2 * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 #include "rangecoder.h"
27 #include "mpegvideo.h"
32 static const int8_t quant3[256]={
33 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
34 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
35 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
36 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
37 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
38 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
39 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
40 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
41 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,
42 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,
43 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,
44 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,
45 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,
46 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,
47 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,
48 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, 0,
50 static const int8_t quant3b[256]={
51 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
52 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
53 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
54 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
55 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
56 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
57 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
58 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
59 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,
60 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,
61 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,
62 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,
63 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,
64 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,
65 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,
66 -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,
68 static const int8_t quant3bA[256]={
69 0, 0, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1,
70 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1,
71 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1,
72 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1,
73 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1,
74 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1,
75 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1,
76 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1,
77 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1,
78 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1,
79 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1,
80 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1,
81 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1,
82 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1,
83 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1,
84 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1,
86 static const int8_t quant5[256]={
87 0, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
88 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
89 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
90 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
91 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
92 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
93 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
94 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
95 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,
96 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,
97 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,
98 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,
99 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,
100 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,
101 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,
102 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-1,-1,-1,
104 static const int8_t quant7[256]={
105 0, 1, 1, 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, 3, 3, 3, 3, 3, 3, 3, 3,
108 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
109 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
110 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
111 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
112 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
113 -3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,
114 -3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,
115 -3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,
116 -3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,
117 -3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-3,
118 -3,-3,-3,-3,-3,-3,-3,-3,-3,-2,-2,-2,-2,-2,-2,-2,
119 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,
120 -2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-2,-1,-1,
122 static const int8_t quant9[256]={
123 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3,
124 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
125 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
126 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
127 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
128 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
129 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
130 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
131 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,
132 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,
133 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,
134 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,
135 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,
136 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,
137 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-3,-3,-3,-3,
138 -3,-3,-3,-3,-3,-3,-3,-3,-3,-3,-2,-2,-2,-2,-1,-1,
140 static const int8_t quant11[256]={
141 0, 1, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 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, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
144 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
145 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
146 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
147 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
148 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
149 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,
150 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,
151 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,
152 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,
153 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,
154 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-4,-4,
155 -4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,-4,
156 -4,-4,-4,-4,-4,-3,-3,-3,-3,-3,-3,-3,-2,-2,-2,-1,
158 static const int8_t quant13[256]={
159 0, 1, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4,
160 4, 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, 5, 5,
162 5, 5, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
163 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
164 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
165 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
166 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
167 -6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,
168 -6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,
169 -6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,
170 -6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,
171 -6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-6,-5,
172 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,
173 -5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,-5,
174 -4,-4,-4,-4,-4,-4,-4,-4,-4,-3,-3,-3,-3,-2,-2,-1,
178 static const uint8_t obmc32[1024]={
179 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,
180 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 4, 4, 4, 4, 4, 4, 4, 4, 0, 0, 0, 0, 0,
181 0, 0, 0, 4, 4, 4, 4, 8, 8, 12, 12, 12, 16, 16, 16, 16, 16, 16, 16, 16, 12, 12, 12, 8, 8, 4, 4, 4, 4, 0, 0, 0,
182 0, 0, 4, 4, 8, 8, 12, 16, 16, 20, 24, 24, 28, 28, 32, 32, 32, 32, 28, 28, 24, 24, 20, 16, 16, 12, 8, 8, 4, 4, 0, 0,
183 0, 0, 4, 8, 8, 12, 16, 24, 28, 32, 36, 40, 44, 48, 48, 48, 48, 48, 48, 44, 40, 36, 32, 28, 24, 16, 12, 8, 8, 4, 0, 0,
184 0, 4, 4, 8, 12, 20, 24, 32, 40, 44, 52, 56, 60, 64, 68, 72, 72, 68, 64, 60, 56, 52, 44, 40, 32, 24, 20, 12, 8, 4, 4, 0,
185 0, 4, 4, 12, 16, 24, 32, 40, 52, 60, 68, 76, 80, 88, 88, 92, 92, 88, 88, 80, 76, 68, 60, 52, 40, 32, 24, 16, 12, 4, 4, 0,
186 0, 4, 8, 16, 24, 32, 40, 52, 64, 76, 84, 92,100,108,112,116,116,112,108,100, 92, 84, 76, 64, 52, 40, 32, 24, 16, 8, 4, 0,
187 0, 4, 8, 16, 28, 40, 52, 64, 76, 88,100,112,124,132,136,140,140,136,132,124,112,100, 88, 76, 64, 52, 40, 28, 16, 8, 4, 0,
188 0, 4, 12, 20, 32, 44, 60, 76, 88,104,120,132,144,152,160,164,164,160,152,144,132,120,104, 88, 76, 60, 44, 32, 20, 12, 4, 0,
189 0, 4, 12, 24, 36, 48, 68, 84,100,120,136,152,164,176,180,184,184,180,176,164,152,136,120,100, 84, 68, 48, 36, 24, 12, 4, 0,
190 0, 4, 12, 24, 40, 56, 76, 92,112,132,152,168,180,192,204,208,208,204,192,180,168,152,132,112, 92, 76, 56, 40, 24, 12, 4, 0,
191 0, 4, 16, 28, 44, 60, 80,100,124,144,164,180,196,208,220,224,224,220,208,196,180,164,144,124,100, 80, 60, 44, 28, 16, 4, 0,
192 0, 8, 16, 28, 48, 64, 88,108,132,152,176,192,208,224,232,240,240,232,224,208,192,176,152,132,108, 88, 64, 48, 28, 16, 8, 0,
193 0, 4, 16, 32, 48, 68, 88,112,136,160,180,204,220,232,244,248,248,244,232,220,204,180,160,136,112, 88, 68, 48, 32, 16, 4, 0,
194 1, 8, 16, 32, 48, 72, 92,116,140,164,184,208,224,240,248,255,255,248,240,224,208,184,164,140,116, 92, 72, 48, 32, 16, 8, 1,
195 1, 8, 16, 32, 48, 72, 92,116,140,164,184,208,224,240,248,255,255,248,240,224,208,184,164,140,116, 92, 72, 48, 32, 16, 8, 1,
196 0, 4, 16, 32, 48, 68, 88,112,136,160,180,204,220,232,244,248,248,244,232,220,204,180,160,136,112, 88, 68, 48, 32, 16, 4, 0,
197 0, 8, 16, 28, 48, 64, 88,108,132,152,176,192,208,224,232,240,240,232,224,208,192,176,152,132,108, 88, 64, 48, 28, 16, 8, 0,
198 0, 4, 16, 28, 44, 60, 80,100,124,144,164,180,196,208,220,224,224,220,208,196,180,164,144,124,100, 80, 60, 44, 28, 16, 4, 0,
199 0, 4, 12, 24, 40, 56, 76, 92,112,132,152,168,180,192,204,208,208,204,192,180,168,152,132,112, 92, 76, 56, 40, 24, 12, 4, 0,
200 0, 4, 12, 24, 36, 48, 68, 84,100,120,136,152,164,176,180,184,184,180,176,164,152,136,120,100, 84, 68, 48, 36, 24, 12, 4, 0,
201 0, 4, 12, 20, 32, 44, 60, 76, 88,104,120,132,144,152,160,164,164,160,152,144,132,120,104, 88, 76, 60, 44, 32, 20, 12, 4, 0,
202 0, 4, 8, 16, 28, 40, 52, 64, 76, 88,100,112,124,132,136,140,140,136,132,124,112,100, 88, 76, 64, 52, 40, 28, 16, 8, 4, 0,
203 0, 4, 8, 16, 24, 32, 40, 52, 64, 76, 84, 92,100,108,112,116,116,112,108,100, 92, 84, 76, 64, 52, 40, 32, 24, 16, 8, 4, 0,
204 0, 4, 4, 12, 16, 24, 32, 40, 52, 60, 68, 76, 80, 88, 88, 92, 92, 88, 88, 80, 76, 68, 60, 52, 40, 32, 24, 16, 12, 4, 4, 0,
205 0, 4, 4, 8, 12, 20, 24, 32, 40, 44, 52, 56, 60, 64, 68, 72, 72, 68, 64, 60, 56, 52, 44, 40, 32, 24, 20, 12, 8, 4, 4, 0,
206 0, 0, 4, 8, 8, 12, 16, 24, 28, 32, 36, 40, 44, 48, 48, 48, 48, 48, 48, 44, 40, 36, 32, 28, 24, 16, 12, 8, 8, 4, 0, 0,
207 0, 0, 4, 4, 8, 8, 12, 16, 16, 20, 24, 24, 28, 28, 32, 32, 32, 32, 28, 28, 24, 24, 20, 16, 16, 12, 8, 8, 4, 4, 0, 0,
208 0, 0, 0, 4, 4, 4, 4, 8, 8, 12, 12, 12, 16, 16, 16, 16, 16, 16, 16, 16, 12, 12, 12, 8, 8, 4, 4, 4, 4, 0, 0, 0,
209 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 4, 4, 4, 4, 4, 4, 4, 4, 0, 0, 0, 0, 0,
210 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,
213 static const uint8_t obmc16[256]={
214 0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 0, 0, 0, 0, 0, 0,
215 0, 4, 4, 8, 16, 20, 20, 24, 24, 20, 20, 16, 8, 4, 4, 0,
216 0, 4, 16, 24, 36, 44, 52, 60, 60, 52, 44, 36, 24, 16, 4, 0,
217 0, 8, 24, 44, 60, 80, 96,104,104, 96, 80, 60, 44, 24, 8, 0,
218 0, 16, 36, 60, 92,116,136,152,152,136,116, 92, 60, 36, 16, 0,
219 0, 20, 44, 80,116,152,180,196,196,180,152,116, 80, 44, 20, 0,
220 4, 20, 52, 96,136,180,212,228,228,212,180,136, 96, 52, 20, 4,
221 4, 24, 60,104,152,196,228,248,248,228,196,152,104, 60, 24, 4,
222 4, 24, 60,104,152,196,228,248,248,228,196,152,104, 60, 24, 4,
223 4, 20, 52, 96,136,180,212,228,228,212,180,136, 96, 52, 20, 4,
224 0, 20, 44, 80,116,152,180,196,196,180,152,116, 80, 44, 20, 0,
225 0, 16, 36, 60, 92,116,136,152,152,136,116, 92, 60, 36, 16, 0,
226 0, 8, 24, 44, 60, 80, 96,104,104, 96, 80, 60, 44, 24, 8, 0,
227 0, 4, 16, 24, 36, 44, 52, 60, 60, 52, 44, 36, 24, 16, 4, 0,
228 0, 4, 4, 8, 16, 20, 20, 24, 24, 20, 20, 16, 8, 4, 4, 0,
229 0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 0, 0, 0, 0, 0, 0,
233 static const uint8_t obmc32[1024]={
234 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8, 8, 4, 4, 4, 4, 4, 4, 4, 4, 0, 0, 0, 0,
235 0, 4, 4, 4, 8, 8, 8, 12, 12, 16, 16, 16, 20, 20, 20, 24, 24, 20, 20, 20, 16, 16, 16, 12, 12, 8, 8, 8, 4, 4, 4, 0,
236 0, 4, 8, 8, 12, 12, 16, 20, 20, 24, 28, 28, 32, 32, 36, 40, 40, 36, 32, 32, 28, 28, 24, 20, 20, 16, 12, 12, 8, 8, 4, 0,
237 0, 4, 8, 12, 16, 20, 24, 28, 28, 32, 36, 40, 44, 48, 52, 56, 56, 52, 48, 44, 40, 36, 32, 28, 28, 24, 20, 16, 12, 8, 4, 0,
238 4, 8, 12, 16, 20, 24, 28, 32, 40, 44, 48, 52, 56, 60, 64, 68, 68, 64, 60, 56, 52, 48, 44, 40, 32, 28, 24, 20, 16, 12, 8, 4,
239 4, 8, 12, 20, 24, 32, 36, 40, 48, 52, 56, 64, 68, 76, 80, 84, 84, 80, 76, 68, 64, 56, 52, 48, 40, 36, 32, 24, 20, 12, 8, 4,
240 4, 8, 16, 24, 28, 36, 44, 48, 56, 60, 68, 76, 80, 88, 96,100,100, 96, 88, 80, 76, 68, 60, 56, 48, 44, 36, 28, 24, 16, 8, 4,
241 4, 12, 20, 28, 32, 40, 48, 56, 64, 72, 80, 88, 92,100,108,116,116,108,100, 92, 88, 80, 72, 64, 56, 48, 40, 32, 28, 20, 12, 4,
242 4, 12, 20, 28, 40, 48, 56, 64, 72, 80, 88, 96,108,116,124,132,132,124,116,108, 96, 88, 80, 72, 64, 56, 48, 40, 28, 20, 12, 4,
243 4, 16, 24, 32, 44, 52, 60, 72, 80, 92,100,108,120,128,136,148,148,136,128,120,108,100, 92, 80, 72, 60, 52, 44, 32, 24, 16, 4,
244 4, 16, 28, 36, 48, 56, 68, 80, 88,100,112,120,132,140,152,164,164,152,140,132,120,112,100, 88, 80, 68, 56, 48, 36, 28, 16, 4,
245 4, 16, 28, 40, 52, 64, 76, 88, 96,108,120,132,144,156,168,180,180,168,156,144,132,120,108, 96, 88, 76, 64, 52, 40, 28, 16, 4,
246 8, 20, 32, 44, 56, 68, 80, 92,108,120,132,144,156,168,180,192,192,180,168,156,144,132,120,108, 92, 80, 68, 56, 44, 32, 20, 8,
247 8, 20, 32, 48, 60, 76, 88,100,116,128,140,156,168,184,196,208,208,196,184,168,156,140,128,116,100, 88, 76, 60, 48, 32, 20, 8,
248 8, 20, 36, 52, 64, 80, 96,108,124,136,152,168,180,196,212,224,224,212,196,180,168,152,136,124,108, 96, 80, 64, 52, 36, 20, 8,
249 8, 24, 40, 56, 68, 84,100,116,132,148,164,180,192,208,224,240,240,224,208,192,180,164,148,132,116,100, 84, 68, 56, 40, 24, 8,
250 8, 24, 40, 56, 68, 84,100,116,132,148,164,180,192,208,224,240,240,224,208,192,180,164,148,132,116,100, 84, 68, 56, 40, 24, 8,
251 8, 20, 36, 52, 64, 80, 96,108,124,136,152,168,180,196,212,224,224,212,196,180,168,152,136,124,108, 96, 80, 64, 52, 36, 20, 8,
252 8, 20, 32, 48, 60, 76, 88,100,116,128,140,156,168,184,196,208,208,196,184,168,156,140,128,116,100, 88, 76, 60, 48, 32, 20, 8,
253 8, 20, 32, 44, 56, 68, 80, 92,108,120,132,144,156,168,180,192,192,180,168,156,144,132,120,108, 92, 80, 68, 56, 44, 32, 20, 8,
254 4, 16, 28, 40, 52, 64, 76, 88, 96,108,120,132,144,156,168,180,180,168,156,144,132,120,108, 96, 88, 76, 64, 52, 40, 28, 16, 4,
255 4, 16, 28, 36, 48, 56, 68, 80, 88,100,112,120,132,140,152,164,164,152,140,132,120,112,100, 88, 80, 68, 56, 48, 36, 28, 16, 4,
256 4, 16, 24, 32, 44, 52, 60, 72, 80, 92,100,108,120,128,136,148,148,136,128,120,108,100, 92, 80, 72, 60, 52, 44, 32, 24, 16, 4,
257 4, 12, 20, 28, 40, 48, 56, 64, 72, 80, 88, 96,108,116,124,132,132,124,116,108, 96, 88, 80, 72, 64, 56, 48, 40, 28, 20, 12, 4,
258 4, 12, 20, 28, 32, 40, 48, 56, 64, 72, 80, 88, 92,100,108,116,116,108,100, 92, 88, 80, 72, 64, 56, 48, 40, 32, 28, 20, 12, 4,
259 4, 8, 16, 24, 28, 36, 44, 48, 56, 60, 68, 76, 80, 88, 96,100,100, 96, 88, 80, 76, 68, 60, 56, 48, 44, 36, 28, 24, 16, 8, 4,
260 4, 8, 12, 20, 24, 32, 36, 40, 48, 52, 56, 64, 68, 76, 80, 84, 84, 80, 76, 68, 64, 56, 52, 48, 40, 36, 32, 24, 20, 12, 8, 4,
261 4, 8, 12, 16, 20, 24, 28, 32, 40, 44, 48, 52, 56, 60, 64, 68, 68, 64, 60, 56, 52, 48, 44, 40, 32, 28, 24, 20, 16, 12, 8, 4,
262 0, 4, 8, 12, 16, 20, 24, 28, 28, 32, 36, 40, 44, 48, 52, 56, 56, 52, 48, 44, 40, 36, 32, 28, 28, 24, 20, 16, 12, 8, 4, 0,
263 0, 4, 8, 8, 12, 12, 16, 20, 20, 24, 28, 28, 32, 32, 36, 40, 40, 36, 32, 32, 28, 28, 24, 20, 20, 16, 12, 12, 8, 8, 4, 0,
264 0, 4, 4, 4, 8, 8, 8, 12, 12, 16, 16, 16, 20, 20, 20, 24, 24, 20, 20, 20, 16, 16, 16, 12, 12, 8, 8, 8, 4, 4, 4, 0,
265 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8, 8, 4, 4, 4, 4, 4, 4, 4, 4, 0, 0, 0, 0,
268 static const uint8_t obmc16[256]={
269 0, 4, 4, 8, 8, 12, 12, 16, 16, 12, 12, 8, 8, 4, 4, 0,
270 4, 8, 16, 20, 28, 32, 40, 44, 44, 40, 32, 28, 20, 16, 8, 4,
271 4, 16, 24, 36, 44, 56, 64, 76, 76, 64, 56, 44, 36, 24, 16, 4,
272 8, 20, 36, 48, 64, 76, 92,104,104, 92, 76, 64, 48, 36, 20, 8,
273 8, 28, 44, 64, 80,100,116,136,136,116,100, 80, 64, 44, 28, 8,
274 12, 32, 56, 76,100,120,144,164,164,144,120,100, 76, 56, 32, 12,
275 12, 40, 64, 92,116,144,168,196,196,168,144,116, 92, 64, 40, 12,
276 16, 44, 76,104,136,164,196,224,224,196,164,136,104, 76, 44, 16,
277 16, 44, 76,104,136,164,196,224,224,196,164,136,104, 76, 44, 16,
278 12, 40, 64, 92,116,144,168,196,196,168,144,116, 92, 64, 40, 12,
279 12, 32, 56, 76,100,120,144,164,164,144,120,100, 76, 56, 32, 12,
280 8, 28, 44, 64, 80,100,116,136,136,116,100, 80, 64, 44, 28, 8,
281 8, 20, 36, 48, 64, 76, 92,104,104, 92, 76, 64, 48, 36, 20, 8,
282 4, 16, 24, 36, 44, 56, 64, 76, 76, 64, 56, 44, 36, 24, 16, 4,
283 4, 8, 16, 20, 28, 32, 40, 44, 44, 40, 32, 28, 20, 16, 8, 4,
284 0, 4, 4, 8, 8, 12, 12, 16, 16, 12, 12, 8, 8, 4, 4, 0,
288 static const uint8_t obmc32[1024]={
289 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,
290 0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 8, 4, 4, 8, 4, 4, 4, 4, 4, 4, 4, 4, 0, 0, 0, 0, 0, 0,
291 0, 0, 0, 4, 4, 4, 4, 8, 8, 12, 12, 12, 12, 16, 16, 16, 16, 16, 16, 12, 12, 12, 12, 8, 8, 4, 4, 4, 4, 0, 0, 0,
292 0, 0, 4, 4, 4, 8, 8, 12, 16, 20, 20, 24, 28, 28, 28, 28, 28, 28, 28, 28, 24, 20, 20, 16, 12, 8, 8, 4, 4, 4, 0, 0,
293 0, 0, 4, 4, 8, 12, 16, 20, 24, 28, 36, 40, 44, 44, 48, 48, 48, 48, 44, 44, 40, 36, 28, 24, 20, 16, 12, 8, 4, 4, 0, 0,
294 0, 0, 4, 8, 12, 20, 24, 32, 36, 44, 48, 56, 60, 64, 68, 68, 68, 68, 64, 60, 56, 48, 44, 36, 32, 24, 20, 12, 8, 4, 0, 0,
295 0, 4, 4, 8, 16, 24, 32, 40, 48, 60, 68, 76, 80, 84, 88, 92, 92, 88, 84, 80, 76, 68, 60, 48, 40, 32, 24, 16, 8, 4, 4, 0,
296 0, 4, 8, 12, 20, 32, 40, 52, 64, 76, 84, 96,104,108,112,116,116,112,108,104, 96, 84, 76, 64, 52, 40, 32, 20, 12, 8, 4, 0,
297 0, 4, 8, 16, 24, 36, 48, 64, 76, 92,104,116,124,132,136,140,140,136,132,124,116,104, 92, 76, 64, 48, 36, 24, 16, 8, 4, 0,
298 0, 4, 12, 20, 28, 44, 60, 76, 92,104,120,136,148,156,160,164,164,160,156,148,136,120,104, 92, 76, 60, 44, 28, 20, 12, 4, 0,
299 0, 4, 12, 20, 36, 48, 68, 84,104,120,140,152,168,176,184,188,188,184,176,168,152,140,120,104, 84, 68, 48, 36, 20, 12, 4, 0,
300 0, 4, 12, 24, 36, 56, 76, 96,116,136,152,172,184,196,204,208,208,204,196,184,172,152,136,116, 96, 76, 56, 36, 24, 12, 4, 0,
301 0, 4, 12, 24, 44, 60, 80,104,124,148,168,184,200,212,224,228,228,224,212,200,184,168,148,124,104, 80, 60, 44, 24, 12, 4, 0,
302 0, 4, 12, 28, 44, 64, 84,108,132,156,176,196,212,228,236,240,240,236,228,212,196,176,156,132,108, 84, 64, 44, 28, 12, 4, 0,
303 0, 4, 16, 28, 48, 68, 88,112,136,160,184,204,224,236,244,252,252,244,236,224,204,184,160,136,112, 88, 68, 48, 28, 16, 4, 0,
304 1, 4, 16, 28, 48, 68, 92,116,140,164,188,208,228,240,252,255,255,252,240,228,208,188,164,140,116, 92, 68, 48, 28, 16, 4, 1,
305 1, 4, 16, 28, 48, 68, 92,116,140,164,188,208,228,240,252,255,255,252,240,228,208,188,164,140,116, 92, 68, 48, 28, 16, 4, 1,
306 0, 4, 16, 28, 48, 68, 88,112,136,160,184,204,224,236,244,252,252,244,236,224,204,184,160,136,112, 88, 68, 48, 28, 16, 4, 0,
307 0, 4, 12, 28, 44, 64, 84,108,132,156,176,196,212,228,236,240,240,236,228,212,196,176,156,132,108, 84, 64, 44, 28, 12, 4, 0,
308 0, 4, 12, 24, 44, 60, 80,104,124,148,168,184,200,212,224,228,228,224,212,200,184,168,148,124,104, 80, 60, 44, 24, 12, 4, 0,
309 0, 4, 12, 24, 36, 56, 76, 96,116,136,152,172,184,196,204,208,208,204,196,184,172,152,136,116, 96, 76, 56, 36, 24, 12, 4, 0,
310 0, 4, 12, 20, 36, 48, 68, 84,104,120,140,152,168,176,184,188,188,184,176,168,152,140,120,104, 84, 68, 48, 36, 20, 12, 4, 0,
311 0, 4, 12, 20, 28, 44, 60, 76, 92,104,120,136,148,156,160,164,164,160,156,148,136,120,104, 92, 76, 60, 44, 28, 20, 12, 4, 0,
312 0, 4, 8, 16, 24, 36, 48, 64, 76, 92,104,116,124,132,136,140,140,136,132,124,116,104, 92, 76, 64, 48, 36, 24, 16, 8, 4, 0,
313 0, 4, 8, 12, 20, 32, 40, 52, 64, 76, 84, 96,104,108,112,116,116,112,108,104, 96, 84, 76, 64, 52, 40, 32, 20, 12, 8, 4, 0,
314 0, 4, 4, 8, 16, 24, 32, 40, 48, 60, 68, 76, 80, 84, 88, 92, 92, 88, 84, 80, 76, 68, 60, 48, 40, 32, 24, 16, 8, 4, 4, 0,
315 0, 0, 4, 8, 12, 20, 24, 32, 36, 44, 48, 56, 60, 64, 68, 68, 68, 68, 64, 60, 56, 48, 44, 36, 32, 24, 20, 12, 8, 4, 0, 0,
316 0, 0, 4, 4, 8, 12, 16, 20, 24, 28, 36, 40, 44, 44, 48, 48, 48, 48, 44, 44, 40, 36, 28, 24, 20, 16, 12, 8, 4, 4, 0, 0,
317 0, 0, 4, 4, 4, 8, 8, 12, 16, 20, 20, 24, 28, 28, 28, 28, 28, 28, 28, 28, 24, 20, 20, 16, 12, 8, 8, 4, 4, 4, 0, 0,
318 0, 0, 0, 4, 4, 4, 4, 8, 8, 12, 12, 12, 12, 16, 16, 16, 16, 16, 16, 12, 12, 12, 12, 8, 8, 4, 4, 4, 4, 0, 0, 0,
319 0, 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 8, 4, 4, 8, 4, 4, 4, 4, 4, 4, 4, 4, 0, 0, 0, 0, 0, 0,
320 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,
323 static const uint8_t obmc16[256]={
324 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 0, 0, 0, 0, 0,
325 0, 0, 4, 8, 12, 16, 20, 20, 20, 20, 16, 12, 8, 4, 0, 0,
326 0, 4, 12, 24, 32, 44, 52, 56, 56, 52, 44, 32, 24, 12, 4, 0,
327 0, 8, 24, 40, 60, 80, 96,104,104, 96, 80, 60, 40, 24, 8, 0,
328 0, 12, 32, 64, 92,120,140,152,152,140,120, 92, 64, 32, 12, 0,
329 4, 16, 44, 80,120,156,184,196,196,184,156,120, 80, 44, 16, 4,
330 4, 20, 52, 96,140,184,216,232,232,216,184,140, 96, 52, 20, 4,
331 0, 20, 56,104,152,196,232,252,252,232,196,152,104, 56, 20, 0,
332 0, 20, 56,104,152,196,232,252,252,232,196,152,104, 56, 20, 0,
333 4, 20, 52, 96,140,184,216,232,232,216,184,140, 96, 52, 20, 4,
334 4, 16, 44, 80,120,156,184,196,196,184,156,120, 80, 44, 16, 4,
335 0, 12, 32, 64, 92,120,140,152,152,140,120, 92, 64, 32, 12, 0,
336 0, 8, 24, 40, 60, 80, 96,104,104, 96, 80, 60, 40, 24, 8, 0,
337 0, 4, 12, 24, 32, 44, 52, 56, 56, 52, 44, 32, 24, 12, 4, 0,
338 0, 0, 4, 8, 12, 16, 20, 20, 20, 20, 16, 12, 8, 4, 0, 0,
339 0, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 0, 0, 0, 0, 0,
345 static const uint8_t obmc8[64]={
346 4, 12, 20, 28, 28, 20, 12, 4,
347 12, 36, 60, 84, 84, 60, 36, 12,
348 20, 60,100,140,140,100, 60, 20,
349 28, 84,140,196,196,140, 84, 28,
350 28, 84,140,196,196,140, 84, 28,
351 20, 60,100,140,140,100, 60, 20,
352 12, 36, 60, 84, 84, 60, 36, 12,
353 4, 12, 20, 28, 28, 20, 12, 4,
358 static const uint8_t obmc4[16]={
366 static const uint8_t *obmc_tab[4]={
367 obmc32, obmc16, obmc8, obmc4
370 static int scale_mv_ref[MAX_REF_FRAMES][MAX_REF_FRAMES];
372 typedef struct BlockNode{
378 //#define TYPE_SPLIT 1
379 #define BLOCK_INTRA 1
381 //#define TYPE_NOCOLOR 4
382 uint8_t level; //FIXME merge into type?
