Mercurial > libavcodec.hg
annotate snow.c @ 5646:adc9ec0f7385 libavcodec
Let H.264 decoder skip deblocking across slices
if running with multiple threads and CODEC_FLAGS2_FAST is set.
Thus, it may decode the slices in parallel to gain speed.
Patch by Andreas ªÓman: [andreas olebyn nu]
author | benoit |
---|---|
date | Fri, 07 Sep 2007 18:27:13 +0000 |
parents | 473cada682a1 |
children | cd26ab6e3953 |
rev | line source |
---|---|
2138 | 1 /* |
2 * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at> | |
3 * | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3920
diff
changeset
|
4 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3920
diff
changeset
|
5 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3920
diff
changeset
|
6 * FFmpeg is free software; you can redistribute it and/or |
2138 | 7 * modify it under the terms of the GNU Lesser General Public |
8 * License as published by the Free Software Foundation; either | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3920
diff
changeset
|
9 * version 2.1 of the License, or (at your option) any later version. |
2138 | 10 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3920
diff
changeset
|
11 * FFmpeg is distributed in the hope that it will be useful, |
2138 | 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. | |
15 * | |
16 * You should have received a copy of the GNU Lesser General Public | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3920
diff
changeset
|
17 * License along with FFmpeg; if not, write to the Free Software |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
3035
diff
changeset
|
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
2138 | 19 */ |
20 | |
21 #include "avcodec.h" | |
22 #include "dsputil.h" | |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
23 #include "snow.h" |
2335 | 24 |
25 #include "rangecoder.h" | |
2138 | 26 |
27 #include "mpegvideo.h" | |
28 | |
29 #undef NDEBUG | |
30 #include <assert.h> | |
31 | |
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, | |
49 }; | |
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, | |
67 }; | |
2596 | 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, | |
85 }; | |
2138 | 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, | |
103 }; | |
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, | |
121 }; | |
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, | |
139 }; | |
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, | |
157 }; | |
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, | |
175 }; | |
176 | |
177 #if 0 //64*cubic | |
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, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, | |
181 0, 0, 0, 1, 1, 1, 1, 2, 2, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, 3, 3, 3, 2, 2, 1, 1, 1, 1, 0, 0, 0, | |
182 0, 0, 1, 1, 2, 2, 3, 4, 4, 5, 6, 6, 7, 7, 8, 8, 8, 8, 7, 7, 6, 6, 5, 4, 4, 3, 2, 2, 1, 1, 0, 0, | |
183 0, 0, 1, 2, 2, 3, 4, 6, 7, 8, 9,10,11,12,12,12,12,12,12,11,10, 9, 8, 7, 6, 4, 3, 2, 2, 1, 0, 0, | |
184 0, 1, 1, 2, 3, 5, 6, 8,10,11,13,14,15,16,17,18,18,17,16,15,14,13,11,10, 8, 6, 5, 3, 2, 1, 1, 0, | |
185 0, 1, 1, 3, 4, 6, 8,10,13,15,17,19,20,22,22,23,23,22,22,20,19,17,15,13,10, 8, 6, 4, 3, 1, 1, 0, | |
186 0, 1, 2, 4, 6, 8,10,13,16,19,21,23,25,27,28,29,29,28,27,25,23,21,19,16,13,10, 8, 6, 4, 2, 1, 0, | |
187 0, 1, 2, 4, 7,10,13,16,19,22,25,28,31,33,34,35,35,34,33,31,28,25,22,19,16,13,10, 7, 4, 2, 1, 0, | |
188 0, 1, 3, 5, 8,11,15,19,22,26,30,33,36,38,40,41,41,40,38,36,33,30,26,22,19,15,11, 8, 5, 3, 1, 0, | |
189 0, 1, 3, 6, 9,12,17,21,25,30,34,38,41,44,45,46,46,45,44,41,38,34,30,25,21,17,12, 9, 6, 3, 1, 0, | |
190 0, 1, 3, 6,10,14,19,23,28,33,38,42,45,48,51,52,52,51,48,45,42,38,33,28,23,19,14,10, 6, 3, 1, 0, | |
191 0, 1, 4, 7,11,15,20,25,31,36,41,45,49,52,55,56,56,55,52,49,45,41,36,31,25,20,15,11, 7, 4, 1, 0, | |
192 0, 2, 4, 7,12,16,22,27,33,38,44,48,52,56,58,60,60,58,56,52,48,44,38,33,27,22,16,12, 7, 4, 2, 0, | |
193 0, 1, 4, 8,12,17,22,28,34,40,45,51,55,58,61,62,62,61,58,55,51,45,40,34,28,22,17,12, 8, 4, 1, 0, | |
194 0, 2, 4, 8,12,18,23,29,35,41,46,52,56,60,62,64,64,62,60,56,52,46,41,35,29,23,18,12, 8, 4, 2, 0, | |
195 0, 2, 4, 8,12,18,23,29,35,41,46,52,56,60,62,64,64,62,60,56,52,46,41,35,29,23,18,12, 8, 4, 2, 0, | |
196 0, 1, 4, 8,12,17,22,28,34,40,45,51,55,58,61,62,62,61,58,55,51,45,40,34,28,22,17,12, 8, 4, 1, 0, | |
197 0, 2, 4, 7,12,16,22,27,33,38,44,48,52,56,58,60,60,58,56,52,48,44,38,33,27,22,16,12, 7, 4, 2, 0, | |
198 0, 1, 4, 7,11,15,20,25,31,36,41,45,49,52,55,56,56,55,52,49,45,41,36,31,25,20,15,11, 7, 4, 1, 0, | |
199 0, 1, 3, 6,10,14,19,23,28,33,38,42,45,48,51,52,52,51,48,45,42,38,33,28,23,19,14,10, 6, 3, 1, 0, | |
200 0, 1, 3, 6, 9,12,17,21,25,30,34,38,41,44,45,46,46,45,44,41,38,34,30,25,21,17,12, 9, 6, 3, 1, 0, | |
201 0, 1, 3, 5, 8,11,15,19,22,26,30,33,36,38,40,41,41,40,38,36,33,30,26,22,19,15,11, 8, 5, 3, 1, 0, | |
202 0, 1, 2, 4, 7,10,13,16,19,22,25,28,31,33,34,35,35,34,33,31,28,25,22,19,16,13,10, 7, 4, 2, 1, 0, | |
203 0, 1, 2, 4, 6, 8,10,13,16,19,21,23,25,27,28,29,29,28,27,25,23,21,19,16,13,10, 8, 6, 4, 2, 1, 0, | |
204 0, 1, 1, 3, 4, 6, 8,10,13,15,17,19,20,22,22,23,23,22,22,20,19,17,15,13,10, 8, 6, 4, 3, 1, 1, 0, | |
205 0, 1, 1, 2, 3, 5, 6, 8,10,11,13,14,15,16,17,18,18,17,16,15,14,13,11,10, 8, 6, 5, 3, 2, 1, 1, 0, | |
206 0, 0, 1, 2, 2, 3, 4, 6, 7, 8, 9,10,11,12,12,12,12,12,12,11,10, 9, 8, 7, 6, 4, 3, 2, 2, 1, 0, 0, | |
207 0, 0, 1, 1, 2, 2, 3, 4, 4, 5, 6, 6, 7, 7, 8, 8, 8, 8, 7, 7, 6, 6, 5, 4, 4, 3, 2, 2, 1, 1, 0, 0, | |
208 0, 0, 0, 1, 1, 1, 1, 2, 2, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, 3, 3, 3, 2, 2, 1, 1, 1, 1, 0, 0, 0, | |
209 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, | |
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, | |
211 //error:0.000022 | |
212 }; | |
213 static const uint8_t obmc16[256]={ | |
214 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, | |
215 0, 1, 1, 2, 4, 5, 5, 6, 6, 5, 5, 4, 2, 1, 1, 0, | |
216 0, 1, 4, 6, 9,11,13,15,15,13,11, 9, 6, 4, 1, 0, | |
217 0, 2, 6,11,15,20,24,26,26,24,20,15,11, 6, 2, 0, | |
218 0, 4, 9,15,23,29,34,38,38,34,29,23,15, 9, 4, 0, | |
219 0, 5,11,20,29,38,45,49,49,45,38,29,20,11, 5, 0, | |
220 1, 5,13,24,34,45,53,57,57,53,45,34,24,13, 5, 1, | |
221 1, 6,15,26,38,49,57,62,62,57,49,38,26,15, 6, 1, | |
222 1, 6,15,26,38,49,57,62,62,57,49,38,26,15, 6, 1, | |
223 1, 5,13,24,34,45,53,57,57,53,45,34,24,13, 5, 1, | |
224 0, 5,11,20,29,38,45,49,49,45,38,29,20,11, 5, 0, | |
225 0, 4, 9,15,23,29,34,38,38,34,29,23,15, 9, 4, 0, | |
226 0, 2, 6,11,15,20,24,26,26,24,20,15,11, 6, 2, 0, | |
227 0, 1, 4, 6, 9,11,13,15,15,13,11, 9, 6, 4, 1, 0, | |
228 0, 1, 1, 2, 4, 5, 5, 6, 6, 5, 5, 4, 2, 1, 1, 0, | |
229 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, | |
230 //error:0.000033 | |
231 }; | |
232 #elif 1 // 64*linear | |
233 static const uint8_t obmc32[1024]={ | |
3206 | 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, | |
2138 | 266 //error:0.000020 |
267 }; | |
268 static const uint8_t obmc16[256]={ | |
3206 | 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, | |
2138 | 285 //error:0.000015 |
286 }; | |
287 #else //64*cos | |
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, 1, 1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, | |
291 0, 0, 0, 1, 1, 1, 1, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 3, 3, 3, 3, 2, 2, 1, 1, 1, 1, 0, 0, 0, | |
292 0, 0, 1, 1, 1, 2, 2, 3, 4, 5, 5, 6, 7, 7, 7, 7, 7, 7, 7, 7, 6, 5, 5, 4, 3, 2, 2, 1, 1, 1, 0, 0, | |
293 0, 0, 1, 1, 2, 3, 4, 5, 6, 7, 9,10,11,11,12,12,12,12,11,11,10, 9, 7, 6, 5, 4, 3, 2, 1, 1, 0, 0, | |
294 0, 0, 1, 2, 3, 5, 6, 8, 9,11,12,14,15,16,17,17,17,17,16,15,14,12,11, 9, 8, 6, 5, 3, 2, 1, 0, 0, | |
295 0, 1, 1, 2, 4, 6, 8,10,12,15,17,19,20,21,22,23,23,22,21,20,19,17,15,12,10, 8, 6, 4, 2, 1, 1, 0, | |
296 0, 1, 2, 3, 5, 8,10,13,16,19,21,24,26,27,28,29,29,28,27,26,24,21,19,16,13,10, 8, 5, 3, 2, 1, 0, | |
297 0, 1, 2, 4, 6, 9,12,16,19,23,26,29,31,33,34,35,35,34,33,31,29,26,23,19,16,12, 9, 6, 4, 2, 1, 0, | |
298 0, 1, 3, 5, 7,11,15,19,23,26,30,34,37,39,40,41,41,40,39,37,34,30,26,23,19,15,11, 7, 5, 3, 1, 0, | |
299 0, 1, 3, 5, 9,12,17,21,26,30,35,38,42,44,46,47,47,46,44,42,38,35,30,26,21,17,12, 9, 5, 3, 1, 0, | |
300 0, 1, 3, 6, 9,14,19,24,29,34,38,43,46,49,51,52,52,51,49,46,43,38,34,29,24,19,14, 9, 6, 3, 1, 0, | |
301 0, 1, 3, 6,11,15,20,26,31,37,42,46,50,53,56,57,57,56,53,50,46,42,37,31,26,20,15,11, 6, 3, 1, 0, | |
302 0, 1, 3, 7,11,16,21,27,33,39,44,49,53,57,59,60,60,59,57,53,49,44,39,33,27,21,16,11, 7, 3, 1, 0, | |
303 0, 1, 4, 7,12,17,22,28,34,40,46,51,56,59,61,63,63,61,59,56,51,46,40,34,28,22,17,12, 7, 4, 1, 0, | |
304 0, 1, 4, 7,12,17,23,29,35,41,47,52,57,60,63,64,64,63,60,57,52,47,41,35,29,23,17,12, 7, 4, 1, 0, | |
