,12,10, 7, 5, 3, 1, | |
267 1, 3, 5, 7, 8,10,12,14,16,18,20,22,23,25,27,29,29,27,25,23,22,20,18,16,14,12,10, 8, 7, 5, 3, 1, | |
268 1, 2, 4, 6, 7, 9,11,12,14,15,17,19,20,22,24,25,25,24,22,20,19,17,15,14,12,11, 9, 7, 6, 4, 2, 1, | |
269 1, 2, 3, 5, 6, 8, 9,10,12,13,14,16,17,19,20,21,21,20,19,17,16,14,13,12,10, 9, 8, 6, 5, 3, 2, 1, | |
270 1, 2, 3, 4, 5, 6, 7, 8,10,11,12,13,14,15,16,17,17,16,15,14,13,12,11,10, 8, 7, 6, 5, 4, 3, 2, 1, | |
271 0, 1, 2, 3, 4, 5, 6, 7, 7, 8, 9,10,11,12,13,14,14,13,12,11,10, 9, 8, 7, 7, 6, 5, 4, 3, 2, 1, 0, | |
272 0, 1, 2, 2, 3, 3, 4, 5, 5, 6, 7, 7, 8, 8, 9,10,10, 9, 8, 8, 7, 7, 6, 5, 5, 4, 3, 3, 2, 2, 1, 0, | |
273 0, 1, 1, 1, 2, 2, 2, 3, 3, 4, 4, 4, 5, 5, 5, 6, 6, 5, 5, 5, 4, 4, 4, 3, 3, 2, 2, 2, 1, 1, 1, 0, | |
274 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, | |
275 //error:0.000020 | |
276 }; | |
277 static const uint8_t obmc16[256]={ | |
278 0, 1, 1, 2, 2, 3, 3, 4, 4, 3, 3, 2, 2, 1, 1, 0, | |
279 1, 2, 4, 5, 7, 8,10,11,11,10, 8, 7, 5, 4, 2, 1, | |
280 1, 4, 6, 9,11,14,16,19,19,16,14,11, 9, 6, 4, 1, | |
281 2, 5, 9,12,16,19,23,26,26,23,19,16,12, 9, 5, 2, | |
282 2, 7,11,16,20,25,29,34,34,29,25,20,16,11, 7, 2, | |
283 3, 8,14,19,25,30,36,41,41,36,30,25,19,14, 8, 3, | |
284 3,10,16,23,29,36,42,49,49,42,36,29,23,16,10, 3, | |
285 4,11,19,26,34,41,49,56,56,49,41,34,26,19,11, 4, | |
286 4,11,19,26,34,41,49,56,56,49,41,34,26,19,11, 4, | |
287 3,10,16,23,29,36,42,49,49,42,36,29,23,16,10, 3, | |
288 3, 8,14,19,25,30,36,41,41,36,30,25,19,14, 8, 3, | |
289 2, 7,11,16,20,25,29,34,34,29,25,20,16,11, 7, 2, | |
290 2, 5, 9,12,16,19,23,26,26,23,19,16,12, 9, 5, 2, | |
291 1, 4, 6, 9,11,14,16,19,19,16,14,11, 9, 6, 4, 1, | |
292 1, 2, 4, 5, 7, 8,10,11,11,10, 8, 7, 5, 4, 2, 1, | |
293 0, 1, 1, 2, 2, 3, 3, 4, 4, 3, 3, 2, 2, 1, 1, 0, | |
294 //error:0.000015 | |
295 }; | |
296 #else //64*cos | |
297 static const uint8_t obmc32[1024]={ | |
298 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, | |
299 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, | |
300 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, | |
301 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, | |
302 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, | |
303 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, | |
304 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, | |
305 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, | |
306 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, | |
307 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, | |
308 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, | |
309 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, | |
310 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, | |
311 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, | |
312 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, | |
313 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, | |
314 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, | |
315 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, | |
316 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, | |
317 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, | |
318 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, | |
319 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, | |
320 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, | |
321 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, | |
322 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, | |
323 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, | |
324 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, | |
325 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, | |
326 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, | |
327 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, | |
328 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, | |
329 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, | |
330 //error:0.000022 | |
331 }; | |
332 static const uint8_t obmc16[256]={ | |
333 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, | |
334 0, 0, 1, 2, 3, 4, 5, 5, 5, 5, 4, 3, 2, 1, 0, 0, | |
335 0, 1, 3, 6, 8,11,13,14,14,13,11, 8, 6, 3, 1, 0, | |
336 0, 2, 6,10,15,20,24,26,26,24,20,15,10, 6, 2, 0, | |
337 0, 3, 8,16,23,30,35,38,38,35,30,23,16, 8, 3, 0, | |
338 1, 4,11,20,30,39,46,49,49,46,39,30,20,11, 4, 1, | |
339 1, 5,13,24,35,46,54,58,58,54,46,35,24,13, 5, 1, | |
340 0, 5,14,26,38,49,58,63,63,58,49,38,26,14, 5, 0, | |
341 0, 5,14,26,38,49,58,63,63,58,49,38,26,14, 5, 0, | |
342 1, 5,13,24,35,46,54,58,58,54,46,35,24,13, 5, 1, | |
343 1, 4,11,20,30,39,46,49,49,46,39,30,20,11, 4, 1, | |
344 0, 3, 8,16,23,30,35,38,38,35,30,23,16, 8, 3, 0, | |
345 0, 2, 6,10,15,20,24,26,26,24,20,15,10, 6, 2, 0, | |
346 0, 1, 3, 6, 8,11,13,14,14,13,11, 8, 6, 3, 1, 0, | |
347 0, 0, 1, 2, 3, 4, 5, 5, 5, 5, 4, 3, 2, 1, 0, 0, | |
348 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, | |
349 //error:0.000022 | |
350 }; | |
351 #endif | |
352 | |
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 //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
|
354 static const uint8_t obmc8[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
|
355 1, 3, 5, 7, 7, 5, 3, 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
|
356 3, 9,15,21,21,15, 9, 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
|
357 5,15,25,35,35,25,15, 5, |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
358 7,21,35,49,49,35,21, 7, |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
359 7,21,35,49,49,35,21, 7, |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
360 5,15,25,35,35,25,15, 5, |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
361 3, 9,15,21,21,15, 9, 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
|
362 1, 3, 5, 7, 7, 5, 3, 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
|
363 //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
|
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 |
70b27300a496
quad 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 //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
|
367 static const uint8_t obmc4[16]={ |
70b27300a496
quad 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 4,12,12, 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
|
369 12,36,36,12, |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
370 12,36,36,12, |
70b27300a496
quad 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 4,12,12, 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
|
372 //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
|
373 }; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
374 |
70b27300a496
quad 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 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
|
376 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
|
377 }; |
70b27300a496
quad 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 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
379 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
|
380 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
|
381 int16_t 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
|
382 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
|
383 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
|
384 //#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
|
385 #define BLOCK_INTRA 1 |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
386 #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
|
387 //#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
|
388 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
|
389 }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
|
390 |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
391 static const BlockNode null_block= { //FIXME add border maybe |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
392 .color= {128,128,128}, |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
393 .mx= 0, |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
394 .my= 0, |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
395 .type= 0, |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
396 .level= 0, |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
397 }; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
398 |
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
|
399 #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
|
400 #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
|
401 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
402 typedef struct x_and_coeff{ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
403 int16_t x; |
2596 | 404 uint16_t coeff; |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
405 } x_and_coeff; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
406 |
2138 | 407 typedef struct SubBand{ |
408 int level; | |
409 int stride; | |
410 int width; | |
411 int height; | |
412 int qlog; ///< log(qscale)/log[2^(1/6)] | |
413 DWTELEM *buf; | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
414 int buf_x_offset; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
415 int buf_y_offset; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
416 int stride_line; ///< Stride measured in lines, not pixels. |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
417 x_and_coeff * x_coeff; |
2138 | 418 struct SubBand *parent; |
419 uint8_t state[/*7*2*/ 7 + 512][32]; | |
420 }SubBand; | |
421 | |
422 typedef struct Plane{ | |
423 int width; | |
424 int height; | |
425 SubBand band[MAX_DECOMPOSITIONS][4]; | |
426 }Plane; | |
427 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
428 /** Used to minimize the amount of memory used in order to optimize cache performance. **/ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
429 typedef struct { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
430 DWTELEM * * line; ///< For use by idwt and predict_slices. |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
431 DWTELEM * * data_stack; ///< Used for internal purposes. |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
432 int data_stack_top; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
433 int line_count; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
434 int line_width; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
435 int data_count; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
436 DWTELEM * base_buffer; ///< Buffer that this structure is caching. |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
437 } slice_buffer; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
438 |
2138 | 439 typedef struct SnowContext{ |
440 // 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) | |
441 | |
442 AVCodecContext *avctx; | |
2335 | 443 RangeCoder c; |
2138 | 444 DSPContext dsp; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
445 AVFrame new_picture; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
446 AVFrame input_picture; ///< new_picture with the internal linesizes |
2138 | 447 AVFrame current_picture; |
448 AVFrame last_picture; | |
449 AVFrame mconly_picture; | |
450 // uint8_t q_context[16]; | |
451 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
|
452 uint8_t block_state[128 + 32*128]; |
2138 | 453 int keyframe; |
2199 | 454 int always_reset; |
2138 | 455 int version; |
456 int spatial_decomposition_type; | |
457 int temporal_decomposition_type; | |
458 int spatial_decomposition_count; | |
459 int temporal_decomposition_count; | |
460 DWTELEM *spatial_dwt_buffer; | |
461 int colorspace_type; | |
462 int chroma_h_shift; | |
463 int chroma_v_shift; | |
464 int spatial_scalability; | |
465 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
|
466 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
|
467 int lambda2; |
2138 | 468 int mv_scale; |
469 int qbias; | |
470 #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
|
471 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
|
472 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
|
473 int block_max_depth; |
2138 | 474 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
|
475 BlockNode *block; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
476 #define ME_CACHE_SIZE 1024 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
477 int me_cache[ME_CACHE_SIZE]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
478 int me_cache_generation; |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
479 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
|
480 |
2138 | 481 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) |
482 }SnowContext; | |
483 | |
2562 | 484 typedef struct { |
485 DWTELEM *b0; | |
486 DWTELEM *b1; | |
487 DWTELEM *b2; | |
488 DWTELEM *b3; | |
489 int y; | |
490 } dwt_compose_t; | |
491 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
492 #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
|
493 //#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
|
494 |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
495 static void iterative_me(SnowContext *s); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
496 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
497 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
|
498 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
499 int i; |
2967 | 500 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
501 buf->base_buffer = base_buffer; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
502 buf->line_count = line_count; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
503 buf->line_width = line_width; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
504 buf->data_count = max_allocated_lines; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
505 buf->line = (DWTELEM * *) av_mallocz (sizeof(DWTELEM *) * line_count); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
506 buf->data_stack = (DWTELEM * *) av_malloc (sizeof(DWTELEM *) * max_allocated_lines); |
2967 | 507 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
508 for (i = 0; i < max_allocated_lines; i++) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
509 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
510 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
|
511 } |
2967 | 512 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
513 buf->data_stack_top = max_allocated_lines - 1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
514 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
515 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
516 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
|
517 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
518 int offset; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
519 DWTELEM * buffer; |
2967 | 520 |
521 // av_log(NULL, AV_LOG_DEBUG, "Cache hit: %d\n", line); | |
522 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
523 assert(buf->data_stack_top >= 0); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
524 // assert(!buf->line[line]); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
525 if (buf->line[line]) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
526 return buf->line[line]; |
2967 | 527 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
528 offset = buf->line_width * line; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
529 buffer = buf->data_stack[buf->data_stack_top]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
530 buf->data_stack_top--; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
531 buf->line[line] = buffer; |
2967 | 532 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
533 // av_log(NULL, AV_LOG_DEBUG, "slice_buffer_load_line: line: %d remaining: %d\n", line, buf->data_stack_top + 1); |
2967 | 534 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
535 return buffer; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
536 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
537 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
538 static void slice_buffer_release(slice_buffer * buf, int line) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
539 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
540 int offset; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
541 DWTELEM * buffer; |
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 assert(line >= 0 && line < buf->line_count); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
544 assert(buf->line[line]); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
545 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
546 offset = buf->line_width * line; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
547 buffer = buf->line[line]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
548 buf->data_stack_top++; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
549 buf->data_stack[buf->data_stack_top] = buffer; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
550 buf->line[line] = NULL; |
2967 | 551 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
552 // 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
|
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 static void slice_buffer_flush(slice_buffer * buf) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
556 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
557 int i; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
558 for (i = 0; i < buf->line_count; i++) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
559 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
560 if (buf->line[i]) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
561 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
562 // 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
|
563 slice_buffer_release(buf, i); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
564 } |
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 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
567 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
568 static void slice_buffer_destroy(slice_buffer * buf) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
569 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
570 int i; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
571 slice_buffer_flush(buf); |
2967 | 572 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
573 for (i = buf->data_count - 1; i >= 0; i--) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
574 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
575 assert(buf->data_stack[i]); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
576 av_free(buf->data_stack[i]); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
577 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
578 assert(buf->data_stack); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
579 av_free(buf->data_stack); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
580 assert(buf->line); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
581 av_free(buf->line); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
582 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
583 |
2979 | 584 #ifdef __sgi |
2368
a7ac68734a91
fix for build on IRIX by (Michel Bardiaux {mbardiaux peaktime be})
michael
parents:
2335
diff
changeset
|
585 // Avoid a name clash on SGI IRIX |
2979 | 586 #undef qexp |
2368
a7ac68734a91
fix for build on IRIX by (Michel Bardiaux {mbardiaux peaktime be})
michael
parents:
2335
diff
changeset
|
587 #endif |
2246 | 588 #define QEXPSHIFT (7-FRAC_BITS+8) //FIXME try to change this to 0 |
2600 | 589 static uint8_t qexp[QROOT]; |
2138 | 590 |
591 static inline int mirror(int v, int m){ | |
2998 | 592 while((unsigned)v > (unsigned)m){ |
593 v=-v; | |
594 if(v<0) v+= 2*m; | |
595 } | |
596 return v; | |
2138 | 597 } |
598 | |
2335 | 599 static inline void put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed){ |
2138 | 600 int i; |
601 | |
602 if(v){ | |
603 const int a= ABS(v); | |
604 const int e= av_log2(a); | |
605 #if 1 | |
2967 | 606 const int el= FFMIN(e, 10); |
2335 | 607 put_rac(c, state+0, 0); |
2138 | 608 |
609 for(i=0; i<el; i++){ | |
2335 | 610 put_rac(c, state+1+i, 1); //1..10 |
2138 | 611 } |
612 for(; i<e; i++){ | |
2335 | 613 put_rac(c, state+1+9, 1); //1..10 |
2138 | 614 } |
2335 | 615 put_rac(c, state+1+FFMIN(i,9), 0); |
2138 | 616 |
617 for(i=e-1; i>=el; i--){ | |
2335 | 618 put_rac(c, state+22+9, (a>>i)&1); //22..31 |
2138 | 619 } |
620 for(; i>=0; i--){ | |
2335 | 621 put_rac(c, state+22+i, (a>>i)&1); //22..31 |
2138 | 622 } |
623 | |
624 if(is_signed) | |
2335 | 625 put_rac(c, state+11 + el, v < 0); //11..21 |
2138 | 626 #else |
2967 | 627 |
2335 | 628 put_rac(c, state+0, 0); |
2138 | 629 if(e<=9){ |
630 for(i=0; i<e; i++){ | |
2335 | 631 put_rac(c, state+1+i, 1); //1..10 |
2138 | 632 } |
2335 | 633 put_rac(c, state+1+i, 0); |
2138 | 634 |
635 for(i=e-1; i>=0; i--){ | |
2335 | 636 put_rac(c, state+22+i, (a>>i)&1); //22..31 |
2138 | 637 } |
638 | |
639 if(is_signed) | |
2335 | 640 put_rac(c, state+11 + e, v < 0); //11..21 |
2138 | 641 }else{ |
642 for(i=0; i<e; i++){ | |
2335 | 643 put_rac(c, state+1+FFMIN(i,9), 1); //1..10 |
2138 | 644 } |
2335 | 645 put_rac(c, state+1+FFMIN(i,9), 0); |
2138 | 646 |
647 for(i=e-1; i>=0; i--){ | |
2335 | 648 put_rac(c, state+22+FFMIN(i,9), (a>>i)&1); //22..31 |
2138 | 649 } |
650 | |
651 if(is_signed) | |
2335 | 652 put_rac(c, state+11 + FFMIN(e,10), v < 0); //11..21 |
2138 | 653 } |
654 #endif | |
655 }else{ | |
2335 | 656 put_rac(c, state+0, 1); |
2138 | 657 } |
658 } | |
659 | |
2335 | 660 static inline int get_symbol(RangeCoder *c, uint8_t *state, int is_signed){ |
661 if(get_rac(c, state+0)) | |
2138 | 662 return 0; |
663 else{ | |
2240
c46fed9b7575
simplify getsymbol patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2232
diff
changeset
|
664 int i, e, a; |
c46fed9b7575
simplify getsymbol patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2232
diff
changeset
|
665 e= 0; |
2335 | 666 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
|
667 e++; |
2138 | 668 } |
669 | |
2240
c46fed9b7575
simplify getsymbol patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2232
diff
changeset
|
670 a= 1; |
c46fed9b7575
simplify getsymbol patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2232
diff
changeset
|
671 for(i=e-1; i>=0; i--){ |
2335 | 672 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
|
673 } |
c46fed9b7575
simplify getsymbol patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2232
diff
changeset
|
674 |
2335 | 675 if(is_signed && get_rac(c, state+11 + FFMIN(e,10))) //11..21 |
2138 | 676 return -a; |
677 else | |
678 return a; | |
679 } | |
680 } | |
681 | |
2335 | 682 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
|
683 int i; |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
684 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
|
685 |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
686 assert(v>=0); |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
687 assert(log2>=-4); |
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
688 |
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
689 while(v >= r){ |
2335 | 690 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
|
691 v -= r; |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
692 log2++; |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
693 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
|
694 } |
2335 | 695 put_rac(c, state+4+log2, 0); |
2967 | 696 |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
697 for(i=log2-1; i>=0; i--){ |
2335 | 698 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
|
699 } |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
700 } |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
701 |
2335 | 702 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
|
703 int i; |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
704 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
|
705 int v=0; |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
706 |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
707 assert(log2>=-4); |
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
708 |
2335 | 709 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
|
710 v+= r; |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
711 log2++; |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
712 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
|
713 } |
2967 | 714 |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
715 for(i=log2-1; i>=0; i--){ |
2335 | 716 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
|
717 } |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
718 |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
719 return v; |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
720 } |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
721 |
2138 | 722 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){ |
723 const int mirror_left= !highpass; | |
724 const int mirror_right= (width&1) ^ highpass; | |
725 const int w= (width>>1) - 1 + (highpass & width); | |
726 int i; | |
727 | |
728 #define LIFT(src, ref, inv) ((src) + ((inv) ? - (ref) : + (ref))) | |
729 if(mirror_left){ | |
730 dst[0] = LIFT(src[0], ((mul*2*ref[0]+add)>>shift), inverse); | |
731 dst += dst_step; | |
732 src += src_step; | |
733 } | |
2967 | 734 |
2138 | 735 for(i=0; i<w; i++){ |
736 dst[i*dst_step] = LIFT(src[i*src_step], ((mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add)>>shift), inverse); | |
737 } | |
2967 | 738 |
2138 | 739 if(mirror_right){ |
740 dst[w*dst_step] = LIFT(src[w*src_step], ((mul*2*ref[w*ref_step]+add)>>shift), inverse); | |
741 } | |
742 } | |
743 | |
744 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){ | |
745 const int mirror_left= !highpass; | |
746 const int mirror_right= (width&1) ^ highpass; | |
747 const int w= (width>>1) - 1 + (highpass & width); | |
748 int i; | |
749 | |
750 if(mirror_left){ | |
751 int r= 3*2*ref[0]; | |
752 r += r>>4; | |
753 r += r>>8; | |
754 dst[0] = LIFT(src[0], ((r+add)>>shift), inverse); | |
755 dst += dst_step; | |
756 src += src_step; | |
757 } | |
2967 | 758 |
2138 | 759 for(i=0; i<w; i++){ |
760 int r= 3*(ref[i*ref_step] + ref[(i+1)*ref_step]); | |
761 r += r>>4; | |
762 r += r>>8; | |
763 dst[i*dst_step] = LIFT(src[i*src_step], ((r+add)>>shift), inverse); | |
764 } | |
2967 | 765 |
2138 | 766 if(mirror_right){ |
767 int r= 3*2*ref[w*ref_step]; | |
768 r += r>>4; | |
769 r += r>>8; | |
770 dst[w*dst_step] = LIFT(src[w*src_step], ((r+add)>>shift), inverse); | |
771 } | |
772 } | |
773 | |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
774 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
|
775 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
|
776 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
|
777 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
|
778 int i; |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
779 |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
780 assert(shift == 4); |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
781 #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
|
782 if(mirror_left){ |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
783 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
|
784 dst += dst_step; |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
785 src += src_step; |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
786 } |
2967 | 787 |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
788 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
|
789 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
|