385 static const BlockNode null_block= { //FIXME add border maybe
386 .color= {128,128,128},
394 #define LOG2_MB_SIZE 4
395 #define MB_SIZE (1<<LOG2_MB_SIZE)
396 #define ENCODER_EXTRA_BITS 4
399 typedef struct x_and_coeff{
404 typedef struct SubBand{
409 int qlog; ///< log(qscale)/log[2^(1/6)]
414 int stride_line; ///< Stride measured in lines, not pixels.
415 x_and_coeff * x_coeff;
416 struct SubBand *parent;
417 uint8_t state[/*7*2*/ 7 + 512][32];
420 typedef struct Plane{
423 SubBand band[MAX_DECOMPOSITIONS][4];
426 int8_t hcoeff[HTAPS_MAX/2];
431 int8_t last_hcoeff[HTAPS_MAX/2];
435 typedef struct SnowContext{
436 // MpegEncContext m; // needed for motion estimation, should not be used for anything else, the idea is to eventually make the motion estimation independent of MpegEncContext, so this will be removed then (FIXME/XXX)
438 AVCodecContext *avctx;
442 AVFrame input_picture; ///< new_picture with the internal linesizes
443 AVFrame current_picture;
444 AVFrame last_picture[MAX_REF_FRAMES];
445 uint8_t *halfpel_plane[MAX_REF_FRAMES][4][4];
446 AVFrame mconly_picture;
447 // uint8_t q_context[16];
448 uint8_t header_state[32];
449 uint8_t block_state[128 + 32*128];
453 int spatial_decomposition_type;
454 int last_spatial_decomposition_type;
455 int temporal_decomposition_type;
456 int spatial_decomposition_count;
457 int last_spatial_decomposition_count;
458 int temporal_decomposition_count;
461 int16_t (*ref_mvs[MAX_REF_FRAMES])[2];
462 uint32_t *ref_scores[MAX_REF_FRAMES];
463 DWTELEM *spatial_dwt_buffer;
464 IDWTELEM *spatial_idwt_buffer;
468 int spatial_scalability;
478 #define QBIAS_SHIFT 3
482 int last_block_max_depth;
483 Plane plane[MAX_PLANES];
485 #define ME_CACHE_SIZE 1024
486 int me_cache[ME_CACHE_SIZE];
487 int me_cache_generation;
490 MpegEncContext m; // needed for motion estimation, should not be used for anything else, the idea is to eventually make the motion estimation independent of MpegEncContext, so this will be removed then (FIXME/XXX)
501 #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)))
502 //#define slice_buffer_get_line(slice_buf, line_num) (slice_buffer_load_line((slice_buf), (line_num)))
504 static void iterative_me(SnowContext *s);
506 static void slice_buffer_init(slice_buffer * buf, int line_count, int max_allocated_lines, int line_width, IDWTELEM * base_buffer)
510 buf->base_buffer = base_buffer;
511 buf->line_count = line_count;
512 buf->line_width = line_width;
513 buf->data_count = max_allocated_lines;
514 buf->line = av_mallocz (sizeof(IDWTELEM *) * line_count);
515 buf->data_stack = av_malloc (sizeof(IDWTELEM *) * max_allocated_lines);
517 for (i = 0; i < max_allocated_lines; i++)
519 buf->data_stack[i] = av_malloc (sizeof(IDWTELEM) * line_width);
522 buf->data_stack_top = max_allocated_lines - 1;
525 static IDWTELEM * slice_buffer_load_line(slice_buffer * buf, int line)
530 assert(buf->data_stack_top >= 0);
531 // assert(!buf->line[line]);
533 return buf->line[line];
535 offset = buf->line_width * line;
536 buffer = buf->data_stack[buf->data_stack_top];
537 buf->data_stack_top--;
538 buf->line[line] = buffer;
543 static void slice_buffer_release(slice_buffer * buf, int line)
548 assert(line >= 0 && line < buf->line_count);
549 assert(buf->line[line]);
551 offset = buf->line_width * line;
552 buffer = buf->line[line];
553 buf->data_stack_top++;
554 buf->data_stack[buf->data_stack_top] = buffer;
555 buf->line[line] = NULL;
558 static void slice_buffer_flush(slice_buffer * buf)
561 for (i = 0; i < buf->line_count; i++)
564 slice_buffer_release(buf, i);
568 static void slice_buffer_destroy(slice_buffer * buf)
571 slice_buffer_flush(buf);
573 for (i = buf->data_count - 1; i >= 0; i--)
575 av_freep(&buf->data_stack[i]);
577 av_freep(&buf->data_stack);
578 av_freep(&buf->line);
582 // Avoid a name clash on SGI IRIX
585 #define QEXPSHIFT (7-FRAC_BITS+8) //FIXME try to change this to 0
586 static uint8_t qexp[QROOT];
588 static inline int mirror(int v, int m){
589 while((unsigned)v > (unsigned)m){
596 static inline void put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed){
600 const int a= FFABS(v);
601 const int e= av_log2(a);
603 const int el= FFMIN(e, 10);
604 put_rac(c, state+0, 0);
607 put_rac(c, state+1+i, 1); //1..10
610 put_rac(c, state+1+9, 1); //1..10
612 put_rac(c, state+1+FFMIN(i,9), 0);
614 for(i=e-1; i>=el; i--){
615 put_rac(c, state+22+9, (a>>i)&1); //22..31
618 put_rac(c, state+22+i, (a>>i)&1); //22..31
622 put_rac(c, state+11 + el, v < 0); //11..21
625 put_rac(c, state+0, 0);
628 put_rac(c, state+1+i, 1); //1..10
630 put_rac(c, state+1+i, 0);
632 for(i=e-1; i>=0; i--){
633 put_rac(c, state+22+i, (a>>i)&1); //22..31
637 put_rac(c, state+11 + e, v < 0); //11..21
640 put_rac(c, state+1+FFMIN(i,9), 1); //1..10
642 put_rac(c, state+1+FFMIN(i,9), 0);
644 for(i=e-1; i>=0; i--){
645 put_rac(c, state+22+FFMIN(i,9), (a>>i)&1); //22..31
649 put_rac(c, state+11 + FFMIN(e,10), v < 0); //11..21
653 put_rac(c, state+0, 1);
657 static inline int get_symbol(RangeCoder *c, uint8_t *state, int is_signed){
658 if(get_rac(c, state+0))
663 while(get_rac(c, state+1 + FFMIN(e,9))){ //1..10
668 for(i=e-1; i>=0; i--){
669 a += a + get_rac(c, state+22 + FFMIN(i,9)); //22..31
672 if(is_signed && get_rac(c, state+11 + FFMIN(e,10))) //11..21
679 static inline void put_symbol2(RangeCoder *c, uint8_t *state, int v, int log2){
681 int r= log2>=0 ? 1<<log2 : 1;
687 put_rac(c, state+4+log2, 1);
692 put_rac(c, state+4+log2, 0);
694 for(i=log2-1; i>=0; i--){
695 put_rac(c, state+31-i, (v>>i)&1);
699 static inline int get_symbol2(RangeCoder *c, uint8_t *state, int log2){
701 int r= log2>=0 ? 1<<log2 : 1;
706 while(get_rac(c, state+4+log2)){
712 for(i=log2-1; i>=0; i--){
713 v+= get_rac(c, state+31-i)<<i;
719 static av_always_inline void
720 lift(DWTELEM *dst, DWTELEM *src, DWTELEM *ref,
721 int dst_step, int src_step, int ref_step,
722 int width, int mul, int add, int shift,
723 int highpass, int inverse){
724 const int mirror_left= !highpass;
725 const int mirror_right= (width&1) ^ highpass;
726 const int w= (width>>1) - 1 + (highpass & width);
729 #define LIFT(src, ref, inv) ((src) + ((inv) ? - (ref) : + (ref)))
731 dst[0] = LIFT(src[0], ((mul*2*ref[0]+add)>>shift), inverse);
738 LIFT(src[i*src_step],
739 ((mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add)>>shift),
745 LIFT(src[w*src_step],
746 ((mul*2*ref[w*ref_step]+add)>>shift),
751 static av_always_inline void
752 inv_lift(IDWTELEM *dst, IDWTELEM *src, IDWTELEM *ref,
753 int dst_step, int src_step, int ref_step,
754 int width, int mul, int add, int shift,
755 int highpass, int inverse){
756 const int mirror_left= !highpass;
757 const int mirror_right= (width&1) ^ highpass;
758 const int w= (width>>1) - 1 + (highpass & width);
761 #define LIFT(src, ref, inv) ((src) + ((inv) ? - (ref) : + (ref)))
763 dst[0] = LIFT(src[0], ((mul*2*ref[0]+add)>>shift), inverse);
770 LIFT(src[i*src_step],
771 ((mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add)>>shift),
777 LIFT(src[w*src_step],
778 ((mul*2*ref[w*ref_step]+add)>>shift),
784 static av_always_inline void
785 liftS(DWTELEM *dst, DWTELEM *src, DWTELEM *ref,
786 int dst_step, int src_step, int ref_step,
787 int width, int mul, int add, int shift,
788 int highpass, int inverse){
789 const int mirror_left= !highpass;
790 const int mirror_right= (width&1) ^ highpass;
791 const int w= (width>>1) - 1 + (highpass & width);
795 #define LIFTS(src, ref, inv) \
797 (src) + (((ref) + 4*(src))>>shift): \
798 -((-16*(src) + (ref) + add/4 + 1 + (5<<25))/(5*4) - (1<<23)))
800 dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse);
807 LIFTS(src[i*src_step],
808 mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add,
814 LIFTS(src[w*src_step], mul*2*ref[w*ref_step]+add, inverse);
817 static av_always_inline void
818 inv_liftS(IDWTELEM *dst, IDWTELEM *src, IDWTELEM *ref,
819 int dst_step, int src_step, int ref_step,
820 int width, int mul, int add, int shift,
821 int highpass, int inverse){
822 const int mirror_left= !highpass;
823 const int mirror_right= (width&1) ^ highpass;
824 const int w= (width>>1) - 1 + (highpass & width);
828 #define LIFTS(src, ref, inv) \
830 (src) + (((ref) + 4*(src))>>shift): \
831 -((-16*(src) + (ref) + add/4 + 1 + (5<<25))/(5*4) - (1<<23)))
833 dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse);
840 LIFTS(src[i*src_step],
841 mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add,
847 LIFTS(src[w*src_step], mul*2*ref[w*ref_step]+add, inverse);
852 static void horizontal_decompose53i(DWTELEM *b, int width){
854 const int width2= width>>1;
856 const int w2= (width+1)>>1;
858 for(x=0; x<width2; x++){
860 temp[x+w2]= b[2*x + 1];
874 for(x=1; x+1<width2; x+=2){
878 A2 += (A1 + A3 + 2)>>2;
882 A1= temp[x+1+width2];
885 A4 += (A1 + A3 + 2)>>2;
891 A2 += (A1 + A3 + 2)>>2;
896 lift(b+w2, temp+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 0);
897 lift(b , temp , b+w2, 1, 1, 1, width, 1, 2, 2, 0, 0);
901 static void vertical_decompose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
904 for(i=0; i<width; i++){
905 b1[i] -= (b0[i] + b2[i])>>1;
909 static void vertical_decompose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
912 for(i=0; i<width; i++){
913 b1[i] += (b0[i] + b2[i] + 2)>>2;
917 static void spatial_decompose53i(DWTELEM *buffer, int width, int height, int stride){
919 DWTELEM *b0= buffer + mirror(-2-1, height-1)*stride;
920 DWTELEM *b1= buffer + mirror(-2 , height-1)*stride;
922 for(y=-2; y<height; y+=2){
923 DWTELEM *b2= buffer + mirror(y+1, height-1)*stride;
924 DWTELEM *b3= buffer + mirror(y+2, height-1)*stride;
926 if(y+1<(unsigned)height) horizontal_decompose53i(b2, width);
927 if(y+2<(unsigned)height) horizontal_decompose53i(b3, width);
929 if(y+1<(unsigned)height) vertical_decompose53iH0(b1, b2, b3, width);
930 if(y+0<(unsigned)height) vertical_decompose53iL0(b0, b1, b2, width);
937 static void horizontal_decompose97i(DWTELEM *b, int width){
939 const int w2= (width+1)>>1;
941 lift (temp+w2, b +1, b , 1, 2, 2, width, W_AM, W_AO, W_AS, 1, 1);
942 liftS(temp , b , temp+w2, 1, 2, 1, width, W_BM, W_BO, W_BS, 0, 0);
943 lift (b +w2, temp+w2, temp , 1, 1, 1, width, W_CM, W_CO, W_CS, 1, 0);
944 lift (b , temp , b +w2, 1, 1, 1, width, W_DM, W_DO, W_DS, 0, 0);
948 static void vertical_decompose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
951 for(i=0; i<width; i++){
952 b1[i] -= (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
956 static void vertical_decompose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
959 for(i=0; i<width; i++){
960 b1[i] += (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS;
964 static void vertical_decompose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
967 for(i=0; i<width; i++){
969 b1[i] -= (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS;
971 b1[i] = (16*4*b1[i] - 4*(b0[i] + b2[i]) + W_BO*5 + (5<<27)) / (5*16) - (1<<23);
976 static void vertical_decompose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
979 for(i=0; i<width; i++){
980 b1[i] += (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS;
984 static void spatial_decompose97i(DWTELEM *buffer, int width, int height, int stride){
986 DWTELEM *b0= buffer + mirror(-4-1, height-1)*stride;
987 DWTELEM *b1= buffer + mirror(-4 , height-1)*stride;
988 DWTELEM *b2= buffer + mirror(-4+1, height-1)*stride;
989 DWTELEM *b3= buffer + mirror(-4+2, height-1)*stride;
991 for(y=-4; y<height; y+=2){
992 DWTELEM *b4= buffer + mirror(y+3, height-1)*stride;
993 DWTELEM *b5= buffer + mirror(y+4, height-1)*stride;
995 if(y+3<(unsigned)height) horizontal_decompose97i(b4, width);
996 if(y+4<(unsigned)height) horizontal_decompose97i(b5, width);
998 if(y+3<(unsigned)height) vertical_decompose97iH0(b3, b4, b5, width);
999 if(y+2<(unsigned)height) vertical_decompose97iL0(b2, b3, b4, width);
1000 if(y+1<(unsigned)height) vertical_decompose97iH1(b1, b2, b3, width);
1001 if(y+0<(unsigned)height) vertical_decompose97iL1(b0, b1, b2, width);
1010 void ff_spatial_dwt(DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
1013 for(level=0; level<decomposition_count; level++){
1015 case DWT_97: spatial_decompose97i(buffer, width>>level, height>>level, stride<<level); break;
1016 case DWT_53: spatial_decompose53i(buffer, width>>level, height>>level, stride<<level); break;
1021 static void horizontal_compose53i(IDWTELEM *b, int width){
1022 IDWTELEM temp[width];
1023 const int width2= width>>1;
1024 const int w2= (width+1)>>1;
1036 for(x=1; x+1<width2; x+=2){
1040 A2 += (A1 + A3 + 2)>>2;
1044 A1= temp[x+1+width2];
1047 A4 += (A1 + A3 + 2)>>2;
1053 A2 += (A1 + A3 + 2)>>2;
1057 inv_lift(temp , b , b+w2, 1, 1, 1, width, 1, 2, 2, 0, 1);
1058 inv_lift(temp+w2, b+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 1);
1060 for(x=0; x<width2; x++){
1062 b[2*x + 1]= temp[x+w2];
1068 static void vertical_compose53iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
1071 for(i=0; i<width; i++){
1072 b1[i] += (b0[i] + b2[i])>>1;
1076 static void vertical_compose53iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
1079 for(i=0; i<width; i++){
1080 b1[i] -= (b0[i] + b2[i] + 2)>>2;
1084 static void spatial_compose53i_buffered_init(dwt_compose_t *cs, slice_buffer * sb, int height, int stride_line){
1085 cs->b0 = slice_buffer_get_line(sb, mirror(-1-1, height-1) * stride_line);
1086 cs->b1 = slice_buffer_get_line(sb, mirror(-1 , height-1) * stride_line);
1090 static void spatial_compose53i_init(dwt_compose_t *cs, IDWTELEM *buffer, int height, int stride){
1091 cs->b0 = buffer + mirror(-1-1, height-1)*stride;
1092 cs->b1 = buffer + mirror(-1 , height-1)*stride;
1096 static void spatial_compose53i_dy_buffered(dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line){
1099 IDWTELEM *b0= cs->b0;
1100 IDWTELEM *b1= cs->b1;
1101 IDWTELEM *b2= slice_buffer_get_line(sb, mirror(y+1, height-1) * stride_line);
1102 IDWTELEM *b3= slice_buffer_get_line(sb, mirror(y+2, height-1) * stride_line);
1104 if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width);
1105 if(y+0<(unsigned)height) vertical_compose53iH0(b0, b1, b2, width);
1107 if(y-1<(unsigned)height) horizontal_compose53i(b0, width);
1108 if(y+0<(unsigned)height) horizontal_compose53i(b1, width);
1115 static void spatial_compose53i_dy(dwt_compose_t *cs, IDWTELEM *buffer, int width, int height, int stride){
1117 IDWTELEM *b0= cs->b0;
1118 IDWTELEM *b1= cs->b1;
1119 IDWTELEM *b2= buffer + mirror(y+1, height-1)*stride;
1120 IDWTELEM *b3= buffer + mirror(y+2, height-1)*stride;
1122 if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width);
1123 if(y+0<(unsigned)height) vertical_compose53iH0(b0, b1, b2, width);
1125 if(y-1<(unsigned)height) horizontal_compose53i(b0, width);
1126 if(y+0<(unsigned)height) horizontal_compose53i(b1, width);
1133 static void av_unused spatial_compose53i(IDWTELEM *buffer, int width, int height, int stride){
1135 spatial_compose53i_init(&cs, buffer, height, stride);
1136 while(cs.y <= height)
1137 spatial_compose53i_dy(&cs, buffer, width, height, stride);
1141 void ff_snow_horizontal_compose97i(IDWTELEM *b, int width){
1142 IDWTELEM temp[width];
1143 const int w2= (width+1)>>1;
1145 inv_lift (temp , b , b +w2, 1, 1, 1, width, W_DM, W_DO, W_DS, 0, 1);
1146 inv_lift (temp+w2, b +w2, temp , 1, 1, 1, width, W_CM, W_CO, W_CS, 1, 1);
1147 inv_liftS(b , temp , temp+w2, 2, 1, 1, width, W_BM, W_BO, W_BS, 0, 1);
1148 inv_lift (b+1 , temp+w2, b , 2, 1, 2, width, W_AM, W_AO, W_AS, 1, 0);
1151 static void vertical_compose97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
1154 for(i=0; i<width; i++){
1155 b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