305 0, 1, 4, 7,12,17,23,29,35,41,47,52,57,60,63,64,64,63,60,57,52,47,41,35,29,23,17,12, 7, 4, 1, 0, | |
306 0, 1, 4, 7,12,17,22,28,34,40,46,51,56,59,61,63,63,61,59,56,51,46,40,34,28,22,17,12, 7, 4, 1, 0, | |
307 0, 1, 3, 7,11,16,21,27,33,39,44,49,53,57,59,60,60,59,57,53,49,44,39,33,27,21,16,11, 7, 3, 1, 0, | |
308 0, 1, 3, 6,11,15,20,26,31,37,42,46,50,53,56,57,57,56,53,50,46,42,37,31,26,20,15,11, 6, 3, 1, 0, | |
309 0, 1, 3, 6, 9,14,19,24,29,34,38,43,46,49,51,52,52,51,49,46,43,38,34,29,24,19,14, 9, 6, 3, 1, 0, | |
310 0, 1, 3, 5, 9,12,17,21,26,30,35,38,42,44,46,47,47,46,44,42,38,35,30,26,21,17,12, 9, 5, 3, 1, 0, | |
311 0, 1, 3, 5, 7,11,15,19,23,26,30,34,37,39,40,41,41,40,39,37,34,30,26,23,19,15,11, 7, 5, 3, 1, 0, | |
312 0, 1, 2, 4, 6, 9,12,16,19,23,26,29,31,33,34,35,35,34,33,31,29,26,23,19,16,12, 9, 6, 4, 2, 1, 0, | |
313 0, 1, 2, 3, 5, 8,10,13,16,19,21,24,26,27,28,29,29,28,27,26,24,21,19,16,13,10, 8, 5, 3, 2, 1, 0, | |
314 0, 1, 1, 2, 4, 6, 8,10,12,15,17,19,20,21,22,23,23,22,21,20,19,17,15,12,10, 8, 6, 4, 2, 1, 1, 0, | |
315 0, 0, 1, 2, 3, 5, 6, 8, 9,11,12,14,15,16,17,17,17,17,16,15,14,12,11, 9, 8, 6, 5, 3, 2, 1, 0, 0, | |
316 0, 0, 1, 1, 2, 3, 4, 5, 6, 7, 9,10,11,11,12,12,12,12,11,11,10, 9, 7, 6, 5, 4, 3, 2, 1, 1, 0, 0, | |
317 0, 0, 1, 1, 1, 2, 2, 3, 4, 5, 5, 6, 7, 7, 7, 7, 7, 7, 7, 7, 6, 5, 5, 4, 3, 2, 2, 1, 1, 1, 0, 0, | |
318 0, 0, 0, 1, 1, 1, 1, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 3, 3, 3, 3, 2, 2, 1, 1, 1, 1, 0, 0, 0, | |
319 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, | |
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, | |
321 //error:0.000022 | |
322 }; | |
323 static const uint8_t obmc16[256]={ | |
324 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, | |
325 0, 0, 1, 2, 3, 4, 5, 5, 5, 5, 4, 3, 2, 1, 0, 0, | |
326 0, 1, 3, 6, 8,11,13,14,14,13,11, 8, 6, 3, 1, 0, | |
327 0, 2, 6,10,15,20,24,26,26,24,20,15,10, 6, 2, 0, | |
328 0, 3, 8,16,23,30,35,38,38,35,30,23,16, 8, 3, 0, | |
329 1, 4,11,20,30,39,46,49,49,46,39,30,20,11, 4, 1, | |
330 1, 5,13,24,35,46,54,58,58,54,46,35,24,13, 5, 1, | |
331 0, 5,14,26,38,49,58,63,63,58,49,38,26,14, 5, 0, | |
332 0, 5,14,26,38,49,58,63,63,58,49,38,26,14, 5, 0, | |
333 1, 5,13,24,35,46,54,58,58,54,46,35,24,13, 5, 1, | |
334 1, 4,11,20,30,39,46,49,49,46,39,30,20,11, 4, 1, | |
335 0, 3, 8,16,23,30,35,38,38,35,30,23,16, 8, 3, 0, | |
336 0, 2, 6,10,15,20,24,26,26,24,20,15,10, 6, 2, 0, | |
337 0, 1, 3, 6, 8,11,13,14,14,13,11, 8, 6, 3, 1, 0, | |
338 0, 0, 1, 2, 3, 4, 5, 5, 5, 5, 4, 3, 2, 1, 0, 0, | |
339 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, | |
340 //error:0.000022 | |
341 }; | |
342 #endif | |
343 | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
344 //linear *64 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
345 static const uint8_t obmc8[64]={ |
3206 | 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, | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
354 //error:0.000000 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
355 }; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
356 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
357 //linear *64 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
358 static const uint8_t obmc4[16]={ |
3206 | 359 16, 48, 48, 16, |
360 48,144,144, 48, | |
361 48,144,144, 48, | |
362 16, 48, 48, 16, | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
363 //error:0.000000 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
364 }; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
365 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
366 static const uint8_t *obmc_tab[4]={ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
367 obmc32, obmc16, obmc8, obmc4 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
368 }; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
369 |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
370 static int scale_mv_ref[MAX_REF_FRAMES][MAX_REF_FRAMES]; |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
371 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
372 typedef struct BlockNode{ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
373 int16_t mx; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
374 int16_t my; |
3314 | 375 uint8_t ref; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
376 uint8_t color[3]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
377 uint8_t type; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
378 //#define TYPE_SPLIT 1 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
379 #define BLOCK_INTRA 1 |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
380 #define BLOCK_OPT 2 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
381 //#define TYPE_NOCOLOR 4 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
382 uint8_t level; //FIXME merge into type? |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
383 }BlockNode; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
384 |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
385 static const BlockNode null_block= { //FIXME add border maybe |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
386 .color= {128,128,128}, |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
387 .mx= 0, |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
388 .my= 0, |
3314 | 389 .ref= 0, |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
390 .type= 0, |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
391 .level= 0, |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
392 }; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
393 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
394 #define LOG2_MB_SIZE 4 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
395 #define MB_SIZE (1<<LOG2_MB_SIZE) |
5575 | 396 #define ENCODER_EXTRA_BITS 4 |
5634
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
397 #define HTAPS 6 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
398 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
399 typedef struct x_and_coeff{ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
400 int16_t x; |
2596 | 401 uint16_t coeff; |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
402 } x_and_coeff; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
403 |
2138 | 404 typedef struct SubBand{ |
405 int level; | |
406 int stride; | |
407 int width; | |
408 int height; | |
409 int qlog; ///< log(qscale)/log[2^(1/6)] | |
410 DWTELEM *buf; | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
411 IDWTELEM *ibuf; |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
412 int buf_x_offset; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
413 int buf_y_offset; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
414 int stride_line; ///< Stride measured in lines, not pixels. |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
415 x_and_coeff * x_coeff; |
2138 | 416 struct SubBand *parent; |
417 uint8_t state[/*7*2*/ 7 + 512][32]; | |
418 }SubBand; | |
419 | |
420 typedef struct Plane{ | |
421 int width; | |
422 int height; | |
423 SubBand band[MAX_DECOMPOSITIONS][4]; | |
424 }Plane; | |
425 | |
426 typedef struct SnowContext{ | |
4588
fc155ff94878
cosmetics: Fix another common typo, dependAnt --> dependEnt.
diego
parents:
4494
diff
changeset
|
427 // MpegEncContext m; // needed for motion estimation, should not be used for anything else, the idea is to make the motion estimation eventually independent of MpegEncContext, so this will be removed then (FIXME/XXX) |
2138 | 428 |
429 AVCodecContext *avctx; | |
2335 | 430 RangeCoder c; |
2138 | 431 DSPContext dsp; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
432 AVFrame new_picture; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
433 AVFrame input_picture; ///< new_picture with the internal linesizes |
2138 | 434 AVFrame current_picture; |
3314 | 435 AVFrame last_picture[MAX_REF_FRAMES]; |
5633
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
436 uint8_t *halfpel_plane[MAX_REF_FRAMES][4][4]; |
2138 | 437 AVFrame mconly_picture; |
438 // uint8_t q_context[16]; | |
439 uint8_t header_state[32]; | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
440 uint8_t block_state[128 + 32*128]; |
2138 | 441 int keyframe; |
2199 | 442 int always_reset; |
2138 | 443 int version; |
444 int spatial_decomposition_type; | |
4331
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
445 int last_spatial_decomposition_type; |
2138 | 446 int temporal_decomposition_type; |
447 int spatial_decomposition_count; | |
448 int temporal_decomposition_count; | |
3314 | 449 int max_ref_frames; |
450 int ref_frames; | |
451 int16_t (*ref_mvs[MAX_REF_FRAMES])[2]; | |
452 uint32_t *ref_scores[MAX_REF_FRAMES]; | |
2138 | 453 DWTELEM *spatial_dwt_buffer; |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