790 } |
2967 | 791 |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
792 if(mirror_right){ |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
793 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
|
794 } |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
795 } |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
796 |
2138 | 797 |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
798 static void inplace_lift(DWTELEM *dst, int width, int *coeffs, int n, int shift, int start, int inverse){ |
2138 | 799 int x, i; |
2967 | 800 |
2138 | 801 for(x=start; x<width; x+=2){ |
802 int64_t sum=0; | |
803 | |
804 for(i=0; i<n; i++){ | |
805 int x2= x + 2*i - n + 1; | |
806 if (x2< 0) x2= -x2; | |
807 else if(x2>=width) x2= 2*width-x2-2; | |
808 sum += coeffs[i]*(int64_t)dst[x2]; | |
809 } | |
810 if(inverse) dst[x] -= (sum + (1<<shift)/2)>>shift; | |
811 else dst[x] += (sum + (1<<shift)/2)>>shift; | |
812 } | |
813 } | |
814 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
815 static void inplace_liftV(DWTELEM *dst, int width, int height, int stride, int *coeffs, int n, int shift, int start, int inverse){ |
2138 | 816 int x, y, i; |
817 for(y=start; y<height; y+=2){ | |
818 for(x=0; x<width; x++){ | |
819 int64_t sum=0; | |
2967 | 820 |
2138 | 821 for(i=0; i<n; i++){ |
822 int y2= y + 2*i - n + 1; | |
823 if (y2< 0) y2= -y2; | |
824 else if(y2>=height) y2= 2*height-y2-2; | |
825 sum += coeffs[i]*(int64_t)dst[x + y2*stride]; | |
826 } | |
827 if(inverse) dst[x + y*stride] -= (sum + (1<<shift)/2)>>shift; | |
828 else dst[x + y*stride] += (sum + (1<<shift)/2)>>shift; | |
829 } | |
830 } | |
831 } | |
832 | |
833 #define SCALEX 1 | |
834 #define LX0 0 | |
835 #define LX1 1 | |
836 | |
2139 | 837 #if 0 // more accurate 9/7 |
2138 | 838 #define N1 2 |
839 #define SHIFT1 14 | |
840 #define COEFFS1 (int[]){-25987,-25987} | |
841 #define N2 2 | |
842 #define SHIFT2 19 | |
843 #define COEFFS2 (int[]){-27777,-27777} | |
844 #define N3 2 | |
845 #define SHIFT3 15 | |
846 #define COEFFS3 (int[]){28931,28931} | |
847 #define N4 2 | |
848 #define SHIFT4 15 | |
849 #define COEFFS4 (int[]){14533,14533} | |
850 #elif 1 // 13/7 CRF | |
851 #define N1 4 | |
852 #define SHIFT1 4 | |
853 #define COEFFS1 (int[]){1,-9,-9,1} | |
2139 | 854 #define N2 4 |
2138 | 855 #define SHIFT2 4 |
856 #define COEFFS2 (int[]){-1,5,5,-1} | |
857 #define N3 0 | |
858 #define SHIFT3 1 | |
859 #define COEFFS3 NULL | |
860 #define N4 0 | |
861 #define SHIFT4 1 | |
862 #define COEFFS4 NULL | |
863 #elif 1 // 3/5 | |
864 #define LX0 1 | |
865 #define LX1 0 | |
866 #define SCALEX 0.5 | |
867 #define N1 2 | |
868 #define SHIFT1 1 | |
869 #define COEFFS1 (int[]){1,1} | |
870 #define N2 2 | |
871 #define SHIFT2 2 | |
872 #define COEFFS2 (int[]){-1,-1} | |
873 #define N3 0 | |
874 #define SHIFT3 0 | |
875 #define COEFFS3 NULL | |
876 #define N4 0 | |
877 #define SHIFT4 0 | |
878 #define COEFFS4 NULL | |
2967 | 879 #elif 1 // 11/5 |
2138 | 880 #define N1 0 |
881 #define SHIFT1 1 | |
882 #define COEFFS1 NULL | |
883 #define N2 2 | |
884 #define SHIFT2 2 | |
885 #define COEFFS2 (int[]){-1,-1} | |
886 #define N3 2 | |
887 #define SHIFT3 0 | |
888 #define COEFFS3 (int[]){-1,-1} | |
889 #define N4 4 | |
890 #define SHIFT4 7 | |
891 #define COEFFS4 (int[]){-5,29,29,-5} | |
892 #define SCALEX 4 | |
893 #elif 1 // 9/7 CDF | |
894 #define N1 2 | |
895 #define SHIFT1 7 | |
896 #define COEFFS1 (int[]){-203,-203} | |
897 #define N2 2 | |
898 #define SHIFT2 12 | |
899 #define COEFFS2 (int[]){-217,-217} | |
900 #define N3 2 | |
901 #define SHIFT3 7 | |
902 #define COEFFS3 (int[]){113,113} | |
903 #define N4 2 | |
904 #define SHIFT4 9 | |
905 #define COEFFS4 (int[]){227,227} | |
906 #define SCALEX 1 | |
907 #elif 1 // 7/5 CDF | |
908 #define N1 0 | |
909 #define SHIFT1 1 | |
910 #define COEFFS1 NULL | |
911 #define N2 2 | |
912 #define SHIFT2 2 | |
913 #define COEFFS2 (int[]){-1,-1} | |
914 #define N3 2 | |
915 #define SHIFT3 0 | |
916 #define COEFFS3 (int[]){-1,-1} | |
917 #define N4 2 | |
918 #define SHIFT4 4 | |
919 #define COEFFS4 (int[]){3,3} | |
920 #elif 1 // 9/7 MN | |
921 #define N1 4 | |
922 #define SHIFT1 4 | |
923 #define COEFFS1 (int[]){1,-9,-9,1} | |
924 #define N2 2 | |
925 #define SHIFT2 2 | |
926 #define COEFFS2 (int[]){1,1} | |
927 #define N3 0 | |
928 #define SHIFT3 1 | |
929 #define COEFFS3 NULL | |
930 #define N4 0 | |
931 #define SHIFT4 1 | |
932 #define COEFFS4 NULL | |
933 #else // 13/7 CRF | |
934 #define N1 4 | |
935 #define SHIFT1 4 | |
936 #define COEFFS1 (int[]){1,-9,-9,1} | |
937 #define N2 4 | |
938 #define SHIFT2 4 | |
939 #define COEFFS2 (int[]){-1,5,5,-1} | |
940 #define N3 0 | |
941 #define SHIFT3 1 | |
942 #define COEFFS3 NULL | |
943 #define N4 0 | |
944 #define SHIFT4 1 | |
945 #define COEFFS4 NULL | |
946 #endif | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
947 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
|
948 DWTELEM temp[width]; |
2138 | 949 const int width2= width>>1; |
950 const int w2= (width+1)>>1; | |
2893 | 951 int x; |
2138 | 952 |
953 inplace_lift(b, width, COEFFS1, N1, SHIFT1, LX1, 0); | |
954 inplace_lift(b, width, COEFFS2, N2, SHIFT2, LX0, 0); | |
955 inplace_lift(b, width, COEFFS3, N3, SHIFT3, LX1, 0); | |
956 inplace_lift(b, width, COEFFS4, N4, SHIFT4, LX0, 0); | |
2967 | 957 |
2138 | 958 for(x=0; x<width2; x++){ |
959 temp[x ]= b[2*x ]; | |
960 temp[x+w2]= b[2*x + 1]; | |
961 } | |
962 if(width&1) | |
963 temp[x ]= b[2*x ]; | |
964 memcpy(b, temp, width*sizeof(int)); | |
965 } | |
966 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
967 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
|
968 DWTELEM temp[width]; |
2138 | 969 const int width2= width>>1; |
2893 | 970 int x; |
2138 | 971 const int w2= (width+1)>>1; |
972 | |
973 memcpy(temp, b, width*sizeof(int)); | |
974 for(x=0; x<width2; x++){ | |
975 b[2*x ]= temp[x ]; | |
976 b[2*x + 1]= temp[x+w2]; | |
977 } | |
978 if(width&1) | |
979 b[2*x ]= temp[x ]; | |
980 | |
981 inplace_lift(b, width, COEFFS4, N4, SHIFT4, LX0, 1); | |
982 inplace_lift(b, width, COEFFS3, N3, SHIFT3, LX1, 1); | |
983 inplace_lift(b, width, COEFFS2, N2, SHIFT2, LX0, 1); | |
984 inplace_lift(b, width, COEFFS1, N1, SHIFT1, LX1, 1); | |
985 } | |
986 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
987 static void spatial_decomposeX(DWTELEM *buffer, int width, int height, int stride){ |
2138 | 988 int x, y; |
2967 | 989 |
2138 | 990 for(y=0; y<height; y++){ |
991 for(x=0; x<width; x++){ | |
992 buffer[y*stride + x] *= SCALEX; | |
993 } | |
994 } | |
995 | |
996 for(y=0; y<height; y++){ | |
997 horizontal_decomposeX(buffer + y*stride, width); | |
998 } | |
2967 | 999 |
2138 | 1000 inplace_liftV(buffer, width, height, stride, COEFFS1, N1, SHIFT1, LX1, 0); |
1001 inplace_liftV(buffer, width, height, stride, COEFFS2, N2, SHIFT2, LX0, 0); | |
1002 inplace_liftV(buffer, width, height, stride, COEFFS3, N3, SHIFT3, LX1, 0); | |
2967 | 1003 inplace_liftV(buffer, width, height, stride, COEFFS4, N4, SHIFT4, LX0, 0); |
2138 | 1004 } |
1005 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1006 static void spatial_composeX(DWTELEM *buffer, int width, int height, int stride){ |
2138 | 1007 int x, y; |
2967 | 1008 |
2138 | 1009 inplace_liftV(buffer, width, height, stride, COEFFS4, N4, SHIFT4, LX0, 1); |
1010 inplace_liftV(buffer, width, height, stride, COEFFS3, N3, SHIFT3, LX1, 1); | |
1011 inplace_liftV(buffer, width, height, stride, COEFFS2, N2, SHIFT2, LX0, 1); | |
1012 inplace_liftV(buffer, width, height, stride, COEFFS1, N1, SHIFT1, LX1, 1); | |
1013 | |
1014 for(y=0; y<height; y++){ | |
1015 horizontal_composeX(buffer + y*stride, width); | |
1016 } | |
1017 | |
1018 for(y=0; y<height; y++){ | |
1019 for(x=0; x<width; x++){ | |
1020 buffer[y*stride + x] /= SCALEX; | |
1021 } | |
1022 } | |
1023 } | |
1024 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1025 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
|
1026 DWTELEM temp[width]; |
2138 | 1027 const int width2= width>>1; |
2893 | 1028 int x; |
2138 | 1029 const int w2= (width+1)>>1; |
1030 | |
1031 for(x=0; x<width2; x++){ | |
1032 temp[x ]= b[2*x ]; | |
1033 temp[x+w2]= b[2*x + 1]; | |
1034 } | |
1035 if(width&1) | |
1036 temp[x ]= b[2*x ]; | |
1037 #if 0 | |
2893 | 1038 { |
1039 int A1,A2,A3,A4; | |
2138 | 1040 A2= temp[1 ]; |
1041 A4= temp[0 ]; | |
1042 A1= temp[0+width2]; | |
1043 A1 -= (A2 + A4)>>1; | |
1044 A4 += (A1 + 1)>>1; | |
1045 b[0+width2] = A1; | |
1046 b[0 ] = A4; | |
1047 for(x=1; x+1<width2; x+=2){ | |
1048 A3= temp[x+width2]; | |
1049 A4= temp[x+1 ]; | |
1050 A3 -= (A2 + A4)>>1; | |
1051 A2 += (A1 + A3 + 2)>>2; | |
1052 b[x+width2] = A3; | |
1053 b[x ] = A2; | |
1054 | |
1055 A1= temp[x+1+width2]; | |
1056 A2= temp[x+2 ]; | |
1057 A1 -= (A2 + A4)>>1; | |
1058 A4 += (A1 + A3 + 2)>>2; | |
1059 b[x+1+width2] = A1; | |
1060 b[x+1 ] = A4; | |
1061 } | |
1062 A3= temp[width-1]; | |
1063 A3 -= A2; | |
1064 A2 += (A1 + A3 + 2)>>2; | |
1065 b[width -1] = A3; | |
1066 b[width2-1] = A2; | |
2893 | 1067 } |
2967 | 1068 #else |
2138 | 1069 lift(b+w2, temp+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 0); |
1070 lift(b , temp , b+w2, 1, 1, 1, width, 1, 2, 2, 0, 0); | |
1071 #endif | |
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_decompose53iH0(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])>>1; | |
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 vertical_decompose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1083 int i; |
2967 | 1084 |
2138 | 1085 for(i=0; i<width; i++){ |
1086 b1[i] += (b0[i] + b2[i] + 2)>>2; | |
1087 } | |
1088 } | |
1089 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1090 static void spatial_decompose53i(DWTELEM *buffer, int width, int height, int stride){ |
2198 | 1091 int y; |
2138 | 1092 DWTELEM *b0= buffer + mirror(-2-1, height-1)*stride; |
1093 DWTELEM *b1= buffer + mirror(-2 , height-1)*stride; | |
2967 | 1094 |
2138 | 1095 for(y=-2; y<height; y+=2){ |
1096 DWTELEM *b2= buffer + mirror(y+1, height-1)*stride; | |
1097 DWTELEM *b3= buffer + mirror(y+2, height-1)*stride; | |
1098 | |
1099 {START_TIMER | |
2998 | 1100 if(y+1<(unsigned)height) horizontal_decompose53i(b2, width); |
1101 if(y+2<(unsigned)height) horizontal_decompose53i(b3, width); | |
2138 | 1102 STOP_TIMER("horizontal_decompose53i")} |
2967 | 1103 |
2138 | 1104 {START_TIMER |
2998 | 1105 if(y+1<(unsigned)height) vertical_decompose53iH0(b1, b2, b3, width); |
1106 if(y+0<(unsigned)height) vertical_decompose53iL0(b0, b1, b2, width); | |
2138 | 1107 STOP_TIMER("vertical_decompose53i*")} |
2967 | 1108 |
2138 | 1109 b0=b2; |
1110 b1=b3; | |
1111 } | |
1112 } | |
1113 | |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1114 #define liftS lift |
2138 | 1115 #define lift5 lift |
1116 #if 1 | |
1117 #define W_AM 3 | |
1118 #define W_AO 0 | |
1119 #define W_AS 1 | |
1120 | |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1121 #undef liftS |
2138 | 1122 #define W_BM 1 |
1123 #define W_BO 8 | |
1124 #define W_BS 4 | |
1125 | |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1126 #define W_CM 1 |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1127 #define W_CO 0 |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1128 #define W_CS 0 |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1129 |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1130 #define W_DM 3 |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1131 #define W_DO 4 |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1132 #define W_DS 3 |
2138 | 1133 #elif 0 |
1134 #define W_AM 55 | |
1135 #define W_AO 16 | |
1136 #define W_AS 5 | |
1137 | |
1138 #define W_BM 3 | |
1139 #define W_BO 32 | |
1140 #define W_BS 6 | |
1141 | |
1142 #define W_CM 127 | |
1143 #define W_CO 64 | |
1144 #define W_CS 7 | |
1145 | |
1146 #define W_DM 7 | |
1147 #define W_DO 8 | |
1148 #define W_DS 4 | |
1149 #elif 0 | |
1150 #define W_AM 97 | |
1151 #define W_AO 32 | |
1152 #define W_AS 6 | |
1153 | |
1154 #define W_BM 63 | |
1155 #define W_BO 512 | |
1156 #define W_BS 10 | |
1157 | |
1158 #define W_CM 13 | |
1159 #define W_CO 8 | |
1160 #define W_CS 4 | |
1161 | |
1162 #define W_DM 15 | |
1163 #define W_DO 16 | |
1164 #define W_DS 5 | |
1165 | |
1166 #else | |
1167 | |
1168 #define W_AM 203 | |
1169 #define W_AO 64 | |
1170 #define W_AS 7 | |
1171 | |
1172 #define W_BM 217 | |
1173 #define W_BO 2048 | |
1174 #define W_BS 12 | |
1175 | |
1176 #define W_CM 113 | |
1177 #define W_CO 64 | |
1178 #define W_CS 7 | |
1179 | |
1180 #define W_DM 227 | |
1181 #define W_DO 128 | |
1182 #define W_DS 9 | |
1183 #endif | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1184 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
|
1185 DWTELEM temp[width]; |
2138 | 1186 const int w2= (width+1)>>1; |
1187 | |
1188 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
|
1189 liftS(temp , b , temp+w2, 1, 2, 1, width, -W_BM, W_BO, W_BS, 0, 0); |
2138 | 1190 lift5(b +w2, temp+w2, temp , 1, 1, 1, width, W_CM, W_CO, W_CS, 1, 0); |
1191 lift (b , temp , b +w2, 1, 1, 1, width, W_DM, W_DO, W_DS, 0, 0); | |
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 static void vertical_decompose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1196 int i; |
2967 | 1197 |
2138 | 1198 for(i=0; i<width; i++){ |
1199 b1[i] -= (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS; | |
1200 } | |
1201 } | |
1202 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1203 static void vertical_decompose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1204 int i; |
2967 | 1205 |
2138 | 1206 for(i=0; i<width; i++){ |
1207 #ifdef lift5 | |
1208 b1[i] += (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS; | |
1209 #else | |
1210 int r= 3*(b0[i] + b2[i]); | |
1211 r+= r>>4; | |
1212 r+= r>>8; | |
1213 b1[i] += (r+W_CO)>>W_CS; | |
1214 #endif | |
1215 } | |
1216 } | |
1217 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1218 static void vertical_decompose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1219 int i; |
2967 | 1220 |
2138 | 1221 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
|
1222 #ifdef liftS |
2138 | 1223 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
|
1224 #else |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1225 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
|
1226 #endif |
2138 | 1227 } |
1228 } | |
1229 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1230 static void vertical_decompose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1231 int i; |
2967 | 1232 |
2138 | 1233 for(i=0; i<width; i++){ |
1234 b1[i] += (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS; | |
1235 } | |
1236 } | |
1237 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1238 static void spatial_decompose97i(DWTELEM *buffer, int width, int height, int stride){ |
2198 | 1239 int y; |
2138 | 1240 DWTELEM *b0= buffer + mirror(-4-1, height-1)*stride; |
1241 DWTELEM *b1= buffer + mirror(-4 , height-1)*stride; | |
1242 DWTELEM *b2= buffer + mirror(-4+1, height-1)*stride; | |
1243 DWTELEM *b3= buffer + mirror(-4+2, height-1)*stride; | |
2967 | 1244 |
2138 | 1245 for(y=-4; y<height; y+=2){ |
1246 DWTELEM *b4= buffer + mirror(y+3, height-1)*stride; | |
1247 DWTELEM *b5= buffer + mirror(y+4, height-1)*stride; | |
1248 | |
1249 {START_TIMER | |
2998 | 1250 if(y+3<(unsigned)height) horizontal_decompose97i(b4, width); |
1251 if(y+4<(unsigned)height) horizontal_decompose97i(b5, width); | |
2138 | 1252 if(width>400){ |
1253 STOP_TIMER("horizontal_decompose97i") | |
1254 }} | |
2967 | 1255 |
2138 | 1256 {START_TIMER |
2998 | 1257 if(y+3<(unsigned)height) vertical_decompose97iH0(b3, b4, b5, width); |
1258 if(y+2<(unsigned)height) vertical_decompose97iL0(b2, b3, b4, width); | |
1259 if(y+1<(unsigned)height) vertical_decompose97iH1(b1, b2, b3, width); | |
1260 if(y+0<(unsigned)height) vertical_decompose97iL1(b0, b1, b2, width); | |
2138 | 1261 |
1262 if(width>400){ | |
1263 STOP_TIMER("vertical_decompose97i") | |
1264 }} | |
2967 | 1265 |
2138 | 1266 b0=b2; |
1267 b1=b3; | |
1268 b2=b4; | |
1269 b3=b5; | |
1270 } | |
1271 } | |
1272 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1273 void ff_spatial_dwt(DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){ |
2138 | 1274 int level; |
2967 | 1275 |
2164 | 1276 for(level=0; level<decomposition_count; level++){ |
1277 switch(type){ | |
2138 | 1278 case 0: spatial_decompose97i(buffer, width>>level, height>>level, stride<<level); break; |
1279 case 1: spatial_decompose53i(buffer, width>>level, height>>level, stride<<level); break; | |
1280 case 2: spatial_decomposeX (buffer, width>>level, height>>level, stride<<level); break; | |
1281 } | |
1282 } | |
1283 } | |
1284 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1285 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
|
1286 DWTELEM temp[width]; |
2138 | 1287 const int width2= width>>1; |
1288 const int w2= (width+1)>>1; | |
2893 | 1289 int x; |
2138 | 1290 |
1291 #if 0 | |
2893 | 1292 int A1,A2,A3,A4; |
2138 | 1293 A2= temp[1 ]; |
1294 A4= temp[0 ]; | |
1295 A1= temp[0+width2]; | |
1296 A1 -= (A2 + A4)>>1; | |
1297 A4 += (A1 + 1)>>1; | |
1298 b[0+width2] = A1; | |
1299 b[0 ] = A4; | |
1300 for(x=1; x+1<width2; x+=2){ | |
1301 A3= temp[x+width2]; | |
1302 A4= temp[x+1 ]; | |
1303 A3 -= (A2 + A4)>>1; | |
1304 A2 += (A1 + A3 + 2)>>2; | |
1305 b[x+width2] = A3; | |
1306 b[x ] = A2; | |
1307 | |
1308 A1= temp[x+1+width2]; | |
1309 A2= temp[x+2 ]; | |
1310 A1 -= (A2 + A4)>>1; | |
1311 A4 += (A1 + A3 + 2)>>2; | |
1312 b[x+1+width2] = A1; | |
1313 b[x+1 ] = A4; | |
1314 } | |
1315 A3= temp[width-1]; | |
1316 A3 -= A2; | |
1317 A2 += (A1 + A3 + 2)>>2; | |
1318 b[width -1] = A3; | |
1319 b[width2-1] = A2; | |
2967 | 1320 #else |
2138 | 1321 lift(temp , b , b+w2, 1, 1, 1, width, 1, 2, 2, 0, 1); |
1322 lift(temp+w2, b+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 1); | |
1323 #endif | |
1324 for(x=0; x<width2; x++){ | |
1325 b[2*x ]= temp[x ]; | |
1326 b[2*x + 1]= temp[x+w2]; | |
1327 } | |
1328 if(width&1) | |
1329 b[2*x ]= temp[x ]; | |
1330 } | |
1331 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1332 static void vertical_compose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1333 int i; |
2967 | 1334 |
2138 | 1335 for(i=0; i<width; i++){ |
1336 b1[i] += (b0[i] + b2[i])>>1; | |
1337 } | |
1338 } | |
1339 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1340 static void vertical_compose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1341 int i; |
2967 | 1342 |
2138 | 1343 for(i=0; i<width; i++){ |
1344 b1[i] -= (b0[i] + b2[i] + 2)>>2; | |
1345 } | |
1346 } | |
1347 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1348 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
|
1349 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
|
1350 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
|
1351 cs->y = -1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1352 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1353 |
2562 | 1354 static void spatial_compose53i_init(dwt_compose_t *cs, DWTELEM *buffer, int height, int stride){ |
1355 cs->b0 = buffer + mirror(-1-1, height-1)*stride; | |
1356 cs->b1 = buffer + mirror(-1 , height-1)*stride; | |
1357 cs->y = -1; | |
1358 } | |
1359 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1360 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
|
1361 int y= cs->y; |
2967 | 1362 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1363 DWTELEM *b0= cs->b0; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1364 DWTELEM *b1= cs->b1; |
3012 | 1365 DWTELEM *b2= slice_buffer_get_line(sb, mirror(y+1, height-1) * stride_line); |
1366 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
|
1367 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1368 {START_TIMER |
2998 | 1369 if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width); |
1370 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
|
1371 STOP_TIMER("vertical_compose53i*")} |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1372 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1373 {START_TIMER |
2998 | 1374 if(y-1<(unsigned)height) horizontal_compose53i(b0, width); |
1375 if(y+0<(unsigned)height) horizontal_compose53i(b1, width); | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1376 STOP_TIMER("horizontal_compose53i")} |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1377 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1378 cs->b0 = b2; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1379 cs->b1 = b3; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1380 cs->y += 2; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1381 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1382 |
2562 | 1383 static void spatial_compose53i_dy(dwt_compose_t *cs, DWTELEM *buffer, int width, int height, int stride){ |
1384 int y= cs->y; | |
1385 DWTELEM *b0= cs->b0; | |
1386 DWTELEM *b1= cs->b1; | |
1387 DWTELEM *b2= buffer + mirror(y+1, height-1)*stride; | |
1388 DWTELEM *b3= buffer + mirror(y+2, height-1)*stride; | |
2138 | 1389 |
1390 {START_TIMER | |
2998 | 1391 if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width); |
1392 if(y+0<(unsigned)height) vertical_compose53iH0(b0, b1, b2, width); | |
2138 | 1393 STOP_TIMER("vertical_compose53i*")} |
1394 | |
1395 {START_TIMER | |
2998 | 1396 if(y-1<(unsigned)height) horizontal_compose53i(b0, width); |
1397 if(y+0<(unsigned)height) horizontal_compose53i(b1, width); | |
2138 | 1398 STOP_TIMER("horizontal_compose53i")} |
1399 | |
2562 | 1400 cs->b0 = b2; |
1401 cs->b1 = b3; | |
1402 cs->y += 2; | |
1403 } | |
1404 | |
1405 static void spatial_compose53i(DWTELEM *buffer, int width, int height, int stride){ | |
1406 dwt_compose_t cs; | |
1407 spatial_compose53i_init(&cs, buffer, height, stride); | |
1408 while(cs.y <= height) | |
1409 spatial_compose53i_dy(&cs, buffer, width, height, stride); | |
2967 | 1410 } |
1411 | |
1412 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1413 static void horizontal_compose97i(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
|
1414 DWTELEM temp[width]; |
2138 | 1415 const int w2= (width+1)>>1; |
1416 | |
1417 lift (temp , b , b +w2, 1, 1, 1, width, W_DM, W_DO, W_DS, 0, 1); | |
1418 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
|
1419 liftS(b , temp , temp+w2, 2, 1, 1, width, -W_BM, W_BO, W_BS, 0, 1); |
2138 | 1420 lift (b+1 , temp+w2, b , 2, 1, 2, width, -W_AM, W_AO, W_AS, 1, 1); |
1421 } | |
1422 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1423 static void vertical_compose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1424 int i; |
2967 | 1425 |
2138 | 1426 for(i=0; i<width; i++){ |
1427 b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS; | |
1428 } | |
1429 } | |
1430 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1431 static void vertical_compose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1432 int i; |
2967 | 1433 |
2138 | 1434 for(i=0; i<width; i++){ |
1435 #ifdef lift5 | |
1436 b1[i] -= (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS; | |
1437 #else | |
1438 int r= 3*(b0[i] + b2[i]); | |
1439 r+= r>>4; | |
1440 r+= r>>8; | |
1441 b1[i] -= (r+W_CO)>>W_CS; | |
1442 #endif | |
1443 } | |
1444 } | |
1445 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1446 static void vertical_compose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1447 int i; |
2967 | 1448 |
2138 | 1449 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
|
1450 #ifdef liftS |
2138 | 1451 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
|
1452 #else |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1453 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
|
1454 #endif |
2138 | 1455 } |
1456 } | |
1457 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1458 static void vertical_compose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 1459 int i; |
2967 | 1460 |
2138 | 1461 for(i=0; i<width; i++){ |
1462 b1[i] -= (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS; | |
1463 } | |
1464 } | |
1465 | |
2592 | 1466 static void vertical_compose97i(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width){ |
1467 int i; | |
2967 | 1468 |
2592 | 1469 for(i=0; i<width; i++){ |
2893 | 1470 #ifndef lift5 |
2592 | 1471 int r; |
2893 | 1472 #endif |
2592 | 1473 b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS; |
1474 #ifdef lift5 | |
1475 b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS; | |
1476 #else | |
1477 r= 3*(b2[i] + b4[i]); | |
1478 r+= r>>4; | |
1479 r+= r>>8; | |
1480 b3[i] -= (r+W_CO)>>W_CS; | |
1481 #endif | |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1482 #ifdef liftS |
2592 | 1483 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
|
1484 #else |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1485 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
|
1486 #endif |
2592 | 1487 b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS; |
1488 } | |
1489 } | |
1490 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1491 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
|
1492 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
|
1493 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
|
1494 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
|
1495 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
|
1496 cs->y = -3; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1497 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1498 |
2562 | 1499 static void spatial_compose97i_init(dwt_compose_t *cs, DWTELEM *buffer, int height, int stride){ |
1500 cs->b0 = buffer + mirror(-3-1, height-1)*stride; | |
1501 cs->b1 = buffer + mirror(-3 , height-1)*stride; | |
1502 cs->b2 = buffer + mirror(-3+1, height-1)*stride; | |
1503 cs->b3 = buffer + mirror(-3+2, height-1)*stride; | |
1504 cs->y = -3; | |
1505 } | |
2138 | 1506 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1507 static void spatial_compose97i_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
|
1508 int y = cs->y; |
2967 | 1509 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1510 DWTELEM *b0= cs->b0; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1511 DWTELEM *b1= cs->b1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1512 DWTELEM *b2= cs->b2; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1513 DWTELEM *b3= cs->b3; |
3012 | 1514 DWTELEM *b4= slice_buffer_get_line(sb, mirror(y + 3, height - 1) * stride_line); |
1515 DWTELEM *b5= slice_buffer_get_line(sb, mirror(y + 4, height - 1) * stride_line); | |
2967 | 1516 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1517 {START_TIMER |
2592 | 1518 if(y>0 && y+4<height){ |
1519 vertical_compose97i(b0, b1, b2, b3, b4, b5, width); | |
1520 }else{ | |
2998 | 1521 if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width); |
1522 if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width); | |
1523 if(y+1<(unsigned)height) vertical_compose97iL0(b1, b2, b3, width); | |
1524 if(y+0<(unsigned)height) vertical_compose97iH0(b0, b1, b2, width); | |
2592 | 1525 } |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1526 if(width>400){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1527 STOP_TIMER("vertical_compose97i")}} |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1528 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1529 {START_TIMER |
2998 | 1530 if(y-1<(unsigned)height) horizontal_compose97i(b0, width); |
1531 if(y+0<(unsigned)height) horizontal_compose97i(b1, width); | |