1159 static void vertical_compose97iH1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
1162 for(i=0; i<width; i++){
1163 b1[i] -= (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS;
1167 static void vertical_compose97iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
1170 for(i=0; i<width; i++){
1172 b1[i] += (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS;
1174 b1[i] += (W_BM*(b0[i] + b2[i])+4*b1[i]+W_BO)>>W_BS;
1179 static void vertical_compose97iL1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
1182 for(i=0; i<width; i++){
1183 b1[i] -= (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS;
1187 void ff_snow_vertical_compose97i(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){
1190 for(i=0; i<width; i++){
1191 b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;
1192 b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS;
1194 b2[i] += (W_BM*(b1[i] + b3[i])+W_BO)>>W_BS;
1196 b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS;
1198 b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
1202 static void spatial_compose97i_buffered_init(dwt_compose_t *cs, slice_buffer * sb, int height, int stride_line){
1203 cs->b0 = slice_buffer_get_line(sb, mirror(-3-1, height-1) * stride_line);
1204 cs->b1 = slice_buffer_get_line(sb, mirror(-3 , height-1) * stride_line);
1205 cs->b2 = slice_buffer_get_line(sb, mirror(-3+1, height-1) * stride_line);
1206 cs->b3 = slice_buffer_get_line(sb, mirror(-3+2, height-1) * stride_line);
1210 static void spatial_compose97i_init(dwt_compose_t *cs, IDWTELEM *buffer, int height, int stride){
1211 cs->b0 = buffer + mirror(-3-1, height-1)*stride;
1212 cs->b1 = buffer + mirror(-3 , height-1)*stride;
1213 cs->b2 = buffer + mirror(-3+1, height-1)*stride;
1214 cs->b3 = buffer + mirror(-3+2, height-1)*stride;
1218 static void spatial_compose97i_dy_buffered(DSPContext *dsp, dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line){
1221 IDWTELEM *b0= cs->b0;
1222 IDWTELEM *b1= cs->b1;
1223 IDWTELEM *b2= cs->b2;
1224 IDWTELEM *b3= cs->b3;
1225 IDWTELEM *b4= slice_buffer_get_line(sb, mirror(y + 3, height - 1) * stride_line);
1226 IDWTELEM *b5= slice_buffer_get_line(sb, mirror(y + 4, height - 1) * stride_line);
1228 if(y>0 && y+4<height){
1229 dsp->vertical_compose97i(b0, b1, b2, b3, b4, b5, width);
1231 if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width);
1232 if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width);
1233 if(y+1<(unsigned)height) vertical_compose97iL0(b1, b2, b3, width);
1234 if(y+0<(unsigned)height) vertical_compose97iH0(b0, b1, b2, width);
1237 if(y-1<(unsigned)height) dsp->horizontal_compose97i(b0, width);
1238 if(y+0<(unsigned)height) dsp->horizontal_compose97i(b1, width);
1247 static void spatial_compose97i_dy(dwt_compose_t *cs, IDWTELEM *buffer, int width, int height, int stride){
1249 IDWTELEM *b0= cs->b0;
1250 IDWTELEM *b1= cs->b1;
1251 IDWTELEM *b2= cs->b2;
1252 IDWTELEM *b3= cs->b3;
1253 IDWTELEM *b4= buffer + mirror(y+3, height-1)*stride;
1254 IDWTELEM *b5= buffer + mirror(y+4, height-1)*stride;
1256 if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width);
1257 if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width);
1258 if(y+1<(unsigned)height) vertical_compose97iL0(b1, b2, b3, width);
1259 if(y+0<(unsigned)height) vertical_compose97iH0(b0, b1, b2, width);
1261 if(y-1<(unsigned)height) ff_snow_horizontal_compose97i(b0, width);
1262 if(y+0<(unsigned)height) ff_snow_horizontal_compose97i(b1, width);
1271 static void av_unused spatial_compose97i(IDWTELEM *buffer, int width, int height, int stride){
1273 spatial_compose97i_init(&cs, buffer, height, stride);
1274 while(cs.y <= height)
1275 spatial_compose97i_dy(&cs, buffer, width, height, stride);
1278 static void ff_spatial_idwt_buffered_init(dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line, int type, int decomposition_count){
1280 for(level=decomposition_count-1; level>=0; level--){
1282 case DWT_97: spatial_compose97i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break;
1283 case DWT_53: spatial_compose53i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break;
1288 static void ff_spatial_idwt_init(dwt_compose_t *cs, IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
1290 for(level=decomposition_count-1; level>=0; level--){
1292 case DWT_97: spatial_compose97i_init(cs+level, buffer, height>>level, stride<<level); break;
1293 case DWT_53: spatial_compose53i_init(cs+level, buffer, height>>level, stride<<level); break;
1298 static void ff_spatial_idwt_slice(dwt_compose_t *cs, IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count, int y){
1299 const int support = type==1 ? 3 : 5;
1303 for(level=decomposition_count-1; level>=0; level--){
1304 while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){
1306 case DWT_97: spatial_compose97i_dy(cs+level, buffer, width>>level, height>>level, stride<<level);
1308 case DWT_53: spatial_compose53i_dy(cs+level, buffer, width>>level, height>>level, stride<<level);
1315 static void ff_spatial_idwt_buffered_slice(DSPContext *dsp, dwt_compose_t *cs, slice_buffer * slice_buf, int width, int height, int stride_line, int type, int decomposition_count, int y){
1316 const int support = type==1 ? 3 : 5;
1320 for(level=decomposition_count-1; level>=0; level--){
1321 while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){
1323 case DWT_97: spatial_compose97i_dy_buffered(dsp, cs+level, slice_buf, width>>level, height>>level, stride_line<<level);
1325 case DWT_53: spatial_compose53i_dy_buffered(cs+level, slice_buf, width>>level, height>>level, stride_line<<level);
1332 static void ff_spatial_idwt(IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
1333 dwt_compose_t cs[MAX_DECOMPOSITIONS];
1335 ff_spatial_idwt_init(cs, buffer, width, height, stride, type, decomposition_count);
1336 for(y=0; y<height; y+=4)
1337 ff_spatial_idwt_slice(cs, buffer, width, height, stride, type, decomposition_count, y);
1340 static int encode_subband_c0run(SnowContext *s, SubBand *b, IDWTELEM *src, IDWTELEM *parent, int stride, int orientation){
1341 const int w= b->width;
1342 const int h= b->height;
1354 int /*ll=0, */l=0, lt=0, t=0, rt=0;
1355 v= src[x + y*stride];
1358 t= src[x + (y-1)*stride];
1360 lt= src[x - 1 + (y-1)*stride];
1363 rt= src[x + 1 + (y-1)*stride];
1367 l= src[x - 1 + y*stride];
1369 if(orientation==1) ll= src[y + (x-2)*stride];
1370 else ll= src[x - 2 + y*stride];
1376 if(px<b->parent->width && py<b->parent->height)
1377 p= parent[px + py*2*stride];
1379 if(!(/*ll|*/l|lt|t|rt|p)){
1381 runs[run_index++]= run;
1389 max_index= run_index;
1390 runs[run_index++]= run;
1392 run= runs[run_index++];
1394 put_symbol2(&s->c, b->state[30], max_index, 0);
1395 if(run_index <= max_index)
1396 put_symbol2(&s->c, b->state[1], run, 3);
1399 if(s->c.bytestream_end - s->c.bytestream < w*40){
1400 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
1405 int /*ll=0, */l=0, lt=0, t=0, rt=0;
1406 v= src[x + y*stride];
1409 t= src[x + (y-1)*stride];
1411 lt= src[x - 1 + (y-1)*stride];
1414 rt= src[x + 1 + (y-1)*stride];
1418 l= src[x - 1 + y*stride];
1420 if(orientation==1) ll= src[y + (x-2)*stride];
1421 else ll= src[x - 2 + y*stride];
1427 if(px<b->parent->width && py<b->parent->height)
1428 p= parent[px + py*2*stride];
1430 if(/*ll|*/l|lt|t|rt|p){
1431 int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p));
1433 put_rac(&s->c, &b->state[0][context], !!v);
1436 run= runs[run_index++];
1438 if(run_index <= max_index)
1439 put_symbol2(&s->c, b->state[1], run, 3);
1447 int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p));
1448 int l2= 2*FFABS(l) + (l<0);
1449 int t2= 2*FFABS(t) + (t<0);
1451 put_symbol2(&s->c, b->state[context + 2], FFABS(v)-1, context-4);
1452 put_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l2&0xFF] + 3*quant3bA[t2&0xFF]], v<0);
1460 static int encode_subband(SnowContext *s, SubBand *b, IDWTELEM *src, IDWTELEM *parent, int stride, int orientation){
1461 // encode_subband_qtree(s, b, src, parent, stride, orientation);
1462 // encode_subband_z0run(s, b, src, parent, stride, orientation);
1463 return encode_subband_c0run(s, b, src, parent, stride, orientation);
1464 // encode_subband_dzr(s, b, src, parent, stride, orientation);
1467 static inline void unpack_coeffs(SnowContext *s, SubBand *b, SubBand * parent, int orientation){
1468 const int w= b->width;
1469 const int h= b->height;
1474 x_and_coeff *xc= b->x_coeff;
1475 x_and_coeff *prev_xc= NULL;
1476 x_and_coeff *prev2_xc= xc;
1477 x_and_coeff *parent_xc= parent ? parent->x_coeff : NULL;
1478 x_and_coeff *prev_parent_xc= parent_xc;
1480 runs= get_symbol2(&s->c, b->state[30], 0);
1481 if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
1486 int lt=0, t=0, rt=0;
1488 if(y && prev_xc->x == 0){
1500 if(prev_xc->x == x + 1)
1506 if(x>>1 > parent_xc->x){
1509 if(x>>1 == parent_xc->x){
1510 p= parent_xc->coeff;
1513 if(/*ll|*/l|lt|t|rt|p){
1514 int context= av_log2(/*FFABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1));
1516 v=get_rac(&s->c, &b->state[0][context]);
1518 v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1);
1519 v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l&0xFF] + 3*quant3bA[t&0xFF]]);
1526 if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
1528 v= 2*(get_symbol2(&s->c, b->state[0 + 2], 0-4) + 1);
1529 v+=get_rac(&s->c, &b->state[0][16 + 1 + 3]);
1538 if(y) max_run= FFMIN(run, prev_xc->x - x - 2);
1539 else max_run= FFMIN(run, w-x-1);
1541 max_run= FFMIN(max_run, 2*parent_xc->x - x - 1);
1547 (xc++)->x= w+1; //end marker
1553 while(parent_xc->x != parent->width+1)
1556 prev_parent_xc= parent_xc;
1558 parent_xc= prev_parent_xc;
1563 (xc++)->x= w+1; //end marker
1567 static inline void decode_subband_slice_buffered(SnowContext *s, SubBand *b, slice_buffer * sb, int start_y, int h, int save_state[1]){
1568 const int w= b->width;
1570 const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
1571 int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
1572 int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
1575 if(b->ibuf == s->spatial_idwt_buffer || s->qlog == LOSSLESS_QLOG){
1580 /* If we are on the second or later slice, restore our index. */
1582 new_index = save_state[0];
1585 for(y=start_y; y<h; y++){
1588 IDWTELEM * line = slice_buffer_get_line(sb, y * b->stride_line + b->buf_y_offset) + b->buf_x_offset;
1589 memset(line, 0, b->width*sizeof(IDWTELEM));
1590 v = b->x_coeff[new_index].coeff;
1591 x = b->x_coeff[new_index++].x;
1594 register int t= ( (v>>1)*qmul + qadd)>>QEXPSHIFT;
1595 register int u= -(v&1);
1596 line[x] = (t^u) - u;
1598 v = b->x_coeff[new_index].coeff;
1599 x = b->x_coeff[new_index++].x;
1603 /* Save our variables for the next slice. */
1604 save_state[0] = new_index;
1609 static void reset_contexts(SnowContext *s){ //FIXME better initial contexts
1610 int plane_index, level, orientation;
1612 for(plane_index=0; plane_index<3; plane_index++){
1613 for(level=0; level<MAX_DECOMPOSITIONS; level++){
1614 for(orientation=level ? 1:0; orientation<4; orientation++){
1615 memset(s->plane[plane_index].band[level][orientation].state, MID_STATE, sizeof(s->plane[plane_index].band[level][orientation].state));
1619 memset(s->header_state, MID_STATE, sizeof(s->header_state));
1620 memset(s->block_state, MID_STATE, sizeof(s->block_state));
1623 static int alloc_blocks(SnowContext *s){
1624 int w= -((-s->avctx->width )>>LOG2_MB_SIZE);
1625 int h= -((-s->avctx->height)>>LOG2_MB_SIZE);
1630 s->block= av_mallocz(w * h * sizeof(BlockNode) << (s->block_max_depth*2));
1634 static inline void copy_rac_state(RangeCoder *d, RangeCoder *s){
1635 uint8_t *bytestream= d->bytestream;
1636 uint8_t *bytestream_start= d->bytestream_start;
1638 d->bytestream= bytestream;
1639 d->bytestream_start= bytestream_start;
1642 //near copy & paste from dsputil, FIXME
1643 static int pix_sum(uint8_t * pix, int line_size, int w)
1648 for (i = 0; i < w; i++) {
1649 for (j = 0; j < w; j++) {
1653 pix += line_size - w;
1658 //near copy & paste from dsputil, FIXME
1659 static int pix_norm1(uint8_t * pix, int line_size, int w)
1662 uint32_t *sq = ff_squareTbl + 256;
1665 for (i = 0; i < w; i++) {
1666 for (j = 0; j < w; j ++) {
1670 pix += line_size - w;
1675 static inline void set_blocks(SnowContext *s, int level, int x, int y, int l, int cb, int cr, int mx, int my, int ref, int type){
1676 const int w= s->b_width << s->block_max_depth;
1677 const int rem_depth= s->block_max_depth - level;
1678 const int index= (x + y*w) << rem_depth;
1679 const int block_w= 1<<rem_depth;
1692 for(j=0; j<block_w; j++){
1693 for(i=0; i<block_w; i++){
1694 s->block[index + i + j*w]= block;
1699 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){
1700 const int offset[3]= {
1702 ((y*c->uvstride + x)>>1),
1703 ((y*c->uvstride + x)>>1),
1707 c->src[0][i]= src [i];
1708 c->ref[0][i]= ref [i] + offset[i];
1713 static inline void pred_mv(SnowContext *s, int *mx, int *my, int ref,
1714 const BlockNode *left, const BlockNode *top, const BlockNode *tr){
1715 if(s->ref_frames == 1){
1716 *mx = mid_pred(left->mx, top->mx, tr->mx);
1717 *my = mid_pred(left->my, top->my, tr->my);
1719 const int *scale = scale_mv_ref[ref];
1720 *mx = mid_pred((left->mx * scale[left->ref] + 128) >>8,
1721 (top ->mx * scale[top ->ref] + 128) >>8,
1722 (tr ->mx * scale[tr ->ref] + 128) >>8);
1723 *my = mid_pred((left->my * scale[left->ref] + 128) >>8,
1724 (top ->my * scale[top ->ref] + 128) >>8,
1725 (tr ->my * scale[tr ->ref] + 128) >>8);
1732 #define P_TOPRIGHT P[3]
1733 #define P_MEDIAN P[4]
1735 #define FLAG_QPEL 1 //must be 1
1737 static int encode_q_branch(SnowContext *s, int level, int x, int y){
1738 uint8_t p_buffer[1024];
1739 uint8_t i_buffer[1024];
1740 uint8_t p_state[sizeof(s->block_state)];
1741 uint8_t i_state[sizeof(s->block_state)];
1743 uint8_t *pbbak= s->c.bytestream;
1744 uint8_t *pbbak_start= s->c.bytestream_start;
1745 int score, score2, iscore, i_len, p_len, block_s, sum, base_bits;
1746 const int w= s->b_width << s->block_max_depth;
1747 const int h= s->b_height << s->block_max_depth;
1748 const int rem_depth= s->block_max_depth - level;
1749 const int index= (x + y*w) << rem_depth;
1750 const int block_w= 1<<(LOG2_MB_SIZE - level);
1751 int trx= (x+1)<<rem_depth;
1752 int try= (y+1)<<rem_depth;
1753 const BlockNode *left = x ? &s->block[index-1] : &null_block;
1754 const BlockNode *top = y ? &s->block[index-w] : &null_block;
1755 const BlockNode *right = trx<w ? &s->block[index+1] : &null_block;
1756 const BlockNode *bottom= try<h ? &s->block[index+w] : &null_block;
1757 const BlockNode *tl = y && x ? &s->block[index-w-1] : left;
1758 const BlockNode *tr = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
1759 int pl = left->color[0];
1760 int pcb= left->color[1];
1761 int pcr= left->color[2];
1765 const int stride= s->current_picture.linesize[0];
1766 const int uvstride= s->current_picture.linesize[1];
1767 uint8_t *current_data[3]= { s->input_picture.data[0] + (x + y* stride)*block_w,
1768 s->input_picture.data[1] + (x + y*uvstride)*block_w/2,
1769 s->input_picture.data[2] + (x + y*uvstride)*block_w/2};
1771 int16_t last_mv[3][2];
1772 int qpel= !!(s->avctx->flags & CODEC_FLAG_QPEL); //unused
1773 const int shift= 1+qpel;
1774 MotionEstContext *c= &s->m.me;
1775 int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
1776 int mx_context= av_log2(2*FFABS(left->mx - top->mx));
1777 int my_context= av_log2(2*FFABS(left->my - top->my));
1778 int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
1779 int ref, best_ref, ref_score, ref_mx, ref_my;
1781 assert(sizeof(s->block_state) >= 256);
1783 set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA);
1787 // clip predictors / edge ?