454 IDWTELEM *spatial_idwt_buffer; |
2138 | 455 int colorspace_type; |
456 int chroma_h_shift; | |
457 int chroma_v_shift; | |
458 int spatial_scalability; | |
459 int qlog; | |
4331
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
460 int last_qlog; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
461 int lambda; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
462 int lambda2; |
3313 | 463 int pass1_rc; |
2138 | 464 int mv_scale; |
4331
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
465 int last_mv_scale; |
2138 | 466 int qbias; |
4331
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
467 int last_qbias; |
2138 | 468 #define QBIAS_SHIFT 3 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
469 int b_width; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
470 int b_height; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
471 int block_max_depth; |
4331
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
472 int last_block_max_depth; |
2138 | 473 Plane plane[MAX_PLANES]; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
474 BlockNode *block; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
475 #define ME_CACHE_SIZE 1024 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
476 int me_cache[ME_CACHE_SIZE]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
477 int me_cache_generation; |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
478 slice_buffer sb; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
479 |
4588
fc155ff94878
cosmetics: Fix another common typo, dependAnt --> dependEnt.
diego
parents:
4494
diff
changeset
|
480 MpegEncContext m; // needed for motion estimation, should not be used for anything else, the idea is to make the motion estimation eventually independent of MpegEncContext, so this will be removed then (FIXME/XXX) |
2138 | 481 }SnowContext; |
482 | |
2562 | 483 typedef struct { |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
484 IDWTELEM *b0; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
485 IDWTELEM *b1; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
486 IDWTELEM *b2; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
487 IDWTELEM *b3; |
2562 | 488 int y; |
489 } dwt_compose_t; | |
490 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
491 #define slice_buffer_get_line(slice_buf, line_num) ((slice_buf)->line[line_num] ? (slice_buf)->line[line_num] : slice_buffer_load_line((slice_buf), (line_num))) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
492 //#define slice_buffer_get_line(slice_buf, line_num) (slice_buffer_load_line((slice_buf), (line_num))) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
493 |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
494 static void iterative_me(SnowContext *s); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
495 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
496 static void slice_buffer_init(slice_buffer * buf, int line_count, int max_allocated_lines, int line_width, IDWTELEM * base_buffer) |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
497 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
498 int i; |
2967 | 499 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
500 buf->base_buffer = base_buffer; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
501 buf->line_count = line_count; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
502 buf->line_width = line_width; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
503 buf->data_count = max_allocated_lines; |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
504 buf->line = av_mallocz (sizeof(IDWTELEM *) * line_count); |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
505 buf->data_stack = av_malloc (sizeof(IDWTELEM *) * max_allocated_lines); |
2967 | 506 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
507 for (i = 0; i < max_allocated_lines; i++) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
508 { |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
509 buf->data_stack[i] = av_malloc (sizeof(IDWTELEM) * line_width); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
510 } |
2967 | 511 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
512 buf->data_stack_top = max_allocated_lines - 1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
513 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
514 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
515 static IDWTELEM * slice_buffer_load_line(slice_buffer * buf, int line) |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
516 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
517 int offset; |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
518 IDWTELEM * buffer; |
2967 | 519 |
520 // av_log(NULL, AV_LOG_DEBUG, "Cache hit: %d\n", line); | |
521 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
522 assert(buf->data_stack_top >= 0); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
523 // assert(!buf->line[line]); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
524 if (buf->line[line]) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
525 return buf->line[line]; |
2967 | 526 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
527 offset = buf->line_width * line; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
528 buffer = buf->data_stack[buf->data_stack_top]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
529 buf->data_stack_top--; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
530 buf->line[line] = buffer; |
2967 | 531 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
532 // av_log(NULL, AV_LOG_DEBUG, "slice_buffer_load_line: line: %d remaining: %d\n", line, buf->data_stack_top + 1); |
2967 | 533 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
534 return buffer; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
535 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
536 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
537 static void slice_buffer_release(slice_buffer * buf, int line) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
538 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
539 int offset; |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
540 IDWTELEM * buffer; |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
541 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
542 assert(line >= 0 && line < buf->line_count); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
543 assert(buf->line[line]); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
544 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
545 offset = buf->line_width * line; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
546 buffer = buf->line[line]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
547 buf->data_stack_top++; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
548 buf->data_stack[buf->data_stack_top] = buffer; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
549 buf->line[line] = NULL; |
2967 | 550 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
551 // av_log(NULL, AV_LOG_DEBUG, "slice_buffer_release: line: %d remaining: %d\n", line, buf->data_stack_top + 1); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
552 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
553 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
554 static void slice_buffer_flush(slice_buffer * buf) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
555 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
556 int i; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
557 for (i = 0; i < buf->line_count; i++) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
558 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
559 if (buf->line[i]) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
560 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
561 // av_log(NULL, AV_LOG_DEBUG, "slice_buffer_flush: line: %d \n", i); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
562 slice_buffer_release(buf, i); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
563 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
564 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
565 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
566 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
567 static void slice_buffer_destroy(slice_buffer * buf) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
568 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
569 int i; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
570 slice_buffer_flush(buf); |
2967 | 571 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
572 for (i = buf->data_count - 1; i >= 0; i--) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
573 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