3012 | 1532 if(width>400 && y+0<(unsigned)height){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1533 STOP_TIMER("horizontal_compose97i")}} |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1534 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1535 cs->b0=b2; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1536 cs->b1=b3; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1537 cs->b2=b4; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1538 cs->b3=b5; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1539 cs->y += 2; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1540 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1541 |
2562 | 1542 static void spatial_compose97i_dy(dwt_compose_t *cs, DWTELEM *buffer, int width, int height, int stride){ |
1543 int y = cs->y; | |
1544 DWTELEM *b0= cs->b0; | |
1545 DWTELEM *b1= cs->b1; | |
1546 DWTELEM *b2= cs->b2; | |
1547 DWTELEM *b3= cs->b3; | |
1548 DWTELEM *b4= buffer + mirror(y+3, height-1)*stride; | |
1549 DWTELEM *b5= buffer + mirror(y+4, height-1)*stride; | |
2138 | 1550 |
1551 {START_TIMER | |
2998 | 1552 if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width); |
1553 if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width); | |
1554 if(y+1<(unsigned)height) vertical_compose97iL0(b1, b2, b3, width); | |
1555 if(y+0<(unsigned)height) vertical_compose97iH0(b0, b1, b2, width); | |
2138 | 1556 if(width>400){ |
1557 STOP_TIMER("vertical_compose97i")}} | |
1558 | |
1559 {START_TIMER | |
2998 | 1560 if(y-1<(unsigned)height) horizontal_compose97i(b0, width); |
1561 if(y+0<(unsigned)height) horizontal_compose97i(b1, width); | |
2138 | 1562 if(width>400 && b0 <= b2){ |
1563 STOP_TIMER("horizontal_compose97i")}} | |
2562 | 1564 |
1565 cs->b0=b2; | |
1566 cs->b1=b3; | |
1567 cs->b2=b4; | |
1568 cs->b3=b5; | |
1569 cs->y += 2; | |
1570 } | |
1571 | |
1572 static void spatial_compose97i(DWTELEM *buffer, int width, int height, int stride){ | |
1573 dwt_compose_t cs; | |
1574 spatial_compose97i_init(&cs, buffer, height, stride); | |
1575 while(cs.y <= height) | |
1576 spatial_compose97i_dy(&cs, buffer, width, height, stride); | |
1577 } | |
1578 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1579 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){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1580 int level; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1581 for(level=decomposition_count-1; level>=0; level--){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1582 switch(type){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1583 case 0: spatial_compose97i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1584 case 1: spatial_compose53i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1585 /* not slicified yet */ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1586 case 2: /*spatial_composeX(buffer, width>>level, height>>level, stride<<level); break;*/ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1587 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
|
1588 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1589 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1590 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1591 |
2562 | 1592 void ff_spatial_idwt_init(dwt_compose_t *cs, DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){ |
1593 int level; | |
1594 for(level=decomposition_count-1; level>=0; level--){ | |
1595 switch(type){ | |
1596 case 0: spatial_compose97i_init(cs+level, buffer, height>>level, stride<<level); break; | |
1597 case 1: spatial_compose53i_init(cs+level, buffer, height>>level, stride<<level); break; | |
1598 /* not slicified yet */ | |
1599 case 2: spatial_composeX(buffer, width>>level, height>>level, stride<<level); break; | |
1600 } | |
1601 } | |
1602 } | |
1603 | |
1604 void ff_spatial_idwt_slice(dwt_compose_t *cs, DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count, int y){ | |
1605 const int support = type==1 ? 3 : 5; | |
1606 int level; | |
1607 if(type==2) return; | |
1608 | |
1609 for(level=decomposition_count-1; level>=0; level--){ | |
1610 while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){ | |
1611 switch(type){ | |
1612 case 0: spatial_compose97i_dy(cs+level, buffer, width>>level, height>>level, stride<<level); | |
1613 break; | |
1614 case 1: spatial_compose53i_dy(cs+level, buffer, width>>level, height>>level, stride<<level); | |
1615 break; | |
1616 case 2: break; | |
1617 } | |
1618 } | |
2138 | 1619 } |
1620 } | |
1621 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1622 void ff_spatial_idwt_buffered_slice(dwt_compose_t *cs, slice_buffer * slice_buf, int width, int height, int stride_line, int type, int decomposition_count, int y){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1623 const int support = type==1 ? 3 : 5; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1624 int level; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1625 if(type==2) return; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1626 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1627 for(level=decomposition_count-1; level>=0; level--){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1628 while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1629 switch(type){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1630 case 0: spatial_compose97i_dy_buffered(cs+level, slice_buf, width>>level, height>>level, stride_line<<level); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1631 break; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1632 case 1: spatial_compose53i_dy_buffered(cs+level, slice_buf, width>>level, height>>level, stride_line<<level); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1633 break; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1634 case 2: break; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1635 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1636 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1637 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1638 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1639 |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1640 void ff_spatial_idwt(DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){ |
2562 | 1641 if(type==2){ |
1642 int level; | |
1643 for(level=decomposition_count-1; level>=0; level--) | |
1644 spatial_composeX (buffer, width>>level, height>>level, stride<<level); | |
1645 }else{ | |
1646 dwt_compose_t cs[MAX_DECOMPOSITIONS]; | |
1647 int y; | |
1648 ff_spatial_idwt_init(cs, buffer, width, height, stride, type, decomposition_count); | |
1649 for(y=0; y<height; y+=4) | |
1650 ff_spatial_idwt_slice(cs, buffer, width, height, stride, type, decomposition_count, y); | |
2138 | 1651 } |
1652 } | |
1653 | |
2422 | 1654 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
|
1655 const int w= b->width; |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1656 const int h= b->height; |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1657 int x, y; |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1658 |
2138 | 1659 if(1){ |
1660 int run=0; | |
2149 | 1661 int runs[w*h]; |
2138 | 1662 int run_index=0; |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1663 int max_index; |
2967 | 1664 |
2138 | 1665 for(y=0; y<h; y++){ |
1666 for(x=0; x<w; x++){ | |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1667 int v, p=0; |
2144 | 1668 int /*ll=0, */l=0, lt=0, t=0, rt=0; |
2149 | 1669 v= src[x + y*stride]; |
2138 | 1670 |
1671 if(y){ | |
2149 | 1672 t= src[x + (y-1)*stride]; |
2138 | 1673 if(x){ |
2149 | 1674 lt= src[x - 1 + (y-1)*stride]; |
2138 | 1675 } |
1676 if(x + 1 < w){ | |
2149 | 1677 rt= src[x + 1 + (y-1)*stride]; |
2138 | 1678 } |
1679 } | |
1680 if(x){ | |
2149 | 1681 l= src[x - 1 + y*stride]; |
2144 | 1682 /*if(x > 1){ |
1683 if(orientation==1) ll= src[y + (x-2)*stride]; | |
1684 else ll= src[x - 2 + y*stride]; | |
2138 | 1685 }*/ |
1686 } | |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1687 if(parent){ |
2149 | 1688 int px= x>>1; |
1689 int py= y>>1; | |
2967 | 1690 if(px<b->parent->width && py<b->parent->height) |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1691 p= parent[px + py*2*stride]; |
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1692 } |
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1693 if(!(/*ll|*/l|lt|t|rt|p)){ |
2138 | 1694 if(v){ |
1695 runs[run_index++]= run; | |
1696 run=0; | |
1697 }else{ | |
1698 run++; | |
1699 } | |
1700 } | |
1701 } | |
1702 } | |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1703 max_index= run_index; |
2138 | 1704 runs[run_index++]= run; |
1705 run_index=0; | |
1706 run= runs[run_index++]; | |
1707 | |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1708 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
|
1709 if(run_index <= max_index) |
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1710 put_symbol2(&s->c, b->state[1], run, 3); |
2967 | 1711 |
2138 | 1712 for(y=0; y<h; y++){ |
2435
c89ac0e70c66
10l patch by (matthieu castet <castet.matthieu free fr>)
michael
parents:
2422
diff
changeset
|
1713 if(s->c.bytestream_end - s->c.bytestream < w*40){ |
2422 | 1714 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n"); |
1715 return -1; | |
1716 } | |
2138 | 1717 for(x=0; x<w; x++){ |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1718 int v, p=0; |
2144 | 1719 int /*ll=0, */l=0, lt=0, t=0, rt=0; |
2149 | 1720 v= src[x + y*stride]; |
2138 | 1721 |
1722 if(y){ | |
2149 | 1723 t= src[x + (y-1)*stride]; |
2138 | 1724 if(x){ |
2149 | 1725 lt= src[x - 1 + (y-1)*stride]; |
2138 | 1726 } |
1727 if(x + 1 < w){ | |
2149 | 1728 rt= src[x + 1 + (y-1)*stride]; |
2138 | 1729 } |
1730 } | |
1731 if(x){ | |
2149 | 1732 l= src[x - 1 + y*stride]; |
2144 | 1733 /*if(x > 1){ |
1734 if(orientation==1) ll= src[y + (x-2)*stride]; | |
1735 else ll= src[x - 2 + y*stride]; | |
2138 | 1736 }*/ |
1737 } | |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1738 if(parent){ |
2149 | 1739 int px= x>>1; |
1740 int py= y>>1; | |
2967 | 1741 if(px<b->parent->width && py<b->parent->height) |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1742 p= parent[px + py*2*stride]; |
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1743 } |
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1744 if(/*ll|*/l|lt|t|rt|p){ |
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1745 int context= av_log2(/*ABS(ll) + */3*ABS(l) + ABS(lt) + 2*ABS(t) + ABS(rt) + ABS(p)); |
2144 | 1746 |
2335 | 1747 put_rac(&s->c, &b->state[0][context], !!v); |
2138 | 1748 }else{ |
1749 if(!run){ | |
1750 run= runs[run_index++]; | |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1751 |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1752 if(run_index <= max_index) |
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1753 put_symbol2(&s->c, b->state[1], run, 3); |
2138 | 1754 assert(v); |
1755 }else{ | |
1756 run--; | |
1757 assert(!v); | |
1758 } | |
1759 } | |
1760 if(v){ | |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1761 int context= av_log2(/*ABS(ll) + */3*ABS(l) + ABS(lt) + 2*ABS(t) + ABS(rt) + ABS(p)); |
2596 | 1762 int l2= 2*ABS(l) + (l<0); |
1763 int t2= 2*ABS(t) + (t<0); | |
2144 | 1764 |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
1765 put_symbol2(&s->c, b->state[context + 2], ABS(v)-1, context-4); |
2596 | 1766 put_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l2&0xFF] + 3*quant3bA[t2&0xFF]], v<0); |
2138 | 1767 } |
1768 } | |
1769 } | |
1770 } | |
2422 | 1771 return 0; |
2138 | 1772 } |
1773 | |
2967 | 1774 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
|
1775 // 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
|
1776 // encode_subband_z0run(s, b, src, parent, stride, orientation); |
2422 | 1777 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
|
1778 // 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
|
1779 } |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1780 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1781 static inline void unpack_coeffs(SnowContext *s, SubBand *b, SubBand * parent, int orientation){ |
2138 | 1782 const int w= b->width; |
1783 const int h= b->height; | |
1784 int x,y; | |
2967 | 1785 |
2138 | 1786 if(1){ |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1787 int run, runs; |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1788 x_and_coeff *xc= b->x_coeff; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1789 x_and_coeff *prev_xc= NULL; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1790 x_and_coeff *prev2_xc= xc; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1791 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
|
1792 x_and_coeff *prev_parent_xc= parent_xc; |
2138 | 1793 |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1794 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
|
1795 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
|
1796 else run= INT_MAX; |
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1797 |
2138 | 1798 for(y=0; y<h; y++){ |
2193 | 1799 int v=0; |
1800 int lt=0, t=0, rt=0; | |
1801 | |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1802 if(y && prev_xc->x == 0){ |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1803 rt= prev_xc->coeff; |
2193 | 1804 } |
2138 | 1805 for(x=0; x<w; x++){ |
2193 | 1806 int p=0; |
1807 const int l= v; | |
2967 | 1808 |
2193 | 1809 lt= t; t= rt; |
1810 | |
2194 | 1811 if(y){ |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1812 if(prev_xc->x <= x) |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1813 prev_xc++; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1814 if(prev_xc->x == x + 1) |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1815 rt= prev_xc->coeff; |
2194 | 1816 else |
1817 rt=0; | |
1818 } | |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1819 if(parent_xc){ |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1820 if(x>>1 > parent_xc->x){ |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1821 parent_xc++; |
2192 | 1822 } |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1823 if(x>>1 == parent_xc->x){ |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1824 p= parent_xc->coeff; |
2194 | 1825 } |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1826 } |
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1827 if(/*ll|*/l|lt|t|rt|p){ |
2596 | 1828 int context= av_log2(/*ABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1)); |
2144 | 1829 |
2335 | 1830 v=get_rac(&s->c, &b->state[0][context]); |
2605 | 1831 if(v){ |
1832 v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1); | |
1833 v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l&0xFF] + 3*quant3bA[t&0xFF]]); | |
2967 | 1834 |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1835 xc->x=x; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1836 (xc++)->coeff= v; |
2605 | 1837 } |
2138 | 1838 }else{ |
1839 if(!run){ | |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1840 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
|
1841 else run= INT_MAX; |
2605 | 1842 v= 2*(get_symbol2(&s->c, b->state[0 + 2], 0-4) + 1); |
1843 v+=get_rac(&s->c, &b->state[0][16 + 1 + 3]); | |
2967 | 1844 |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1845 xc->x=x; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1846 (xc++)->coeff= v; |
2138 | 1847 }else{ |
2606 | 1848 int max_run; |
2138 | 1849 run--; |
1850 v=0; | |
2191 | 1851 |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1852 if(y) max_run= FFMIN(run, prev_xc->x - x - 2); |
2606 | 1853 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
|
1854 if(parent_xc) |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1855 max_run= FFMIN(max_run, 2*parent_xc->x - x - 1); |
2606 | 1856 x+= max_run; |
1857 run-= max_run; | |
2138 | 1858 } |
1859 } | |
2192 | 1860 } |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1861 (xc++)->x= w+1; //end marker |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1862 prev_xc= prev2_xc; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1863 prev2_xc= xc; |
2967 | 1864 |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1865 if(parent_xc){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1866 if(y&1){ |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1867 while(parent_xc->x != parent->width+1) |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1868 parent_xc++; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1869 parent_xc++; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1870 prev_parent_xc= parent_xc; |
2192 | 1871 }else{ |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1872 parent_xc= prev_parent_xc; |
2138 | 1873 } |
1874 } | |
1875 } | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1876 |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1877 (xc++)->x= w+1; //end marker |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1878 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1879 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1880 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1881 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
|
1882 const int w= b->width; |
2893 | 1883 int y; |
2600 | 1884 const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16); |
1885 int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1886 int qadd= (s->qbias*qmul)>>QBIAS_SHIFT; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1887 int new_index = 0; |
2967 | 1888 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1889 START_TIMER |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1890 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1891 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
|
1892 qadd= 0; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1893 qmul= 1<<QEXPSHIFT; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1894 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1895 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1896 /* 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
|
1897 if (start_y != 0) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1898 new_index = save_state[0]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1899 |
2967 | 1900 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1901 for(y=start_y; y<h; y++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1902 int x = 0; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1903 int v; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1904 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
|
1905 memset(line, 0, b->width*sizeof(DWTELEM)); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1906 v = b->x_coeff[new_index].coeff; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1907 x = b->x_coeff[new_index++].x; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1908 while(x < w) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1909 { |
2596 | 1910 register int t= ( (v>>1)*qmul + qadd)>>QEXPSHIFT; |
1911 register int u= -(v&1); | |
1912 line[x] = (t^u) - u; | |
1913 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1914 v = b->x_coeff[new_index].coeff; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1915 x = b->x_coeff[new_index++].x; |
2138 | 1916 } |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1917 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1918 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
|
1919 STOP_TIMER("decode_subband") |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1920 } |
2967 | 1921 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1922 /* Save our variables for the next slice. */ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1923 save_state[0] = new_index; |
2967 | 1924 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1925 return; |
2138 | 1926 } |
1927 | |
1928 static void reset_contexts(SnowContext *s){ | |
1929 int plane_index, level, orientation; | |
1930 | |
2199 | 1931 for(plane_index=0; plane_index<3; plane_index++){ |
2138 | 1932 for(level=0; level<s->spatial_decomposition_count; level++){ |
1933 for(orientation=level ? 1:0; orientation<4; orientation++){ | |
2335 | 1934 memset(s->plane[plane_index].band[level][orientation].state, MID_STATE, sizeof(s->plane[plane_index].band[level][orientation].state)); |
2138 | 1935 } |
1936 } | |
1937 } | |
2335 | 1938 memset(s->header_state, MID_STATE, sizeof(s->header_state)); |
1939 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
|
1940 } |
70b27300a496
quad 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 |
70b27300a496
quad 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 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
|
1943 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
|
1944 int h= -((-s->avctx->height)>>LOG2_MB_SIZE); |
2967 | 1945 |
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
|
1946 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
|
1947 s->b_height= h; |
2967 | 1948 |
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
|
1949 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
|
1950 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
|
1951 } |
70b27300a496
quad 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 |
2335 | 1953 static inline void copy_rac_state(RangeCoder *d, RangeCoder *s){ |
1954 uint8_t *bytestream= d->bytestream; | |
1955 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
|
1956 *d= *s; |
2335 | 1957 d->bytestream= bytestream; |
1958 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
|
1959 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1960 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1961 //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
|
1962 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
|
1963 { |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1964 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
|
1965 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1966 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
|
1967 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
|
1968 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
|
1969 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
|
1970 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
|
1971 } |
70b27300a496
quad 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 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
|
1973 } |
70b27300a496
quad 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 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
|
1975 } |
70b27300a496
quad 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 |
70b27300a496
quad 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 //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
|
1978 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
|
1979 { |
70b27300a496
quad 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 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
|
1981 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
|
1982 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1983 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
|
1984 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
|
1985 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
|
1986 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
|
1987 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
|
1988 } |
70b27300a496
quad 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 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
|
1990 } |
70b27300a496
quad 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 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
|
1992 } |
70b27300a496
quad 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 |
70b27300a496
quad 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 static inline void set_blocks(SnowContext *s, int level, int x, int y, int l, int cb, int cr, int mx, int my, 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
|
1995 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
|
1996 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
|
1997 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
|
1998 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
|
1999 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
|
2000 int i,j; |
2967 | 2001 |
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
|
2002 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
|
2003 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
|
2004 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
|
2005 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
|
2006 block.my= 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
|
2007 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
|
2008 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
|
2009 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2010 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
|
2011 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
|
2012 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
|
2013 } |
70b27300a496
quad 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 } |
70b27300a496
quad 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 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2016 |
70b27300a496
quad 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 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
|
2018 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
|
2019 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
|
2020 ((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
|
2021 ((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
|
2022 }; |
70b27300a496
quad 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 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
|
2024 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
|
2025 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
|
2026 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
|
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 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
|
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 |
70b27300a496
quad 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 //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
|
2032 #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
|
2033 #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
|
2034 #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
|
2035 #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
|
2036 #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
|
2037 #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
|
2038 |
70b27300a496
quad 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 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
|
2040 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
|
2041 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
|
2042 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
|
2043 uint8_t i_state[sizeof(s->block_state)]; |
2335 | 2044 RangeCoder pc, ic; |
2045 uint8_t *pbbak= s->c.bytestream; | |
2046 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
|
2047 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
|
2048 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
|
2049 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
|
2050 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
|
2051 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
|
2052 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
|
2053 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