1789 P_LEFT[0]= left->mx;
1790 P_LEFT[1]= left->my;
1793 P_TOPRIGHT[0]= tr->mx;
1794 P_TOPRIGHT[1]= tr->my;
1796 last_mv[0][0]= s->block[index].mx;
1797 last_mv[0][1]= s->block[index].my;
1798 last_mv[1][0]= right->mx;
1799 last_mv[1][1]= right->my;
1800 last_mv[2][0]= bottom->mx;
1801 last_mv[2][1]= bottom->my;
1808 assert(c-> stride == stride);
1809 assert(c->uvstride == uvstride);
1811 c->penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
1812 c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
1813 c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
1814 c->current_mv_penalty= c->mv_penalty[s->m.f_code=1] + MAX_MV;
1816 c->xmin = - x*block_w - 16+2;
1817 c->ymin = - y*block_w - 16+2;
1818 c->xmax = - (x+1)*block_w + (w<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-2;
1819 c->ymax = - (y+1)*block_w + (h<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-2;
1821 if(P_LEFT[0] > (c->xmax<<shift)) P_LEFT[0] = (c->xmax<<shift);
1822 if(P_LEFT[1] > (c->ymax<<shift)) P_LEFT[1] = (c->ymax<<shift);
1823 if(P_TOP[0] > (c->xmax<<shift)) P_TOP[0] = (c->xmax<<shift);
1824 if(P_TOP[1] > (c->ymax<<shift)) P_TOP[1] = (c->ymax<<shift);
1825 if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
1826 if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift); //due to pmx no clip
1827 if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);
1829 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
1830 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
1833 c->pred_x= P_LEFT[0];
1834 c->pred_y= P_LEFT[1];
1836 c->pred_x = P_MEDIAN[0];
1837 c->pred_y = P_MEDIAN[1];
1842 for(ref=0; ref<s->ref_frames; ref++){
1843 init_ref(c, current_data, s->last_picture[ref].data, NULL, block_w*x, block_w*y, 0);
1845 ref_score= ff_epzs_motion_search(&s->m, &ref_mx, &ref_my, P, 0, /*ref_index*/ 0, last_mv,
1846 (1<<16)>>shift, level-LOG2_MB_SIZE+4, block_w);
1848 assert(ref_mx >= c->xmin);
1849 assert(ref_mx <= c->xmax);
1850 assert(ref_my >= c->ymin);
1851 assert(ref_my <= c->ymax);
1853 ref_score= c->sub_motion_search(&s->m, &ref_mx, &ref_my, ref_score, 0, 0, level-LOG2_MB_SIZE+4, block_w);
1854 ref_score= ff_get_mb_score(&s->m, ref_mx, ref_my, 0, 0, level-LOG2_MB_SIZE+4, block_w, 0);
1855 ref_score+= 2*av_log2(2*ref)*c->penalty_factor;
1856 if(s->ref_mvs[ref]){
1857 s->ref_mvs[ref][index][0]= ref_mx;
1858 s->ref_mvs[ref][index][1]= ref_my;
1859 s->ref_scores[ref][index]= ref_score;
1861 if(score > ref_score){
1868 //FIXME if mb_cmp != SSE then intra cannot be compared currently and mb_penalty vs. lambda2
1871 base_bits= get_rac_count(&s->c) - 8*(s->c.bytestream - s->c.bytestream_start);
1873 pc.bytestream_start=
1874 pc.bytestream= p_buffer; //FIXME end/start? and at the other stoo
1875 memcpy(p_state, s->block_state, sizeof(s->block_state));
1877 if(level!=s->block_max_depth)
1878 put_rac(&pc, &p_state[4 + s_context], 1);
1879 put_rac(&pc, &p_state[1 + left->type + top->type], 0);
1880 if(s->ref_frames > 1)
1881 put_symbol(&pc, &p_state[128 + 1024 + 32*ref_context], best_ref, 0);
1882 pred_mv(s, &pmx, &pmy, best_ref, left, top, tr);
1883 put_symbol(&pc, &p_state[128 + 32*(mx_context + 16*!!best_ref)], mx - pmx, 1);
1884 put_symbol(&pc, &p_state[128 + 32*(my_context + 16*!!best_ref)], my - pmy, 1);
1885 p_len= pc.bytestream - pc.bytestream_start;
1886 score += (s->lambda2*(get_rac_count(&pc)-base_bits))>>FF_LAMBDA_SHIFT;
1888 block_s= block_w*block_w;
1889 sum = pix_sum(current_data[0], stride, block_w);
1890 l= (sum + block_s/2)/block_s;
1891 iscore = pix_norm1(current_data[0], stride, block_w) - 2*l*sum + l*l*block_s;
1893 block_s= block_w*block_w>>2;
1894 sum = pix_sum(current_data[1], uvstride, block_w>>1);
1895 cb= (sum + block_s/2)/block_s;
1896 // iscore += pix_norm1(¤t_mb[1][0], uvstride, block_w>>1) - 2*cb*sum + cb*cb*block_s;
1897 sum = pix_sum(current_data[2], uvstride, block_w>>1);
1898 cr= (sum + block_s/2)/block_s;
1899 // iscore += pix_norm1(¤t_mb[2][0], uvstride, block_w>>1) - 2*cr*sum + cr*cr*block_s;
1902 ic.bytestream_start=
1903 ic.bytestream= i_buffer; //FIXME end/start? and at the other stoo
1904 memcpy(i_state, s->block_state, sizeof(s->block_state));
1905 if(level!=s->block_max_depth)
1906 put_rac(&ic, &i_state[4 + s_context], 1);
1907 put_rac(&ic, &i_state[1 + left->type + top->type], 1);
1908 put_symbol(&ic, &i_state[32], l-pl , 1);
1909 put_symbol(&ic, &i_state[64], cb-pcb, 1);
1910 put_symbol(&ic, &i_state[96], cr-pcr, 1);
1911 i_len= ic.bytestream - ic.bytestream_start;
1912 iscore += (s->lambda2*(get_rac_count(&ic)-base_bits))>>FF_LAMBDA_SHIFT;
1914 // assert(score==256*256*256*64-1);
1915 assert(iscore < 255*255*256 + s->lambda2*10);
1916 assert(iscore >= 0);
1917 assert(l>=0 && l<=255);
1918 assert(pl>=0 && pl<=255);
1921 int varc= iscore >> 8;
1922 int vard= score >> 8;
1923 if (vard <= 64 || vard < varc)
1924 c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
1926 c->scene_change_score+= s->m.qscale;
1929 if(level!=s->block_max_depth){
1930 put_rac(&s->c, &s->block_state[4 + s_context], 0);
1931 score2 = encode_q_branch(s, level+1, 2*x+0, 2*y+0);
1932 score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+0);
1933 score2+= encode_q_branch(s, level+1, 2*x+0, 2*y+1);
1934 score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+1);
1935 score2+= s->lambda2>>FF_LAMBDA_SHIFT; //FIXME exact split overhead
1937 if(score2 < score && score2 < iscore)
1942 pred_mv(s, &pmx, &pmy, 0, left, top, tr);
1943 memcpy(pbbak, i_buffer, i_len);
1945 s->c.bytestream_start= pbbak_start;
1946 s->c.bytestream= pbbak + i_len;
1947 set_blocks(s, level, x, y, l, cb, cr, pmx, pmy, 0, BLOCK_INTRA);
1948 memcpy(s->block_state, i_state, sizeof(s->block_state));
1951 memcpy(pbbak, p_buffer, p_len);
1953 s->c.bytestream_start= pbbak_start;
1954 s->c.bytestream= pbbak + p_len;
1955 set_blocks(s, level, x, y, pl, pcb, pcr, mx, my, best_ref, 0);
1956 memcpy(s->block_state, p_state, sizeof(s->block_state));
1961 static av_always_inline int same_block(BlockNode *a, BlockNode *b){
1962 if((a->type&BLOCK_INTRA) && (b->type&BLOCK_INTRA)){
1963 return !((a->color[0] - b->color[0]) | (a->color[1] - b->color[1]) | (a->color[2] - b->color[2]));
1965 return !((a->mx - b->mx) | (a->my - b->my) | (a->ref - b->ref) | ((a->type ^ b->type)&BLOCK_INTRA));
1969 static void encode_q_branch2(SnowContext *s, int level, int x, int y){
1970 const int w= s->b_width << s->block_max_depth;
1971 const int rem_depth= s->block_max_depth - level;
1972 const int index= (x + y*w) << rem_depth;
1973 int trx= (x+1)<<rem_depth;
1974 BlockNode *b= &s->block[index];
1975 const BlockNode *left = x ? &s->block[index-1] : &null_block;
1976 const BlockNode *top = y ? &s->block[index-w] : &null_block;
1977 const BlockNode *tl = y && x ? &s->block[index-w-1] : left;
1978 const BlockNode *tr = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
1979 int pl = left->color[0];
1980 int pcb= left->color[1];
1981 int pcr= left->color[2];
1983 int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
1984 int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 16*!!b->ref;
1985 int my_context= av_log2(2*FFABS(left->my - top->my)) + 16*!!b->ref;
1986 int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
1989 set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA);
1993 if(level!=s->block_max_depth){
1994 if(same_block(b,b+1) && same_block(b,b+w) && same_block(b,b+w+1)){
1995 put_rac(&s->c, &s->block_state[4 + s_context], 1);
1997 put_rac(&s->c, &s->block_state[4 + s_context], 0);
1998 encode_q_branch2(s, level+1, 2*x+0, 2*y+0);
1999 encode_q_branch2(s, level+1, 2*x+1, 2*y+0);
2000 encode_q_branch2(s, level+1, 2*x+0, 2*y+1);
2001 encode_q_branch2(s, level+1, 2*x+1, 2*y+1);
2005 if(b->type & BLOCK_INTRA){
2006 pred_mv(s, &pmx, &pmy, 0, left, top, tr);
2007 put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 1);
2008 put_symbol(&s->c, &s->block_state[32], b->color[0]-pl , 1);
2009 put_symbol(&s->c, &s->block_state[64], b->color[1]-pcb, 1);
2010 put_symbol(&s->c, &s->block_state[96], b->color[2]-pcr, 1);
2011 set_blocks(s, level, x, y, b->color[0], b->color[1], b->color[2], pmx, pmy, 0, BLOCK_INTRA);
2013 pred_mv(s, &pmx, &pmy, b->ref, left, top, tr);
2014 put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 0);
2015 if(s->ref_frames > 1)
2016 put_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], b->ref, 0);
2017 put_symbol(&s->c, &s->block_state[128 + 32*mx_context], b->mx - pmx, 1);
2018 put_symbol(&s->c, &s->block_state[128 + 32*my_context], b->my - pmy, 1);
2019 set_blocks(s, level, x, y, pl, pcb, pcr, b->mx, b->my, b->ref, 0);
2023 static void decode_q_branch(SnowContext *s, int level, int x, int y){
2024 const int w= s->b_width << s->block_max_depth;
2025 const int rem_depth= s->block_max_depth - level;
2026 const int index= (x + y*w) << rem_depth;
2027 int trx= (x+1)<<rem_depth;
2028 const BlockNode *left = x ? &s->block[index-1] : &null_block;
2029 const BlockNode *top = y ? &s->block[index-w] : &null_block;
2030 const BlockNode *tl = y && x ? &s->block[index-w-1] : left;
2031 const BlockNode *tr = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
2032 int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
2035 set_blocks(s, level, x, y, null_block.color[0], null_block.color[1], null_block.color[2], null_block.mx, null_block.my, null_block.ref, BLOCK_INTRA);
2039 if(level==s->block_max_depth || get_rac(&s->c, &s->block_state[4 + s_context])){
2041 int l = left->color[0];
2042 int cb= left->color[1];
2043 int cr= left->color[2];
2045 int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
2046 int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 0*av_log2(2*FFABS(tr->mx - top->mx));
2047 int my_context= av_log2(2*FFABS(left->my - top->my)) + 0*av_log2(2*FFABS(tr->my - top->my));
2049 type= get_rac(&s->c, &s->block_state[1 + left->type + top->type]) ? BLOCK_INTRA : 0;
2052 pred_mv(s, &mx, &my, 0, left, top, tr);
2053 l += get_symbol(&s->c, &s->block_state[32], 1);
2054 cb+= get_symbol(&s->c, &s->block_state[64], 1);
2055 cr+= get_symbol(&s->c, &s->block_state[96], 1);
2057 if(s->ref_frames > 1)
2058 ref= get_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], 0);
2059 pred_mv(s, &mx, &my, ref, left, top, tr);
2060 mx+= get_symbol(&s->c, &s->block_state[128 + 32*(mx_context + 16*!!ref)], 1);
2061 my+= get_symbol(&s->c, &s->block_state[128 + 32*(my_context + 16*!!ref)], 1);
2063 set_blocks(s, level, x, y, l, cb, cr, mx, my, ref, type);
2065 decode_q_branch(s, level+1, 2*x+0, 2*y+0);
2066 decode_q_branch(s, level+1, 2*x+1, 2*y+0);
2067 decode_q_branch(s, level+1, 2*x+0, 2*y+1);
2068 decode_q_branch(s, level+1, 2*x+1, 2*y+1);
2072 static void encode_blocks(SnowContext *s, int search){
2077 if(s->avctx->me_method == ME_ITER && !s->keyframe && search)
2081 if(s->c.bytestream_end - s->c.bytestream < w*MB_SIZE*MB_SIZE*3){ //FIXME nicer limit
2082 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
2086 if(s->avctx->me_method == ME_ITER || !search)
2087 encode_q_branch2(s, 0, x, y);
2089 encode_q_branch (s, 0, x, y);
2094 static void decode_blocks(SnowContext *s){
2101 decode_q_branch(s, 0, x, y);
2106 static void mc_block(Plane *p, uint8_t *dst, const uint8_t *src, uint8_t *tmp, int stride, int b_w, int b_h, int dx, int dy){
2107 const static uint8_t weight[64]={
2118 const static uint8_t brane[256]={
2119 0x00,0x01,0x01,0x01,0x01,0x01,0x01,0x01,0x11,0x12,0x12,0x12,0x12,0x12,0x12,0x12,
2120 0x04,0x05,0xcc,0xcc,0xcc,0xcc,0xcc,0x41,0x15,0x16,0xcc,0xcc,0xcc,0xcc,0xcc,0x52,
2121 0x04,0xcc,0x05,0xcc,0xcc,0xcc,0x41,0xcc,0x15,0xcc,0x16,0xcc,0xcc,0xcc,0x52,0xcc,
2122 0x04,0xcc,0xcc,0x05,0xcc,0x41,0xcc,0xcc,0x15,0xcc,0xcc,0x16,0xcc,0x52,0xcc,0xcc,
2123 0x04,0xcc,0xcc,0xcc,0x41,0xcc,0xcc,0xcc,0x15,0xcc,0xcc,0xcc,0x16,0xcc,0xcc,0xcc,
2124 0x04,0xcc,0xcc,0x41,0xcc,0x05,0xcc,0xcc,0x15,0xcc,0xcc,0x52,0xcc,0x16,0xcc,0xcc,
2125 0x04,0xcc,0x41,0xcc,0xcc,0xcc,0x05,0xcc,0x15,0xcc,0x52,0xcc,0xcc,0xcc,0x16,0xcc,
2126 0x04,0x41,0xcc,0xcc,0xcc,0xcc,0xcc,0x05,0x15,0x52,0xcc,0xcc,0xcc,0xcc,0xcc,0x16,
2127 0x44,0x45,0x45,0x45,0x45,0x45,0x45,0x45,0x55,0x56,0x56,0x56,0x56,0x56,0x56,0x56,
2128 0x48,0x49,0xcc,0xcc,0xcc,0xcc,0xcc,0x85,0x59,0x5A,0xcc,0xcc,0xcc,0xcc,0xcc,0x96,
2129 0x48,0xcc,0x49,0xcc,0xcc,0xcc,0x85,0xcc,0x59,0xcc,0x5A,0xcc,0xcc,0xcc,0x96,0xcc,
2130 0x48,0xcc,0xcc,0x49,0xcc,0x85,0xcc,0xcc,0x59,0xcc,0xcc,0x5A,0xcc,0x96,0xcc,0xcc,
2131 0x48,0xcc,0xcc,0xcc,0x49,0xcc,0xcc,0xcc,0x59,0xcc,0xcc,0xcc,0x96,0xcc,0xcc,0xcc,
2132 0x48,0xcc,0xcc,0x85,0xcc,0x49,0xcc,0xcc,0x59,0xcc,0xcc,0x96,0xcc,0x5A,0xcc,0xcc,
2133 0x48,0xcc,0x85,0xcc,0xcc,0xcc,0x49,0xcc,0x59,0xcc,0x96,0xcc,0xcc,0xcc,0x5A,0xcc,
2134 0x48,0x85,0xcc,0xcc,0xcc,0xcc,0xcc,0x49,0x59,0x96,0xcc,0xcc,0xcc,0xcc,0xcc,0x5A,
2137 const static uint8_t needs[16]={
2145 int16_t tmpIt [64*(32+HTAPS_MAX)];
2146 uint8_t tmp2t[3][stride*(32+HTAPS_MAX)];
2147 int16_t *tmpI= tmpIt;
2148 uint8_t *tmp2= tmp2t[0];
2149 const uint8_t *hpel[11];
2150 assert(dx<16 && dy<16);
2151 r= brane[dx + 16*dy]&15;
2152 l= brane[dx + 16*dy]>>4;
2154 b= needs[l] | needs[r];
2155 if(p && !p->diag_mc)
2159 for(y=0; y < b_h+HTAPS_MAX-1; y++){
2160 for(x=0; x < b_w; x++){
2161 int a_1=src[x + HTAPS_MAX/2-4];
2162 int a0= src[x + HTAPS_MAX/2-3];
2163 int a1= src[x + HTAPS_MAX/2-2];
2164 int a2= src[x + HTAPS_MAX/2-1];
2165 int a3= src[x + HTAPS_MAX/2+0];
2166 int a4= src[x + HTAPS_MAX/2+1];
2167 int a5= src[x + HTAPS_MAX/2+2];
2168 int a6= src[x + HTAPS_MAX/2+3];
2170 if(!p || p->fast_mc){
2171 am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5);
2175 am= p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6);
2180 if(am&(~255)) am= ~(am>>31);
2189 src += HTAPS_MAX/2 - 1;
2193 for(y=0; y < b_h; y++){
2194 for(x=0; x < b_w+1; x++){
2195 int a_1=src[x + (HTAPS_MAX/2-4)*stride];
2196 int a0= src[x + (HTAPS_MAX/2-3)*stride];
2197 int a1= src[x + (HTAPS_MAX/2-2)*stride];
2198 int a2= src[x + (HTAPS_MAX/2-1)*stride];
2199 int a3= src[x + (HTAPS_MAX/2+0)*stride];
2200 int a4= src[x + (HTAPS_MAX/2+1)*stride];
2201 int a5= src[x + (HTAPS_MAX/2+2)*stride];
2202 int a6= src[x + (HTAPS_MAX/2+3)*stride];
2204 if(!p || p->fast_mc)
2205 am= (20*(a2+a3) - 5*(a1+a4) + (a0+a5) + 16)>>5;
2207 am= (p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6) + 32)>>6;
2209 if(am&(~255)) am= ~(am>>31);
2217 src += stride*(HTAPS_MAX/2 - 1);
2221 for(y=0; y < b_h; y++){
2222 for(x=0; x < b_w; x++){
2223 int a_1=tmpI[x + (HTAPS_MAX/2-4)*64];
2224 int a0= tmpI[x + (HTAPS_MAX/2-3)*64];
2225 int a1= tmpI[x + (HTAPS_MAX/2-2)*64];
2226 int a2= tmpI[x + (HTAPS_MAX/2-1)*64];
2227 int a3= tmpI[x + (HTAPS_MAX/2+0)*64];
2228 int a4= tmpI[x + (HTAPS_MAX/2+1)*64];
2229 int a5= tmpI[x + (HTAPS_MAX/2+2)*64];
2230 int a6= tmpI[x + (HTAPS_MAX/2+3)*64];
2232 if(!p || p->fast_mc)
2233 am= (20*(a2+a3) - 5*(a1+a4) + (a0+a5) + 512)>>10;
2235 am= (p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6) + 2048)>>12;
2236 if(am&(~255)) am= ~(am>>31);
2245 hpel[ 1]= tmp2t[0] + stride*(HTAPS_MAX/2-1);
2250 hpel[ 6]= tmp2t[1] + 1;
2252 hpel[ 8]= src + stride;
2253 hpel[ 9]= hpel[1] + stride;
2254 hpel[10]= hpel[8] + 1;
2257 const uint8_t *src1= hpel[dx/8 + dy/8*4 ];
2258 const uint8_t *src2= hpel[dx/8 + dy/8*4+1];
2259 const uint8_t *src3= hpel[dx/8 + dy/8*4+4];
2260 const uint8_t *src4= hpel[dx/8 + dy/8*4+5];
2263 for(y=0; y < b_h; y++){
2264 for(x=0; x < b_w; x++){
2265 dst[x]= ((8-dx)*(8-dy)*src1[x] + dx*(8-dy)*src2[x]+
2266 (8-dx)* dy *src3[x] + dx* dy *src4[x]+32)>>6;
2275 const uint8_t *src1= hpel[l];
2276 const uint8_t *src2= hpel[r];
2277 int a= weight[((dx&7) + (8*(dy&7)))];
2279 for(y=0; y < b_h; y++){
2280 for(x=0; x < b_w; x++){
2281 dst[x]= (a*src1[x] + b*src2[x] + 4)>>3;
2290 #define mca(dx,dy,b_w)\
2291 static void mc_block_hpel ## dx ## dy ## b_w(uint8_t *dst, const uint8_t *src, int stride, int h){\
2292 uint8_t tmp[stride*(b_w+HTAPS_MAX-1)];\
2294 mc_block(NULL, dst, src-(HTAPS_MAX/2-1)-(HTAPS_MAX/2-1)*stride, tmp, stride, b_w, b_w, dx, dy);\
2306 static void pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, int stride, int sx, int sy, int b_w, int b_h, BlockNode *block, int plane_index, int w, int h){