574 assert(buf->data_stack[i]); |
3190 | 575 av_freep(&buf->data_stack[i]); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
576 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
577 assert(buf->data_stack); |
3190 | 578 av_freep(&buf->data_stack); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
579 assert(buf->line); |
3190 | 580 av_freep(&buf->line); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
581 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
582 |
2979 | 583 #ifdef __sgi |
2368
a7ac68734a91
fix for build on IRIX by (Michel Bardiaux {mbardiaux peaktime be})
michael
parents:
2335
diff
changeset
|
584 // Avoid a name clash on SGI IRIX |
2979 | 585 #undef qexp |
2368
a7ac68734a91
fix for build on IRIX by (Michel Bardiaux {mbardiaux peaktime be})
michael
parents:
2335
diff
changeset
|
586 #endif |
2246 | 587 #define QEXPSHIFT (7-FRAC_BITS+8) //FIXME try to change this to 0 |
2600 | 588 static uint8_t qexp[QROOT]; |
2138 | 589 |
590 static inline int mirror(int v, int m){ | |
2998 | 591 while((unsigned)v > (unsigned)m){ |
592 v=-v; | |
593 if(v<0) v+= 2*m; | |
594 } | |
595 return v; | |
2138 | 596 } |
597 | |
2335 | 598 static inline void put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed){ |
2138 | 599 int i; |
600 | |
601 if(v){ | |
4001 | 602 const int a= FFABS(v); |
2138 | 603 const int e= av_log2(a); |
604 #if 1 | |
2967 | 605 const int el= FFMIN(e, 10); |
2335 | 606 put_rac(c, state+0, 0); |
2138 | 607 |
608 for(i=0; i<el; i++){ | |
2335 | 609 put_rac(c, state+1+i, 1); //1..10 |
2138 | 610 } |
611 for(; i<e; i++){ | |
2335 | 612 put_rac(c, state+1+9, 1); //1..10 |
2138 | 613 } |
2335 | 614 put_rac(c, state+1+FFMIN(i,9), 0); |
2138 | 615 |
616 for(i=e-1; i>=el; i--){ | |
2335 | 617 put_rac(c, state+22+9, (a>>i)&1); //22..31 |
2138 | 618 } |
619 for(; i>=0; i--){ | |
2335 | 620 put_rac(c, state+22+i, (a>>i)&1); //22..31 |
2138 | 621 } |
622 | |
623 if(is_signed) | |
2335 | 624 put_rac(c, state+11 + el, v < 0); //11..21 |
2138 | 625 #else |
2967 | 626 |
2335 | 627 put_rac(c, state+0, 0); |
2138 | 628 if(e<=9){ |
629 for(i=0; i<e; i++){ | |
2335 | 630 put_rac(c, state+1+i, 1); //1..10 |
2138 | 631 } |
2335 | 632 put_rac(c, state+1+i, 0); |
2138 | 633 |
634 for(i=e-1; i>=0; i--){ | |
2335 | 635 put_rac(c, state+22+i, (a>>i)&1); //22..31 |
2138 | 636 } |
637 | |
638 if(is_signed) | |
2335 | 639 put_rac(c, state+11 + e, v < 0); //11..21 |
2138 | 640 }else{ |
641 for(i=0; i<e; i++){ | |
2335 | 642 put_rac(c, state+1+FFMIN(i,9), 1); //1..10 |
2138 | 643 } |
2335 | 644 put_rac(c, state+1+FFMIN(i,9), 0); |
2138 | 645 |
646 for(i=e-1; i>=0; i--){ | |
2335 | 647 put_rac(c, state+22+FFMIN(i,9), (a>>i)&1); //22..31 |
2138 | 648 } |
649 | |
650 if(is_signed) | |
2335 | 651 put_rac(c, state+11 + FFMIN(e,10), v < 0); //11..21 |
2138 | 652 } |
653 #endif | |
654 }else{ | |
2335 | 655 put_rac(c, state+0, 1); |
2138 | 656 } |
657 } | |
658 | |
2335 | 659 static inline int get_symbol(RangeCoder *c, uint8_t *state, int is_signed){ |
660 if(get_rac(c, state+0)) | |
2138 | 661 return 0; |
662 else{ | |
2240
c46fed9b7575
simplify getsymbol patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2232
diff
changeset
|
663 int i, e, a; |
c46fed9b7575
simplify getsymbol patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2232
diff
changeset
|
664 e= 0; |
2335 | 665 while(get_rac(c, state+1 + FFMIN(e,9))){ //1..10 |
2240
c46fed9b7575
simplify getsymbol patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2232
diff
changeset
|
666 e++; |
2138 | 667 } |
668 | |
2240
c46fed9b7575
simplify getsymbol patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2232
diff
changeset
|
669 a= 1; |
c46fed9b7575
simplify getsymbol patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2232
diff
changeset
|
670 for(i=e-1; i>=0; i--){ |
2335 | 671 a += a + get_rac(c, state+22 + FFMIN(i,9)); //22..31 |
2240
c46fed9b7575
simplify getsymbol patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2232
diff
changeset
|
672 } |
c46fed9b7575
simplify getsymbol patch by (Loren Merritt <lorenm at u dot washington dot edu>)
michael
parents:
2232
diff
changeset
|
673 |
2335 | 674 if(is_signed && get_rac(c, state+11 + FFMIN(e,10))) //11..21 |
2138 | 675 return -a; |
676 else | |
677 return a; | |
678 } | |
679 } | |
680 | |
2335 | 681 static inline void put_symbol2(RangeCoder *c, uint8_t *state, int v, int log2){ |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
682 int i; |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
683 int r= log2>=0 ? 1<<log2 : 1; |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
684 |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
685 assert(v>=0); |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
686 assert(log2>=-4); |
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
687 |
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
688 while(v >= r){ |
2335 | 689 put_rac(c, state+4+log2, 1); |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
690 v -= r; |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
691 log2++; |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
692 if(log2>0) r+=r; |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
693 } |
2335 | 694 put_rac(c, state+4+log2, 0); |
2967 | 695 |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
696 for(i=log2-1; i>=0; i--){ |
2335 | 697 put_rac(c, state+31-i, (v>>i)&1); |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
698 } |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
699 } |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
700 |
2335 | 701 static inline int get_symbol2(RangeCoder *c, uint8_t *state, int log2){ |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
702 int i; |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
703 int r= log2>=0 ? 1<<log2 : 1; |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
704 int v=0; |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
705 |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
706 assert(log2>=-4); |
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
707 |
2335 | 708 while(get_rac(c, state+4+log2)){ |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
709 v+= r; |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
710 log2++; |
2159
7f42295c1517
improved magnitude coding, 0.2% lower bitrate (foreman@352x288 qscale 1 and 8)
michael
parents:
2156
diff
changeset
|
711 if(log2>0) r+=r; |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
712 } |
2967 | 713 |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
714 for(i=log2-1; i>=0; i--){ |
2335 | 715 v+= get_rac(c, state+31-i)<<i; |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
716 } |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
717 |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
718 return v; |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
719 } |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
720 |
5627 | 721 static av_always_inline void |
722 lift(DWTELEM *dst, DWTELEM *src, DWTELEM *ref, | |
723 int dst_step, int src_step, int ref_step, | |
724 int width, int mul, int add, int shift, | |
725 int highpass, int inverse){ | |
2138 | 726 const int mirror_left= !highpass; |
727 const int mirror_right= (width&1) ^ highpass; | |
728 const int w= (width>>1) - 1 + (highpass & width); | |
729 int i; | |
730 | |
731 #define LIFT(src, ref, inv) ((src) + ((inv) ? - (ref) : + (ref))) | |
732 if(mirror_left){ | |
733 dst[0] = LIFT(src[0], ((mul*2*ref[0]+add)>>shift), inverse); | |
734 dst += dst_step; | |
735 src += src_step; | |
736 } | |
2967 | 737 |
2138 | 738 for(i=0; i<w; i++){ |
5627 | 739 dst[i*dst_step] = |
740 LIFT(src[i*src_step], | |
741 ((mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add)>>shift), | |
742 inverse); | |
2138 | 743 } |
2967 | 744 |
2138 | 745 if(mirror_right){ |
5627 | 746 dst[w*dst_step] = |
747 LIFT(src[w*src_step], | |
748 ((mul*2*ref[w*ref_step]+add)>>shift), | |
749 inverse); | |
2138 | 750 } |
751 } | |
752 | |
5627 | 753 static av_always_inline void |
754 inv_lift(IDWTELEM *dst, IDWTELEM *src, IDWTELEM *ref, | |
755 int dst_step, int src_step, int ref_step, | |
756 int width, int mul, int add, int shift, | |
757 int highpass, int inverse){ | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
758 const int mirror_left= !highpass; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
759 const int mirror_right= (width&1) ^ highpass; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
760 const int w= (width>>1) - 1 + (highpass & width); |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
761 int i; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
762 |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