|
2054 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
|
2055 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
|
2056 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
|
2057 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
|
2058 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
|
2059 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
|
2060 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
|
2061 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
|
2062 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
|
2063 int pcr= 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
|
2064 int pmx= 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
|
2065 int pmy= mid_pred(left->my, top->my, tr->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
|
2066 int mx=0, my=0; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2067 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
|
2068 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
|
2069 const int uvstride= s->current_picture.linesize[1]; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2070 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
|
2071 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
|
2072 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
|
2073 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
|
2074 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
|
2075 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
|
2076 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
|
2077 MotionEstContext *c= &s->m.me; |
70b27300a496
quad 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 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
|
2079 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
|
2080 int s_context= 2*left->level + 2*top->level + tl->level + tr->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
|
2081 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2082 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
|
2083 if(s->keyframe){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2084 set_blocks(s, level, x, y, pl, pcb, pcr, pmx, pmy, BLOCK_INTRA); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2085 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
|
2086 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2087 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2088 // 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
|
2089 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2090 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
|
2091 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
|
2092 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
|
2093 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
|
2094 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
|
2095 P_TOPRIGHT[1]= tr->my; |
2967 | 2096 |
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
|
2097 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
|
2098 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
|
2099 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
|
2100 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
|
2101 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
|
2102 last_mv[2][1]= bottom->my; |
2967 | 2103 |
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
|
2104 s->m.mb_stride=2; |
2967 | 2105 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
|
2106 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
|
2107 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
|
2108 |
70b27300a496
quad 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 init_ref(c, current_data, s->last_picture.data, NULL, block_w*x, block_w*y, 0); |
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 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
|
2112 assert(s->m.me.uvstride == uvstride); |
2967 | 2113 |
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
|
2114 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
|
2115 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
|
2116 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
|
2117 c->current_mv_penalty= c->mv_penalty[s->m.f_code=1] + MAX_MV; |
2967 | 2118 |
2206 | 2119 c->xmin = - x*block_w - 16+2; |
2120 c->ymin = - y*block_w - 16+2; | |
2121 c->xmax = - (x+1)*block_w + (w<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-2; | |
2122 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
|
2123 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2124 if(P_LEFT[0] > (c->xmax<<shift)) P_LEFT[0] = (c->xmax<<shift); |
2967 | 2125 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
|
2126 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
|
2127 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
|
2128 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
|
2129 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
|
2130 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
|
2131 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2132 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
|
2133 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
|
2134 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2135 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
|
2136 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
|
2137 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
|
2138 } 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
|
2139 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
|
2140 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
|
2141 } |
70b27300a496
quad 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 |
2967 | 2143 score= ff_epzs_motion_search(&s->m, &mx, &my, P, 0, /*ref_index*/ 0, last_mv, |
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
|
2144 (1<<16)>>shift, level-LOG2_MB_SIZE+4, block_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
|
2145 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2146 assert(mx >= c->xmin); |
70b27300a496
quad 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 assert(mx <= c->xmax); |
70b27300a496
quad 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 assert(my >= c->ymin); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2149 assert(my <= c->ymax); |
2967 | 2150 |
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 score= s->m.me.sub_motion_search(&s->m, &mx, &my, score, 0, 0, level-LOG2_MB_SIZE+4, block_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
|
2152 score= ff_get_mb_score(&s->m, mx, my, 0, 0, level-LOG2_MB_SIZE+4, block_w, 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
|
2153 //FIXME if mb_cmp != SSE then intra cant be compared currently and mb_penalty vs. lambda2 |
2967 | 2154 |
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
|
2155 // 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
|
2156 pc= s->c; |
2335 | 2157 pc.bytestream_start= |
2158 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
|
2159 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
|
2160 |
70b27300a496
quad 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 if(level!=s->block_max_depth) |
2335 | 2162 put_rac(&pc, &p_state[4 + s_context], 1); |
2163 put_rac(&pc, &p_state[1 + left->type + top->type], 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
|
2164 put_symbol(&pc, &p_state[128 + 32*mx_context], mx - pmx, 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
|
2165 put_symbol(&pc, &p_state[128 + 32*my_context], my - pmy, 1); |
2335 | 2166 p_len= pc.bytestream - pc.bytestream_start; |
2167 score += (s->lambda2*(p_len*8 | |
2168 + (pc.outstanding_count - s->c.outstanding_count)*8 | |
2169 + (-av_log2(pc.range) + av_log2(s->c.range)) | |
2170 ))>>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
|
2171 |
70b27300a496
quad 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 block_s= block_w*block_w; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2173 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
|
2174 l= (sum + block_s/2)/block_s; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2175 iscore = pix_norm1(current_data[0], stride, block_w) - 2*l*sum + l*l*block_s; |
2967 | 2176 |
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 block_s= block_w*block_w>>2; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2178 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
|
2179 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
|
2180 // 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
|
2181 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
|
2182 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
|
2183 // 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
|
2184 |
70b27300a496
quad 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 ic= s->c; |
2335 | 2186 ic.bytestream_start= |
2187 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
|
2188 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
|
2189 if(level!=s->block_max_depth) |
2335 | 2190 put_rac(&ic, &i_state[4 + s_context], 1); |
2191 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
|
2192 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
|
2193 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
|
2194 put_symbol(&ic, &i_state[96], cr-pcr, 1); |
2335 | 2195 i_len= ic.bytestream - ic.bytestream_start; |
2196 iscore += (s->lambda2*(i_len*8 | |
2197 + (ic.outstanding_count - s->c.outstanding_count)*8 | |
2198 + (-av_log2(ic.range) + av_log2(s->c.range)) | |
2199 ))>>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
|
2200 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2201 // 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
|
2202 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
|
2203 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
|
2204 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
|
2205 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
|
2206 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2207 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
|
2208 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
|
2209 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
|
2210 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
|
2211 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
|
2212 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
|
2213 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
|
2214 } |
2967 | 2215 |
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
|
2216 if(level!=s->block_max_depth){ |
2335 | 2217 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
|
2218 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
|
2219 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
|
2220 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
|
2221 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
|
2222 score2+= s->lambda2>>FF_LAMBDA_SHIFT; //FIXME exact split overhead |
2967 | 2223 |
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
|
2224 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
|
2225 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
|
2226 } |
2967 | 2227 |
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
|
2228 if(iscore < score){ |
2335 | 2229 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
|
2230 s->c= ic; |
2335 | 2231 s->c.bytestream_start= pbbak_start; |
2232 s->c.bytestream= pbbak + 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
|
2233 set_blocks(s, level, x, y, l, cb, cr, pmx, pmy, BLOCK_INTRA); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2234 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
|
2235 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
|
2236 }else{ |
2335 | 2237 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
|
2238 s->c= pc; |
2335 | 2239 s->c.bytestream_start= pbbak_start; |
2240 s->c.bytestream= pbbak + 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
|
2241 set_blocks(s, level, x, y, pl, pcb, pcr, mx, my, 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
|
2242 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
|
2243 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
|
2244 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2245 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2246 |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2247 static always_inline int same_block(BlockNode *a, BlockNode *b){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2248 if((a->type&BLOCK_INTRA) && (b->type&BLOCK_INTRA)){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2249 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
|
2250 }else{ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2251 return !((a->mx - b->mx) | (a->my - b->my) | ((a->type ^ b->type)&BLOCK_INTRA)); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2252 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2253 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2254 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2255 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
|
2256 const int w= s->b_width << s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2257 const int rem_depth= s->block_max_depth - level; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2258 const int index= (x + y*w) << rem_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2259 int trx= (x+1)<<rem_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2260 BlockNode *b= &s->block[index]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2261 BlockNode *left = x ? &s->block[index-1] : &null_block; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2262 BlockNode *top = y ? &s->block[index-w] : &null_block; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2263 BlockNode *tl = y && x ? &s->block[index-w-1] : left; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2264 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
|
2265 int pl = left->color[0]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2266 int pcb= left->color[1]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2267 int pcr= left->color[2]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2268 int pmx= mid_pred(left->mx, top->mx, tr->mx); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2269 int pmy= mid_pred(left->my, top->my, tr->my); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2270 int mx_context= av_log2(2*ABS(left->mx - top->mx)); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2271 int my_context= av_log2(2*ABS(left->my - top->my)); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2272 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
|
2273 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2274 if(s->keyframe){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2275 set_blocks(s, level, x, y, pl, pcb, pcr, pmx, pmy, BLOCK_INTRA); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2276 return; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2277 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2278 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2279 if(level!=s->block_max_depth){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2280 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
|
2281 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
|
2282 }else{ |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2283 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
|
2284 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
|
2285 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
|
2286 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
|
2287 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
|
2288 return; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2289 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2290 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2291 if(b->type & BLOCK_INTRA){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2292 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
|
2293 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
|
2294 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
|
2295 put_symbol(&s->c, &s->block_state[96], b->color[2]-pcr, 1); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2296 set_blocks(s, level, x, y, b->color[0], b->color[1], b->color[2], pmx, pmy, BLOCK_INTRA); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2297 }else{ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2298 put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 0); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2299 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
|
2300 put_symbol(&s->c, &s->block_state[128 + 32*my_context], b->my - pmy, 1); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2301 set_blocks(s, level, x, y, pl, pcb, pcr, b->mx, b->my, 0); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2302 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2303 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2304 |
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
|
2305 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
|
2306 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
|
2307 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
|
2308 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
|
2309 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
|
2310 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
|
2311 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
|
2312 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
|
2313 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
|
2314 int s_context= 2*left->level + 2*top->level + tl->level + tr->level; |
2967 | 2315 |
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
|
2316 if(s->keyframe){ |
70b27300a496
quad 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 set_blocks(s, level, x, y, null_block.color[0], null_block.color[1], null_block.color[2], null_block.mx, null_block.my, BLOCK_INTRA); |
70b27300a496
quad 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 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
|
2319 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2320 |
2335 | 2321 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
|
2322 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
|
2323 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
|
2324 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
|
2325 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
|
2326 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
|
2327 int my= mid_pred(left->my, top->my, tr->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
|
2328 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
|
2329 int my_context= av_log2(2*ABS(left->my - top->my)) + 0*av_log2(2*ABS(tr->my - top->my)); |
2967 | 2330 |
2335 | 2331 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
|
2332 |
70b27300a496
quad 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 if(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
|
2334 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
|
2335 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
|
2336 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
|
2337 }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
|
2338 mx+= get_symbol(&s->c, &s->block_state[128 + 32*mx_context], 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
|
2339 my+= get_symbol(&s->c, &s->block_state[128 + 32*my_context], 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
|
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 set_blocks(s, level, x, y, l, cb, cr, mx, my, 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
|
2342 }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
|
2343 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
|
2344 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
|
2345 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
|
2346 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
|
2347 } |
70b27300a496
quad 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 } |
70b27300a496
quad 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 |
70b27300a496
quad 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 static void encode_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
|
2351 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
|
2352 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
|
2353 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
|
2354 |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2355 if(s->avctx->me_method == ME_ITER && !s->keyframe) |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2356 iterative_me(s); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2357 |
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
|
2358 for(y=0; y<h; y++){ |
2435
c89ac0e70c66
10l patch by (matthieu castet <castet.matthieu free fr>)
michael
parents:
2422
diff
changeset
|
2359 if(s->c.bytestream_end - s->c.bytestream < w*MB_SIZE*MB_SIZE*3){ //FIXME nicer limit |
2422 | 2360 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n"); |
2361 return; | |
2362 } | |
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
|
2363 for(x=0; x<w; x++){ |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2364 if(s->avctx->me_method == ME_ITER) |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2365 encode_q_branch2(s, 0, x, y); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2366 else |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2367 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
|
2368 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2369 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2370 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2371 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2372 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
|
2373 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
|
2374 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
|
2375 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
|
2376 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2377 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
|
2378 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
|
2379 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
|
2380 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2381 } |
2138 | 2382 } |
2383 | |
2384 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){ | |
2385 int x, y; | |
2221 | 2386 START_TIMER |
2138 | 2387 for(y=0; y < b_h+5; y++){ |
2388 for(x=0; x < b_w; x++){ | |
2221 | 2389 int a0= src[x ]; |
2390 int a1= src[x + 1]; | |
2391 int a2= src[x + 2]; | |
2392 int a3= src[x + 3]; | |
2393 int a4= src[x + 4]; | |
2394 int a5= src[x + 5]; | |
2138 | 2395 // int am= 9*(a1+a2) - (a0+a3); |
2396 int am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5); | |
2397 // int am= 18*(a2+a3) - 2*(a1+a4); | |
2398 // int aL= (-7*a0 + 105*a1 + 35*a2 - 5*a3)>>3; | |
2399 // int aR= (-7*a3 + 105*a2 + 35*a1 - 5*a0)>>3; | |
2400 | |
2401 // if(b_w==16) am= 8*(a1+a2); | |
2402 | |
2610
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2403 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
|
2404 else am = ( am*(16-dx) + 32*a3*(dx-8) + 128)>>8; |
2967 | 2405 |
2610
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2406 /* 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
|
2407 if(am&(~255)) am= ~(am>>31); |
2967 | 2408 |
2610
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2409 tmp[x] = am; |
2138 | 2410 |
2411 /* if (dx< 4) tmp[x + y*stride]= (16*a1*( 4-dx) + aL* dx + 32)>>6; | |
2412 else if(dx< 8) tmp[x + y*stride]= ( aL*( 8-dx) + am*(dx- 4) + 32)>>6; | |
2413 else if(dx<12) tmp[x + y*stride]= ( am*(12-dx) + aR*(dx- 8) + 32)>>6; | |
2414 else tmp[x + y*stride]= ( aR*(16-dx) + 16*a2*(dx-12) + 32)>>6;*/ | |
2415 } | |
2221 | 2416 tmp += stride; |
2417 src += stride; | |
2138 | 2418 } |
2221 | 2419 tmp -= (b_h+5)*stride; |
2967 | 2420 |
2138 | 2421 for(y=0; y < b_h; y++){ |
2422 for(x=0; x < b_w; x++){ | |
2221 | 2423 int a0= tmp[x + 0*stride]; |
2424 int a1= tmp[x + 1*stride]; | |
2425 int a2= tmp[x + 2*stride]; | |
2426 int a3= tmp[x + 3*stride]; | |
2427 int a4= tmp[x + 4*stride]; | |
2428 int a5= tmp[x + 5*stride]; | |
2138 | 2429 int am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5); |
2430 // int am= 18*(a2+a3) - 2*(a1+a4); | |
2431 /* int aL= (-7*a0 + 105*a1 + 35*a2 - 5*a3)>>3; | |
2432 int aR= (-7*a3 + 105*a2 + 35*a1 - 5*a0)>>3;*/ | |
2967 | 2433 |
2138 | 2434 // if(b_w==16) am= 8*(a1+a2); |
2435 | |
2610
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2436 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
|
2437 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
|
2438 |
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2439 if(am&(~255)) am= ~(am>>31); |
2967 | 2440 |
2610
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2441 dst[x] = am; |
2138 | 2442 /* if (dy< 4) tmp[x + y*stride]= (16*a1*( 4-dy) + aL* dy + 32)>>6; |
2443 else if(dy< 8) tmp[x + y*stride]= ( aL*( 8-dy) + am*(dy- 4) + 32)>>6; | |
2444 else if(dy<12) tmp[x + y*stride]= ( am*(12-dy) + aR*(dy- 8) + 32)>>6; | |
2445 else tmp[x + y*stride]= ( aR*(16-dy) + 16*a2*(dy-12) + 32)>>6;*/ | |
2446 } | |
2221 | 2447 dst += stride; |
2448 tmp += stride; | |
2138 | 2449 } |
2221 | 2450 STOP_TIMER("mc_block") |
2138 | 2451 } |
2452 | |
2453 #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
|
2454 static void mc_block_hpel ## dx ## dy ## b_w(uint8_t *dst, uint8_t *src, int stride, int h){\ |
2138 | 2455 uint8_t tmp[stride*(b_w+5)];\ |
2456 assert(h==b_w);\ | |
2457 mc_block(dst, src-2-2*stride, tmp, stride, b_w, b_w, dx, dy);\ | |
2458 } | |
2459 | |
2460 mca( 0, 0,16) | |
2461 mca( 8, 0,16) | |
2462 mca( 0, 8,16) | |
2463 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
|
2464 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
|
2465 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
|
2466 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
|
2467 mca( 8, 8,8) |
2138 | 2468 |
2206 | 2469 static void pred_block(SnowContext *s, uint8_t *dst, uint8_t *src, 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
|
2470 if(block->type & BLOCK_INTRA){ |
2206 | 2471 int x, y; |
3018 | 2472 const int color = block->color[plane_index]; |
2473 const int color4= color*0x01010101; | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2474 if(b_w==32){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2475 for(y=0; y < b_h; y++){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2476 *(uint32_t*)&dst[0 + y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2477 *(uint32_t*)&dst[4 + y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2478 *(uint32_t*)&dst[8 + y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2479 *(uint32_t*)&dst[12+ y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2480 *(uint32_t*)&dst[16+ y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2481 *(uint32_t*)&dst[20+ y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2482 *(uint32_t*)&dst[24+ y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2483 *(uint32_t*)&dst[28+ y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2484 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2485 }else if(b_w==16){ |
3018 | 2486 for(y=0; y < b_h; y++){ |
2487 *(uint32_t*)&dst[0 + y*stride]= color4; | |
2488 *(uint32_t*)&dst[4 + y*stride]= color4; | |
2489 *(uint32_t*)&dst[8 + y*stride]= color4; | |
2490 *(uint32_t*)&dst[12+ y*stride]= color4; | |
2491 } | |