2307 if(block->type & BLOCK_INTRA){
2309 const int color = block->color[plane_index];
2310 const int color4= color*0x01010101;
2312 for(y=0; y < b_h; y++){
2313 *(uint32_t*)&dst[0 + y*stride]= color4;
2314 *(uint32_t*)&dst[4 + y*stride]= color4;
2315 *(uint32_t*)&dst[8 + y*stride]= color4;
2316 *(uint32_t*)&dst[12+ y*stride]= color4;
2317 *(uint32_t*)&dst[16+ y*stride]= color4;
2318 *(uint32_t*)&dst[20+ y*stride]= color4;
2319 *(uint32_t*)&dst[24+ y*stride]= color4;
2320 *(uint32_t*)&dst[28+ y*stride]= color4;
2323 for(y=0; y < b_h; y++){
2324 *(uint32_t*)&dst[0 + y*stride]= color4;
2325 *(uint32_t*)&dst[4 + y*stride]= color4;
2326 *(uint32_t*)&dst[8 + y*stride]= color4;
2327 *(uint32_t*)&dst[12+ y*stride]= color4;
2330 for(y=0; y < b_h; y++){
2331 *(uint32_t*)&dst[0 + y*stride]= color4;
2332 *(uint32_t*)&dst[4 + y*stride]= color4;
2335 for(y=0; y < b_h; y++){
2336 *(uint32_t*)&dst[0 + y*stride]= color4;
2339 for(y=0; y < b_h; y++){
2340 for(x=0; x < b_w; x++){
2341 dst[x + y*stride]= color;
2346 uint8_t *src= s->last_picture[block->ref].data[plane_index];
2347 const int scale= plane_index ? s->mv_scale : 2*s->mv_scale;
2348 int mx= block->mx*scale;
2349 int my= block->my*scale;
2350 const int dx= mx&15;
2351 const int dy= my&15;
2352 const int tab_index= 3 - (b_w>>2) + (b_w>>4);
2353 sx += (mx>>4) - (HTAPS_MAX/2-1);
2354 sy += (my>>4) - (HTAPS_MAX/2-1);
2355 src += sx + sy*stride;
2356 if( (unsigned)sx >= w - b_w - (HTAPS_MAX-2)
2357 || (unsigned)sy >= h - b_h - (HTAPS_MAX-2)){
2358 ff_emulated_edge_mc(tmp + MB_SIZE, src, stride, b_w+HTAPS_MAX-1, b_h+HTAPS_MAX-1, sx, sy, w, h);
2361 // assert(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h);
2362 // assert(!(b_w&(b_w-1)));
2363 assert(b_w>1 && b_h>1);
2364 assert((tab_index>=0 && tab_index<4) || b_w==32);
2365 if((dx&3) || (dy&3) || !(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h) || (b_w&(b_w-1)) || !s->plane[plane_index].fast_mc )
2366 mc_block(&s->plane[plane_index], dst, src, tmp, stride, b_w, b_h, dx, dy);
2369 for(y=0; y<b_h; y+=16){
2370 s->dsp.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + y*stride, src + 3 + (y+3)*stride,stride);
2371 s->dsp.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + 16 + y*stride, src + 19 + (y+3)*stride,stride);
2374 s->dsp.put_h264_qpel_pixels_tab[tab_index ][dy+(dx>>2)](dst,src + 3 + 3*stride,stride);
2375 else if(b_w==2*b_h){
2376 s->dsp.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst ,src + 3 + 3*stride,stride);
2377 s->dsp.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst+b_h,src + 3 + b_h + 3*stride,stride);
2380 s->dsp.put_h264_qpel_pixels_tab[tab_index ][dy+(dx>>2)](dst ,src + 3 + 3*stride ,stride);
2381 s->dsp.put_h264_qpel_pixels_tab[tab_index ][dy+(dx>>2)](dst+b_w*stride,src + 3 + 3*stride+b_w*stride,stride);
2386 void ff_snow_inner_add_yblock(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
2387 int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){
2390 for(y=0; y<b_h; y++){
2391 //FIXME ugly misuse of obmc_stride
2392 const uint8_t *obmc1= obmc + y*obmc_stride;
2393 const uint8_t *obmc2= obmc1+ (obmc_stride>>1);
2394 const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
2395 const uint8_t *obmc4= obmc3+ (obmc_stride>>1);
2396 dst = slice_buffer_get_line(sb, src_y + y);
2397 for(x=0; x<b_w; x++){
2398 int v= obmc1[x] * block[3][x + y*src_stride]
2399 +obmc2[x] * block[2][x + y*src_stride]
2400 +obmc3[x] * block[1][x + y*src_stride]
2401 +obmc4[x] * block[0][x + y*src_stride];
2403 v <<= 8 - LOG2_OBMC_MAX;
2405 v >>= 8 - FRAC_BITS;
2408 v += dst[x + src_x];
2409 v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS;
2410 if(v&(~255)) v= ~(v>>31);
2411 dst8[x + y*src_stride] = v;
2413 dst[x + src_x] -= v;
2419 //FIXME name cleanup (b_w, block_w, b_width stuff)
2420 static av_always_inline void add_yblock(SnowContext *s, int sliced, slice_buffer *sb, IDWTELEM *dst, uint8_t *dst8, const uint8_t *obmc, int src_x, int src_y, int b_w, int b_h, int w, int h, int dst_stride, int src_stride, int obmc_stride, int b_x, int b_y, int add, int offset_dst, int plane_index){
2421 const int b_width = s->b_width << s->block_max_depth;
2422 const int b_height= s->b_height << s->block_max_depth;
2423 const int b_stride= b_width;
2424 BlockNode *lt= &s->block[b_x + b_y*b_stride];
2425 BlockNode *rt= lt+1;
2426 BlockNode *lb= lt+b_stride;
2427 BlockNode *rb= lb+1;
2429 int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride;
2430 uint8_t tmp[src_stride*7*MB_SIZE]; //FIXME align
2437 }else if(b_x + 1 >= b_width){
2444 }else if(b_y + 1 >= b_height){
2449 if(src_x<0){ //FIXME merge with prev & always round internal width up to *16
2452 if(!sliced && !offset_dst)
2455 }else if(src_x + b_w > w){
2459 obmc -= src_y*obmc_stride;
2461 if(!sliced && !offset_dst)
2462 dst -= src_y*dst_stride;
2464 }else if(src_y + b_h> h){
2468 if(b_w<=0 || b_h<=0) return;
2470 assert(src_stride > 2*MB_SIZE + 5);
2472 if(!sliced && offset_dst)
2473 dst += src_x + src_y*dst_stride;
2474 dst8+= src_x + src_y*src_stride;
2475 // src += src_x + src_y*src_stride;
2477 ptmp= tmp + 3*tmp_step;
2480 pred_block(s, block[0], tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h);
2482 if(same_block(lt, rt)){
2487 pred_block(s, block[1], tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h);
2490 if(same_block(lt, lb)){
2492 }else if(same_block(rt, lb)){
2497 pred_block(s, block[2], tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h);
2500 if(same_block(lt, rb) ){
2502 }else if(same_block(rt, rb)){
2504 }else if(same_block(lb, rb)){
2508 pred_block(s, block[3], tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h);
2511 for(y=0; y<b_h; y++){
2512 for(x=0; x<b_w; x++){
2513 int v= obmc [x + y*obmc_stride] * block[3][x + y*src_stride] * (256/OBMC_MAX);
2514 if(add) dst[x + y*dst_stride] += v;
2515 else dst[x + y*dst_stride] -= v;
2518 for(y=0; y<b_h; y++){
2519 uint8_t *obmc2= obmc + (obmc_stride>>1);
2520 for(x=0; x<b_w; x++){
2521 int v= obmc2[x + y*obmc_stride] * block[2][x + y*src_stride] * (256/OBMC_MAX);
2522 if(add) dst[x + y*dst_stride] += v;
2523 else dst[x + y*dst_stride] -= v;
2526 for(y=0; y<b_h; y++){
2527 uint8_t *obmc3= obmc + obmc_stride*(obmc_stride>>1);
2528 for(x=0; x<b_w; x++){
2529 int v= obmc3[x + y*obmc_stride] * block[1][x + y*src_stride] * (256/OBMC_MAX);
2530 if(add) dst[x + y*dst_stride] += v;
2531 else dst[x + y*dst_stride] -= v;
2534 for(y=0; y<b_h; y++){
2535 uint8_t *obmc3= obmc + obmc_stride*(obmc_stride>>1);
2536 uint8_t *obmc4= obmc3+ (obmc_stride>>1);
2537 for(x=0; x<b_w; x++){
2538 int v= obmc4[x + y*obmc_stride] * block[0][x + y*src_stride] * (256/OBMC_MAX);
2539 if(add) dst[x + y*dst_stride] += v;
2540 else dst[x + y*dst_stride] -= v;
2545 s->dsp.inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
2547 for(y=0; y<b_h; y++){
2548 //FIXME ugly misuse of obmc_stride
2549 const uint8_t *obmc1= obmc + y*obmc_stride;
2550 const uint8_t *obmc2= obmc1+ (obmc_stride>>1);
2551 const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
2552 const uint8_t *obmc4= obmc3+ (obmc_stride>>1);
2553 for(x=0; x<b_w; x++){
2554 int v= obmc1[x] * block[3][x + y*src_stride]
2555 +obmc2[x] * block[2][x + y*src_stride]
2556 +obmc3[x] * block[1][x + y*src_stride]
2557 +obmc4[x] * block[0][x + y*src_stride];
2559 v <<= 8 - LOG2_OBMC_MAX;
2561 v >>= 8 - FRAC_BITS;
2564 v += dst[x + y*dst_stride];
2565 v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS;
2566 if(v&(~255)) v= ~(v>>31);
2567 dst8[x + y*src_stride] = v;
2569 dst[x + y*dst_stride] -= v;
2576 static av_always_inline void predict_slice_buffered(SnowContext *s, slice_buffer * sb, IDWTELEM * old_buffer, int plane_index, int add, int mb_y){
2577 Plane *p= &s->plane[plane_index];
2578 const int mb_w= s->b_width << s->block_max_depth;
2579 const int mb_h= s->b_height << s->block_max_depth;
2581 int block_size = MB_SIZE >> s->block_max_depth;
2582 int block_w = plane_index ? block_size/2 : block_size;
2583 const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth];
2584 int obmc_stride= plane_index ? block_size : 2*block_size;
2585 int ref_stride= s->current_picture.linesize[plane_index];
2586 uint8_t *dst8= s->current_picture.data[plane_index];
2590 if(s->keyframe || (s->avctx->debug&512)){
2595 for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++)
2597 // DWTELEM * line = slice_buffer_get_line(sb, y);
2598 IDWTELEM * line = sb->line[y];
2601 // int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
2602 int v= line[x] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
2604 if(v&(~255)) v= ~(v>>31);
2605 dst8[x + y*ref_stride]= v;
2609 for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++)
2611 // DWTELEM * line = slice_buffer_get_line(sb, y);
2612 IDWTELEM * line = sb->line[y];
2615 line[x] -= 128 << FRAC_BITS;
2616 // buf[x + y*w]-= 128<<FRAC_BITS;
2624 for(mb_x=0; mb_x<=mb_w; mb_x++){
2625 add_yblock(s, 1, sb, old_buffer, dst8, obmc,
2626 block_w*mb_x - block_w/2,
2627 block_w*mb_y - block_w/2,
2630 w, ref_stride, obmc_stride,
2632 add, 0, plane_index);
2636 static av_always_inline void predict_slice(SnowContext *s, IDWTELEM *buf, int plane_index, int add, int mb_y){
2637 Plane *p= &s->plane[plane_index];
2638 const int mb_w= s->b_width << s->block_max_depth;
2639 const int mb_h= s->b_height << s->block_max_depth;
2641 int block_size = MB_SIZE >> s->block_max_depth;
2642 int block_w = plane_index ? block_size/2 : block_size;
2643 const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth];
2644 const int obmc_stride= plane_index ? block_size : 2*block_size;
2645 int ref_stride= s->current_picture.linesize[plane_index];
2646 uint8_t *dst8= s->current_picture.data[plane_index];
2650 if(s->keyframe || (s->avctx->debug&512)){
2655 for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
2657 int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
2659 if(v&(~255)) v= ~(v>>31);
2660 dst8[x + y*ref_stride]= v;
2664 for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
2666 buf[x + y*w]-= 128<<FRAC_BITS;
2674 for(mb_x=0; mb_x<=mb_w; mb_x++){
2675 add_yblock(s, 0, NULL, buf, dst8, obmc,
2676 block_w*mb_x - block_w/2,
2677 block_w*mb_y - block_w/2,
2680 w, ref_stride, obmc_stride,
2682 add, 1, plane_index);
2686 static av_always_inline void predict_plane(SnowContext *s, IDWTELEM *buf, int plane_index, int add){
2687 const int mb_h= s->b_height << s->block_max_depth;
2689 for(mb_y=0; mb_y<=mb_h; mb_y++)
2690 predict_slice(s, buf, plane_index, add, mb_y);
2693 static int get_dc(SnowContext *s, int mb_x, int mb_y, int plane_index){
2695 Plane *p= &s->plane[plane_index];
2696 const int block_size = MB_SIZE >> s->block_max_depth;
2697 const int block_w = plane_index ? block_size/2 : block_size;
2698 const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth];
2699 const int obmc_stride= plane_index ? block_size : 2*block_size;
2700 const int ref_stride= s->current_picture.linesize[plane_index];
2701 uint8_t *src= s-> input_picture.data[plane_index];
2702 IDWTELEM *dst= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4; //FIXME change to unsigned
2703 const int b_stride = s->b_width << s->block_max_depth;
2704 const int w= p->width;
2705 const int h= p->height;
2706 int index= mb_x + mb_y*b_stride;
2707 BlockNode *b= &s->block[index];
2708 BlockNode backup= *b;
2712 b->type|= BLOCK_INTRA;
2713 b->color[plane_index]= 0;
2714 memset(dst, 0, obmc_stride*obmc_stride*sizeof(IDWTELEM));
2717 int mb_x2= mb_x + (i &1) - 1;
2718 int mb_y2= mb_y + (i>>1) - 1;
2719 int x= block_w*mb_x2 + block_w/2;
2720 int y= block_w*mb_y2 + block_w/2;
2722 add_yblock(s, 0, NULL, dst + ((i&1)+(i>>1)*obmc_stride)*block_w, NULL, obmc,
2723 x, y, block_w, block_w, w, h, obmc_stride, ref_stride, obmc_stride, mb_x2, mb_y2, 0, 0, plane_index);
2725 for(y2= FFMAX(y, 0); y2<FFMIN(h, y+block_w); y2++){
2726 for(x2= FFMAX(x, 0); x2<FFMIN(w, x+block_w); x2++){
2727 int index= x2-(block_w*mb_x - block_w/2) + (y2-(block_w*mb_y - block_w/2))*obmc_stride;
2728 int obmc_v= obmc[index];
2730 if(y<0) obmc_v += obmc[index + block_w*obmc_stride];
2731 if(x<0) obmc_v += obmc[index + block_w];
2732 if(y+block_w>h) obmc_v += obmc[index - block_w*obmc_stride];
2733 if(x+block_w>w) obmc_v += obmc[index - block_w];
2734 //FIXME precalculate this or simplify it somehow else
2736 d = -dst[index] + (1<<(FRAC_BITS-1));
2738 ab += (src[x2 + y2*ref_stride] - (d>>FRAC_BITS)) * obmc_v;
2739 aa += obmc_v * obmc_v; //FIXME precalculate this
2745 return av_clip(((ab<<LOG2_OBMC_MAX) + aa/2)/aa, 0, 255); //FIXME we should not need clipping
2748 static inline int get_block_bits(SnowContext *s, int x, int y, int w){
2749 const int b_stride = s->b_width << s->block_max_depth;
2750 const int b_height = s->b_height<< s->block_max_depth;
2751 int index= x + y*b_stride;
2752 const BlockNode *b = &s->block[index];
2753 const BlockNode *left = x ? &s->block[index-1] : &null_block;
2754 const BlockNode *top = y ? &s->block[index-b_stride] : &null_block;
2755 const BlockNode *tl = y && x ? &s->block[index-b_stride-1] : left;
2756 const BlockNode *tr = y && x+w<b_stride ? &s->block[index-b_stride+w] : tl;
2758 // int mx_context= av_log2(2*FFABS(left->mx - top->mx));
2759 // int my_context= av_log2(2*FFABS(left->my - top->my));
2761 if(x<0 || x>=b_stride || y>=b_height)
2768 00001XXXX 15-30 8-15
2770 //FIXME try accurate rate
2771 //FIXME intra and inter predictors if surrounding blocks are not the same type
2772 if(b->type & BLOCK_INTRA){
2773 return 3+2*( av_log2(2*FFABS(left->color[0] - b->color[0]))
2774 + av_log2(2*FFABS(left->color[1] - b->color[1]))
2775 + av_log2(2*FFABS(left->color[2] - b->color[2])));
2777 pred_mv(s, &dmx, &dmy, b->ref, left, top, tr);
2780 return 2*(1 + av_log2(2*FFABS(dmx)) //FIXME kill the 2* can be merged in lambda
2781 + av_log2(2*FFABS(dmy))
2782 + av_log2(2*b->ref));
2786 static int get_block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index, const uint8_t *obmc_edged){
2787 Plane *p= &s->plane[plane_index];
2788 const int block_size = MB_SIZE >> s->block_max_depth;
2789 const int block_w = plane_index ? block_size/2 : block_size;
2790 const int obmc_stride= plane_index ? block_size : 2*block_size;
2791 const int ref_stride= s->current_picture.linesize[plane_index];
2792 uint8_t *dst= s->current_picture.data[plane_index];
2793 uint8_t *src= s-> input_picture.data[plane_index];
2794 IDWTELEM *pred= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4;
2795 uint8_t cur[ref_stride*2*MB_SIZE]; //FIXME alignment
2796 uint8_t tmp[ref_stride*(2*MB_SIZE+HTAPS_MAX-1)];
2797 const int b_stride = s->b_width << s->block_max_depth;
2798 const int b_height = s->b_height<< s->block_max_depth;
2799 const int w= p->width;
2800 const int h= p->height;
2803 const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp);
2804 int sx= block_w*mb_x - block_w/2;
2805 int sy= block_w*mb_y - block_w/2;
2806 int x0= FFMAX(0,-sx);
2807 int y0= FFMAX(0,-sy);
2808 int x1= FFMIN(block_w*2, w-sx);
2809 int y1= FFMIN(block_w*2, h-sy);
2812 pred_block(s, cur, tmp, ref_stride, sx, sy, block_w*2, block_w*2, &s->block[mb_x + mb_y*b_stride], plane_index, w, h);
2814 for(y=y0; y<y1; y++){
2815 const uint8_t *obmc1= obmc_edged + y*obmc_stride;
2816 const IDWTELEM *pred1 = pred + y*obmc_stride;
2817 uint8_t *cur1 = cur + y*ref_stride;
2818 uint8_t *dst1 = dst + sx + (sy+y)*ref_stride;
2819 for(x=x0; x<x1; x++){
2820 #if FRAC_BITS >= LOG2_OBMC_MAX
2821 int v = (cur1[x] * obmc1[x]) << (FRAC_BITS - LOG2_OBMC_MAX);
2823 int v = (cur1[x] * obmc1[x] + (1<<(LOG2_OBMC_MAX - FRAC_BITS-1))) >> (LOG2_OBMC_MAX - FRAC_BITS);
2825 v = (v + pred1[x]) >> FRAC_BITS;
2826 if(v&(~255)) v= ~(v>>31);
2831 /* copy the regions where obmc[] = (uint8_t)256 */
2832 if(LOG2_OBMC_MAX == 8
2833 && (mb_x == 0 || mb_x == b_stride-1)
2834 && (mb_y == 0 || mb_y == b_height-1)){
2843 for(y=y0; y<y1; y++)
2844 memcpy(dst + sx+x0 + (sy+y)*ref_stride, cur + x0 + y*ref_stride, x1-x0);
2848 /* FIXME rearrange dsputil to fit 32x32 cmp functions */
2849 /* FIXME check alignment of the cmp wavelet vs the encoding wavelet */
2850 /* FIXME cmps overlap but do not cover the wavelet's whole support.