763 #define LIFT(src, ref, inv) ((src) + ((inv) ? - (ref) : + (ref))) |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
764 if(mirror_left){ |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
765 dst[0] = LIFT(src[0], ((mul*2*ref[0]+add)>>shift), inverse); |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
766 dst += dst_step; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
767 src += src_step; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
768 } |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
769 |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
770 for(i=0; i<w; i++){ |
5627 | 771 dst[i*dst_step] = |
772 LIFT(src[i*src_step], | |
773 ((mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add)>>shift), | |
774 inverse); | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
775 } |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
776 |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
777 if(mirror_right){ |
5627 | 778 dst[w*dst_step] = |
779 LIFT(src[w*src_step], | |
780 ((mul*2*ref[w*ref_step]+add)>>shift), | |
781 inverse); | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
782 } |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
783 } |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
784 |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
785 #ifndef liftS |
5627 | 786 static av_always_inline void |
787 liftS(DWTELEM *dst, DWTELEM *src, DWTELEM *ref, | |
788 int dst_step, int src_step, int ref_step, | |
789 int width, int mul, int add, int shift, | |
790 int highpass, int inverse){ | |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
791 const int mirror_left= !highpass; |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
792 const int mirror_right= (width&1) ^ highpass; |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
793 const int w= (width>>1) - 1 + (highpass & width); |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
794 int i; |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
795 |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
796 assert(shift == 4); |
5627 | 797 #define LIFTS(src, ref, inv) \ |
798 ((inv) ? \ | |
799 (src) + (((ref) + 4*(src))>>shift): \ | |
800 -((-16*(src) + (ref) + add/4 + 1 + (5<<25))/(5*4) - (1<<23))) | |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
801 if(mirror_left){ |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
802 dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse); |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
803 dst += dst_step; |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
804 src += src_step; |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
805 } |
2967 | 806 |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
807 for(i=0; i<w; i++){ |
5627 | 808 dst[i*dst_step] = |
809 LIFTS(src[i*src_step], | |
810 mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add, | |
811 inverse); | |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
812 } |
2967 | 813 |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
814 if(mirror_right){ |
5627 | 815 dst[w*dst_step] = |
816 LIFTS(src[w*src_step], mul*2*ref[w*ref_step]+add, inverse); | |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
817 } |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
818 } |
5627 | 819 static av_always_inline void |
820 inv_liftS(IDWTELEM *dst, IDWTELEM *src, IDWTELEM *ref, | |
821 int dst_step, int src_step, int ref_step, | |
822 int width, int mul, int add, int shift, | |
823 int highpass, int inverse){ | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
824 const int mirror_left= !highpass; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
825 const int mirror_right= (width&1) ^ highpass; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
826 const int w= (width>>1) - 1 + (highpass & width); |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
827 int i; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
828 |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
829 assert(shift == 4); |
5627 | 830 #define LIFTS(src, ref, inv) \ |
831 ((inv) ? \ | |
832 (src) + (((ref) + 4*(src))>>shift): \ | |
833 -((-16*(src) + (ref) + add/4 + 1 + (5<<25))/(5*4) - (1<<23))) | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
834 if(mirror_left){ |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
835 dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse); |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
836 dst += dst_step; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
837 src += src_step; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
838 } |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
839 |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
840 for(i=0; i<w; i++){ |
5627 | 841 dst[i*dst_step] = |
842 LIFTS(src[i*src_step], | |
843 mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add, | |
844 inverse); | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
845 } |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
846 |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
847 if(mirror_right){ |
5627 | 848 dst[w*dst_step] = |
849 LIFTS(src[w*src_step], mul*2*ref[w*ref_step]+add, inverse); | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
850 } |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
851 } |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
852 #endif |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
853 |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
854 static void horizontal_decompose53i(DWTELEM *b, int width){ |
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
855 DWTELEM temp[width]; |
2138 | 856 const int width2= width>>1; |
2893 | 857 int x; |
2138 | 858 const int w2= (width+1)>>1; |
859 | |
860 for(x=0; x<width2; x++){ | |
861 temp[x ]= b[2*x ]; | |
862 temp[x+w2]= b[2*x + 1]; | |
863 } | |
864 if(width&1) | |
865 temp[x ]= b[2*x ]; | |
866 #if 0 | |
2893 | 867 { |
868 int A1,A2,A3,A4; | |
2138 | 869 A2= temp[1 ]; |
870 A4= temp[0 ]; | |
871 A1= temp[0+width2]; | |
872 A1 -= (A2 + A4)>>1; | |
873 A4 += (A1 + 1)>>1; | |
874 b[0+width2] = A1; | |
875 b[0 ] = A4; | |
876 for(x=1; x+1<width2; x+=2){ | |
877 A3= temp[x+width2]; | |
878 A4= temp[x+1 ]; | |
879 A3 -= (A2 + A4)>>1; | |
880 A2 += (A1 + A3 + 2)>>2; | |
881 b[x+width2] = A3; | |
882 b[x ] = A2; | |
883 | |
884 A1= temp[x+1+width2]; | |
885 A2= temp[x+2 ]; | |
886 A1 -= (A2 + A4)>>1; | |
887 A4 += (A1 + A3 + 2)>>2; | |
888 b[x+1+width2] = A1; | |
889 b[x+1 ] = A4; | |
890 } | |
891 A3= temp[width-1]; | |
892 A3 -= A2; | |
893 A2 += (A1 + A3 + 2)>>2; | |
894 b[width -1] = A3; | |
895 b[width2-1] = A2; | |
2893 | 896 } |
2967 | 897 #else |
2138 | 898 lift(b+w2, temp+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 0); |
899 lift(b , temp , b+w2, 1, 1, 1, width, 1, 2, 2, 0, 0); | |
900 #endif | |
901 } | |
902 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
903 static void vertical_decompose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 904 int i; |
2967 | 905 |
2138 | 906 for(i=0; i<width; i++){ |
907 b1[i] -= (b0[i] + b2[i])>>1; | |
908 } | |
909 } | |
910 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
911 static void vertical_decompose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 912 int i; |
2967 | 913 |
2138 | 914 for(i=0; i<width; i++){ |
915 b1[i] += (b0[i] + b2[i] + 2)>>2; | |
916 } | |
917 } | |
918 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
919 static void spatial_decompose53i(DWTELEM *buffer, int width, int height, int stride){ |
2198 | 920 int y; |
2138 | 921 DWTELEM *b0= buffer + mirror(-2-1, height-1)*stride; |
922 DWTELEM *b1= buffer + mirror(-2 , height-1)*stride; | |
2967 | 923 |
2138 | 924 for(y=-2; y<height; y+=2){ |
925 DWTELEM *b2= buffer + mirror(y+1, height-1)*stride; | |
926 DWTELEM *b3= buffer + mirror(y+2, height-1)*stride; | |
927 | |
928 {START_TIMER | |
2998 | 929 if(y+1<(unsigned)height) horizontal_decompose53i(b2, width); |
930 if(y+2<(unsigned)height) horizontal_decompose53i(b3, width); | |
2138 | 931 STOP_TIMER("horizontal_decompose53i")} |
2967 | 932 |
2138 | 933 {START_TIMER |
2998 | 934 if(y+1<(unsigned)height) vertical_decompose53iH0(b1, b2, b3, width); |
935 if(y+0<(unsigned)height) vertical_decompose53iL0(b0, b1, b2, width); | |
2138 | 936 STOP_TIMER("vertical_decompose53i*")} |
2967 | 937 |
2138 | 938 b0=b2; |
939 b1=b3; | |
940 } | |
941 } | |
942 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
943 static void horizontal_decompose97i(DWTELEM *b, int width){ |
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
944 DWTELEM temp[width]; |
2138 | 945 const int w2= (width+1)>>1; |
946 | |
5565
93082c591c8b
Change rounding of the horizontal DWT to match the vertical one.