2492 }else if(b_w==8){ | |
2493 for(y=0; y < b_h; y++){ | |
2494 *(uint32_t*)&dst[0 + y*stride]= color4; | |
2495 *(uint32_t*)&dst[4 + y*stride]= color4; | |
2496 } | |
2497 }else if(b_w==4){ | |
2498 for(y=0; y < b_h; y++){ | |
2499 *(uint32_t*)&dst[0 + y*stride]= color4; | |
2500 } | |
2501 }else{ | |
2502 for(y=0; y < b_h; y++){ | |
2503 for(x=0; x < b_w; x++){ | |
2504 dst[x + y*stride]= color; | |
2505 } | |
2138 | 2506 } |
2507 } | |
2508 }else{ | |
2206 | 2509 const int scale= plane_index ? s->mv_scale : 2*s->mv_scale; |
2510 int mx= block->mx*scale; | |
2511 int my= block->my*scale; | |
2223 | 2512 const int dx= mx&15; |
2513 const int dy= my&15; | |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3018
diff
changeset
|
2514 const int tab_index= 3 - (b_w>>2) + (b_w>>4); |
2206 | 2515 sx += (mx>>4) - 2; |
2516 sy += (my>>4) - 2; | |
2517 src += sx + sy*stride; | |
2518 if( (unsigned)sx >= w - b_w - 4 | |
2519 || (unsigned)sy >= h - b_h - 4){ | |
2520 ff_emulated_edge_mc(tmp + MB_SIZE, src, stride, b_w+5, b_h+5, sx, sy, w, h); | |
2521 src= tmp + MB_SIZE; | |
2138 | 2522 } |
3018 | 2523 assert(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h); |
2524 assert(!(b_w&(b_w-1))); | |
2525 assert(b_w>1 && b_h>1); | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2526 assert(tab_index>=0 && tab_index<4 || b_w==32); |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3018
diff
changeset
|
2527 if((dx&3) || (dy&3)) |
2223 | 2528 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
|
2529 else if(b_w==32){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2530 int y; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2531 for(y=0; y<b_h; y+=16){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2532 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
|
2533 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
|
2534 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2535 }else if(b_w==b_h) |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3018
diff
changeset
|
2536 s->dsp.put_h264_qpel_pixels_tab[tab_index ][dy+(dx>>2)](dst,src + 2 + 2*stride,stride); |
3018 | 2537 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
|
2538 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
|
2539 s->dsp.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst+b_h,src + 2 + b_h + 2*stride,stride); |
3018 | 2540 }else{ |
2541 assert(2*b_w==b_h); | |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3018
diff
changeset
|
2542 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
|
2543 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 | 2544 } |
2138 | 2545 } |
2546 } | |
2547 | |
2206 | 2548 //FIXME name clenup (b_w, block_w, b_width stuff) |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2549 static always_inline void add_yblock_buffered(SnowContext *s, slice_buffer * sb, DWTELEM *old_dst, uint8_t *dst8, uint8_t *src, 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 plane_index){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2550 DWTELEM * dst = NULL; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2551 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
|
2552 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
|
2553 const int b_stride= b_width; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2554 BlockNode *lt= &s->block[b_x + b_y*b_stride]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2555 BlockNode *rt= lt+1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2556 BlockNode *lb= lt+b_stride; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2557 BlockNode *rb= lb+1; |
2967 | 2558 uint8_t *block[4]; |
2842 | 2559 int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride; |
2560 uint8_t tmp[src_stride*7*MB_SIZE]; //FIXME align | |
2561 uint8_t *ptmp; | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2562 int x,y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2563 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2564 if(b_x<0){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2565 lt= rt; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2566 lb= rb; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2567 }else if(b_x + 1 >= b_width){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2568 rt= lt; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2569 rb= lb; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2570 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2571 if(b_y<0){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2572 lt= lb; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2573 rt= rb; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2574 }else if(b_y + 1 >= b_height){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2575 lb= lt; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2576 rb= rt; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2577 } |
2967 | 2578 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2579 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
|
2580 obmc -= src_x; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2581 b_w += src_x; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2582 src_x=0; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2583 }else if(src_x + b_w > w){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2584 b_w = w - src_x; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2585 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2586 if(src_y<0){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2587 obmc -= src_y*obmc_stride; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2588 b_h += src_y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2589 src_y=0; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2590 }else if(src_y + b_h> h){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2591 b_h = h - src_y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2592 } |
2967 | 2593 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2594 if(b_w<=0 || b_h<=0) return; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2595 |
2842 | 2596 assert(src_stride > 2*MB_SIZE + 5); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2597 // old_dst += src_x + src_y*dst_stride; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2598 dst8+= src_x + src_y*src_stride; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2599 // src += src_x + src_y*src_stride; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2600 |
2842 | 2601 ptmp= tmp + 3*tmp_step; |
2602 block[0]= ptmp; | |
2603 ptmp+=tmp_step; | |
2967 | 2604 pred_block(s, block[0], src, tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2605 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2606 if(same_block(lt, rt)){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2607 block[1]= block[0]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2608 }else{ |
2842 | 2609 block[1]= ptmp; |
2610 ptmp+=tmp_step; | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2611 pred_block(s, block[1], src, tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2612 } |
2967 | 2613 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2614 if(same_block(lt, lb)){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2615 block[2]= block[0]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2616 }else if(same_block(rt, lb)){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2617 block[2]= block[1]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2618 }else{ |
2842 | 2619 block[2]= ptmp; |
2620 ptmp+=tmp_step; | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2621 pred_block(s, block[2], src, tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2622 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2623 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2624 if(same_block(lt, rb) ){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2625 block[3]= block[0]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2626 }else if(same_block(rt, rb)){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2627 block[3]= block[1]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2628 }else if(same_block(lb, rb)){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2629 block[3]= block[2]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2630 }else{ |
2842 | 2631 block[3]= ptmp; |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2632 pred_block(s, block[3], src, tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2633 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2634 #if 0 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2635 for(y=0; y<b_h; y++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2636 for(x=0; x<b_w; x++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2637 int v= obmc [x + y*obmc_stride] * block[3][x + y*src_stride] * (256/OBMC_MAX); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2638 if(add) dst[x + y*dst_stride] += v; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2639 else dst[x + y*dst_stride] -= v; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2640 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2641 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2642 for(y=0; y<b_h; y++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2643 uint8_t *obmc2= obmc + (obmc_stride>>1); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2644 for(x=0; x<b_w; x++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2645 int v= obmc2[x + y*obmc_stride] * block[2][x + y*src_stride] * (256/OBMC_MAX); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2646 if(add) dst[x + y*dst_stride] += v; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2647 else dst[x + y*dst_stride] -= v; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2648 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2649 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2650 for(y=0; y<b_h; y++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2651 uint8_t *obmc3= obmc + obmc_stride*(obmc_stride>>1); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2652 for(x=0; x<b_w; x++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2653 int v= obmc3[x + y*obmc_stride] * block[1][x + y*src_stride] * (256/OBMC_MAX); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2654 if(add) dst[x + y*dst_stride] += v; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2655 else dst[x + y*dst_stride] -= v; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2656 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2657 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2658 for(y=0; y<b_h; y++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2659 uint8_t *obmc3= obmc + obmc_stride*(obmc_stride>>1); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2660 uint8_t *obmc4= obmc3+ (obmc_stride>>1); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2661 for(x=0; x<b_w; x++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2662 int v= obmc4[x + y*obmc_stride] * block[0][x + y*src_stride] * (256/OBMC_MAX); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2663 if(add) dst[x + y*dst_stride] += v; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2664 else dst[x + y*dst_stride] -= v; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2665 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2666 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2667 #else |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2668 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2669 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2670 START_TIMER |
2967 | 2671 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2672 for(y=0; y<b_h; y++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2673 //FIXME ugly missue of obmc_stride |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2674 uint8_t *obmc1= obmc + y*obmc_stride; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2675 uint8_t *obmc2= obmc1+ (obmc_stride>>1); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2676 uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2677 uint8_t *obmc4= obmc3+ (obmc_stride>>1); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2678 dst = slice_buffer_get_line(sb, src_y + y); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2679 for(x=0; x<b_w; x++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2680 int v= obmc1[x] * block[3][x + y*src_stride] |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2681 +obmc2[x] * block[2][x + y*src_stride] |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2682 +obmc3[x] * block[1][x + y*src_stride] |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2683 +obmc4[x] * block[0][x + y*src_stride]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2684 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2685 v <<= 8 - LOG2_OBMC_MAX; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2686 if(FRAC_BITS != 8){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2687 v += 1<<(7 - FRAC_BITS); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2688 v >>= 8 - FRAC_BITS; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2689 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2690 if(add){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2691 // v += old_dst[x + y*dst_stride]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2692 v += dst[x + src_x]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2693 v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2694 if(v&(~255)) v= ~(v>>31); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2695 dst8[x + y*src_stride] = v; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2696 }else{ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2697 // old_dst[x + y*dst_stride] -= v; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2698 dst[x + src_x] -= v; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2699 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2700 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2701 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2702 STOP_TIMER("Inner add y block") |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2703 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2704 #endif |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2705 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2706 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2707 //FIXME name clenup (b_w, block_w, b_width stuff) |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2708 static always_inline void add_yblock(SnowContext *s, DWTELEM *dst, uint8_t *dst8, uint8_t *src, 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){ |
2206 | 2709 const int b_width = s->b_width << s->block_max_depth; |
2710 const int b_height= s->b_height << s->block_max_depth; | |
2711 const int b_stride= b_width; | |
2712 BlockNode *lt= &s->block[b_x + b_y*b_stride]; | |
2713 BlockNode *rt= lt+1; | |
2714 BlockNode *lb= lt+b_stride; | |
2715 BlockNode *rb= lb+1; | |
2967 | 2716 uint8_t *block[4]; |
2842 | 2717 int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride; |
2718 uint8_t tmp[src_stride*7*MB_SIZE]; //FIXME align | |
2719 uint8_t *ptmp; | |
2206 | 2720 int x,y; |
2721 | |
2722 if(b_x<0){ | |
2723 lt= rt; | |
2724 lb= rb; | |
2725 }else if(b_x + 1 >= b_width){ | |
2726 rt= lt; | |
2727 rb= lb; | |
2728 } | |
2729 if(b_y<0){ | |
2730 lt= lb; | |
2731 rt= rb; | |
2732 }else if(b_y + 1 >= b_height){ | |
2733 lb= lt; | |
2734 rb= rt; | |
2735 } | |
2967 | 2736 |
2206 | 2737 if(src_x<0){ //FIXME merge with prev & always round internal width upto *16 |
2738 obmc -= src_x; | |
2739 b_w += src_x; | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2740 if(!offset_dst) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2741 dst -= src_x; |
2206 | 2742 src_x=0; |
2743 }else if(src_x + b_w > w){ | |
2744 b_w = w - src_x; | |
2745 } | |
2746 if(src_y<0){ | |
2747 obmc -= src_y*obmc_stride; | |
2748 b_h += src_y; | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2749 if(!offset_dst) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2750 dst -= src_y*dst_stride; |
2206 | 2751 src_y=0; |
2752 }else if(src_y + b_h> h){ | |
2753 b_h = h - src_y; | |
2754 } | |
2967 | 2755 |
2206 | 2756 if(b_w<=0 || b_h<=0) return; |
2757 | |
2842 | 2758 assert(src_stride > 2*MB_SIZE + 5); |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2759 if(offset_dst) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2760 dst += src_x + src_y*dst_stride; |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2761 dst8+= src_x + src_y*src_stride; |
2206 | 2762 // src += src_x + src_y*src_stride; |
2763 | |
2842 | 2764 ptmp= tmp + 3*tmp_step; |
2765 block[0]= ptmp; | |
2766 ptmp+=tmp_step; | |
2967 | 2767 pred_block(s, block[0], src, tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h); |
2206 | 2768 |
2769 if(same_block(lt, rt)){ | |
2770 block[1]= block[0]; | |
2771 }else{ | |
2842 | 2772 block[1]= ptmp; |
2773 ptmp+=tmp_step; | |
2206 | 2774 pred_block(s, block[1], src, tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h); |
2775 } | |
2967 | 2776 |
2206 | 2777 if(same_block(lt, lb)){ |
2778 block[2]= block[0]; | |
2779 }else if(same_block(rt, lb)){ | |
2780 block[2]= block[1]; | |
2781 }else{ | |
2842 | 2782 block[2]= ptmp; |
2783 ptmp+=tmp_step; | |
2206 | 2784 pred_block(s, block[2], src, tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h); |
2785 } | |
2786 | |
2787 if(same_block(lt, rb) ){ | |
2788 block[3]= block[0]; | |
2789 }else if(same_block(rt, rb)){ | |
2790 block[3]= block[1]; | |
2791 }else if(same_block(lb, rb)){ | |
2792 block[3]= block[2]; | |
2793 }else{ | |
2842 | 2794 block[3]= ptmp; |
2206 | 2795 pred_block(s, block[3], src, tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h); |
2796 } | |
2797 #if 0 | |
2798 for(y=0; y<b_h; y++){ | |
2799 for(x=0; x<b_w; x++){ | |
2800 int v= obmc [x + y*obmc_stride] * block[3][x + y*src_stride] * (256/OBMC_MAX); | |
2801 if(add) dst[x + y*dst_stride] += v; | |
2802 else dst[x + y*dst_stride] -= v; | |
2803 } | |
2804 } | |
2805 for(y=0; y<b_h; y++){ | |
2806 uint8_t *obmc2= obmc + (obmc_stride>>1); | |
2807 for(x=0; x<b_w; x++){ | |
2808 int v= obmc2[x + y*obmc_stride] * block[2][x + y*src_stride] * (256/OBMC_MAX); | |
2809 if(add) dst[x + y*dst_stride] += v; | |
2810 else dst[x + y*dst_stride] -= v; | |
2811 } | |
2812 } | |
2813 for(y=0; y<b_h; y++){ | |
2814 uint8_t *obmc3= obmc + obmc_stride*(obmc_stride>>1); | |
2815 for(x=0; x<b_w; x++){ | |
2816 int v= obmc3[x + y*obmc_stride] * block[1][x + y*src_stride] * (256/OBMC_MAX); | |
2817 if(add) dst[x + y*dst_stride] += v; | |
2818 else dst[x + y*dst_stride] -= v; | |
2819 } | |
2820 } | |
2821 for(y=0; y<b_h; y++){ | |
2822 uint8_t *obmc3= obmc + obmc_stride*(obmc_stride>>1); | |
2823 uint8_t *obmc4= obmc3+ (obmc_stride>>1); | |
2824 for(x=0; x<b_w; x++){ | |
2825 int v= obmc4[x + y*obmc_stride] * block[0][x + y*src_stride] * (256/OBMC_MAX); | |
2826 if(add) dst[x + y*dst_stride] += v; | |
2827 else dst[x + y*dst_stride] -= v; | |
2828 } | |
2829 } | |
2830 #else | |
2831 for(y=0; y<b_h; y++){ | |
2832 //FIXME ugly missue of obmc_stride | |
2833 uint8_t *obmc1= obmc + y*obmc_stride; | |
2834 uint8_t *obmc2= obmc1+ (obmc_stride>>1); | |
2835 uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1); | |
2836 uint8_t *obmc4= obmc3+ (obmc_stride>>1); | |
2837 for(x=0; x<b_w; x++){ | |
2838 int v= obmc1[x] * block[3][x + y*src_stride] | |
2839 +obmc2[x] * block[2][x + y*src_stride] | |
2840 +obmc3[x] * block[1][x + y*src_stride] | |
2841 +obmc4[x] * block[0][x + y*src_stride]; | |
2967 | 2842 |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2843 v <<= 8 - LOG2_OBMC_MAX; |
2246 | 2844 if(FRAC_BITS != 8){ |
2845 v += 1<<(7 - FRAC_BITS); | |
2846 v >>= 8 - FRAC_BITS; | |
2847 } | |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2848 if(add){ |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2849 v += dst[x + y*dst_stride]; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2850 v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2851 if(v&(~255)) v= ~(v>>31); |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2852 dst8[x + y*src_stride] = v; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2853 }else{ |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2854 dst[x + y*dst_stride] -= v; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2855 } |
2206 | 2856 } |
2857 } | |
2858 #endif | |
2859 } | |
2860 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2861 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
|
2862 Plane *p= &s->plane[plane_index]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2863 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
|
2864 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
|
2865 int x, y, mb_x; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2866 int block_size = MB_SIZE >> s->block_max_depth; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2867 int block_w = plane_index ? block_size/2 : block_size; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2868 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
|
2869 int obmc_stride= plane_index ? block_size : 2*block_size; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2870 int ref_stride= s->current_picture.linesize[plane_index]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2871 uint8_t *ref = s->last_picture.data[plane_index]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2872 uint8_t *dst8= s->current_picture.data[plane_index]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2873 int w= p->width; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2874 int h= p->height; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2875 START_TIMER |
2967 | 2876 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2877 if(s->keyframe || (s->avctx->debug&512)){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2878 if(mb_y==mb_h) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2879 return; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2880 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2881 if(add){ |
2604
b7e6c3d31c65
Snow segfault bug in revision 1.43 patch by (Yartrebo /yartrebo earthlink net/)
michael
parents:
2602
diff
changeset
|
2882 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
|
2883 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2884 // DWTELEM * line = slice_buffer_get_line(sb, y); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2885 DWTELEM * line = sb->line[y]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2886 for(x=0; x<w; x++) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2887 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2888 // 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
|
2889 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
|
2890 v >>= FRAC_BITS; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2891 if(v&(~255)) v= ~(v>>31); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2892 dst8[x + y*ref_stride]= v; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2893 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2894 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2895 }else{ |
2604
b7e6c3d31c65
Snow segfault bug in revision 1.43 patch by (Yartrebo /yartrebo earthlink net/)
michael
parents:
2602
diff
changeset
|
2896 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
|
2897 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2898 // DWTELEM * line = slice_buffer_get_line(sb, y); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2899 DWTELEM * line = sb->line[y]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2900 for(x=0; x<w; x++) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2901 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2902 line[x] -= 128 << FRAC_BITS; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2903 // buf[x + y*w]-= 128<<FRAC_BITS; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2904 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2905 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2906 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2907 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2908 return; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2909 } |
2967 | 2910 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2911 for(mb_x=0; mb_x<=mb_w; mb_x++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2912 START_TIMER |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2913 |
2967 | 2914 add_yblock_buffered(s, sb, old_buffer, dst8, ref, obmc, |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2915 block_w*mb_x - block_w/2, |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2916 block_w*mb_y - block_w/2, |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2917 block_w, block_w, |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2918 w, h, |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2919 w, ref_stride, obmc_stride, |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2920 mb_x - 1, mb_y - 1, |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2921 add, plane_index); |
2967 | 2922 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2923 STOP_TIMER("add_yblock") |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2924 } |
2967 | 2925 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2926 STOP_TIMER("predict_slice") |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2927 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2928 |
2562 | 2929 static always_inline void predict_slice(SnowContext *s, DWTELEM *buf, int plane_index, int add, int mb_y){ |
2138 | 2930 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
|
2931 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
|
2932 const int mb_h= s->b_height << s->block_max_depth; |
2562 | 2933 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
|
2934 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
|
2935 int block_w = plane_index ? block_size/2 : block_size; |
2206 | 2936 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
|
2937 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
|
2938 int ref_stride= s->current_picture.linesize[plane_index]; |
2138 | 2939 uint8_t *ref = s->last_picture.data[plane_index]; |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2940 uint8_t *dst8= s->current_picture.data[plane_index]; |
2138 | 2941 int w= p->width; |
2942 int h= p->height; | |
2197 | 2943 START_TIMER |
2967 | 2944 |
2206 | 2945 if(s->keyframe || (s->avctx->debug&512)){ |
2562 | 2946 if(mb_y==mb_h) |
2947 return; | |
2948 | |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2949 if(add){ |
2604
b7e6c3d31c65
Snow segfault bug in revision 1.43 patch by (Yartrebo /yartrebo earthlink net/)
michael
parents:
2602
diff
changeset
|
2950 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
|
2951 for(x=0; x<w; x++){ |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2952 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
|
2953 v >>= FRAC_BITS; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2954 if(v&(~255)) v= ~(v>>31); |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2955 dst8[x + y*ref_stride]= v; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2956 } |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2957 } |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2958 }else{ |
2604
b7e6c3d31c65
Snow segfault bug in revision 1.43 patch by (Yartrebo /yartrebo earthlink net/)
michael
parents:
2602
diff
changeset
|