2851 * So improving the score of one block is not strictly guaranteed
2852 * to improve the score of the whole frame, thus iterative motion
2853 * estimation does not always converge. */
2854 if(s->avctx->me_cmp == FF_CMP_W97)
2855 distortion = w97_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32);
2856 else if(s->avctx->me_cmp == FF_CMP_W53)
2857 distortion = w53_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32);
2861 int off = sx+16*(i&1) + (sy+16*(i>>1))*ref_stride;
2862 distortion += s->dsp.me_cmp[0](&s->m, src + off, dst + off, ref_stride, 16);
2867 distortion = s->dsp.me_cmp[0](&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, block_w*2);
2876 rate += get_block_bits(s, mb_x + (i&1) - (i>>1), mb_y + (i>>1), 1);
2878 if(mb_x == b_stride-2)
2879 rate += get_block_bits(s, mb_x + 1, mb_y + 1, 1);
2881 return distortion + rate*penalty_factor;
2884 static int get_4block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index){
2886 Plane *p= &s->plane[plane_index];
2887 const int block_size = MB_SIZE >> s->block_max_depth;
2888 const int block_w = plane_index ? block_size/2 : block_size;
2889 const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth];
2890 const int obmc_stride= plane_index ? block_size : 2*block_size;
2891 const int ref_stride= s->current_picture.linesize[plane_index];
2892 uint8_t *dst= s->current_picture.data[plane_index];
2893 uint8_t *src= s-> input_picture.data[plane_index];
2894 //FIXME zero_dst is const but add_yblock changes dst if add is 0 (this is never the case for dst=zero_dst
2895 // const has only been removed from zero_dst to suppress a warning
2896 static IDWTELEM zero_dst[4096]; //FIXME
2897 const int b_stride = s->b_width << s->block_max_depth;
2898 const int w= p->width;
2899 const int h= p->height;
2902 const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp);
2905 int mb_x2= mb_x + (i%3) - 1;
2906 int mb_y2= mb_y + (i/3) - 1;
2907 int x= block_w*mb_x2 + block_w/2;
2908 int y= block_w*mb_y2 + block_w/2;
2910 add_yblock(s, 0, NULL, zero_dst, dst, obmc,
2911 x, y, block_w, block_w, w, h, /*dst_stride*/0, ref_stride, obmc_stride, mb_x2, mb_y2, 1, 1, plane_index);
2913 //FIXME find a cleaner/simpler way to skip the outside stuff
2914 for(y2= y; y2<0; y2++)
2915 memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w);
2916 for(y2= h; y2<y+block_w; y2++)
2917 memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w);
2919 for(y2= y; y2<y+block_w; y2++)
2920 memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, -x);
2923 for(y2= y; y2<y+block_w; y2++)
2924 memcpy(dst + w + y2*ref_stride, src + w + y2*ref_stride, x+block_w - w);
2927 assert(block_w== 8 || block_w==16);
2928 distortion += s->dsp.me_cmp[block_w==8](&s->m, src + x + y*ref_stride, dst + x + y*ref_stride, ref_stride, block_w);
2932 BlockNode *b= &s->block[mb_x+mb_y*b_stride];
2933 int merged= same_block(b,b+1) && same_block(b,b+b_stride) && same_block(b,b+b_stride+1);
2941 rate = get_block_bits(s, mb_x, mb_y, 2);
2942 for(i=merged?4:0; i<9; i++){
2943 static const int dxy[9][2] = {{0,0},{1,0},{0,1},{1,1},{2,0},{2,1},{-1,2},{0,2},{1,2}};
2944 rate += get_block_bits(s, mb_x + dxy[i][0], mb_y + dxy[i][1], 1);
2947 return distortion + rate*penalty_factor;
2950 static av_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){
2951 const int b_stride= s->b_width << s->block_max_depth;
2952 BlockNode *block= &s->block[mb_x + mb_y * b_stride];
2953 BlockNode backup= *block;
2954 int rd, index, value;
2956 assert(mb_x>=0 && mb_y>=0);
2957 assert(mb_x<b_stride);
2960 block->color[0] = p[0];
2961 block->color[1] = p[1];
2962 block->color[2] = p[2];
2963 block->type |= BLOCK_INTRA;
2965 index= (p[0] + 31*p[1]) & (ME_CACHE_SIZE-1);
2966 value= s->me_cache_generation + (p[0]>>10) + (p[1]<<6) + (block->ref<<12);
2967 if(s->me_cache[index] == value)
2969 s->me_cache[index]= value;
2973 block->type &= ~BLOCK_INTRA;
2976 rd= get_block_rd(s, mb_x, mb_y, 0, obmc_edged);
2988 /* special case for int[2] args we discard afterwards,
2989 * fixes compilation problem with gcc 2.95 */
2990 static av_always_inline int check_block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, const uint8_t *obmc_edged, int *best_rd){
2991 int p[2] = {p0, p1};
2992 return check_block(s, mb_x, mb_y, p, 0, obmc_edged, best_rd);
2995 static av_always_inline int check_4block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, int ref, int *best_rd){
2996 const int b_stride= s->b_width << s->block_max_depth;
2997 BlockNode *block= &s->block[mb_x + mb_y * b_stride];
2998 BlockNode backup[4]= {block[0], block[1], block[b_stride], block[b_stride+1]};
2999 int rd, index, value;
3001 assert(mb_x>=0 && mb_y>=0);
3002 assert(mb_x<b_stride);
3003 assert(((mb_x|mb_y)&1) == 0);
3005 index= (p0 + 31*p1) & (ME_CACHE_SIZE-1);
3006 value= s->me_cache_generation + (p0>>10) + (p1<<6) + (block->ref<<12);
3007 if(s->me_cache[index] == value)
3009 s->me_cache[index]= value;
3014 block->type &= ~BLOCK_INTRA;
3015 block[1]= block[b_stride]= block[b_stride+1]= *block;
3017 rd= get_4block_rd(s, mb_x, mb_y, 0);
3024 block[0]= backup[0];
3025 block[1]= backup[1];
3026 block[b_stride]= backup[2];
3027 block[b_stride+1]= backup[3];
3032 static void iterative_me(SnowContext *s){
3033 int pass, mb_x, mb_y;
3034 const int b_width = s->b_width << s->block_max_depth;
3035 const int b_height= s->b_height << s->block_max_depth;
3036 const int b_stride= b_width;
3040 RangeCoder r = s->c;
3041 uint8_t state[sizeof(s->block_state)];
3042 memcpy(state, s->block_state, sizeof(s->block_state));
3043 for(mb_y= 0; mb_y<s->b_height; mb_y++)
3044 for(mb_x= 0; mb_x<s->b_width; mb_x++)
3045 encode_q_branch(s, 0, mb_x, mb_y);
3047 memcpy(s->block_state, state, sizeof(s->block_state));
3050 for(pass=0; pass<25; pass++){
3053 for(mb_y= 0; mb_y<b_height; mb_y++){
3054 for(mb_x= 0; mb_x<b_width; mb_x++){
3055 int dia_change, i, j, ref;
3056 int best_rd= INT_MAX, ref_rd;
3057 BlockNode backup, ref_b;
3058 const int index= mb_x + mb_y * b_stride;
3059 BlockNode *block= &s->block[index];
3060 BlockNode *tb = mb_y ? &s->block[index-b_stride ] : NULL;
3061 BlockNode *lb = mb_x ? &s->block[index -1] : NULL;
3062 BlockNode *rb = mb_x+1<b_width ? &s->block[index +1] : NULL;
3063 BlockNode *bb = mb_y+1<b_height ? &s->block[index+b_stride ] : NULL;
3064 BlockNode *tlb= mb_x && mb_y ? &s->block[index-b_stride-1] : NULL;
3065 BlockNode *trb= mb_x+1<b_width && mb_y ? &s->block[index-b_stride+1] : NULL;
3066 BlockNode *blb= mb_x && mb_y+1<b_height ? &s->block[index+b_stride-1] : NULL;
3067 BlockNode *brb= mb_x+1<b_width && mb_y+1<b_height ? &s->block[index+b_stride+1] : NULL;
3068 const int b_w= (MB_SIZE >> s->block_max_depth);
3069 uint8_t obmc_edged[b_w*2][b_w*2];
3071 if(pass && (block->type & BLOCK_OPT))
3073 block->type |= BLOCK_OPT;
3077 if(!s->me_cache_generation)
3078 memset(s->me_cache, 0, sizeof(s->me_cache));
3079 s->me_cache_generation += 1<<22;
3081 //FIXME precalculate
3084 memcpy(obmc_edged, obmc_tab[s->block_max_depth], b_w*b_w*4);
3086 for(y=0; y<b_w*2; y++)
3087 memset(obmc_edged[y], obmc_edged[y][0] + obmc_edged[y][b_w-1], b_w);
3088 if(mb_x==b_stride-1)
3089 for(y=0; y<b_w*2; y++)
3090 memset(obmc_edged[y]+b_w, obmc_edged[y][b_w] + obmc_edged[y][b_w*2-1], b_w);
3092 for(x=0; x<b_w*2; x++)
3093 obmc_edged[0][x] += obmc_edged[b_w-1][x];
3094 for(y=1; y<b_w; y++)
3095 memcpy(obmc_edged[y], obmc_edged[0], b_w*2);
3097 if(mb_y==b_height-1){
3098 for(x=0; x<b_w*2; x++)
3099 obmc_edged[b_w*2-1][x] += obmc_edged[b_w][x];
3100 for(y=b_w; y<b_w*2-1; y++)
3101 memcpy(obmc_edged[y], obmc_edged[b_w*2-1], b_w*2);
3105 //skip stuff outside the picture
3106 if(mb_x==0 || mb_y==0 || mb_x==b_width-1 || mb_y==b_height-1)
3108 uint8_t *src= s-> input_picture.data[0];
3109 uint8_t *dst= s->current_picture.data[0];
3110 const int stride= s->current_picture.linesize[0];
3111 const int block_w= MB_SIZE >> s->block_max_depth;
3112 const int sx= block_w*mb_x - block_w/2;
3113 const int sy= block_w*mb_y - block_w/2;
3114 const int w= s->plane[0].width;
3115 const int h= s->plane[0].height;
3119 memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2);
3120 for(y=h; y<sy+block_w*2; y++)
3121 memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2);
3123 for(y=sy; y<sy+block_w*2; y++)
3124 memcpy(dst + sx + y*stride, src + sx + y*stride, -sx);
3126 if(sx+block_w*2 > w){
3127 for(y=sy; y<sy+block_w*2; y++)
3128 memcpy(dst + w + y*stride, src + w + y*stride, sx+block_w*2 - w);
3132 // intra(black) = neighbors' contribution to the current block
3134 color[i]= get_dc(s, mb_x, mb_y, i);
3136 // get previous score (cannot be cached due to OBMC)
3137 if(pass > 0 && (block->type&BLOCK_INTRA)){
3138 int color0[3]= {block->color[0], block->color[1], block->color[2]};
3139 check_block(s, mb_x, mb_y, color0, 1, *obmc_edged, &best_rd);
3141 check_block_inter(s, mb_x, mb_y, block->mx, block->my, *obmc_edged, &best_rd);
3145 for(ref=0; ref < s->ref_frames; ref++){
3146 int16_t (*mvr)[2]= &s->ref_mvs[ref][index];
3147 if(s->ref_scores[ref][index] > s->ref_scores[ref_b.ref][index]*3/2) //FIXME tune threshold
3152 check_block_inter(s, mb_x, mb_y, mvr[0][0], mvr[0][1], *obmc_edged, &best_rd);
3153 check_block_inter(s, mb_x, mb_y, 0, 0, *obmc_edged, &best_rd);
3155 check_block_inter(s, mb_x, mb_y, mvr[-b_stride][0], mvr[-b_stride][1], *obmc_edged, &best_rd);
3157 check_block_inter(s, mb_x, mb_y, mvr[-1][0], mvr[-1][1], *obmc_edged, &best_rd);
3159 check_block_inter(s, mb_x, mb_y, mvr[1][0], mvr[1][1], *obmc_edged, &best_rd);
3161 check_block_inter(s, mb_x, mb_y, mvr[b_stride][0], mvr[b_stride][1], *obmc_edged, &best_rd);
3164 //FIXME avoid subpel interpolation / round to nearest integer
3167 for(i=0; i<FFMAX(s->avctx->dia_size, 1); i++){
3169 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my+(4*j), *obmc_edged, &best_rd);
3170 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my-(4*j), *obmc_edged, &best_rd);
3171 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my-(4*j), *obmc_edged, &best_rd);
3172 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my+(4*j), *obmc_edged, &best_rd);
3178 static const int square[8][2]= {{+1, 0},{-1, 0},{ 0,+1},{ 0,-1},{+1,+1},{-1,-1},{+1,-1},{-1,+1},};
3181 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+square[i][0], block->my+square[i][1], *obmc_edged, &best_rd);
3183 //FIXME or try the standard 2 pass qpel or similar
3185 mvr[0][0]= block->mx;
3186 mvr[0][1]= block->my;
3187 if(ref_rd > best_rd){
3195 check_block(s, mb_x, mb_y, color, 1, *obmc_edged, &best_rd);
3196 //FIXME RD style color selection
3198 if(!same_block(block, &backup)){
3199 if(tb ) tb ->type &= ~BLOCK_OPT;
3200 if(lb ) lb ->type &= ~BLOCK_OPT;
3201 if(rb ) rb ->type &= ~BLOCK_OPT;
3202 if(bb ) bb ->type &= ~BLOCK_OPT;
3203 if(tlb) tlb->type &= ~BLOCK_OPT;
3204 if(trb) trb->type &= ~BLOCK_OPT;
3205 if(blb) blb->type &= ~BLOCK_OPT;
3206 if(brb) brb->type &= ~BLOCK_OPT;
3211 av_log(NULL, AV_LOG_ERROR, "pass:%d changed:%d\n", pass, change);
3216 if(s->block_max_depth == 1){
3218 for(mb_y= 0; mb_y<b_height; mb_y+=2){
3219 for(mb_x= 0; mb_x<b_width; mb_x+=2){
3221 int best_rd, init_rd;
3222 const int index= mb_x + mb_y * b_stride;
3225 b[0]= &s->block[index];
3227 b[2]= b[0]+b_stride;
3229 if(same_block(b[0], b[1]) &&
3230 same_block(b[0], b[2]) &&
3231 same_block(b[0], b[3]))
3234 if(!s->me_cache_generation)
3235 memset(s->me_cache, 0, sizeof(s->me_cache));
3236 s->me_cache_generation += 1<<22;
3238 init_rd= best_rd= get_4block_rd(s, mb_x, mb_y, 0);
3240 //FIXME more multiref search?
3241 check_4block_inter(s, mb_x, mb_y,
3242 (b[0]->mx + b[1]->mx + b[2]->mx + b[3]->mx + 2) >> 2,
3243 (b[0]->my + b[1]->my + b[2]->my + b[3]->my + 2) >> 2, 0, &best_rd);
3246 if(!(b[i]->type&BLOCK_INTRA))
3247 check_4block_inter(s, mb_x, mb_y, b[i]->mx, b[i]->my, b[i]->ref, &best_rd);
3249 if(init_rd != best_rd)
3253 av_log(NULL, AV_LOG_ERROR, "pass:4mv changed:%d\n", change*4);
3257 static void quantize(SnowContext *s, SubBand *b, IDWTELEM *dst, DWTELEM *src, int stride, int bias){
3258 const int w= b->width;
3259 const int h= b->height;
3260 const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
3261 const int qmul= qexp[qlog&(QROOT-1)]<<((qlog>>QSHIFT) + ENCODER_EXTRA_BITS);
3262 int x,y, thres1, thres2;
3264 if(s->qlog == LOSSLESS_QLOG){
3267 dst[x + y*stride]= src[x + y*stride];
3271 bias= bias ? 0 : (3*qmul)>>3;
3272 thres1= ((qmul - bias)>>QEXPSHIFT) - 1;
3278 int i= src[x + y*stride];
3280 if((unsigned)(i+thres1) > thres2){
3283 i/= qmul; //FIXME optimize
3284 dst[x + y*stride]= i;
3288 i/= qmul; //FIXME optimize
3289 dst[x + y*stride]= -i;
3292 dst[x + y*stride]= 0;
3298 int i= src[x + y*stride];
3300 if((unsigned)(i+thres1) > thres2){
3303 i= (i + bias) / qmul; //FIXME optimize
3304 dst[x + y*stride]= i;
3308 i= (i + bias) / qmul; //FIXME optimize
3309 dst[x + y*stride]= -i;
3312 dst[x + y*stride]= 0;
3318 static void dequantize_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, IDWTELEM *src, int stride, int start_y, int end_y){
3319 const int w= b->width;
3320 const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
3321 const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
3322 const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
3325 if(s->qlog == LOSSLESS_QLOG) return;
3327 for(y=start_y; y<end_y; y++){
3328 // DWTELEM * line = slice_buffer_get_line_from_address(sb, src + (y * stride));
3329 IDWTELEM * line = slice_buffer_get_line(sb, (y * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;
3333 line[x]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias
3335 line[x]= (( i*qmul + qadd)>>(QEXPSHIFT));
3341 static void dequantize(SnowContext *s, SubBand *b, IDWTELEM *src, int stride){
3342 const int w= b->width;
3343 const int h= b->height;
3344 const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
3345 const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
3346 const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
3349 if(s->qlog == LOSSLESS_QLOG) return;
3353 int i= src[x + y*stride];
3355 src[x + y*stride]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias
3357 src[x + y*stride]= (( i*qmul + qadd)>>(QEXPSHIFT));
3363 static void decorrelate(SnowContext *s, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median){
3364 const int w= b->width;
3365 const int h= b->height;
3368 for(y=h-1; y>=0; y--){
3369 for(x=w-1; x>=0; x--){
3370 int i= x + y*stride;
3374 if(y && x+1<w) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]);
3375 else src[i] -= src[i - 1];
3377 if(y) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]);
3378 else src[i] -= src[i - 1];
3381 if(y) src[i] -= src[i - stride];
3387 static void correlate_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median, int start_y, int end_y){
3388 const int w= b->width;
3391 IDWTELEM * line=0; // silence silly "could be used without having been initialized" warning
3395 line = slice_buffer_get_line(sb, ((start_y - 1) * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;
3397 for(y=start_y; y<end_y; y++){
3399 // line = slice_buffer_get_line_from_address(sb, src + (y * stride));
3400 line = slice_buffer_get_line(sb, (y * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;
3404 if(y && x+1<w) line[x] += mid_pred(line[x - 1], prev[x], prev[x + 1]);
3405 else line[x] += line[x - 1];
3407 if(y) line[x] += mid_pred(line[x - 1], prev[x], line[x - 1] + prev[x] - prev[x - 1]);
3408 else line[x] += line[x - 1];
3411 if(y) line[x] += prev[x];
3417 static void correlate(SnowContext *s, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median){
3418 const int w= b->width;
3419 const int h= b->height;
3424 int i= x + y*stride;
3428 if(y && x+1<w) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]);
3429 else src[i] += src[i - 1];
3431 if(y) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]);
3432 else src[i] += src[i - 1];
3435 if(y) src[i] += src[i - stride];
3441 static void encode_qlogs(SnowContext *s){
3442 int plane_index, level, orientation;
3444 for(plane_index=0; plane_index<2; plane_index++){
3445 for(level=0; level<s->spatial_decomposition_count; level++){
3446 for(orientation=level ? 1:0; orientation<4; orientation++){
3447 if(orientation==2) continue;
3448 put_symbol(&s->c, s->header_state, s->plane[plane_index].band[level][orientation].qlog, 1);
3454 static void encode_header(SnowContext *s){
3458 memset(kstate, MID_STATE, sizeof(kstate));
3460 put_rac(&s->c, kstate, s->keyframe);
3461 if(s->keyframe || s->always_reset){
3463 s->last_spatial_decomposition_type=
3467 s->last_block_max_depth= 0;
3468 for(plane_index=0; plane_index<2; plane_index++){
3469 Plane *p= &s->plane[plane_index];
3472 memset(p->last_hcoeff, 0, sizeof(p->last_hcoeff));
3476 put_symbol(&s->c, s->header_state, s->version, 0);
3477 put_rac(&s->c, s->header_state, s->always_reset);
3478 put_symbol(&s->c, s->header_state, s->temporal_decomposition_type, 0);
3479 put_symbol(&s->c, s->header_state, s->temporal_decomposition_count, 0);
3480 put_symbol(&s->c, s->header_state, s->spatial_decomposition_count, 0);
3481 put_symbol(&s->c, s->header_state, s->colorspace_type, 0);
3482 put_symbol(&s->c, s->header_state, s->chroma_h_shift, 0);
3483 put_symbol(&s->c, s->header_state, s->chroma_v_shift, 0);
3484 put_rac(&s->c, s->header_state, s->spatial_scalability);
3485 // put_rac(&s->c, s->header_state, s->rate_scalability);
3486 put_symbol(&s->c, s->header_state, s->max_ref_frames-1, 0);
3493 for(plane_index=0; plane_index<2; plane_index++){
3494 Plane *p= &s->plane[plane_index];
3495 update_mc |= p->last_htaps != p->htaps;
3496 update_mc |= p->last_diag_mc != p->diag_mc;
3497 update_mc |= !!memcmp(p->last_hcoeff, p->hcoeff, sizeof(p->hcoeff));
3499 put_rac(&s->c, s->header_state, update_mc);
3501 for(plane_index=0; plane_index<2; plane_index++){
3502 Plane *p= &s->plane[plane_index];
3503 put_rac(&s->c, s->header_state, p->diag_mc);
3504 put_symbol(&s->c, s->header_state, p->htaps/2-1, 0);
3505 for(i= p->htaps/2; i; i--)
3506 put_symbol(&s->c, s->header_state, FFABS(p->hcoeff[i]), 0);
3509 if(s->last_spatial_decomposition_count != s->spatial_decomposition_count){
3510 put_rac(&s->c, s->header_state, 1);
3511 put_symbol(&s->c, s->header_state, s->spatial_decomposition_count, 0);
3514 put_rac(&s->c, s->header_state, 0);
3517 put_symbol(&s->c, s->header_state, s->spatial_decomposition_type - s->last_spatial_decomposition_type, 1);
3518 put_symbol(&s->c, s->header_state, s->qlog - s->last_qlog , 1);
3519 put_symbol(&s->c, s->header_state, s->mv_scale - s->last_mv_scale, 1);
3520 put_symbol(&s->c, s->header_state, s->qbias - s->last_qbias , 1);
3521 put_symbol(&s->c, s->header_state, s->block_max_depth - s->last_block_max_depth, 1);
3525 static void update_last_header_values(SnowContext *s){
3529 for(plane_index=0; plane_index<2; plane_index++){
3530 Plane *p= &s->plane[plane_index];
3531 p->last_diag_mc= p->diag_mc;
3532 p->last_htaps = p->htaps;
3533 memcpy(p->last_hcoeff, p->hcoeff, sizeof(p->hcoeff));
3537 s->last_spatial_decomposition_type = s->spatial_decomposition_type;
3538 s->last_qlog = s->qlog;
3539 s->last_qbias = s->qbias;
3540 s->last_mv_scale = s->mv_scale;
3541 s->last_block_max_depth = s->block_max_depth;
3542 s->last_spatial_decomposition_count = s->spatial_decomposition_count;
3545 static void decode_qlogs(SnowContext *s){
3546 int plane_index, level, orientation;
3548 for(plane_index=0; plane_index<3; plane_index++){
3549 for(level=0; level<s->spatial_decomposition_count; level++){
3550 for(orientation=level ? 1:0; orientation<4; orientation++){
3552 if (plane_index==2) q= s->plane[1].band[level][orientation].qlog;
3553 else if(orientation==2) q= s->plane[plane_index].band[level][1].qlog;
3554 else q= get_symbol(&s->c, s->header_state, 1);
3555 s->plane[plane_index].band[level][orientation].qlog= q;
3561 static int decode_header(SnowContext *s){
3565 memset(kstate, MID_STATE, sizeof(kstate));
3567 s->keyframe= get_rac(&s->c, kstate);
3568 if(s->keyframe || s->always_reset){
3570 s->spatial_decomposition_type=
3574 s->block_max_depth= 0;
3577 s->version= get_symbol(&s->c, s->header_state, 0);
3579 av_log(s->avctx, AV_LOG_ERROR, "version %d not supported", s->version);
3582 s->always_reset= get_rac(&s->c, s->header_state);
3583 s->temporal_decomposition_type= get_symbol(&s->c, s->header_state, 0);
3584 s->temporal_decomposition_count= get_symbol(&s->c, s->header_state, 0);
3585 s->spatial_decomposition_count= get_symbol(&s->c, s->header_state, 0);
3586 s->colorspace_type= get_symbol(&s->c, s->header_state, 0);
3587 s->chroma_h_shift= get_symbol(&s->c, s->header_state, 0);
3588 s->chroma_v_shift= get_symbol(&s->c, s->header_state, 0);
3589 s->spatial_scalability= get_rac(&s->c, s->header_state);
3590 // s->rate_scalability= get_rac(&s->c, s->header_state);
3591 s->max_ref_frames= get_symbol(&s->c, s->header_state, 0)+1;
3597 if(get_rac(&s->c, s->header_state)){
3598 for(plane_index=0; plane_index<2; plane_index++){
3599 int htaps, i, sum=0;
3600 Plane *p= &s->plane[plane_index];
3601 p->diag_mc= get_rac(&s->c, s->header_state);
3602 htaps= get_symbol(&s->c, s->header_state, 0)*2 + 2;
3603 if((unsigned)htaps > HTAPS_MAX || htaps==0)
3606 for(i= htaps/2; i; i--){
3607 p->hcoeff[i]= get_symbol(&s->c, s->header_state, 0) * (1-2*(i&1));
3608 sum += p->hcoeff[i];
3610 p->hcoeff[0]= 32-sum;
3612 s->plane[2].diag_mc= s->plane[1].diag_mc;
3613 s->plane[2].htaps = s->plane[1].htaps;
3614 memcpy(s->plane[2].hcoeff, s->plane[1].hcoeff, sizeof(s->plane[1].hcoeff));
3616 if(get_rac(&s->c, s->header_state)){
3617 s->spatial_decomposition_count= get_symbol(&s->c, s->header_state, 0);
3622 s->spatial_decomposition_type+= get_symbol(&s->c, s->header_state, 1);
3623 if(s->spatial_decomposition_type > 1){
3624 av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_type %d not supported", s->spatial_decomposition_type);
3628 s->qlog += get_symbol(&s->c, s->header_state, 1);
3629 s->mv_scale += get_symbol(&s->c, s->header_state, 1);
3630 s->qbias += get_symbol(&s->c, s->header_state, 1);
3631 s->block_max_depth+= get_symbol(&s->c, s->header_state, 1);
3632 if(s->block_max_depth > 1 || s->block_max_depth < 0){
3633 av_log(s->avctx, AV_LOG_ERROR, "block_max_depth= %d is too large", s->block_max_depth);
3634 s->block_max_depth= 0;
3641 static void init_qexp(void){
3645 for(i=0; i<QROOT; i++){
3647 v *= pow(2, 1.0 / QROOT);
3651 static int common_init(AVCodecContext *avctx){
3652 SnowContext *s = avctx->priv_data;
3658 dsputil_init(&s->dsp, avctx);
3661 s->dsp.put_qpel_pixels_tab [0][dy+dx/4]=\
3662 s->dsp.put_no_rnd_qpel_pixels_tab[0][dy+dx/4]=\
3663 s->dsp.put_h264_qpel_pixels_tab[0][dy+dx/4];\
3664 s->dsp.put_qpel_pixels_tab [1][dy+dx/4]=\
3665 s->dsp.put_no_rnd_qpel_pixels_tab[1][dy+dx/4]=\
3666 s->dsp.put_h264_qpel_pixels_tab[1][dy+dx/4];
3685 #define mcfh(dx,dy)\
3686 s->dsp.put_pixels_tab [0][dy/4+dx/8]=\
3687 s->dsp.put_no_rnd_pixels_tab[0][dy/4+dx/8]=\
3688 mc_block_hpel ## dx ## dy ## 16;\
3689 s->dsp.put_pixels_tab [1][dy/4+dx/8]=\
3690 s->dsp.put_no_rnd_pixels_tab[1][dy/4+dx/8]=\
3691 mc_block_hpel ## dx ## dy ## 8;
3701 // dec += FFMAX(s->chroma_h_shift, s->chroma_v_shift);
3703 width= s->avctx->width;
3704 height= s->avctx->height;
3706 s->spatial_idwt_buffer= av_mallocz(width*height*sizeof(IDWTELEM));
3707 s->spatial_dwt_buffer= av_mallocz(width*height*sizeof(DWTELEM)); //FIXME this does not belong here
3709 for(i=0; i<MAX_REF_FRAMES; i++)
3710 for(j=0; j<MAX_REF_FRAMES; j++)
3711 scale_mv_ref[i][j] = 256*(i+1)/(j+1);
3713 s->avctx->get_buffer(s->avctx, &s->mconly_picture);
3718 static int common_init_after_header(AVCodecContext *avctx){
3719 SnowContext *s = avctx->priv_data;
3720 int plane_index, level, orientation;
3722 for(plane_index=0; plane_index<3; plane_index++){
3723 int w= s->avctx->width;
3724 int h= s->avctx->height;
3727 w>>= s->chroma_h_shift;
3728 h>>= s->chroma_v_shift;
3730 s->plane[plane_index].width = w;
3731 s->plane[plane_index].height= h;
3733 for(level=s->spatial_decomposition_count-1; level>=0; level--){
3734 for(orientation=level ? 1 : 0; orientation<4; orientation++){
3735 SubBand *b= &s->plane[plane_index].band[level][orientation];
3737 b->buf= s->spatial_dwt_buffer;
3739 b->stride= s->plane[plane_index].width << (s->spatial_decomposition_count - level);
3740 b->width = (w + !(orientation&1))>>1;
3741 b->height= (h + !(orientation>1))>>1;
3743 b->stride_line = 1 << (s->spatial_decomposition_count - level);
3744 b->buf_x_offset = 0;
3745 b->buf_y_offset = 0;
3749 b->buf_x_offset = (w+1)>>1;
3752 b->buf += b->stride>>1;
3753 b->buf_y_offset = b->stride_line >> 1;
3755 b->ibuf= s->spatial_idwt_buffer + (b->buf - s->spatial_dwt_buffer);
3758 b->parent= &s->plane[plane_index].band[level-1][orientation];
3759 //FIXME avoid this realloc
3760 av_freep(&b->x_coeff);
3761 b->x_coeff=av_mallocz(((b->width+1) * b->height+1)*sizeof(x_and_coeff));
3771 static int qscale2qlog(int qscale){
3772 return rint(QROOT*log(qscale / (float)FF_QP2LAMBDA)/log(2))
3773 + 61*QROOT/8; //<64 >60
3776 static int ratecontrol_1pass(SnowContext *s, AVFrame *pict)
3778 /* Estimate the frame's complexity as a sum of weighted dwt coefficients.