michael
parents:
5551
diff
changeset
|
947 lift (temp+w2, b +1, b , 1, 2, 2, width, W_AM, W_AO, W_AS, 1, 1); |
93082c591c8b
Change rounding of the horizontal DWT to match the vertical one.
michael
parents:
5551
diff
changeset
|
948 liftS(temp , b , temp+w2, 1, 2, 1, width, W_BM, W_BO, W_BS, 0, 0); |
5589
946c2db0a093
cleanup (remove some old experimentation related code)
michael
parents:
5588
diff
changeset
|
949 lift (b +w2, temp+w2, temp , 1, 1, 1, width, W_CM, W_CO, W_CS, 1, 0); |
2138 | 950 lift (b , temp , b +w2, 1, 1, 1, width, W_DM, W_DO, W_DS, 0, 0); |
951 } | |
952 | |
953 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
954 static void vertical_decompose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 955 int i; |
2967 | 956 |
2138 | 957 for(i=0; i<width; i++){ |
958 b1[i] -= (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS; | |
959 } | |
960 } | |
961 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
962 static void vertical_decompose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 963 int i; |
2967 | 964 |
2138 | 965 for(i=0; i<width; i++){ |
966 b1[i] += (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS; | |
967 } | |
968 } | |
969 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
970 static void vertical_decompose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 971 int i; |
2967 | 972 |
2138 | 973 for(i=0; i<width; i++){ |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
974 #ifdef liftS |
2138 | 975 b1[i] -= (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS; |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
976 #else |
5565
93082c591c8b
Change rounding of the horizontal DWT to match the vertical one.
michael
parents:
5551
diff
changeset
|
977 b1[i] = (16*4*b1[i] - 4*(b0[i] + b2[i]) + W_BO*5 + (5<<27)) / (5*16) - (1<<23); |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
978 #endif |
2138 | 979 } |
980 } | |
981 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
982 static void vertical_decompose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ |
2138 | 983 int i; |
2967 | 984 |
2138 | 985 for(i=0; i<width; i++){ |
986 b1[i] += (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS; | |
987 } | |
988 } | |
989 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
990 static void spatial_decompose97i(DWTELEM *buffer, int width, int height, int stride){ |
2198 | 991 int y; |
2138 | 992 DWTELEM *b0= buffer + mirror(-4-1, height-1)*stride; |
993 DWTELEM *b1= buffer + mirror(-4 , height-1)*stride; | |
994 DWTELEM *b2= buffer + mirror(-4+1, height-1)*stride; | |
995 DWTELEM *b3= buffer + mirror(-4+2, height-1)*stride; | |
2967 | 996 |
2138 | 997 for(y=-4; y<height; y+=2){ |
998 DWTELEM *b4= buffer + mirror(y+3, height-1)*stride; | |
999 DWTELEM *b5= buffer + mirror(y+4, height-1)*stride; | |
1000 | |
1001 {START_TIMER | |
2998 | 1002 if(y+3<(unsigned)height) horizontal_decompose97i(b4, width); |
1003 if(y+4<(unsigned)height) horizontal_decompose97i(b5, width); | |
2138 | 1004 if(width>400){ |
1005 STOP_TIMER("horizontal_decompose97i") | |
1006 }} | |
2967 | 1007 |
2138 | 1008 {START_TIMER |
2998 | 1009 if(y+3<(unsigned)height) vertical_decompose97iH0(b3, b4, b5, width); |
1010 if(y+2<(unsigned)height) vertical_decompose97iL0(b2, b3, b4, width); | |
1011 if(y+1<(unsigned)height) vertical_decompose97iH1(b1, b2, b3, width); | |
1012 if(y+0<(unsigned)height) vertical_decompose97iL1(b0, b1, b2, width); | |
2138 | 1013 |
1014 if(width>400){ | |
1015 STOP_TIMER("vertical_decompose97i") | |
1016 }} | |
2967 | 1017 |
2138 | 1018 b0=b2; |
1019 b1=b3; | |
1020 b2=b4; | |
1021 b3=b5; | |
1022 } | |
1023 } | |
1024 | |
2241
c26038875ebc
consistent use of types patch by (D Richard Felker III <dalias at aerifal dot cx>)
michael
parents:
2240
diff
changeset
|
1025 void ff_spatial_dwt(DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){ |
2138 | 1026 int level; |
2967 | 1027 |
2164 | 1028 for(level=0; level<decomposition_count; level++){ |
1029 switch(type){ | |
3326 | 1030 case DWT_97: spatial_decompose97i(buffer, width>>level, height>>level, stride<<level); break; |
1031 case DWT_53: spatial_decompose53i(buffer, width>>level, height>>level, stride<<level); break; | |
2138 | 1032 } |
1033 } | |
1034 } | |
1035 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1036 static void horizontal_compose53i(IDWTELEM *b, int width){ |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1037 IDWTELEM temp[width]; |
2138 | 1038 const int width2= width>>1; |
1039 const int w2= (width+1)>>1; | |
2893 | 1040 int x; |
2138 | 1041 |
1042 #if 0 | |
2893 | 1043 int A1,A2,A3,A4; |
2138 | 1044 A2= temp[1 ]; |
1045 A4= temp[0 ]; | |
1046 A1= temp[0+width2]; | |
1047 A1 -= (A2 + A4)>>1; | |
1048 A4 += (A1 + 1)>>1; | |
1049 b[0+width2] = A1; | |
1050 b[0 ] = A4; | |
1051 for(x=1; x+1<width2; x+=2){ | |
1052 A3= temp[x+width2]; | |
1053 A4= temp[x+1 ]; | |
1054 A3 -= (A2 + A4)>>1; | |
1055 A2 += (A1 + A3 + 2)>>2; | |
1056 b[x+width2] = A3; | |
1057 b[x ] = A2; | |
1058 | |
1059 A1= temp[x+1+width2]; | |
1060 A2= temp[x+2 ]; | |
1061 A1 -= (A2 + A4)>>1; | |
1062 A4 += (A1 + A3 + 2)>>2; | |
1063 b[x+1+width2] = A1; | |
1064 b[x+1 ] = A4; | |
1065 } | |
1066 A3= temp[width-1]; | |
1067 A3 -= A2; | |
1068 A2 += (A1 + A3 + 2)>>2; | |
1069 b[width -1] = A3; | |
1070 b[width2-1] = A2; | |
2967 | 1071 #else |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1072 inv_lift(temp , b , b+w2, 1, 1, 1, width, 1, 2, 2, 0, 1); |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1073 inv_lift(temp+w2, b+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 1); |
2138 | 1074 #endif |
1075 for(x=0; x<width2; x++){ | |
1076 b[2*x ]= temp[x ]; | |
1077 b[2*x + 1]= temp[x+w2]; | |
1078 } | |
1079 if(width&1) | |
1080 b[2*x ]= temp[x ]; | |
1081 } | |
1082 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1083 static void vertical_compose53iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){ |
2138 | 1084 int i; |
2967 | 1085 |
2138 | 1086 for(i=0; i<width; i++){ |
1087 b1[i] += (b0[i] + b2[i])>>1; | |
1088 } | |
1089 } | |
1090 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1091 static void vertical_compose53iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){ |
2138 | 1092 int i; |
2967 | 1093 |
2138 | 1094 for(i=0; i<width; i++){ |
1095 b1[i] -= (b0[i] + b2[i] + 2)>>2; | |
1096 } | |
1097 } | |
1098 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1099 static void spatial_compose53i_buffered_init(dwt_compose_t *cs, slice_buffer * sb, int height, int stride_line){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1100 cs->b0 = slice_buffer_get_line(sb, mirror(-1-1, height-1) * stride_line); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1101 cs->b1 = slice_buffer_get_line(sb, mirror(-1 , height-1) * stride_line); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1102 cs->y = -1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1103 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1104 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1105 static void spatial_compose53i_init(dwt_compose_t *cs, IDWTELEM *buffer, int height, int stride){ |
2562 | 1106 cs->b0 = buffer + mirror(-1-1, height-1)*stride; |
1107 cs->b1 = buffer + mirror(-1 , height-1)*stride; | |
1108 cs->y = -1; | |
1109 } | |
1110 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1111 static void spatial_compose53i_dy_buffered(dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1112 int y= cs->y; |
2967 | 1113 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1114 IDWTELEM *b0= cs->b0; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1115 IDWTELEM *b1= cs->b1; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1116 IDWTELEM *b2= slice_buffer_get_line(sb, mirror(y+1, height-1) * stride_line); |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1117 IDWTELEM *b3= slice_buffer_get_line(sb, mirror(y+2, height-1) * stride_line); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1118 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1119 {START_TIMER |
2998 | 1120 if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width); |
1121 if(y+0<(unsigned)height) vertical_compose53iH0(b0, b1, b2, width); | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1122 STOP_TIMER("vertical_compose53i*")} |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1123 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1124 {START_TIMER |
2998 | 1125 if(y-1<(unsigned)height) horizontal_compose53i(b0, width); |