2959 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
|
2960 for(x=0; x<w; x++){ |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2961 buf[x + y*w]-= 128<<FRAC_BITS; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2962 } |
2206 | 2963 } |
2138 | 2964 } |
2206 | 2965 |
2966 return; | |
2138 | 2967 } |
2967 | 2968 |
2206 | 2969 for(mb_x=0; mb_x<=mb_w; mb_x++){ |
2197 | 2970 START_TIMER |
2206 | 2971 |
2967 | 2972 add_yblock(s, buf, dst8, ref, obmc, |
2206 | 2973 block_w*mb_x - block_w/2, |
2138 | 2974 block_w*mb_y - block_w/2, |
2206 | 2975 block_w, block_w, |
2138 | 2976 w, h, |
2206 | 2977 w, ref_stride, obmc_stride, |
2978 mb_x - 1, mb_y - 1, | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2979 add, 1, plane_index); |
2967 | 2980 |
2206 | 2981 STOP_TIMER("add_yblock") |
2138 | 2982 } |
2967 | 2983 |
2562 | 2984 STOP_TIMER("predict_slice") |
2985 } | |
2986 | |
2987 static always_inline void predict_plane(SnowContext *s, DWTELEM *buf, int plane_index, int add){ | |
2988 const int mb_h= s->b_height << s->block_max_depth; | |
2989 int mb_y; | |
2990 for(mb_y=0; mb_y<=mb_h; mb_y++) | |
2991 predict_slice(s, buf, plane_index, add, mb_y); | |
2138 | 2992 } |
2993 | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2994 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
|
2995 int i, x2, y2; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2996 Plane *p= &s->plane[plane_index]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2997 const int block_size = MB_SIZE >> s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2998 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
|
2999 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
|
3000 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
|
3001 const int ref_stride= s->current_picture.linesize[plane_index]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3002 uint8_t *ref= s-> last_picture.data[plane_index]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3003 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
|
3004 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
|
3005 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
|
3006 const int w= p->width; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3007 const int h= p->height; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3008 int index= mb_x + mb_y*b_stride; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3009 BlockNode *b= &s->block[index]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3010 BlockNode backup= *b; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3011 int ab=0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3012 int aa=0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3013 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3014 b->type|= BLOCK_INTRA; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3015 b->color[plane_index]= 0; |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3016 memset(dst, 0, obmc_stride*obmc_stride*sizeof(DWTELEM)); |
2993
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 for(i=0; i<4; i++){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3019 int mb_x2= mb_x + (i &1) - 1; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3020 int mb_y2= mb_y + (i>>1) - 1; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3021 int x= block_w*mb_x2 + block_w/2; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3022 int y= block_w*mb_y2 + block_w/2; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3023 |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3024 add_yblock(s, dst + ((i&1)+(i>>1)*obmc_stride)*block_w, NULL, ref, obmc, |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3025 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
|
3026 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3027 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
|
3028 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
|
3029 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
|
3030 int obmc_v= obmc[index]; |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3031 int d; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3032 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
|
3033 if(x<0) obmc_v += obmc[index + block_w]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3034 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
|
3035 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
|
3036 //FIXME precalc this or simplify it somehow else |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3037 |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3038 d = -dst[index] + (1<<(FRAC_BITS-1)); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3039 dst[index] = d; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3040 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
|
3041 aa += obmc_v * obmc_v; //FIXME precalclate this |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3042 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3043 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3044 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3045 *b= backup; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3046 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3047 return clip(((ab<<6) + aa/2)/aa, 0, 255); //FIXME we shouldnt need cliping |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3048 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3049 |
3051 | 3050 static inline int get_block_bits(SnowContext *s, int x, int y, int w){ |
3051 const int b_stride = s->b_width << s->block_max_depth; | |
3052 const int b_height = s->b_height<< s->block_max_depth; | |
3053 int index= x + y*b_stride; | |
3054 BlockNode *b = &s->block[index]; | |
3055 BlockNode *left = x ? &s->block[index-1] : &null_block; | |
3056 BlockNode *top = y ? &s->block[index-b_stride] : &null_block; | |
3057 BlockNode *tl = y && x ? &s->block[index-b_stride-1] : left; | |
3058 BlockNode *tr = y && x+w<b_stride ? &s->block[index-b_stride+w] : tl; | |
3059 int dmx, dmy; | |
3060 // int mx_context= av_log2(2*ABS(left->mx - top->mx)); | |
3061 // int my_context= av_log2(2*ABS(left->my - top->my)); | |
3062 | |
3063 if(x<0 || x>=b_stride || y>=b_height) | |
3064 return 0; | |
3065 dmx= b->mx - mid_pred(left->mx, top->mx, tr->mx); | |
3066 dmy= b->my - mid_pred(left->my, top->my, tr->my); | |
3067 /* | |
3068 1 0 0 | |
3069 01X 1-2 1 | |
3070 001XX 3-6 2-3 | |
3071 0001XXX 7-14 4-7 | |
3072 00001XXXX 15-30 8-15 | |
3073 */ | |
3074 //FIXME try accurate rate | |
3075 //FIXME intra and inter predictors if surrounding blocks arent the same type | |
3076 if(b->type & BLOCK_INTRA){ | |
3077 return 3+2*( av_log2(2*ABS(left->color[0] - b->color[0])) | |
3078 + av_log2(2*ABS(left->color[1] - b->color[1])) | |
3079 + av_log2(2*ABS(left->color[2] - b->color[2]))); | |
3080 }else | |
3081 return 2*(1 + av_log2(2*ABS(dmx)) | |
3082 + av_log2(2*ABS(dmy))); //FIXME kill the 2* can be merged in lambda | |
3083 } | |
3084 | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3085 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
|
3086 Plane *p= &s->plane[plane_index]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3087 const int block_size = MB_SIZE >> s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3088 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
|
3089 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
|
3090 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
|
3091 const int ref_stride= s->current_picture.linesize[plane_index]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3092 uint8_t *ref= s-> last_picture.data[plane_index]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3093 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
|
3094 uint8_t *src= s-> input_picture.data[plane_index]; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3095 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
|
3096 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
|
3097 uint8_t tmp[ref_stride*(2*MB_SIZE+5)]; |
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 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
|
3100 const int w= p->width; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3101 const int h= p->height; |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3102 int distortion; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3103 int rate= 0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3104 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
|
3105 int sx= block_w*mb_x - block_w/2; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3106 int sy= block_w*mb_y - block_w/2; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3107 const int x0= FFMAX(0,-sx); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3108 const int y0= FFMAX(0,-sy); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3109 const int x1= FFMIN(block_w*2, w-sx); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3110 const int y1= FFMIN(block_w*2, h-sy); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3111 int i,x,y; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3112 |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3113 pred_block(s, cur, ref, tmp, ref_stride, sx, sy, block_w*2, block_w*2, &s->block[mb_x + mb_y*b_stride], plane_index, w, h); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3114 |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3115 for(y=y0; y<y1; y++){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3116 const uint8_t *obmc1= obmc_edged + y*obmc_stride; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3117 const DWTELEM *pred1 = pred + y*obmc_stride; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3118 uint8_t *cur1 = cur + y*ref_stride; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3119 uint8_t *dst1 = dst + sx + (sy+y)*ref_stride; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3120 for(x=x0; x<x1; x++){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3121 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
|
3122 v = (v + pred1[x]) >> FRAC_BITS; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3123 if(v&(~255)) v= ~(v>>31); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3124 dst1[x] = v; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3125 } |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3126 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3127 |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3128 //FIXME sad/ssd can be broken up, but wavelet cmp should be one 32x32 block |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3129 if(block_w==16){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3130 distortion = 0; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3131 for(i=0; i<4; i++){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3132 int off = sx+16*(i&1) + (sy+16*(i>>1))*ref_stride; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3133 distortion += s->dsp.me_cmp[0](&s->m, src + off, dst + off, ref_stride, 16); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3134 } |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3135 }else{ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3136 assert(block_w==8); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3137 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
|
3138 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3139 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3140 if(plane_index==0){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3141 for(i=0; i<4; i++){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3142 /* ..RRr |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3143 * .RXx. |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3144 * rxx.. |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3145 */ |
3051 | 3146 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
|
3147 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3148 } |
3051 | 3149 return distortion + rate*penalty_factor; |
3150 } | |
3151 | |
3152 static int get_4block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index){ | |
3153 int i, y2; | |
3154 Plane *p= &s->plane[plane_index]; | |
3155 const int block_size = MB_SIZE >> s->block_max_depth; | |
3156 const int block_w = plane_index ? block_size/2 : block_size; | |
3157 const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth]; | |
3158 const int obmc_stride= plane_index ? block_size : 2*block_size; | |
3159 const int ref_stride= s->current_picture.linesize[plane_index]; | |
3160 uint8_t *ref= s-> last_picture.data[plane_index]; | |
3161 uint8_t *dst= s->current_picture.data[plane_index]; | |
3162 uint8_t *src= s-> input_picture.data[plane_index]; | |
3163 const static DWTELEM zero_dst[4096]; //FIXME | |
3164 const int b_stride = s->b_width << s->block_max_depth; | |
3165 const int b_height = s->b_height<< s->block_max_depth; | |
3166 const int w= p->width; | |
3167 const int h= p->height; | |
3168 int distortion= 0; | |
3169 int rate= 0; | |
3170 const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp); | |
3171 | |
3172 for(i=0; i<9; i++){ | |
3173 int mb_x2= mb_x + (i%3) - 1; | |
3174 int mb_y2= mb_y + (i/3) - 1; | |
3175 int x= block_w*mb_x2 + block_w/2; | |
3176 int y= block_w*mb_y2 + block_w/2; | |
3177 | |
3178 add_yblock(s, zero_dst, dst, ref, obmc, | |
3179 x, y, block_w, block_w, w, h, /*dst_stride*/0, ref_stride, obmc_stride, mb_x2, mb_y2, 1, 1, plane_index); | |
3180 | |
3181 //FIXME find a cleaner/simpler way to skip the outside stuff | |
3182 for(y2= y; y2<0; y2++) | |
3183 memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w); | |
3184 for(y2= h; y2<y+block_w; y2++) | |
3185 memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w); | |
3186 if(x<0){ | |
3187 for(y2= y; y2<y+block_w; y2++) | |
3188 memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, -x); | |
3189 } | |
3190 if(x+block_w > w){ | |
3191 for(y2= y; y2<y+block_w; y2++) | |
3192 memcpy(dst + w + y2*ref_stride, src + w + y2*ref_stride, x+block_w - w); | |
3193 } | |
3194 | |
3195 assert(block_w== 8 || block_w==16); | |
3196 distortion += s->dsp.me_cmp[block_w==8](&s->m, src + x + y*ref_stride, dst + x + y*ref_stride, ref_stride, block_w); | |
3197 } | |
3198 | |
3199 if(plane_index==0){ | |
3200 BlockNode *b= &s->block[mb_x+mb_y*b_stride]; | |
3201 int merged= same_block(b,b+1) && same_block(b,b+b_stride) && same_block(b,b+b_stride+1); | |
3202 | |
3203 /* ..RRRr | |
3204 * .RXXx. | |
3205 * .RXXx. | |
3206 * rxxx. | |
3207 */ | |
3208 if(merged) | |
3209 rate = get_block_bits(s, mb_x, mb_y, 2); | |
3210 for(i=merged?4:0; i<9; i++){ | |
3211 static const int dxy[9][2] = {{0,0},{1,0},{0,1},{1,1},{2,0},{2,1},{-1,2},{0,2},{1,2}}; | |
3212 rate += get_block_bits(s, mb_x + dxy[i][0], mb_y + dxy[i][1], 1); | |
3213 } | |
3214 } | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3215 return distortion + rate*penalty_factor; |
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 |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3218 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
|
3219 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
|
3220 BlockNode *block= &s->block[mb_x + mb_y * b_stride]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3221 BlockNode backup= *block; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3222 int rd, index, value; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3223 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3224 assert(mb_x>=0 && mb_y>=0); |
2994 | 3225 assert(mb_x<b_stride); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3226 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3227 if(intra){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3228 block->color[0] = p[0]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3229 block->color[1] = p[1]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3230 block->color[2] = p[2]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3231 block->type |= BLOCK_INTRA; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3232 }else{ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3233 index= (p[0] + 31*p[1]) & (ME_CACHE_SIZE-1); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3234 value= s->me_cache_generation + (p[0]>>10) + (p[1]<<6); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3235 if(s->me_cache[index] == value) |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3236 return 0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3237 s->me_cache[index]= value; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3238 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3239 block->mx= p[0]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3240 block->my= p[1]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3241 block->type &= ~BLOCK_INTRA; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3242 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3243 |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3244 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
|
3245 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3246 //FIXME chroma |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3247 if(rd < *best_rd){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3248 *best_rd= rd; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3249 return 1; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3250 }else{ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3251 *block= backup; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3252 return 0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3253 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3254 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3255 |
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
|
3256 /* special case for int[2] args we discard afterward, fixes compilation prob with gcc 2.95 */ |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3257 static always_inline int check_block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, int intra, 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
|
3258 int p[2] = {p0, p1}; |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3259 return check_block(s, mb_x, mb_y, p, intra, 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
|
3260 } |
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
|
3261 |
3051 | 3262 static always_inline int check_4block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, int *best_rd){ |
3263 const int b_stride= s->b_width << s->block_max_depth; | |
3264 BlockNode *block= &s->block[mb_x + mb_y * b_stride]; | |
3265 BlockNode backup[4]= {block[0], block[1], block[b_stride], block[b_stride+1]}; | |
3266 int rd, index, value; | |
3267 | |
3268 assert(mb_x>=0 && mb_y>=0); | |
3269 assert(mb_x<b_stride); | |
3270 assert(((mb_x|mb_y)&1) == 0); | |
3271 | |
3272 index= (p0 + 31*p1) & (ME_CACHE_SIZE-1); | |
3273 value= s->me_cache_generation + (p0>>10) + (p1<<6); | |
3274 if(s->me_cache[index] == value) | |
3275 return 0; | |
3276 s->me_cache[index]= value; | |
3277 | |
3278 block->mx= p0; | |
3279 block->my= p1; | |
3280 block->type &= ~BLOCK_INTRA; | |
3281 block[1]= block[b_stride]= block[b_stride+1]= *block; | |
3282 | |
3283 rd= get_4block_rd(s, mb_x, mb_y, 0); | |
3284 | |
3285 //FIXME chroma | |
3286 if(rd < *best_rd){ | |
3287 *best_rd= rd; | |
3288 return 1; | |
3289 }else{ | |
3290 block[0]= backup[0]; | |
3291 block[1]= backup[1]; | |
3292 block[b_stride]= backup[2]; | |
3293 block[b_stride+1]= backup[3]; | |
3294 return 0; | |
3295 } | |
3296 } | |
3297 | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3298 static void iterative_me(SnowContext *s){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3299 int pass, mb_x, mb_y; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3300 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
|
3301 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
|
3302 const int b_stride= b_width; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3303 int color[3]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3304 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3305 for(pass=0; pass<50; pass++){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3306 int change= 0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3307 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3308 for(mb_y= 0; mb_y<b_height; mb_y++){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3309 for(mb_x= 0; mb_x<b_width; mb_x++){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3310 int dia_change, i, j; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3311 int best_rd= INT_MAX; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3312 BlockNode backup; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3313 const int index= mb_x + mb_y * b_stride; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3314 BlockNode *block= &s->block[index]; |
3035 | 3315 BlockNode *tb = mb_y ? &s->block[index-b_stride ] : &null_block; |
3316 BlockNode *lb = mb_x ? &s->block[index -1] : &null_block; | |
3317 BlockNode *rb = mb_x+1<b_width ? &s->block[index +1] : &null_block; | |
3318 BlockNode *bb = mb_y+1<b_height ? &s->block[index+b_stride ] : &null_block; | |
3319 BlockNode *tlb= mb_x && mb_y ? &s->block[index-b_stride-1] : &null_block; | |
3320 BlockNode *trb= mb_x+1<b_width && mb_y ? &s->block[index-b_stride+1] : &null_block; | |
3321 BlockNode *blb= mb_x && mb_y+1<b_height ? &s->block[index+b_stride-1] : &null_block; | |
3322 BlockNode *brb= mb_x+1<b_width && mb_y+1<b_height ? &s->block[index+b_stride+1] : &null_block; | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3323 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
|
3324 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
|
3325 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3326 if(pass && (block->type & BLOCK_OPT)) |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3327 continue; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3328 block->type |= BLOCK_OPT; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3329 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3330 backup= *block; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3331 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3332 if(!s->me_cache_generation) |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3333 memset(s->me_cache, 0, sizeof(s->me_cache)); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3334 s->me_cache_generation += 1<<22; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3335 |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3336 //FIXME precalc |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3337 { |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3338 int x, y; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3339 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
|
3340 if(mb_x==0) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3341 for(y=0; y<b_w*2; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3342 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
|
3343 if(mb_x==b_stride-1) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3344 for(y=0; y<b_w*2; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3345 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
|
3346 if(mb_y==0){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3347 for(x=0; x<b_w*2; x++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3348 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
|
3349 for(y=1; y<b_w; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3350 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
|
3351 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3352 if(mb_y==b_height-1){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3353 for(x=0; x<b_w*2; x++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3354 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
|
3355 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
|
3356 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
|
3357 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3358 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3359 |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3360 //skip stuff outside the picture |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3361 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
|
3362 { |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3363 uint8_t *src= s-> input_picture.data[0]; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3364 uint8_t *dst= s->current_picture.data[0]; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3365 const int stride= s->current_picture.linesize[0]; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3366 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
|
3367 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
|
3368 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
|
3369 const int w= s->plane[0].width; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3370 const int h= s->plane[0].height; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3371 int y; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3372 |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3373 for(y=sy; y<0; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3374 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
|
3375 for(y=h; y<sy+block_w*2; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3376 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
|
3377 if(sx<0){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3378 for(y=sy; y<sy+block_w*2; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3379 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
|
3380 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3381 if(sx+block_w*2 > w){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3382 for(y=sy; y<sy+block_w*2; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3383 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
|
3384 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3385 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3386 |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3387 // intra(black) = neighbors' contribution to the current block |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3388 for(i=0; i<3; i++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3389 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
|