3779 * FIXME we know exact mv bits at this point,
3780 * but ratecontrol isn't set up to include them. */
3781 uint32_t coef_sum= 0;
3782 int level, orientation, delta_qlog;
3784 for(level=0; level<s->spatial_decomposition_count; level++){
3785 for(orientation=level ? 1 : 0; orientation<4; orientation++){
3786 SubBand *b= &s->plane[0].band[level][orientation];
3787 IDWTELEM *buf= b->ibuf;
3788 const int w= b->width;
3789 const int h= b->height;
3790 const int stride= b->stride;
3791 const int qlog= av_clip(2*QROOT + b->qlog, 0, QROOT*16);
3792 const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
3793 const int qdiv= (1<<16)/qmul;
3795 //FIXME this is ugly
3798 buf[x+y*stride]= b->buf[x+y*stride];
3800 decorrelate(s, b, buf, stride, 1, 0);
3803 coef_sum+= abs(buf[x+y*stride]) * qdiv >> 16;
3807 /* ugly, ratecontrol just takes a sqrt again */
3808 coef_sum = (uint64_t)coef_sum * coef_sum >> 16;
3809 assert(coef_sum < INT_MAX);
3811 if(pict->pict_type == I_TYPE){
3812 s->m.current_picture.mb_var_sum= coef_sum;
3813 s->m.current_picture.mc_mb_var_sum= 0;
3815 s->m.current_picture.mc_mb_var_sum= coef_sum;
3816 s->m.current_picture.mb_var_sum= 0;
3819 pict->quality= ff_rate_estimate_qscale(&s->m, 1);
3820 if (pict->quality < 0)
3822 s->lambda= pict->quality * 3/2;
3823 delta_qlog= qscale2qlog(pict->quality) - s->qlog;
3824 s->qlog+= delta_qlog;
3828 static void calculate_visual_weight(SnowContext *s, Plane *p){
3829 int width = p->width;
3830 int height= p->height;
3831 int level, orientation, x, y;
3833 for(level=0; level<s->spatial_decomposition_count; level++){
3834 for(orientation=level ? 1 : 0; orientation<4; orientation++){
3835 SubBand *b= &p->band[level][orientation];
3836 IDWTELEM *ibuf= b->ibuf;
3839 memset(s->spatial_idwt_buffer, 0, sizeof(*s->spatial_idwt_buffer)*width*height);
3840 ibuf[b->width/2 + b->height/2*b->stride]= 256*16;
3841 ff_spatial_idwt(s->spatial_idwt_buffer, width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count);
3842 for(y=0; y<height; y++){
3843 for(x=0; x<width; x++){
3844 int64_t d= s->spatial_idwt_buffer[x + y*width]*16;
3849 b->qlog= (int)(log(352256.0/sqrt(error)) / log(pow(2.0, 1.0/QROOT))+0.5);
3859 static void find_sse(SnowContext *s, Plane *p, int *score, int score_stride, IDWTELEM *r0, IDWTELEM *r1, int level, int orientation){
3860 SubBand *b= &p->band[level][orientation];
3864 int step= 1 << (s->spatial_decomposition_count - level);
3871 //FIXME bias for nonzero ?
3873 memset(score, 0, sizeof(*score)*score_stride*((p->height + Q2_STEP-1)/Q2_STEP));
3874 for(y=0; y<p->height; y++){
3875 for(x=0; x<p->width; x++){
3876 int sx= (x-xo + step/2) / step / Q2_STEP;
3877 int sy= (y-yo + step/2) / step / Q2_STEP;
3878 int v= r0[x + y*p->width] - r1[x + y*p->width];
3879 assert(sx>=0 && sy>=0 && sx < score_stride);
3881 score[sx + sy*score_stride] += v*v;
3882 assert(score[sx + sy*score_stride] >= 0);
3887 static void dequantize_all(SnowContext *s, Plane *p, IDWTELEM *buffer, int width, int height){
3888 int level, orientation;
3890 for(level=0; level<s->spatial_decomposition_count; level++){
3891 for(orientation=level ? 1 : 0; orientation<4; orientation++){
3892 SubBand *b= &p->band[level][orientation];
3893 IDWTELEM *dst= buffer + (b->ibuf - s->spatial_idwt_buffer);
3895 dequantize(s, b, dst, b->stride);
3900 static void dwt_quantize(SnowContext *s, Plane *p, DWTELEM *buffer, int width, int height, int stride, int type){
3901 int level, orientation, ys, xs, x, y, pass;
3902 IDWTELEM best_dequant[height * stride];
3903 IDWTELEM idwt2_buffer[height * stride];
3904 const int score_stride= (width + 10)/Q2_STEP;
3905 int best_score[(width + 10)/Q2_STEP * (height + 10)/Q2_STEP]; //FIXME size
3906 int score[(width + 10)/Q2_STEP * (height + 10)/Q2_STEP]; //FIXME size
3907 int threshold= (s->m.lambda * s->m.lambda) >> 6;
3909 //FIXME pass the copy cleanly ?
3911 // memcpy(dwt_buffer, buffer, height * stride * sizeof(DWTELEM));
3912 ff_spatial_dwt(buffer, width, height, stride, type, s->spatial_decomposition_count);
3914 for(level=0; level<s->spatial_decomposition_count; level++){
3915 for(orientation=level ? 1 : 0; orientation<4; orientation++){
3916 SubBand *b= &p->band[level][orientation];
3917 IDWTELEM *dst= best_dequant + (b->ibuf - s->spatial_idwt_buffer);
3918 DWTELEM *src= buffer + (b-> buf - s->spatial_dwt_buffer);
3919 assert(src == b->buf); // code does not depend on this but it is true currently
3921 quantize(s, b, dst, src, b->stride, s->qbias);
3924 for(pass=0; pass<1; pass++){
3925 if(s->qbias == 0) //keyframe
3927 for(level=0; level<s->spatial_decomposition_count; level++){
3928 for(orientation=level ? 1 : 0; orientation<4; orientation++){
3929 SubBand *b= &p->band[level][orientation];
3930 IDWTELEM *dst= idwt2_buffer + (b->ibuf - s->spatial_idwt_buffer);
3931 IDWTELEM *best_dst= best_dequant + (b->ibuf - s->spatial_idwt_buffer);
3933 for(ys= 0; ys<Q2_STEP; ys++){
3934 for(xs= 0; xs<Q2_STEP; xs++){
3935 memcpy(idwt2_buffer, best_dequant, height * stride * sizeof(IDWTELEM));
3936 dequantize_all(s, p, idwt2_buffer, width, height);
3937 ff_spatial_idwt(idwt2_buffer, width, height, stride, type, s->spatial_decomposition_count);
3938 find_sse(s, p, best_score, score_stride, idwt2_buffer, s->spatial_idwt_buffer, level, orientation);
3939 memcpy(idwt2_buffer, best_dequant, height * stride * sizeof(IDWTELEM));
3940 for(y=ys; y<b->height; y+= Q2_STEP){
3941 for(x=xs; x<b->width; x+= Q2_STEP){
3942 if(dst[x + y*b->stride]<0) dst[x + y*b->stride]++;
3943 if(dst[x + y*b->stride]>0) dst[x + y*b->stride]--;
3944 //FIXME try more than just --
3947 dequantize_all(s, p, idwt2_buffer, width, height);
3948 ff_spatial_idwt(idwt2_buffer, width, height, stride, type, s->spatial_decomposition_count);
3949 find_sse(s, p, score, score_stride, idwt2_buffer, s->spatial_idwt_buffer, level, orientation);
3950 for(y=ys; y<b->height; y+= Q2_STEP){
3951 for(x=xs; x<b->width; x+= Q2_STEP){
3952 int score_idx= x/Q2_STEP + (y/Q2_STEP)*score_stride;
3953 if(score[score_idx] <= best_score[score_idx] + threshold){
3954 best_score[score_idx]= score[score_idx];
3955 if(best_dst[x + y*b->stride]<0) best_dst[x + y*b->stride]++;
3956 if(best_dst[x + y*b->stride]>0) best_dst[x + y*b->stride]--;
3957 //FIXME copy instead
3966 memcpy(s->spatial_idwt_buffer, best_dequant, height * stride * sizeof(IDWTELEM)); //FIXME work with that directly instead of copy at the end
3969 #endif /* QUANTIZE2==1 */
3971 static int encode_init(AVCodecContext *avctx)
3973 SnowContext *s = avctx->priv_data;
3976 if(avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL){
3977 av_log(avctx, AV_LOG_ERROR, "This codec is under development, files encoded with it may not be decodable with future versions!!!\n"
3978 "Use vstrict=-2 / -strict -2 to use it anyway.\n");
3982 if(avctx->prediction_method == DWT_97
3983 && (avctx->flags & CODEC_FLAG_QSCALE)
3984 && avctx->global_quality == 0){
3985 av_log(avctx, AV_LOG_ERROR, "The 9/7 wavelet is incompatible with lossless mode.\n");
3989 s->spatial_decomposition_type= avctx->prediction_method; //FIXME add decorrelator type r transform_type
3991 s->chroma_h_shift= 1; //FIXME XXX
3992 s->chroma_v_shift= 1;
3994 s->mv_scale = (avctx->flags & CODEC_FLAG_QPEL) ? 2 : 4;
3995 s->block_max_depth= (avctx->flags & CODEC_FLAG_4MV ) ? 1 : 0;
3997 for(plane_index=0; plane_index<3; plane_index++){
3998 s->plane[plane_index].diag_mc= 1;
3999 s->plane[plane_index].htaps= 6;
4000 s->plane[plane_index].hcoeff[0]= 40;
4001 s->plane[plane_index].hcoeff[1]= -10;
4002 s->plane[plane_index].hcoeff[2]= 2;
4003 s->plane[plane_index].fast_mc= 1;
4012 s->m.flags = avctx->flags;
4013 s->m.bit_rate= avctx->bit_rate;
4015 s->m.me.scratchpad= av_mallocz((avctx->width+64)*2*16*2*sizeof(uint8_t));
4016 s->m.me.map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t));
4017 s->m.me.score_map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t));
4018 s->m.obmc_scratchpad= av_mallocz(MB_SIZE*MB_SIZE*12*sizeof(uint32_t));
4019 h263_encode_init(&s->m); //mv_penalty
4021 s->max_ref_frames = FFMAX(FFMIN(avctx->refs, MAX_REF_FRAMES), 1);
4023 if(avctx->flags&CODEC_FLAG_PASS1){
4024 if(!avctx->stats_out)
4025 avctx->stats_out = av_mallocz(256);
4027 if((avctx->flags&CODEC_FLAG_PASS2) || !(avctx->flags&CODEC_FLAG_QSCALE)){
4028 if(ff_rate_control_init(&s->m) < 0)
4031 s->pass1_rc= !(avctx->flags & (CODEC_FLAG_QSCALE|CODEC_FLAG_PASS2));
4033 avctx->coded_frame= &s->current_picture;
4034 switch(avctx->pix_fmt){
4035 // case PIX_FMT_YUV444P:
4036 // case PIX_FMT_YUV422P:
4037 case PIX_FMT_YUV420P:
4039 // case PIX_FMT_YUV411P:
4040 // case PIX_FMT_YUV410P:
4041 s->colorspace_type= 0;
4043 /* case PIX_FMT_RGB32:
4047 av_log(avctx, AV_LOG_ERROR, "pixel format not supported\n");
4050 // avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_h_shift, &s->chroma_v_shift);
4051 s->chroma_h_shift= 1;
4052 s->chroma_v_shift= 1;
4054 ff_set_cmp(&s->dsp, s->dsp.me_cmp, s->avctx->me_cmp);
4055 ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, s->avctx->me_sub_cmp);
4057 s->avctx->get_buffer(s->avctx, &s->input_picture);
4059 if(s->avctx->me_method == ME_ITER){
4061 int size= s->b_width * s->b_height << 2*s->block_max_depth;
4062 for(i=0; i<s->max_ref_frames; i++){
4063 s->ref_mvs[i]= av_mallocz(size*sizeof(int16_t[2]));
4064 s->ref_scores[i]= av_mallocz(size*sizeof(uint32_t));
4071 #define USE_HALFPEL_PLANE 0
4073 static void halfpel_interpol(SnowContext *s, uint8_t *halfpel[4][4], AVFrame *frame){
4076 assert(!(s->avctx->flags & CODEC_FLAG_EMU_EDGE));
4080 int w= s->avctx->width >>is_chroma;
4081 int h= s->avctx->height >>is_chroma;
4082 int ls= frame->linesize[p];
4083 uint8_t *src= frame->data[p];
4085 halfpel[1][p]= (uint8_t*)av_malloc(ls * (h+2*EDGE_WIDTH)) + EDGE_WIDTH*(1+ls);
4086 halfpel[2][p]= (uint8_t*)av_malloc(ls * (h+2*EDGE_WIDTH)) + EDGE_WIDTH*(1+ls);
4087 halfpel[3][p]= (uint8_t*)av_malloc(ls * (h+2*EDGE_WIDTH)) + EDGE_WIDTH*(1+ls);
4094 halfpel[1][p][i]= (20*(src[i] + src[i+1]) - 5*(src[i-1] + src[i+2]) + (src[i-2] + src[i+3]) + 16 )>>5;
4101 halfpel[2][p][i]= (20*(src[i] + src[i+ls]) - 5*(src[i-ls] + src[i+2*ls]) + (src[i-2*ls] + src[i+3*ls]) + 16 )>>5;
4109 halfpel[3][p][i]= (20*(src[i] + src[i+ls]) - 5*(src[i-ls] + src[i+2*ls]) + (src[i-2*ls] + src[i+3*ls]) + 16 )>>5;
4117 static int frame_start(SnowContext *s){
4119 int w= s->avctx->width; //FIXME round up to x16 ?