1126 if(y+0<(unsigned)height) horizontal_compose53i(b1, width); | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1127 STOP_TIMER("horizontal_compose53i")} |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1128 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1129 cs->b0 = b2; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1130 cs->b1 = b3; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1131 cs->y += 2; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1132 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1133 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1134 static void spatial_compose53i_dy(dwt_compose_t *cs, IDWTELEM *buffer, int width, int height, int stride){ |
2562 | 1135 int y= cs->y; |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1136 IDWTELEM *b0= cs->b0; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1137 IDWTELEM *b1= cs->b1; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1138 IDWTELEM *b2= buffer + mirror(y+1, height-1)*stride; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1139 IDWTELEM *b3= buffer + mirror(y+2, height-1)*stride; |
2138 | 1140 |
1141 {START_TIMER | |
2998 | 1142 if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width); |
1143 if(y+0<(unsigned)height) vertical_compose53iH0(b0, b1, b2, width); | |
2138 | 1144 STOP_TIMER("vertical_compose53i*")} |
1145 | |
1146 {START_TIMER | |
2998 | 1147 if(y-1<(unsigned)height) horizontal_compose53i(b0, width); |
1148 if(y+0<(unsigned)height) horizontal_compose53i(b1, width); | |
2138 | 1149 STOP_TIMER("horizontal_compose53i")} |
1150 | |
2562 | 1151 cs->b0 = b2; |
1152 cs->b1 = b3; | |
1153 cs->y += 2; | |
1154 } | |
1155 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1156 static void spatial_compose53i(IDWTELEM *buffer, int width, int height, int stride){ |
2562 | 1157 dwt_compose_t cs; |
1158 spatial_compose53i_init(&cs, buffer, height, stride); | |
1159 while(cs.y <= height) | |
1160 spatial_compose53i_dy(&cs, buffer, width, height, stride); | |
2967 | 1161 } |
1162 | |
1163 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1164 void ff_snow_horizontal_compose97i(IDWTELEM *b, int width){ |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1165 IDWTELEM temp[width]; |
2138 | 1166 const int w2= (width+1)>>1; |
1167 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1168 inv_lift (temp , b , b +w2, 1, 1, 1, width, W_DM, W_DO, W_DS, 0, 1); |
5589
946c2db0a093
cleanup (remove some old experimentation related code)
michael
parents:
5588
diff
changeset
|
1169 inv_lift (temp+w2, b +w2, temp , 1, 1, 1, width, W_CM, W_CO, W_CS, 1, 1); |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1170 inv_liftS(b , temp , temp+w2, 2, 1, 1, width, W_BM, W_BO, W_BS, 0, 1); |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1171 inv_lift (b+1 , temp+w2, b , 2, 1, 2, width, W_AM, W_AO, W_AS, 1, 0); |
2138 | 1172 } |
1173 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1174 static void vertical_compose97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){ |
2138 | 1175 int i; |
2967 | 1176 |
2138 | 1177 for(i=0; i<width; i++){ |
1178 b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS; | |
1179 } | |
1180 } | |
1181 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1182 static void vertical_compose97iH1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){ |
2138 | 1183 int i; |
2967 | 1184 |
2138 | 1185 for(i=0; i<width; i++){ |
1186 b1[i] -= (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS; | |
1187 } | |
1188 } | |
1189 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1190 static void vertical_compose97iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){ |
2138 | 1191 int i; |
2967 | 1192 |
2138 | 1193 for(i=0; i<width; i++){ |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1194 #ifdef liftS |
2138 | 1195 b1[i] += (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS; |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1196 #else |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1197 b1[i] += (W_BM*(b0[i] + b2[i])+4*b1[i]+W_BO)>>W_BS; |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1198 #endif |
2138 | 1199 } |
1200 } | |
1201 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1202 static void vertical_compose97iL1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){ |
2138 | 1203 int i; |
2967 | 1204 |
2138 | 1205 for(i=0; i<width; i++){ |
1206 b1[i] -= (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS; | |
1207 } | |
1208 } | |
1209 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1210 void ff_snow_vertical_compose97i(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){ |
2592 | 1211 int i; |
2967 | 1212 |
2592 | 1213 for(i=0; i<width; i++){ |
1214 b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS; | |
1215 b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS; | |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1216 #ifdef liftS |
2592 | 1217 b2[i] += (W_BM*(b1[i] + b3[i])+W_BO)>>W_BS; |
2602
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1218 #else |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1219 b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS; |
5ec55feb6fdd
rescale coefficients during IDWT, that way the lifting steps are much simpler and faster
michael
parents:
2601
diff
changeset
|
1220 #endif |
2592 | 1221 b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS; |
1222 } | |
1223 } | |
1224 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1225 static void spatial_compose97i_buffered_init(dwt_compose_t *cs, slice_buffer * sb, int height, int stride_line){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1226 cs->b0 = slice_buffer_get_line(sb, mirror(-3-1, height-1) * stride_line); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1227 cs->b1 = slice_buffer_get_line(sb, mirror(-3 , height-1) * stride_line); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1228 cs->b2 = slice_buffer_get_line(sb, mirror(-3+1, height-1) * stride_line); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1229 cs->b3 = slice_buffer_get_line(sb, mirror(-3+2, height-1) * stride_line); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1230 cs->y = -3; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1231 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1232 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1233 static void spatial_compose97i_init(dwt_compose_t *cs, IDWTELEM *buffer, int height, int stride){ |
2562 | 1234 cs->b0 = buffer + mirror(-3-1, height-1)*stride; |
1235 cs->b1 = buffer + mirror(-3 , height-1)*stride; | |
1236 cs->b2 = buffer + mirror(-3+1, height-1)*stride; | |
1237 cs->b3 = buffer + mirror(-3+2, height-1)*stride; | |
1238 cs->y = -3; | |
1239 } | |
2138 | 1240 |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
1241 static void spatial_compose97i_dy_buffered(DSPContext *dsp, dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1242 int y = cs->y; |
2967 | 1243 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1244 IDWTELEM *b0= cs->b0; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1245 IDWTELEM *b1= cs->b1; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1246 IDWTELEM *b2= cs->b2; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1247 IDWTELEM *b3= cs->b3; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1248 IDWTELEM *b4= slice_buffer_get_line(sb, mirror(y + 3, height - 1) * stride_line); |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1249 IDWTELEM *b5= slice_buffer_get_line(sb, mirror(y + 4, height - 1) * stride_line); |
2967 | 1250 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1251 {START_TIMER |
2592 | 1252 if(y>0 && y+4<height){ |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
1253 dsp->vertical_compose97i(b0, b1, b2, b3, b4, b5, width); |
2592 | 1254 }else{ |
2998 | 1255 if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width); |
1256 if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width); | |
1257 if(y+1<(unsigned)height) vertical_compose97iL0(b1, b2, b3, width); | |
1258 if(y+0<(unsigned)height) vertical_compose97iH0(b0, b1, b2, width); | |
2592 | 1259 } |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1260 if(width>400){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1261 STOP_TIMER("vertical_compose97i")}} |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1262 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1263 {START_TIMER |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
1264 if(y-1<(unsigned)height) dsp->horizontal_compose97i(b0, width); |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