3390 |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3391 // get previous score (cant be cached due to OBMC) |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3392 check_block_inter(s, mb_x, mb_y, block->mx, block->my, 0, *obmc_edged, &best_rd); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3393 check_block_inter(s, mb_x, mb_y, 0, 0, 0, *obmc_edged, &best_rd); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3394 check_block_inter(s, mb_x, mb_y, tb->mx, tb->my, 0, *obmc_edged, &best_rd); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3395 check_block_inter(s, mb_x, mb_y, lb->mx, lb->my, 0, *obmc_edged, &best_rd); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3396 check_block_inter(s, mb_x, mb_y, rb->mx, rb->my, 0, *obmc_edged, &best_rd); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3397 check_block_inter(s, mb_x, mb_y, bb->mx, bb->my, 0, *obmc_edged, &best_rd); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3398 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3399 /* fullpel ME */ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3400 //FIXME avoid subpel interpol / round to nearest integer |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3401 do{ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3402 dia_change=0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3403 for(i=0; i<FFMAX(s->avctx->dia_size, 1); i++){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3404 for(j=0; j<i; j++){ |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3405 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my+(4*j), 0, *obmc_edged, &best_rd); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3406 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my-(4*j), 0, *obmc_edged, &best_rd); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3407 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my-(4*j), 0, *obmc_edged, &best_rd); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3408 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my+(4*j), 0, *obmc_edged, &best_rd); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3409 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3410 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3411 }while(dia_change); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3412 /* subpel ME */ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3413 do{ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3414 static const int square[8][2]= {{+1, 0},{-1, 0},{ 0,+1},{ 0,-1},{+1,+1},{-1,-1},{+1,-1},{-1,+1},}; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3415 dia_change=0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3416 for(i=0; i<8; i++) |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3417 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+square[i][0], block->my+square[i][1], 0, *obmc_edged, &best_rd); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3418 }while(dia_change); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3419 //FIXME or try the standard 2 pass qpel or similar |
2998 | 3420 #if 1 |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3421 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
|
3422 //FIXME RD style color selection |
2998 | 3423 #endif |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3424 if(!same_block(block, &backup)){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3425 if(tb != &null_block) tb ->type &= ~BLOCK_OPT; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3426 if(lb != &null_block) lb ->type &= ~BLOCK_OPT; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3427 if(rb != &null_block) rb ->type &= ~BLOCK_OPT; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3428 if(bb != &null_block) bb ->type &= ~BLOCK_OPT; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3429 if(tlb!= &null_block) tlb->type &= ~BLOCK_OPT; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3430 if(trb!= &null_block) trb->type &= ~BLOCK_OPT; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3431 if(blb!= &null_block) blb->type &= ~BLOCK_OPT; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3432 if(brb!= &null_block) brb->type &= ~BLOCK_OPT; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3433 change ++; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3434 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3435 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3436 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3437 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
|
3438 if(!change) |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3439 break; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3440 } |
3051 | 3441 |
3442 if(s->block_max_depth == 1){ | |
3443 int change= 0; | |
3444 for(mb_y= 0; mb_y<b_height; mb_y+=2){ | |
3445 for(mb_x= 0; mb_x<b_width; mb_x+=2){ | |
3446 int dia_change, i, j; | |
3447 int best_rd, init_rd; | |
3448 const int index= mb_x + mb_y * b_stride; | |
3449 BlockNode *b[4]; | |
3450 | |
3451 b[0]= &s->block[index]; | |
3452 b[1]= b[0]+1; | |
3453 b[2]= b[0]+b_stride; | |
3454 b[3]= b[2]+1; | |
3455 if(same_block(b[0], b[1]) && | |
3456 same_block(b[0], b[2]) && | |
3457 same_block(b[0], b[3])) | |
3458 continue; | |
3459 | |
3460 if(!s->me_cache_generation) | |
3461 memset(s->me_cache, 0, sizeof(s->me_cache)); | |
3462 s->me_cache_generation += 1<<22; | |
3463 | |
3464 init_rd= best_rd= get_4block_rd(s, mb_x, mb_y, 0); | |
3465 | |
3466 check_4block_inter(s, mb_x, mb_y, | |
3467 (b[0]->mx + b[1]->mx + b[2]->mx + b[3]->mx + 2) >> 2, | |
3468 (b[0]->my + b[1]->my + b[2]->my + b[3]->my + 2) >> 2, &best_rd); | |
3469 | |
3470 for(i=0; i<4; i++) | |
3471 if(!(b[i]->type&BLOCK_INTRA)) | |
3472 check_4block_inter(s, mb_x, mb_y, b[i]->mx, b[i]->my, &best_rd); | |
3473 | |
3474 if(init_rd != best_rd) | |
3475 change++; | |
3476 } | |
3477 } | |
3478 av_log(NULL, AV_LOG_ERROR, "pass:4mv changed:%d\n", change*4); | |
3479 } | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3480 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3481 |
2138 | 3482 static void quantize(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int bias){ |
3483 const int level= b->level; | |
3484 const int w= b->width; | |
3485 const int h= b->height; | |
2600 | 3486 const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16); |
3487 const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); | |
2150 | 3488 int x,y, thres1, thres2; |
2893 | 3489 // START_TIMER |
2138 | 3490 |
2161 | 3491 if(s->qlog == LOSSLESS_QLOG) return; |
2967 | 3492 |
2138 | 3493 bias= bias ? 0 : (3*qmul)>>3; |
2150 | 3494 thres1= ((qmul - bias)>>QEXPSHIFT) - 1; |
3495 thres2= 2*thres1; | |
2967 | 3496 |
2138 | 3497 if(!bias){ |
3498 for(y=0; y<h; y++){ | |
3499 for(x=0; x<w; x++){ | |
2150 | 3500 int i= src[x + y*stride]; |
2967 | 3501 |
2150 | 3502 if((unsigned)(i+thres1) > thres2){ |
3503 if(i>=0){ | |
3504 i<<= QEXPSHIFT; | |
3505 i/= qmul; //FIXME optimize | |
3506 src[x + y*stride]= i; | |
3507 }else{ | |
3508 i= -i; | |
3509 i<<= QEXPSHIFT; | |
3510 i/= qmul; //FIXME optimize | |
3511 src[x + y*stride]= -i; | |
3512 } | |
3513 }else | |
3514 src[x + y*stride]= 0; | |
2138 | 3515 } |
3516 } | |
3517 }else{ | |
3518 for(y=0; y<h; y++){ | |
3519 for(x=0; x<w; x++){ | |
2967 | 3520 int i= src[x + y*stride]; |
3521 | |
2150 | 3522 if((unsigned)(i+thres1) > thres2){ |
3523 if(i>=0){ | |
3524 i<<= QEXPSHIFT; | |
3525 i= (i + bias) / qmul; //FIXME optimize | |
3526 src[x + y*stride]= i; | |
3527 }else{ | |
3528 i= -i; | |
3529 i<<= QEXPSHIFT; | |
3530 i= (i + bias) / qmul; //FIXME optimize | |
3531 src[x + y*stride]= -i; | |
3532 } | |
3533 }else | |
3534 src[x + y*stride]= 0; | |
2138 | 3535 } |
3536 } | |
3537 } | |
2150 | 3538 if(level+1 == s->spatial_decomposition_count){ |
3539 // STOP_TIMER("quantize") | |
3540 } | |
2138 | 3541 } |
3542 | |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
3543 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
|
3544 const int w= b->width; |
2600 | 3545 const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16); |
3546 const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3547 const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3548 int x,y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3549 START_TIMER |
2967 | 3550 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3551 if(s->qlog == LOSSLESS_QLOG) return; |
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 // 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
|
3555 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
|
3556 for(x=0; x<w; x++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3557 int i= line[x]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3558 if(i<0){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3559 line[x]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3560 }else if(i>0){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3561 line[x]= (( i*qmul + qadd)>>(QEXPSHIFT)); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3562 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3563 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3564 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3565 if(w > 200 /*level+1 == s->spatial_decomposition_count*/){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3566 STOP_TIMER("dquant") |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3567 } |
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 |
2138 | 3570 static void dequantize(SnowContext *s, SubBand *b, DWTELEM *src, int stride){ |
3571 const int w= b->width; | |
3572 const int h= b->height; | |
2600 | 3573 const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16); |
3574 const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); | |
2138 | 3575 const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT; |
3576 int x,y; | |
2195 | 3577 START_TIMER |
2967 | 3578 |
2161 | 3579 if(s->qlog == LOSSLESS_QLOG) return; |
2967 | 3580 |
2138 | 3581 for(y=0; y<h; y++){ |
3582 for(x=0; x<w; x++){ | |
3583 int i= src[x + y*stride]; | |
3584 if(i<0){ | |
3585 src[x + y*stride]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias | |
3586 }else if(i>0){ | |
3587 src[x + y*stride]= (( i*qmul + qadd)>>(QEXPSHIFT)); | |
3588 } | |
3589 } | |
3590 } | |
2195 | 3591 if(w > 200 /*level+1 == s->spatial_decomposition_count*/){ |
3592 STOP_TIMER("dquant") | |
3593 } | |
2138 | 3594 } |
3595 | |
3596 static void decorrelate(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int inverse, int use_median){ | |
3597 const int w= b->width; | |
3598 const int h= b->height; | |
3599 int x,y; | |
2967 | 3600 |
2138 | 3601 for(y=h-1; y>=0; y--){ |
3602 for(x=w-1; x>=0; x--){ | |
3603 int i= x + y*stride; | |
2967 | 3604 |
2138 | 3605 if(x){ |
3606 if(use_median){ | |
3607 if(y && x+1<w) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]); | |
3608 else src[i] -= src[i - 1]; | |
3609 }else{ | |
3610 if(y) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]); | |
3611 else src[i] -= src[i - 1]; | |
3612 } | |
3613 }else{ | |
3614 if(y) src[i] -= src[i - stride]; | |
3615 } | |
3616 } | |
3617 } | |
3618 } | |
3619 | |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
3620 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
|
3621 const int w= b->width; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3622 int x,y; |
2967 | 3623 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3624 // START_TIMER |
2967 | 3625 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3626 DWTELEM * line; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3627 DWTELEM * prev; |
2967 | 3628 |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
3629 if (start_y != 0) |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
3630 line = slice_buffer_get_line(sb, ((start_y - 1) * b->stride_line) + b->buf_y_offset) + b->buf_x_offset; |
2967 | 3631 |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
3632 for(y=start_y; y<end_y; y++){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3633 prev = line; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3634 // 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
|
3635 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
|
3636 for(x=0; x<w; x++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3637 if(x){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3638 if(use_median){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3639 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
|
3640 else line[x] += line[x - 1]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3641 }else{ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3642 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
|
3643 else line[x] += line[x - 1]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3644 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3645 }else{ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3646 if(y) line[x] += prev[x]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3647 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3648 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3649 } |
2967 | 3650 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3651 // STOP_TIMER("correlate") |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3652 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3653 |
2138 | 3654 static void correlate(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int inverse, int use_median){ |
3655 const int w= b->width; | |
3656 const int h= b->height; | |
3657 int x,y; | |
2967 | 3658 |
2138 | 3659 for(y=0; y<h; y++){ |
3660 for(x=0; x<w; x++){ | |
3661 int i= x + y*stride; | |
2967 | 3662 |
2138 | 3663 if(x){ |
3664 if(use_median){ | |
3665 if(y && x+1<w) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]); | |
3666 else src[i] += src[i - 1]; | |
3667 }else{ | |
3668 if(y) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]); | |
3669 else src[i] += src[i - 1]; | |
3670 } | |
3671 }else{ | |
3672 if(y) src[i] += src[i - stride]; | |
3673 } | |
3674 } | |
3675 } | |
3676 } | |
3677 | |
3678 static void encode_header(SnowContext *s){ | |
3679 int plane_index, level, orientation; | |
2967 | 3680 uint8_t kstate[32]; |
3681 | |
3682 memset(kstate, MID_STATE, sizeof(kstate)); | |
2138 | 3683 |
2335 | 3684 put_rac(&s->c, kstate, s->keyframe); |
2199 | 3685 if(s->keyframe || s->always_reset) |
3686 reset_contexts(s); | |
2138 | 3687 if(s->keyframe){ |
3688 put_symbol(&s->c, s->header_state, s->version, 0); | |
2335 | 3689 put_rac(&s->c, s->header_state, s->always_reset); |
2138 | 3690 put_symbol(&s->c, s->header_state, s->temporal_decomposition_type, 0); |
3691 put_symbol(&s->c, s->header_state, s->temporal_decomposition_count, 0); | |
3692 put_symbol(&s->c, s->header_state, s->spatial_decomposition_count, 0); | |
3693 put_symbol(&s->c, s->header_state, s->colorspace_type, 0); | |
3694 put_symbol(&s->c, s->header_state, s->chroma_h_shift, 0); | |
3695 put_symbol(&s->c, s->header_state, s->chroma_v_shift, 0); | |
2335 | 3696 put_rac(&s->c, s->header_state, s->spatial_scalability); |
3697 // put_rac(&s->c, s->header_state, s->rate_scalability); | |
2138 | 3698 |
3699 for(plane_index=0; plane_index<2; plane_index++){ | |
3700 for(level=0; level<s->spatial_decomposition_count; level++){ | |
3701 for(orientation=level ? 1:0; orientation<4; orientation++){ | |
3702 if(orientation==2) continue; | |
3703 put_symbol(&s->c, s->header_state, s->plane[plane_index].band[level][orientation].qlog, 1); | |
3704 } | |
3705 } | |
3706 } | |
3707 } | |
3708 put_symbol(&s->c, s->header_state, s->spatial_decomposition_type, 0); | |
2967 | 3709 put_symbol(&s->c, s->header_state, s->qlog, 1); |
3710 put_symbol(&s->c, s->header_state, s->mv_scale, 0); | |
2138 | 3711 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
|
3712 put_symbol(&s->c, s->header_state, s->block_max_depth, 0); |
2138 | 3713 } |
3714 | |
3715 static int decode_header(SnowContext *s){ | |
3716 int plane_index, level, orientation; | |
2335 | 3717 uint8_t kstate[32]; |
2138 | 3718 |
2967 | 3719 memset(kstate, MID_STATE, sizeof(kstate)); |
2335 | 3720 |
3721 s->keyframe= get_rac(&s->c, kstate); | |
2199 | 3722 if(s->keyframe || s->always_reset) |
3723 reset_contexts(s); | |
2138 | 3724 if(s->keyframe){ |
3725 s->version= get_symbol(&s->c, s->header_state, 0); | |
3726 if(s->version>0){ | |
3727 av_log(s->avctx, AV_LOG_ERROR, "version %d not supported", s->version); | |
3728 return -1; | |
3729 } | |
2335 | 3730 s->always_reset= get_rac(&s->c, s->header_state); |
2138 | 3731 s->temporal_decomposition_type= get_symbol(&s->c, s->header_state, 0); |
3732 s->temporal_decomposition_count= get_symbol(&s->c, s->header_state, 0); | |
3733 s->spatial_decomposition_count= get_symbol(&s->c, s->header_state, 0); | |
3734 s->colorspace_type= get_symbol(&s->c, s->header_state, 0); | |
3735 s->chroma_h_shift= get_symbol(&s->c, s->header_state, 0); | |
3736 s->chroma_v_shift= get_symbol(&s->c, s->header_state, 0); | |
2335 | 3737 s->spatial_scalability= get_rac(&s->c, s->header_state); |
3738 // s->rate_scalability= get_rac(&s->c, s->header_state); | |
2138 | 3739 |
3740 for(plane_index=0; plane_index<3; plane_index++){ | |
3741 for(level=0; level<s->spatial_decomposition_count; level++){ | |
3742 for(orientation=level ? 1:0; orientation<4; orientation++){ | |
3743 int q; | |
3744 if (plane_index==2) q= s->plane[1].band[level][orientation].qlog; | |
3745 else if(orientation==2) q= s->plane[plane_index].band[level][1].qlog; | |
3746 else q= get_symbol(&s->c, s->header_state, 1); | |
3747 s->plane[plane_index].band[level][orientation].qlog= q; | |
3748 } | |
3749 } | |
3750 } | |
3751 } | |
2967 | 3752 |
2138 | 3753 s->spatial_decomposition_type= get_symbol(&s->c, s->header_state, 0); |
3754 if(s->spatial_decomposition_type > 2){ | |
3755 av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_type %d not supported", s->spatial_decomposition_type); | |
3756 return -1; | |
3757 } | |
2967 | 3758 |
2138 | 3759 s->qlog= get_symbol(&s->c, s->header_state, 1); |
3760 s->mv_scale= get_symbol(&s->c, s->header_state, 0); | |
3761 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
|
3762 s->block_max_depth= get_symbol(&s->c, s->header_state, 0); |
2952 | 3763 if(s->block_max_depth > 1){ |
3764 av_log(s->avctx, AV_LOG_ERROR, "block_max_depth= %d is too large", s->block_max_depth); | |
3765 s->block_max_depth= 0; | |
3766 return -1; | |
3767 } | |
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
|
3768 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3769 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
|
3770 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3771 |
2600 | 3772 static void init_qexp(){ |
3773 int i; | |
3774 double v=128; | |
3775 | |
3776 for(i=0; i<QROOT; i++){ | |
3777 qexp[i]= lrintf(v); | |
2967 | 3778 v *= pow(2, 1.0 / QROOT); |
2600 | 3779 } |
3780 } | |
3781 | |
2138 | 3782 static int common_init(AVCodecContext *avctx){ |
3783 SnowContext *s = avctx->priv_data; | |
3784 int width, height; | |
3785 int level, orientation, plane_index, dec; | |
3786 | |
3787 s->avctx= avctx; | |
2967 | 3788 |
2138 | 3789 dsputil_init(&s->dsp, avctx); |
3790 | |
3791 #define mcf(dx,dy)\ | |
3792 s->dsp.put_qpel_pixels_tab [0][dy+dx/4]=\ | |
3793 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
|
3794 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
|
3795 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
|
3796 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
|
3797 s->dsp.put_h264_qpel_pixels_tab[1][dy+dx/4]; |
2138 | 3798 |
3799 mcf( 0, 0) | |
3800 mcf( 4, 0) | |
3801 mcf( 8, 0) | |
3802 mcf(12, 0) | |
3803 mcf( 0, 4) | |
3804 mcf( 4, 4) | |
3805 mcf( 8, 4) | |
3806 mcf(12, 4) | |
3807 mcf( 0, 8) | |
3808 mcf( 4, 8) | |
3809 mcf( 8, 8) | |
3810 mcf(12, 8) | |
3811 mcf( 0,12) | |
3812 mcf( 4,12) | |
3813 mcf( 8,12) | |
3814 mcf(12,12) | |
3815 | |
3816 #define mcfh(dx,dy)\ | |
3817 s->dsp.put_pixels_tab [0][dy/4+dx/8]=\ | |
3818 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
|
3819 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
|
3820 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
|
3821 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
|
3822 mc_block_hpel ## dx ## dy ## 8; |
2138 | 3823 |
3824 mcfh(0, 0) | |
3825 mcfh(8, 0) | |
3826 mcfh(0, 8) | |
3827 mcfh(8, 8) | |
2600 | 3828 |
3829 if(!qexp[0]) | |
3830 init_qexp(); | |
3831 | |
2138 | 3832 dec= s->spatial_decomposition_count= 5; |
3833 s->spatial_decomposition_type= avctx->prediction_method; //FIXME add decorrelator type r transform_type | |
2967 | 3834 |
2138 | 3835 s->chroma_h_shift= 1; //FIXME XXX |
3836 s->chroma_v_shift= 1; | |
2967 | 3837 |
2138 | 3838 // dec += FFMAX(s->chroma_h_shift, s->chroma_v_shift); |
2967 | 3839 |
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
|
3840 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
|
3841 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
|
3842 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3843 s->spatial_dwt_buffer= av_mallocz(width*height*sizeof(DWTELEM)); |
2967 | 3844 |
2138 | 3845 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
|
3846 s->block_max_depth= (s->avctx->flags & CODEC_FLAG_4MV) ? 1 : 0; |
2967 | 3847 |
3848 for(plane_index=0; plane_index<3; plane_index++){ | |
2138 | 3849 int w= s->avctx->width; |
3850 int h= s->avctx->height; | |
3851 | |
3852 if(plane_index){ | |
3853 w>>= s->chroma_h_shift; | |
3854 h>>= s->chroma_v_shift; | |
3855 } | |
3856 s->plane[plane_index].width = w; | |
3857 s->plane[plane_index].height= h; | |
2160 | 3858 //av_log(NULL, AV_LOG_DEBUG, "%d %d\n", w, h); |
2138 | 3859 for(level=s->spatial_decomposition_count-1; level>=0; level--){ |
3860 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
3861 SubBand *b= &s->plane[plane_index].band[level][orientation]; | |
2967 | 3862 |
2138 | 3863 b->buf= s->spatial_dwt_buffer; |
3864 b->level= level; | |
3865 b->stride= s->plane[plane_index].width << (s->spatial_decomposition_count - level); | |
3866 b->width = (w + !(orientation&1))>>1; | |
3867 b->height= (h + !(orientation>1))>>1; | |
2967 | 3868 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3869 b->stride_line = 1 << (s->spatial_decomposition_count - level); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3870 b->buf_x_offset = 0; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3871 b->buf_y_offset = 0; |
2967 | 3872 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3873 if(orientation&1){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3874 b->buf += (w+1)>>1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3875 b->buf_x_offset = (w+1)>>1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3876 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3877 if(orientation>1){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3878 b->buf += b->stride>>1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3879 b->buf_y_offset = b->stride_line >> 1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3880 } |
2967 | 3881 |
2138 | 3882 if(level) |
3883 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
|
3884 b->x_coeff=av_mallocz(((b->width+1) * b->height+1)*sizeof(x_and_coeff)); |
2138 | 3885 } |
3886 w= (w+1)>>1; | |
3887 h= (h+1)>>1; | |
3888 } | |
3889 } | |
2967 | 3890 |
2138 | 3891 reset_contexts(s); |
2967 | 3892 /* |
2138 | 3893 width= s->width= avctx->width; |
3894 height= s->height= avctx->height; | |
2967 | 3895 |
2138 | 3896 assert(width && height); |
3897 */ | |
3898 s->avctx->get_buffer(s->avctx, &s->mconly_picture); | |
2967 | 3899 |
2138 | 3900 return 0; |
3901 } | |
3902 | |
3903 | |
3904 static void calculate_vissual_weight(SnowContext *s, Plane *p){ | |
3905 int width = p->width; | |
3906 int height= p->height; | |
2198 | 3907 int level, orientation, x, y; |
2138 | 3908 |
3909 for(level=0; level<s->spatial_decomposition_count; level++){ | |
3910 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
3911 SubBand *b= &p->band[level][orientation]; | |
3912 DWTELEM *buf= b->buf; | |
3913 int64_t error=0; | |
2967 | 3914 |
2138 | 3915 memset(s->spatial_dwt_buffer, 0, sizeof(int)*width*height); |
3916 buf[b->width/2 + b->height/2*b->stride]= 256*256; | |
2164 | 3917 ff_spatial_idwt(s->spatial_dwt_buffer, width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count); |
2138 | 3918 for(y=0; y<height; y++){ |
3919 for(x=0; x<width; x++){ | |
3920 int64_t d= s->spatial_dwt_buffer[x + y*width]; | |
3921 error += d*d; | |
3922 } | |
3923 } | |
3924 | |
3925 b->qlog= (int)(log(352256.0/sqrt(error)) / log(pow(2.0, 1.0/QROOT))+0.5); | |
2164 | 3926 // av_log(NULL, AV_LOG_DEBUG, "%d %d %d\n", level, orientation, b->qlog/*, sqrt(error)*/); |
2138 | 3927 } |
3928 } | |
3929 } | |
3930 | |
3931 static int encode_init(AVCodecContext *avctx) | |
3932 { | |
3933 SnowContext *s = avctx->priv_data; | |
2198 | 3934 int plane_index; |
2138 | 3935 |
2658
d1609cfeb1d0
#defines for strict_std_compliance and split between inofficial extensions and non standarized things
michael
parents:
2635
diff
changeset
|
3936 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
|
3937 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
|
3938 "use vstrict=-2 / -strict -2 to use it anyway\n"); |
2151 | 3939 return -1; |
3940 } | |
2967 | 3941 |
2138 | 3942 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
|
3943 alloc_blocks(s); |
2967 | 3944 |
2138 | 3945 s->version=0; |
2967 | 3946 |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3947 s->m.avctx = avctx; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3948 s->m.flags = avctx->flags; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3949 s->m.bit_rate= avctx->bit_rate; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3950 |
2138 | 3951 s->m.me.scratchpad= av_mallocz((avctx->width+64)*2*16*2*sizeof(uint8_t)); |
3952 s->m.me.map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t)); | |
3953 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
|
3954 s->m.obmc_scratchpad= av_mallocz(MB_SIZE*MB_SIZE*12*sizeof(uint32_t)); |
2138 | 3955 h263_encode_init(&s->m); //mv_penalty |
3956 | |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3957 if(avctx->flags&CODEC_FLAG_PASS1){ |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3958 if(!avctx->stats_out) |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3959 avctx->stats_out = av_mallocz(256); |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3960 } |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3961 if(avctx->flags&CODEC_FLAG_PASS2){ |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3962 if(ff_rate_control_init(&s->m) < 0) |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3963 return -1; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3964 } |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3965 |
2138 | 3966 for(plane_index=0; plane_index<3; plane_index++){ |
3967 calculate_vissual_weight(s, &s->plane[plane_index]); | |
3968 } | |
2967 | 3969 |
3970 | |
2138 | 3971 avctx->coded_frame= &s->current_picture; |
3972 switch(avctx->pix_fmt){ | |
3973 // case PIX_FMT_YUV444P: | |
3974 // case PIX_FMT_YUV422P: | |
3975 case PIX_FMT_YUV420P: | |
3976 case PIX_FMT_GRAY8: | |
3977 // case PIX_FMT_YUV411P: | |
3978 // case PIX_FMT_YUV410P: | |
3979 s->colorspace_type= 0; | |
3980 break; | |
3981 /* case PIX_FMT_RGBA32: | |
3982 s->colorspace= 1; | |
3983 break;*/ | |
3984 default: | |
3985 av_log(avctx, AV_LOG_ERROR, "format not supported\n"); | |
3986 return -1; | |
3987 } | |
3988 // avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_h_shift, &s->chroma_v_shift); | |
3989 s->chroma_h_shift= 1; | |
3990 s->chroma_v_shift= 1; | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3991 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3992 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
|
3993 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
|
3994 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3995 s->avctx->get_buffer(s->avctx, &s->input_picture); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3996 |
2138 | 3997 return 0; |
3998 } | |
3999 | |
4000 static int frame_start(SnowContext *s){ | |
4001 AVFrame tmp; | |
2187 | 4002 int w= s->avctx->width; //FIXME round up to x16 ? |
4003 int h= s->avctx->height; | |
2138 | 4004 |
2187 | 4005 if(s->current_picture.data[0]){ |
4006 draw_edges(s->current_picture.data[0], s->current_picture.linesize[0], w , h , EDGE_WIDTH ); | |