4120 int h= s->avctx->height;
4122 if(s->current_picture.data[0]){
4123 draw_edges(s->current_picture.data[0], s->current_picture.linesize[0], w , h , EDGE_WIDTH );
4124 draw_edges(s->current_picture.data[1], s->current_picture.linesize[1], w>>1, h>>1, EDGE_WIDTH/2);
4125 draw_edges(s->current_picture.data[2], s->current_picture.linesize[2], w>>1, h>>1, EDGE_WIDTH/2);
4128 tmp= s->last_picture[s->max_ref_frames-1];
4129 memmove(s->last_picture+1, s->last_picture, (s->max_ref_frames-1)*sizeof(AVFrame));
4130 memmove(s->halfpel_plane+1, s->halfpel_plane, (s->max_ref_frames-1)*sizeof(void*)*4*4);
4131 if(USE_HALFPEL_PLANE && s->current_picture.data[0])
4132 halfpel_interpol(s, s->halfpel_plane[0], &s->current_picture);
4133 s->last_picture[0]= s->current_picture;
4134 s->current_picture= tmp;
4140 for(i=0; i<s->max_ref_frames && s->last_picture[i].data[0]; i++)
4141 if(i && s->last_picture[i-1].key_frame)
4146 s->current_picture.reference= 1;
4147 if(s->avctx->get_buffer(s->avctx, &s->current_picture) < 0){
4148 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
4152 s->current_picture.key_frame= s->keyframe;
4157 static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size, void *data){
4158 SnowContext *s = avctx->priv_data;
4159 RangeCoder * const c= &s->c;
4160 AVFrame *pict = data;
4161 const int width= s->avctx->width;
4162 const int height= s->avctx->height;
4163 int level, orientation, plane_index, i, y;
4164 uint8_t rc_header_bak[sizeof(s->header_state)];
4165 uint8_t rc_block_bak[sizeof(s->block_state)];
4167 ff_init_range_encoder(c, buf, buf_size);
4168 ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
4172 for(y=0; y<(height>>shift); y++)
4173 memcpy(&s->input_picture.data[i][y * s->input_picture.linesize[i]],
4174 &pict->data[i][y * pict->linesize[i]],
4177 s->new_picture = *pict;
4179 s->m.picture_number= avctx->frame_number;
4180 if(avctx->flags&CODEC_FLAG_PASS2){
4182 pict->pict_type= s->m.rc_context.entry[avctx->frame_number].new_pict_type;
4183 s->keyframe= pict->pict_type==FF_I_TYPE;
4184 if(!(avctx->flags&CODEC_FLAG_QSCALE)) {
4185 pict->quality= ff_rate_estimate_qscale(&s->m, 0);
4186 if (pict->quality < 0)
4190 s->keyframe= avctx->gop_size==0 || avctx->frame_number % avctx->gop_size == 0;
4192 pict->pict_type= s->keyframe ? FF_I_TYPE : FF_P_TYPE;
4195 if(s->pass1_rc && avctx->frame_number == 0)
4196 pict->quality= 2*FF_QP2LAMBDA;
4198 s->qlog= qscale2qlog(pict->quality);
4199 s->lambda = pict->quality * 3/2;
4201 if(s->qlog < 0 || (!pict->quality && (avctx->flags & CODEC_FLAG_QSCALE))){
4202 s->qlog= LOSSLESS_QLOG;
4204 }//else keep previous frame's qlog until after motion estimation
4208 s->m.current_picture_ptr= &s->m.current_picture;
4209 if(pict->pict_type == P_TYPE){
4210 int block_width = (width +15)>>4;
4211 int block_height= (height+15)>>4;
4212 int stride= s->current_picture.linesize[0];
4214 assert(s->current_picture.data[0]);
4215 assert(s->last_picture[0].data[0]);
4217 s->m.avctx= s->avctx;
4218 s->m.current_picture.data[0]= s->current_picture.data[0];
4219 s->m. last_picture.data[0]= s->last_picture[0].data[0];
4220 s->m. new_picture.data[0]= s-> input_picture.data[0];
4221 s->m. last_picture_ptr= &s->m. last_picture;
4223 s->m. last_picture.linesize[0]=
4224 s->m. new_picture.linesize[0]=
4225 s->m.current_picture.linesize[0]= stride;
4226 s->m.uvlinesize= s->current_picture.linesize[1];
4228 s->m.height= height;
4229 s->m.mb_width = block_width;
4230 s->m.mb_height= block_height;
4231 s->m.mb_stride= s->m.mb_width+1;
4232 s->m.b8_stride= 2*s->m.mb_width+1;
4234 s->m.pict_type= pict->pict_type;
4235 s->m.me_method= s->avctx->me_method;
4236 s->m.me.scene_change_score=0;
4237 s->m.flags= s->avctx->flags;
4238 s->m.quarter_sample= (s->avctx->flags & CODEC_FLAG_QPEL)!=0;
4239 s->m.out_format= FMT_H263;
4240 s->m.unrestricted_mv= 1;
4242 s->m.lambda = s->lambda;
4243 s->m.qscale= (s->m.lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
4244 s->lambda2= s->m.lambda2= (s->m.lambda*s->m.lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
4246 s->m.dsp= s->dsp; //move
4252 memcpy(rc_header_bak, s->header_state, sizeof(s->header_state));
4253 memcpy(rc_block_bak, s->block_state, sizeof(s->block_state));
4258 if(pict->pict_type == I_TYPE)
4259 s->spatial_decomposition_count= 5;
4261 s->spatial_decomposition_count= 5;
4263 s->m.pict_type = pict->pict_type;
4264 s->qbias= pict->pict_type == P_TYPE ? 2 : 0;
4266 common_init_after_header(avctx);
4268 if(s->last_spatial_decomposition_count != s->spatial_decomposition_count){
4269 for(plane_index=0; plane_index<3; plane_index++){
4270 calculate_visual_weight(s, &s->plane[plane_index]);
4275 s->m.misc_bits = 8*(s->c.bytestream - s->c.bytestream_start);
4276 encode_blocks(s, 1);
4277 s->m.mv_bits = 8*(s->c.bytestream - s->c.bytestream_start) - s->m.misc_bits;
4279 for(plane_index=0; plane_index<3; plane_index++){
4280 Plane *p= &s->plane[plane_index];
4284 // int bits= put_bits_count(&s->c.pb);
4286 if(!(avctx->flags2 & CODEC_FLAG2_MEMC_ONLY)){
4288 if(pict->data[plane_index]) //FIXME gray hack
4291 s->spatial_idwt_buffer[y*w + x]= pict->data[plane_index][y*pict->linesize[plane_index] + x]<<FRAC_BITS;
4294 predict_plane(s, s->spatial_idwt_buffer, plane_index, 0);
4297 && pict->pict_type == P_TYPE
4298 && !(avctx->flags&CODEC_FLAG_PASS2)
4299 && s->m.me.scene_change_score > s->avctx->scenechange_threshold){
4300 ff_init_range_encoder(c, buf, buf_size);
4301 ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
4302 pict->pict_type= FF_I_TYPE;
4304 s->current_picture.key_frame=1;
4308 if(s->qlog == LOSSLESS_QLOG){
4311 s->spatial_dwt_buffer[y*w + x]= (s->spatial_idwt_buffer[y*w + x] + (1<<(FRAC_BITS-1))-1)>>FRAC_BITS;
4317 s->spatial_dwt_buffer[y*w + x]=s->spatial_idwt_buffer[y*w + x]<<ENCODER_EXTRA_BITS;
4323 dwt_quantize(s, p, s->spatial_dwt_buffer, w, h, w, s->spatial_decomposition_type);
4325 ff_spatial_dwt(s->spatial_dwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count);
4327 if(s->pass1_rc && plane_index==0){
4328 int delta_qlog = ratecontrol_1pass(s, pict);
4329 if (delta_qlog <= INT_MIN)
4332 //reordering qlog in the bitstream would eliminate this reset
4333 ff_init_range_encoder(c, buf, buf_size);
4334 memcpy(s->header_state, rc_header_bak, sizeof(s->header_state));
4335 memcpy(s->block_state, rc_block_bak, sizeof(s->block_state));
4337 encode_blocks(s, 0);
4341 for(level=0; level<s->spatial_decomposition_count; level++){
4342 for(orientation=level ? 1 : 0; orientation<4; orientation++){
4343 SubBand *b= &p->band[level][orientation];
4346 quantize(s, b, b->ibuf, b->buf, b->stride, s->qbias);
4348 decorrelate(s, b, b->ibuf, b->stride, pict->pict_type == P_TYPE, 0);
4349 encode_subband(s, b, b->ibuf, b->parent ? b->parent->ibuf : NULL, b->stride, orientation);
4350 assert(b->parent==NULL || b->parent->stride == b->stride*2);
4352 correlate(s, b, b->ibuf, b->stride, 1, 0);
4356 for(level=0; level<s->spatial_decomposition_count; level++){
4357 for(orientation=level ? 1 : 0; orientation<4; orientation++){
4358 SubBand *b= &p->band[level][orientation];
4360 dequantize(s, b, b->ibuf, b->stride);
4364 ff_spatial_idwt(s->spatial_idwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count);
4365 if(s->qlog == LOSSLESS_QLOG){
4368 s->spatial_idwt_buffer[y*w + x]<<=FRAC_BITS;
4372 predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
4375 if(pict->pict_type == I_TYPE){
4378 s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x]=
4379 pict->data[plane_index][y*pict->linesize[plane_index] + x];
4383 memset(s->spatial_idwt_buffer, 0, sizeof(IDWTELEM)*w*h);
4384 predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
4387 if(s->avctx->flags&CODEC_FLAG_PSNR){
4390 if(pict->data[plane_index]) //FIXME gray hack
4393 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];
4397 s->avctx->error[plane_index] += error;
4398 s->current_picture.error[plane_index] = error;
4403 update_last_header_values(s);
4405 if(s->last_picture[s->max_ref_frames-1].data[0]){
4406 avctx->release_buffer(avctx, &s->last_picture[s->max_ref_frames-1]);
4408 if(s->halfpel_plane[s->max_ref_frames-1][1+i/3][i%3])
4409 av_free(s->halfpel_plane[s->max_ref_frames-1][1+i/3][i%3] - EDGE_WIDTH*(1+s->current_picture.linesize[i%3]));
4412 s->current_picture.coded_picture_number = avctx->frame_number;
4413 s->current_picture.pict_type = pict->pict_type;
4414 s->current_picture.quality = pict->quality;
4415 s->m.frame_bits = 8*(s->c.bytestream - s->c.bytestream_start);
4416 s->m.p_tex_bits = s->m.frame_bits - s->m.misc_bits - s->m.mv_bits;
4417 s->m.current_picture.display_picture_number =
4418 s->m.current_picture.coded_picture_number = avctx->frame_number;
4419 s->m.current_picture.quality = pict->quality;
4420 s->m.total_bits += 8*(s->c.bytestream - s->c.bytestream_start);
4422 if (ff_rate_estimate_qscale(&s->m, 0) < 0)
4424 if(avctx->flags&CODEC_FLAG_PASS1)
4425 ff_write_pass1_stats(&s->m);
4426 s->m.last_pict_type = s->m.pict_type;
4427 avctx->frame_bits = s->m.frame_bits;
4428 avctx->mv_bits = s->m.mv_bits;
4429 avctx->misc_bits = s->m.misc_bits;
4430 avctx->p_tex_bits = s->m.p_tex_bits;
4434 return ff_rac_terminate(c);
4437 static void common_end(SnowContext *s){
4438 int plane_index, level, orientation, i;
4440 av_freep(&s->spatial_dwt_buffer);
4441 av_freep(&s->spatial_idwt_buffer);
4443 av_freep(&s->m.me.scratchpad);
4444 av_freep(&s->m.me.map);
4445 av_freep(&s->m.me.score_map);
4446 av_freep(&s->m.obmc_scratchpad);
4448 av_freep(&s->block);
4450 for(i=0; i<MAX_REF_FRAMES; i++){
4451 av_freep(&s->ref_mvs[i]);
4452 av_freep(&s->ref_scores[i]);
4453 if(s->last_picture[i].data[0])
4454 s->avctx->release_buffer(s->avctx, &s->last_picture[i]);
4457 for(plane_index=0; plane_index<3; plane_index++){
4458 for(level=s->spatial_decomposition_count-1; level>=0; level--){
4459 for(orientation=level ? 1 : 0; orientation<4; orientation++){
4460 SubBand *b= &s->plane[plane_index].band[level][orientation];
4462 av_freep(&b->x_coeff);
4468 static int encode_end(AVCodecContext *avctx)
4470 SnowContext *s = avctx->priv_data;
4473 av_free(avctx->stats_out);
4478 static int decode_init(AVCodecContext *avctx)
4480 avctx->pix_fmt= PIX_FMT_YUV420P;
4487 static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, const uint8_t *buf, int buf_size){
4488 SnowContext *s = avctx->priv_data;
4489 RangeCoder * const c= &s->c;
4491 AVFrame *picture = data;
4492 int level, orientation, plane_index, i;
4494 ff_init_range_decoder(c, buf, buf_size);
4495 ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
4497 s->current_picture.pict_type= FF_I_TYPE; //FIXME I vs. P
4498 if(decode_header(s)<0)
4500 common_init_after_header(avctx);
4502 // realloc slice buffer for the case that spatial_decomposition_count changed
4503 slice_buffer_destroy(&s->sb);
4504 slice_buffer_init(&s->sb, s->plane[0].height, (MB_SIZE >> s->block_max_depth) + s->spatial_decomposition_count * 8 + 1, s->plane[0].width, s->spatial_idwt_buffer);
4506 for(plane_index=0; plane_index<3; plane_index++){
4507 Plane *p= &s->plane[plane_index];
4508 p->fast_mc= p->diag_mc && p->htaps==6 && p->hcoeff[0]==40
4509 && p->hcoeff[1]==-10
4513 if(!s->block) alloc_blocks(s);
4516 //keyframe flag duplication mess FIXME
4517 if(avctx->debug&FF_DEBUG_PICT_INFO)
4518 av_log(avctx, AV_LOG_ERROR, "keyframe:%d qlog:%d\n", s->keyframe, s->qlog);
4522 for(plane_index=0; plane_index<3; plane_index++){
4523 Plane *p= &s->plane[plane_index];
4527 int decode_state[MAX_DECOMPOSITIONS][4][1]; /* Stored state info for unpack_coeffs. 1 variable per instance. */
4529 if(s->avctx->debug&2048){
4530 memset(s->spatial_dwt_buffer, 0, sizeof(DWTELEM)*w*h);
4531 predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
4535 int v= s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x];
4536 s->mconly_picture.data[plane_index][y*s->mconly_picture.linesize[plane_index] + x]= v;
4542 for(level=0; level<s->spatial_decomposition_count; level++){
4543 for(orientation=level ? 1 : 0; orientation<4; orientation++){
4544 SubBand *b= &p->band[level][orientation];
4545 unpack_coeffs(s, b, b->parent, orientation);
4551 const int mb_h= s->b_height << s->block_max_depth;
4552 const int block_size = MB_SIZE >> s->block_max_depth;
4553 const int block_w = plane_index ? block_size/2 : block_size;
4555 dwt_compose_t cs[MAX_DECOMPOSITIONS];
4560 ff_spatial_idwt_buffered_init(cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count);
4561 for(mb_y=0; mb_y<=mb_h; mb_y++){
4563 int slice_starty = block_w*mb_y;
4564 int slice_h = block_w*(mb_y+1);
4565 if (!(s->keyframe || s->avctx->debug&512)){
4566 slice_starty = FFMAX(0, slice_starty - (block_w >> 1));
4567 slice_h -= (block_w >> 1);
4570 for(level=0; level<s->spatial_decomposition_count; level++){
4571 for(orientation=level ? 1 : 0; orientation<4; orientation++){
4572 SubBand *b= &p->band[level][orientation];
4575 int our_mb_start = mb_y;
4576 int our_mb_end = (mb_y + 1);
4578 start_y = (mb_y ? ((block_w * our_mb_start) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra: 0);
4579 end_y = (((block_w * our_mb_end) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra);
4580 if (!(s->keyframe || s->avctx->debug&512)){
4581 start_y = FFMAX(0, start_y - (block_w >> (1+s->spatial_decomposition_count - level)));
4582 end_y = FFMAX(0, end_y - (block_w >> (1+s->spatial_decomposition_count - level)));
4584 start_y = FFMIN(b->height, start_y);
4585 end_y = FFMIN(b->height, end_y);
4587 if (start_y != end_y){
4588 if (orientation == 0){
4589 SubBand * correlate_band = &p->band[0][0];
4590 int correlate_end_y = FFMIN(b->height, end_y + 1);
4591 int correlate_start_y = FFMIN(b->height, (start_y ? start_y + 1 : 0));
4592 decode_subband_slice_buffered(s, correlate_band, &s->sb, correlate_start_y, correlate_end_y, decode_state[0][0]);
4593 correlate_slice_buffered(s, &s->sb, correlate_band, correlate_band->ibuf, correlate_band->stride, 1, 0, correlate_start_y, correlate_end_y);
4594 dequantize_slice_buffered(s, &s->sb, correlate_band, correlate_band->ibuf, correlate_band->stride, start_y, end_y);
4597 decode_subband_slice_buffered(s, b, &s->sb, start_y, end_y, decode_state[level][orientation]);
4602 for(; yd<slice_h; yd+=4){
4603 ff_spatial_idwt_buffered_slice(&s->dsp, cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count, yd);
4606 if(s->qlog == LOSSLESS_QLOG){
4607 for(; yq<slice_h && yq<h; yq++){
4608 IDWTELEM * line = slice_buffer_get_line(&s->sb, yq);
4610 line[x] <<= FRAC_BITS;
4615 predict_slice_buffered(s, &s->sb, s->spatial_idwt_buffer, plane_index, 1, mb_y);
4617 y = FFMIN(p->height, slice_starty);
4618 end_y = FFMIN(p->height, slice_h);
4620 slice_buffer_release(&s->sb, y++);
4623 slice_buffer_flush(&s->sb);
4630 if(s->last_picture[s->max_ref_frames-1].data[0]){
4631 avctx->release_buffer(avctx, &s->last_picture[s->max_ref_frames-1]);
4633 if(s->halfpel_plane[s->max_ref_frames-1][1+i/3][i%3])
4634 av_free(s->halfpel_plane[s->max_ref_frames-1][1+i/3][i%3] - EDGE_WIDTH*(1+s->current_picture.linesize[i%3]));
4637 if(!(s->avctx->debug&2048))
4638 *picture= s->current_picture;
4640 *picture= s->mconly_picture;
4642 *data_size = sizeof(AVFrame);
4644 bytes_read= c->bytestream - c->bytestream_start;
4645 if(bytes_read ==0) av_log(s->avctx, AV_LOG_ERROR, "error at end of frame\n"); //FIXME
4650 static int decode_end(AVCodecContext *avctx)
4652 SnowContext *s = avctx->priv_data;
4654 slice_buffer_destroy(&s->sb);
4661 AVCodec snow_decoder = {
4665 sizeof(SnowContext),
4670 0 /*CODEC_CAP_DR1*/ /*| CODEC_CAP_DRAW_HORIZ_BAND*/,
4674 #ifdef CONFIG_SNOW_ENCODER
4675 AVCodec snow_encoder = {
4679 sizeof(SnowContext),
4696 int buffer[2][width*height];
4699 s.spatial_decomposition_count=6;
4700 s.spatial_decomposition_type=1;
4702 printf("testing 5/3 DWT\n");
4703 for(i=0; i<width*height; i++)
4704 buffer[0][i]= buffer[1][i]= random()%54321 - 12345;
4706 ff_spatial_dwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
4707 ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
4709 for(i=0; i<width*height; i++)
4710 if(buffer[0][i]!= buffer[1][i]) printf("fsck: %d %d %d\n",i, buffer[0][i], buffer[1][i]);
4712 printf("testing 9/7 DWT\n");
4713 s.spatial_decomposition_type=0;
4714 for(i=0; i<width*height; i++)
4715 buffer[0][i]= buffer[1][i]= random()%54321 - 12345;
4717 ff_spatial_dwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
4718 ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
4720 for(i=0; i<width*height; i++)
4721 if(FFABS(buffer[0][i] - buffer[1][i])>20) printf("fsck: %d %d %d\n",i, buffer[0][i], buffer[1][i]);
4724 printf("testing AC coder\n");
4725 memset(s.header_state, 0, sizeof(s.header_state));
4726 ff_init_range_encoder(&s.c, buffer[0], 256*256);
4727 ff_init_cabac_states(&s.c, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64);
4729 for(i=-256; i<256; i++){
4730 put_symbol(&s.c, s.header_state, i*i*i/3*FFABS(i), 1);
4732 ff_rac_terminate(&s.c);
4734 memset(s.header_state, 0, sizeof(s.header_state));
4735 ff_init_range_decoder(&s.c, buffer[0], 256*256);
4736 ff_init_cabac_states(&s.c, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64);
4738 for(i=-256; i<256; i++){
4740 j= get_symbol(&s.c, s.header_state, 1);
4741 if(j!=i*i*i/3*FFABS(i)) printf("fsck: %d != %d\n", i, j);
4745 int level, orientation, x, y;
4746 int64_t errors[8][4];
4749 memset(errors, 0, sizeof(errors));
4750 s.spatial_decomposition_count=3;
4751 s.spatial_decomposition_type=0;
4752 for(level=0; level<s.spatial_decomposition_count; level++){
4753 for(orientation=level ? 1 : 0; orientation<4; orientation++){
4754 int w= width >> (s.spatial_decomposition_count-level);
4755 int h= height >> (s.spatial_decomposition_count-level);
4756 int stride= width << (s.spatial_decomposition_count-level);
4757 DWTELEM *buf= buffer[0];
4760 if(orientation&1) buf+=w;
4761 if(orientation>1) buf+=stride>>1;
4763 memset(buffer[0], 0, sizeof(int)*width*height);
4764 buf[w/2 + h/2*stride]= 256*256;
4765 ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
4766 for(y=0; y<height; y++){
4767 for(x=0; x<width; x++){
4768 int64_t d= buffer[0][x + y*width];
4770 if(FFABS(width/2-x)<9 && FFABS(height/2-y)<9 && level==2) printf("%8"PRId64" ", d);
4772 if(FFABS(height/2-y)<9 && level==2) printf("\n");
4774 error= (int)(sqrt(error)+0.5);
4775 errors[level][orientation]= error;
4776 if(g) g=ff_gcd(g, error);
4780 printf("static int const visual_weight[][4]={\n");
4781 for(level=0; level<s.spatial_decomposition_count; level++){
4783 for(orientation=0; orientation<4; orientation++){
4784 printf("%8"PRId64",", errors[level][orientation]/g);
4791 int w= width >> (s.spatial_decomposition_count-level);
4792 //int h= height >> (s.spatial_decomposition_count-level);
4793 int stride= width << (s.spatial_decomposition_count-level);
4794 DWTELEM *buf= buffer[0];
4800 memset(buffer[0], 0, sizeof(int)*width*height);
4802 for(y=0; y<height; y++){
4803 for(x=0; x<width; x++){
4804 int tab[4]={0,2,3,1};
4805 buffer[0][x+width*y]= 256*256*tab[(x&1) + 2*(y&1)];
4808 ff_spatial_dwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
4812 buf[x + y*stride ]=169;
4813 buf[x + y*stride-w]=64;
4816 ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
4818 for(y=0; y<height; y++){
4819 for(x=0; x<width; x++){
4820 int64_t d= buffer[0][x + y*width];
4822 if(FFABS(width/2-x)<9 && FFABS(height/2-y)<9) printf("%8"PRId64" ", d);
4824 if(FFABS(height/2-y)<9) printf("\n");