4007 draw_edges(s->current_picture.data[1], s->current_picture.linesize[1], w>>1, h>>1, EDGE_WIDTH/2); | |
4008 draw_edges(s->current_picture.data[2], s->current_picture.linesize[2], w>>1, h>>1, EDGE_WIDTH/2); | |
4009 } | |
4010 | |
2138 | 4011 tmp= s->last_picture; |
4012 s->last_picture= s->current_picture; | |
4013 s->current_picture= tmp; | |
2967 | 4014 |
2138 | 4015 s->current_picture.reference= 1; |
4016 if(s->avctx->get_buffer(s->avctx, &s->current_picture) < 0){ | |
4017 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n"); | |
4018 return -1; | |
4019 } | |
2967 | 4020 |
2138 | 4021 return 0; |
4022 } | |
4023 | |
4024 static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size, void *data){ | |
4025 SnowContext *s = avctx->priv_data; | |
2335 | 4026 RangeCoder * const c= &s->c; |
2138 | 4027 AVFrame *pict = data; |
4028 const int width= s->avctx->width; | |
4029 const int height= s->avctx->height; | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4030 int level, orientation, plane_index, i, y; |
2138 | 4031 |
2335 | 4032 ff_init_range_encoder(c, buf, buf_size); |
4033 ff_build_rac_states(c, 0.05*(1LL<<32), 256-8); | |
2967 | 4034 |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4035 for(i=0; i<3; i++){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4036 int shift= !!i; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4037 for(y=0; y<(height>>shift); y++) |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4038 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
|
4039 &pict->data[i][y * pict->linesize[i]], |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4040 width>>shift); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4041 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4042 s->new_picture = *pict; |
2138 | 4043 |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4044 if(avctx->flags&CODEC_FLAG_PASS2){ |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4045 s->m.pict_type = |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4046 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
|
4047 s->keyframe= pict->pict_type==FF_I_TYPE; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4048 s->m.picture_number= avctx->frame_number; |
2975 | 4049 pict->quality= ff_rate_estimate_qscale(&s->m, 0); |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4050 }else{ |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4051 s->keyframe= avctx->gop_size==0 || avctx->frame_number % avctx->gop_size == 0; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4052 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
|
4053 } |
2967 | 4054 |
2161 | 4055 if(pict->quality){ |
4056 s->qlog= rint(QROOT*log(pict->quality / (float)FF_QP2LAMBDA)/log(2)); | |
4057 //<64 >60 | |
2600 | 4058 s->qlog += 61*QROOT/8; |
2161 | 4059 }else{ |
4060 s->qlog= LOSSLESS_QLOG; | |
4061 } | |
2138 | 4062 |
4063 frame_start(s); | |
2199 | 4064 s->current_picture.key_frame= s->keyframe; |
2138 | 4065 |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4066 s->m.current_picture_ptr= &s->m.current_picture; |
2138 | 4067 if(pict->pict_type == P_TYPE){ |
4068 int block_width = (width +15)>>4; | |
4069 int block_height= (height+15)>>4; | |
4070 int stride= s->current_picture.linesize[0]; | |
2967 | 4071 |
2138 | 4072 assert(s->current_picture.data[0]); |
4073 assert(s->last_picture.data[0]); | |
2967 | 4074 |
2138 | 4075 s->m.avctx= s->avctx; |
4076 s->m.current_picture.data[0]= s->current_picture.data[0]; | |
4077 s->m. last_picture.data[0]= s-> last_picture.data[0]; | |
4078 s->m. new_picture.data[0]= s-> input_picture.data[0]; | |
4079 s->m. last_picture_ptr= &s->m. last_picture; | |
4080 s->m.linesize= | |
4081 s->m. last_picture.linesize[0]= | |
4082 s->m. new_picture.linesize[0]= | |
4083 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
|
4084 s->m.uvlinesize= s->current_picture.linesize[1]; |
2138 | 4085 s->m.width = width; |
4086 s->m.height= height; | |
4087 s->m.mb_width = block_width; | |
4088 s->m.mb_height= block_height; | |
4089 s->m.mb_stride= s->m.mb_width+1; | |
4090 s->m.b8_stride= 2*s->m.mb_width+1; | |
4091 s->m.f_code=1; | |
4092 s->m.pict_type= pict->pict_type; | |
4093 s->m.me_method= s->avctx->me_method; | |
4094 s->m.me.scene_change_score=0; | |
4095 s->m.flags= s->avctx->flags; | |
4096 s->m.quarter_sample= (s->avctx->flags & CODEC_FLAG_QPEL)!=0; | |
4097 s->m.out_format= FMT_H263; | |
4098 s->m.unrestricted_mv= 1; | |
4099 | |
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
|
4100 s->lambda = s->m.lambda= pict->quality * 3/2; //FIXME bug somewhere else |
2138 | 4101 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
|
4102 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
|
4103 |
2138 | 4104 s->m.dsp= s->dsp; //move |
4105 ff_init_me(&s->m); | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4106 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
|
4107 } |
2967 | 4108 |
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
|
4109 redo_frame: |
2967 | 4110 |
2138 | 4111 s->qbias= pict->pict_type == P_TYPE ? 2 : 0; |
4112 | |
4113 encode_header(s); | |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4114 s->m.misc_bits = 8*(s->c.bytestream - 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
|
4115 encode_blocks(s); |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4116 s->m.mv_bits = 8*(s->c.bytestream - s->c.bytestream_start) - s->m.misc_bits; |
2967 | 4117 |
2138 | 4118 for(plane_index=0; plane_index<3; plane_index++){ |
4119 Plane *p= &s->plane[plane_index]; | |
4120 int w= p->width; | |
4121 int h= p->height; | |
4122 int x, y; | |
2198 | 4123 // int bits= put_bits_count(&s->c.pb); |
2138 | 4124 |
4125 //FIXME optimize | |
4126 if(pict->data[plane_index]) //FIXME gray hack | |
4127 for(y=0; y<h; y++){ | |
4128 for(x=0; x<w; x++){ | |
2246 | 4129 s->spatial_dwt_buffer[y*w + x]= pict->data[plane_index][y*pict->linesize[plane_index] + x]<<FRAC_BITS; |
2138 | 4130 } |
4131 } | |
4132 predict_plane(s, s->spatial_dwt_buffer, plane_index, 0); | |
2967 | 4133 |
4134 if( plane_index==0 | |
4135 && pict->pict_type == P_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
|
4136 && s->m.me.scene_change_score > s->avctx->scenechange_threshold){ |
2335 | 4137 ff_init_range_encoder(c, buf, buf_size); |
4138 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
|
4139 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
|
4140 s->keyframe=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
|
4141 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
|
4142 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
|
4143 } |
2967 | 4144 |
2161 | 4145 if(s->qlog == LOSSLESS_QLOG){ |
4146 for(y=0; y<h; y++){ | |
4147 for(x=0; x<w; x++){ | |
2595 | 4148 s->spatial_dwt_buffer[y*w + x]= (s->spatial_dwt_buffer[y*w + x] + (1<<(FRAC_BITS-1))-1)>>FRAC_BITS; |
2161 | 4149 } |
4150 } | |
4151 } | |
2967 | 4152 |
2164 | 4153 ff_spatial_dwt(s->spatial_dwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count); |
2161 | 4154 |
2138 | 4155 for(level=0; level<s->spatial_decomposition_count; level++){ |
4156 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
4157 SubBand *b= &p->band[level][orientation]; | |
2967 | 4158 |
2138 | 4159 quantize(s, b, b->buf, b->stride, s->qbias); |
4160 if(orientation==0) | |
4161 decorrelate(s, b, b->buf, b->stride, pict->pict_type == P_TYPE, 0); | |
4162 encode_subband(s, b, b->buf, b->parent ? b->parent->buf : NULL, b->stride, orientation); | |
4163 assert(b->parent==NULL || b->parent->stride == b->stride*2); | |
4164 if(orientation==0) | |
4165 correlate(s, b, b->buf, b->stride, 1, 0); | |
4166 } | |
4167 } | |
4168 // av_log(NULL, AV_LOG_DEBUG, "plane:%d bits:%d\n", plane_index, put_bits_count(&s->c.pb) - bits); | |
4169 | |
4170 for(level=0; level<s->spatial_decomposition_count; level++){ | |
4171 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
4172 SubBand *b= &p->band[level][orientation]; | |
4173 | |
4174 dequantize(s, b, b->buf, b->stride); | |
4175 } | |
4176 } | |
2161 | 4177 |
2164 | 4178 ff_spatial_idwt(s->spatial_dwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count); |
2161 | 4179 if(s->qlog == LOSSLESS_QLOG){ |
4180 for(y=0; y<h; y++){ | |
4181 for(x=0; x<w; x++){ | |
2246 | 4182 s->spatial_dwt_buffer[y*w + x]<<=FRAC_BITS; |
2161 | 4183 } |
4184 } | |
4185 } | |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
4186 {START_TIMER |
2138 | 4187 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
|
4188 STOP_TIMER("pred-conv")} |
2138 | 4189 if(s->avctx->flags&CODEC_FLAG_PSNR){ |
4190 int64_t error= 0; | |
2967 | 4191 |
2138 | 4192 if(pict->data[plane_index]) //FIXME gray hack |
4193 for(y=0; y<h; y++){ | |
4194 for(x=0; x<w; x++){ | |
2161 | 4195 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 | 4196 error += d*d; |
4197 } | |
4198 } | |
4199 s->avctx->error[plane_index] += error; | |
2232 | 4200 s->current_picture.error[plane_index] = error; |
2138 | 4201 } |
4202 } | |
4203 | |
4204 if(s->last_picture.data[0]) | |
4205 avctx->release_buffer(avctx, &s->last_picture); | |
4206 | |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4207 s->current_picture.coded_picture_number = avctx->frame_number; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4208 s->current_picture.pict_type = pict->pict_type; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4209 s->current_picture.quality = pict->quality; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4210 if(avctx->flags&CODEC_FLAG_PASS1){ |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4211 s->m.p_tex_bits = 8*(s->c.bytestream - s->c.bytestream_start) - s->m.misc_bits - s->m.mv_bits; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4212 s->m.current_picture.display_picture_number = |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4213 s->m.current_picture.coded_picture_number = avctx->frame_number; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4214 s->m.pict_type = pict->pict_type; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4215 s->m.current_picture.quality = pict->quality; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4216 ff_write_pass1_stats(&s->m); |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4217 } |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4218 if(avctx->flags&CODEC_FLAG_PASS2){ |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4219 s->m.total_bits += 8*(s->c.bytestream - s->c.bytestream_start); |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4220 } |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4221 |
2138 | 4222 emms_c(); |
2967 | 4223 |
2335 | 4224 return ff_rac_terminate(c); |
2138 | 4225 } |
4226 | |
4227 static void common_end(SnowContext *s){ | |
2192 | 4228 int plane_index, level, orientation; |
4229 | |
2138 | 4230 av_freep(&s->spatial_dwt_buffer); |
4231 | |
2967 | 4232 av_freep(&s->m.me.scratchpad); |
2138 | 4233 av_freep(&s->m.me.map); |
4234 av_freep(&s->m.me.score_map); | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
4235 av_freep(&s->m.obmc_scratchpad); |
2967 | 4236 |
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
|
4237 av_freep(&s->block); |
2192 | 4238 |
2967 | 4239 for(plane_index=0; plane_index<3; plane_index++){ |
2192 | 4240 for(level=s->spatial_decomposition_count-1; level>=0; level--){ |
4241 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
4242 SubBand *b= &s->plane[plane_index].band[level][orientation]; | |
2967 | 4243 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4244 av_freep(&b->x_coeff); |
2192 | 4245 } |
4246 } | |
4247 } | |
2138 | 4248 } |
4249 | |
4250 static int encode_end(AVCodecContext *avctx) | |
4251 { | |
4252 SnowContext *s = avctx->priv_data; | |
4253 | |
4254 common_end(s); | |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4255 av_free(avctx->stats_out); |
2138 | 4256 |
4257 return 0; | |
4258 } | |
4259 | |
4260 static int decode_init(AVCodecContext *avctx) | |
4261 { | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4262 SnowContext *s = avctx->priv_data; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4263 int block_size; |
2967 | 4264 |
2635 | 4265 avctx->pix_fmt= PIX_FMT_YUV420P; |
2138 | 4266 |
4267 common_init(avctx); | |
2967 | 4268 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4269 block_size = MB_SIZE >> s->block_max_depth; |
3017 | 4270 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 | 4271 |
2138 | 4272 return 0; |
4273 } | |
4274 | |
4275 static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, uint8_t *buf, int buf_size){ | |
4276 SnowContext *s = avctx->priv_data; | |
2335 | 4277 RangeCoder * const c= &s->c; |
2138 | 4278 int bytes_read; |
4279 AVFrame *picture = data; | |
2198 | 4280 int level, orientation, plane_index; |
2138 | 4281 |
2335 | 4282 ff_init_range_decoder(c, buf, buf_size); |
4283 ff_build_rac_states(c, 0.05*(1LL<<32), 256-8); | |
2138 | 4284 |
4285 s->current_picture.pict_type= FF_I_TYPE; //FIXME I vs. P | |
4286 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
|
4287 if(!s->block) alloc_blocks(s); |
2138 | 4288 |
4289 frame_start(s); | |
4290 //keyframe flag dupliaction mess FIXME | |
4291 if(avctx->debug&FF_DEBUG_PICT_INFO) | |
4292 av_log(avctx, AV_LOG_ERROR, "keyframe:%d qlog:%d\n", s->keyframe, s->qlog); | |
2967 | 4293 |
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
|
4294 decode_blocks(s); |
2138 | 4295 |
4296 for(plane_index=0; plane_index<3; plane_index++){ | |
4297 Plane *p= &s->plane[plane_index]; | |
4298 int w= p->width; | |
4299 int h= p->height; | |
4300 int x, y; | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4301 int decode_state[MAX_DECOMPOSITIONS][4][1]; /* Stored state info for unpack_coeffs. 1 variable per instance. */ |
2967 | 4302 |
2138 | 4303 if(s->avctx->debug&2048){ |
4304 memset(s->spatial_dwt_buffer, 0, sizeof(DWTELEM)*w*h); | |
4305 predict_plane(s, s->spatial_dwt_buffer, plane_index, 1); | |
4306 | |
4307 for(y=0; y<h; y++){ | |
4308 for(x=0; x<w; x++){ | |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
4309 int v= s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x]; |
2138 | 4310 s->mconly_picture.data[plane_index][y*s->mconly_picture.linesize[plane_index] + x]= v; |
4311 } | |
4312 } | |
4313 } | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4314 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4315 { START_TIMER |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4316 for(level=0; level<s->spatial_decomposition_count; level++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4317 for(orientation=level ? 1 : 0; orientation<4; orientation++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4318 SubBand *b= &p->band[level][orientation]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4319 unpack_coeffs(s, b, b->parent, orientation); |
2138 | 4320 } |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4321 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4322 STOP_TIMER("unpack coeffs"); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4323 } |
2138 | 4324 |
2562 | 4325 {START_TIMER |
4326 const int mb_h= s->b_height << s->block_max_depth; | |
4327 const int block_size = MB_SIZE >> s->block_max_depth; | |
4328 const int block_w = plane_index ? block_size/2 : block_size; | |
4329 int mb_y; | |
4330 dwt_compose_t cs[MAX_DECOMPOSITIONS]; | |
4331 int yd=0, yq=0; | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4332 int y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4333 int end_y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4334 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4335 ff_spatial_idwt_buffered_init(cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count); |
2562 | 4336 for(mb_y=0; mb_y<=mb_h; mb_y++){ |
2967 | 4337 |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4338 int slice_starty = block_w*mb_y; |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4339 int slice_h = block_w*(mb_y+1); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4340 if (!(s->keyframe || s->avctx->debug&512)){ |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4341 slice_starty = FFMAX(0, slice_starty - (block_w >> 1)); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4342 slice_h -= (block_w >> 1); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4343 } |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4344 |
2967 | 4345 { |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4346 START_TIMER |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4347 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
|
4348 for(orientation=level ? 1 : 0; orientation<4; orientation++){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4349 SubBand *b= &p->band[level][orientation]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4350 int start_y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4351 int end_y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4352 int our_mb_start = mb_y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4353 int our_mb_end = (mb_y + 1); |
3012 | 4354 const int extra= 3; |
4355 start_y = (mb_y ? ((block_w * our_mb_start) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra: 0); | |
4356 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
|
4357 if (!(s->keyframe || s->avctx->debug&512)){ |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4358 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
|
4359 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
|
4360 } |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4361 start_y = FFMIN(b->height, start_y); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4362 end_y = FFMIN(b->height, end_y); |
2967 | 4363 |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4364 if (start_y != end_y){ |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4365 if (orientation == 0){ |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4366 SubBand * correlate_band = &p->band[0][0]; |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4367 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
|
4368 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
|
4369 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
|
4370 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
|
4371 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
|
4372 } |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4373 else |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4374 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
|
4375 } |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4376 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4377 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4378 STOP_TIMER("decode_subband_slice"); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4379 } |
2967 | 4380 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4381 { START_TIMER |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4382 for(; yd<slice_h; yd+=4){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4383 ff_spatial_idwt_buffered_slice(cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count, yd); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4384 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4385 STOP_TIMER("idwt slice");} |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4386 |
2967 | 4387 |
2161 | 4388 if(s->qlog == LOSSLESS_QLOG){ |
2562 | 4389 for(; yq<slice_h && yq<h; yq++){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4390 DWTELEM * line = slice_buffer_get_line(&s->sb, yq); |
2161 | 4391 for(x=0; x<w; x++){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4392 line[x] <<= FRAC_BITS; |
2161 | 4393 } |
4394 } | |
4395 } | |
2562 | 4396 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4397 predict_slice_buffered(s, &s->sb, s->spatial_dwt_buffer, plane_index, 1, mb_y); |
2967 | 4398 |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4399 y = FFMIN(p->height, slice_starty); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4400 end_y = FFMIN(p->height, slice_h); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4401 while(y < end_y) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4402 slice_buffer_release(&s->sb, y++); |
2562 | 4403 } |
2967 | 4404 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4405 slice_buffer_flush(&s->sb); |
2967 | 4406 |
2562 | 4407 STOP_TIMER("idwt + predict_slices")} |
2138 | 4408 } |
2967 | 4409 |
2138 | 4410 emms_c(); |
4411 | |
4412 if(s->last_picture.data[0]) | |
4413 avctx->release_buffer(avctx, &s->last_picture); | |
4414 | |
2967 | 4415 if(!(s->avctx->debug&2048)) |
2138 | 4416 *picture= s->current_picture; |
4417 else | |
4418 *picture= s->mconly_picture; | |
2967 | 4419 |
2138 | 4420 *data_size = sizeof(AVFrame); |
2967 | 4421 |
2335 | 4422 bytes_read= c->bytestream - c->bytestream_start; |
4423 if(bytes_read ==0) av_log(s->avctx, AV_LOG_ERROR, "error at end of frame\n"); //FIXME | |
2138 | 4424 |
4425 return bytes_read; | |
4426 } | |
4427 | |
4428 static int decode_end(AVCodecContext *avctx) | |
4429 { | |
4430 SnowContext *s = avctx->priv_data; | |
4431 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4432 slice_buffer_destroy(&s->sb); |
2967 | 4433 |
2138 | 4434 common_end(s); |
4435 | |
4436 return 0; | |
4437 } | |
4438 | |
4439 AVCodec snow_decoder = { | |
4440 "snow", | |
4441 CODEC_TYPE_VIDEO, | |
4442 CODEC_ID_SNOW, | |
4443 sizeof(SnowContext), | |
4444 decode_init, | |
4445 NULL, | |
4446 decode_end, | |
4447 decode_frame, | |
4448 0 /*CODEC_CAP_DR1*/ /*| CODEC_CAP_DRAW_HORIZ_BAND*/, | |
4449 NULL | |
4450 }; | |
4451 | |
2408
a6e4da1c28ee
Disable encoders patch by (Gianluigi Tiesi <mplayer netfarm it>)
michael
parents:
2368
diff
changeset
|
4452 #ifdef CONFIG_ENCODERS |
2138 | 4453 AVCodec snow_encoder = { |
4454 "snow", | |
4455 CODEC_TYPE_VIDEO, | |
4456 CODEC_ID_SNOW, | |
4457 sizeof(SnowContext), | |
4458 encode_init, | |
4459 encode_frame, | |
4460 encode_end, | |
4461 }; | |
2408
a6e4da1c28ee
Disable encoders patch by (Gianluigi Tiesi <mplayer netfarm it>)
michael
parents:
2368
diff
changeset
|
4462 #endif |
2138 | 4463 |
4464 | |
4465 #if 0 | |
4466 #undef malloc | |
4467 #undef free | |
4468 #undef printf | |
4469 | |
4470 int main(){ | |
4471 int width=256; | |
4472 int height=256; | |
4473 int buffer[2][width*height]; | |
4474 SnowContext s; | |
4475 int i; | |
4476 s.spatial_decomposition_count=6; | |
4477 s.spatial_decomposition_type=1; | |
2967 | 4478 |
2138 | 4479 printf("testing 5/3 DWT\n"); |
4480 for(i=0; i<width*height; i++) | |
4481 buffer[0][i]= buffer[1][i]= random()%54321 - 12345; | |
2967 | 4482 |
2951 | 4483 ff_spatial_dwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); |
4484 ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); | |
2967 | 4485 |
2138 | 4486 for(i=0; i<width*height; i++) |
4487 if(buffer[0][i]!= buffer[1][i]) printf("fsck: %d %d %d\n",i, buffer[0][i], buffer[1][i]); | |
4488 | |
4489 printf("testing 9/7 DWT\n"); | |
4490 s.spatial_decomposition_type=0; | |
4491 for(i=0; i<width*height; i++) | |
4492 buffer[0][i]= buffer[1][i]= random()%54321 - 12345; | |
2967 | 4493 |
2951 | 4494 ff_spatial_dwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); |
4495 ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); | |
2967 | 4496 |
2138 | 4497 for(i=0; i<width*height; i++) |
2951 | 4498 if(ABS(buffer[0][i] - buffer[1][i])>20) printf("fsck: %d %d %d\n",i, buffer[0][i], buffer[1][i]); |
2967 | 4499 |
2951 | 4500 #if 0 |
2138 | 4501 printf("testing AC coder\n"); |
4502 memset(s.header_state, 0, sizeof(s.header_state)); | |
2335 | 4503 ff_init_range_encoder(&s.c, buffer[0], 256*256); |
2138 | 4504 ff_init_cabac_states(&s.c, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64); |
2967 | 4505 |
2138 | 4506 for(i=-256; i<256; i++){ |
4507 START_TIMER | |
4508 put_symbol(&s.c, s.header_state, i*i*i/3*ABS(i), 1); | |
4509 STOP_TIMER("put_symbol") | |
4510 } | |
2335 | 4511 ff_rac_terminate(&s.c); |
2138 | 4512 |
4513 memset(s.header_state, 0, sizeof(s.header_state)); | |
2335 | 4514 ff_init_range_decoder(&s.c, buffer[0], 256*256); |
2138 | 4515 ff_init_cabac_states(&s.c, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64); |
2967 | 4516 |
2138 | 4517 for(i=-256; i<256; i++){ |
4518 int j; | |
4519 START_TIMER | |
4520 j= get_symbol(&s.c, s.header_state, 1); | |
4521 STOP_TIMER("get_symbol") | |
4522 if(j!=i*i*i/3*ABS(i)) printf("fsck: %d != %d\n", i, j); | |
4523 } | |
2951 | 4524 #endif |
2138 | 4525 { |
4526 int level, orientation, x, y; | |
4527 int64_t errors[8][4]; | |
4528 int64_t g=0; | |
4529 | |
4530 memset(errors, 0, sizeof(errors)); | |
4531 s.spatial_decomposition_count=3; | |
4532 s.spatial_decomposition_type=0; | |
4533 for(level=0; level<s.spatial_decomposition_count; level++){ | |
4534 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
4535 int w= width >> (s.spatial_decomposition_count-level); | |
4536 int h= height >> (s.spatial_decomposition_count-level); | |
4537 int stride= width << (s.spatial_decomposition_count-level); | |
4538 DWTELEM *buf= buffer[0]; | |
4539 int64_t error=0; | |
4540 | |
4541 if(orientation&1) buf+=w; | |
4542 if(orientation>1) buf+=stride>>1; | |
2967 | 4543 |
2138 | 4544 memset(buffer[0], 0, sizeof(int)*width*height); |
4545 buf[w/2 + h/2*stride]= 256*256; | |
2951 | 4546 ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); |
2138 | 4547 for(y=0; y<height; y++){ |
4548 for(x=0; x<width; x++){ | |
4549 int64_t d= buffer[0][x + y*width]; | |
4550 error += d*d; | |
4551 if(ABS(width/2-x)<9 && ABS(height/2-y)<9 && level==2) printf("%8lld ", d); | |
4552 } | |
4553 if(ABS(height/2-y)<9 && level==2) printf("\n"); | |
4554 } | |
4555 error= (int)(sqrt(error)+0.5); | |
4556 errors[level][orientation]= error; | |
4557 if(g) g=ff_gcd(g, error); | |
4558 else g= error; | |
4559 } | |
4560 } | |
4561 printf("static int const visual_weight[][4]={\n"); | |
4562 for(level=0; level<s.spatial_decomposition_count; level++){ | |
4563 printf(" {"); | |
4564 for(orientation=0; orientation<4; orientation++){ | |
4565 printf("%8lld,", errors[level][orientation]/g); | |
4566 } | |
4567 printf("},\n"); | |
4568 } | |
4569 printf("};\n"); | |
4570 { | |
4571 int level=2; | |
4572 int orientation=3; | |
4573 int w= width >> (s.spatial_decomposition_count-level); | |
4574 int h= height >> (s.spatial_decomposition_count-level); | |
4575 int stride= width << (s.spatial_decomposition_count-level); | |
4576 DWTELEM *buf= buffer[0]; | |
4577 int64_t error=0; | |
4578 | |
4579 buf+=w; | |
4580 buf+=stride>>1; | |
2967 | 4581 |
2138 | 4582 memset(buffer[0], 0, sizeof(int)*width*height); |
4583 #if 1 | |
4584 for(y=0; y<height; y++){ | |
4585 for(x=0; x<width; x++){ | |
4586 int tab[4]={0,2,3,1}; | |
4587 buffer[0][x+width*y]= 256*256*tab[(x&1) + 2*(y&1)]; | |
4588 } | |
4589 } | |
2951 | 4590 ff_spatial_dwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); |
2138 | 4591 #else |
4592 for(y=0; y<h; y++){ | |
4593 for(x=0; x<w; x++){ | |
4594 buf[x + y*stride ]=169; | |
4595 buf[x + y*stride-w]=64; | |
4596 } | |
4597 } | |
2951 | 4598 ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); |
2138 | 4599 #endif |
4600 for(y=0; y<height; y++){ | |
4601 for(x=0; x<width; x++){ | |
4602 int64_t d= buffer[0][x + y*width]; | |
4603 error += d*d; | |
4604 if(ABS(width/2-x)<9 && ABS(height/2-y)<9) printf("%8lld ", d); | |
4605 } | |
4606 if(ABS(height/2-y)<9) printf("\n"); | |
4607 } | |
4608 } | |
4609 | |
4610 } | |
4611 return 0; | |
4612 } | |
4613 #endif | |
4614 |