1265 if(y+0<(unsigned)height) dsp->horizontal_compose97i(b1, width); |
3012 | 1266 if(width>400 && y+0<(unsigned)height){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1267 STOP_TIMER("horizontal_compose97i")}} |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1268 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1269 cs->b0=b2; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1270 cs->b1=b3; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1271 cs->b2=b4; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1272 cs->b3=b5; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1273 cs->y += 2; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1274 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1275 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1276 static void spatial_compose97i_dy(dwt_compose_t *cs, IDWTELEM *buffer, int width, int height, int stride){ |
2562 | 1277 int y = cs->y; |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1278 IDWTELEM *b0= cs->b0; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1279 IDWTELEM *b1= cs->b1; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1280 IDWTELEM *b2= cs->b2; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1281 IDWTELEM *b3= cs->b3; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1282 IDWTELEM *b4= buffer + mirror(y+3, height-1)*stride; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1283 IDWTELEM *b5= buffer + mirror(y+4, height-1)*stride; |
2138 | 1284 |
1285 {START_TIMER | |
2998 | 1286 if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width); |
1287 if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width); | |
1288 if(y+1<(unsigned)height) vertical_compose97iL0(b1, b2, b3, width); | |
1289 if(y+0<(unsigned)height) vertical_compose97iH0(b0, b1, b2, width); | |
2138 | 1290 if(width>400){ |
1291 STOP_TIMER("vertical_compose97i")}} | |
1292 | |
1293 {START_TIMER | |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
1294 if(y-1<(unsigned)height) ff_snow_horizontal_compose97i(b0, width); |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
1295 if(y+0<(unsigned)height) ff_snow_horizontal_compose97i(b1, width); |
2138 | 1296 if(width>400 && b0 <= b2){ |
1297 STOP_TIMER("horizontal_compose97i")}} | |
2562 | 1298 |
1299 cs->b0=b2; | |
1300 cs->b1=b3; | |
1301 cs->b2=b4; | |
1302 cs->b3=b5; | |
1303 cs->y += 2; | |
1304 } | |
1305 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1306 static void spatial_compose97i(IDWTELEM *buffer, int width, int height, int stride){ |
2562 | 1307 dwt_compose_t cs; |
1308 spatial_compose97i_init(&cs, buffer, height, stride); | |
1309 while(cs.y <= height) | |
1310 spatial_compose97i_dy(&cs, buffer, width, height, stride); | |
1311 } | |
1312 | |
3075 | 1313 static void ff_spatial_idwt_buffered_init(dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line, int type, int decomposition_count){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1314 int level; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1315 for(level=decomposition_count-1; level>=0; level--){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1316 switch(type){ |
3326 | 1317 case DWT_97: spatial_compose97i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break; |
1318 case DWT_53: spatial_compose53i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break; | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1319 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1320 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1321 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1322 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1323 static void ff_spatial_idwt_init(dwt_compose_t *cs, IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){ |
2562 | 1324 int level; |
1325 for(level=decomposition_count-1; level>=0; level--){ | |
1326 switch(type){ | |
3326 | 1327 case DWT_97: spatial_compose97i_init(cs+level, buffer, height>>level, stride<<level); break; |
1328 case DWT_53: spatial_compose53i_init(cs+level, buffer, height>>level, stride<<level); break; | |
2562 | 1329 } |
1330 } | |
1331 } | |
1332 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1333 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){ |
2562 | 1334 const int support = type==1 ? 3 : 5; |
1335 int level; | |
1336 if(type==2) return; | |
1337 | |
1338 for(level=decomposition_count-1; level>=0; level--){ | |
1339 while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){ | |
1340 switch(type){ | |
3326 | 1341 case DWT_97: spatial_compose97i_dy(cs+level, buffer, width>>level, height>>level, stride<<level); |
2562 | 1342 break; |
3326 | 1343 case DWT_53: spatial_compose53i_dy(cs+level, buffer, width>>level, height>>level, stride<<level); |
2562 | 1344 break; |
1345 } | |
1346 } | |
2138 | 1347 } |
1348 } | |
1349 | |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
1350 static void ff_spatial_idwt_buffered_slice(DSPContext *dsp, dwt_compose_t *cs, slice_buffer * slice_buf, int width, int height, int stride_line, int type, int decomposition_count, int y){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1351 const int support = type==1 ? 3 : 5; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1352 int level; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1353 if(type==2) return; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1354 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1355 for(level=decomposition_count-1; level>=0; level--){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1356 while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1357 switch(type){ |
3326 | 1358 case DWT_97: spatial_compose97i_dy_buffered(dsp, cs+level, slice_buf, width>>level, height>>level, stride_line<<level); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1359 break; |
3326 | 1360 case DWT_53: spatial_compose53i_dy_buffered(cs+level, slice_buf, width>>level, height>>level, stride_line<<level); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1361 break; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1362 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1363 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1364 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1365 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1366 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1367 static void ff_spatial_idwt(IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){ |
2562 | 1368 dwt_compose_t cs[MAX_DECOMPOSITIONS]; |
1369 int y; | |
1370 ff_spatial_idwt_init(cs, buffer, width, height, stride, type, decomposition_count); | |
1371 for(y=0; y<height; y+=4) | |
1372 ff_spatial_idwt_slice(cs, buffer, width, height, stride, type, decomposition_count, y); | |
2138 | 1373 } |
1374 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1375 static int encode_subband_c0run(SnowContext *s, SubBand *b, IDWTELEM *src, IDWTELEM *parent, int stride, int orientation){ |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1376 const int w= b->width; |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1377 const int h= b->height; |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1378 int x, y; |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1379 |
2138 | 1380 if(1){ |
1381 int run=0; | |
2149 | 1382 int runs[w*h]; |
2138 | 1383 int run_index=0; |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1384 int max_index; |
2967 | 1385 |
2138 | 1386 for(y=0; y<h; y++){ |
1387 for(x=0; x<w; x++){ | |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1388 int v, p=0; |
2144 | 1389 int /*ll=0, */l=0, lt=0, t=0, rt=0; |
2149 | 1390 v= src[x + y*stride]; |
2138 | 1391 |
1392 if(y){ | |
2149 | 1393 t= src[x + (y-1)*stride]; |
2138 | 1394 if(x){ |
2149 | 1395 lt= src[x - 1 + (y-1)*stride]; |
2138 | 1396 } |
1397 if(x + 1 < w){ | |
2149 | 1398 rt= src[x + 1 + (y-1)*stride]; |
2138 | 1399 } |
1400 } | |
1401 if(x){ | |
2149 | 1402 l= src[x - 1 + y*stride]; |
2144 | 1403 /*if(x > 1){ |
1404 if(orientation==1) ll= src[y + (x-2)*stride]; | |
1405 else ll= src[x - 2 + y*stride]; | |
2138 | 1406 }*/ |
1407 } | |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1408 if(parent){ |
2149 | 1409 int px= x>>1; |
1410 int py= y>>1; | |
2967 | 1411 if(px<b->parent->width && py<b->parent->height) |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1412 p= parent[px + py*2*stride]; |
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1413 } |
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1414 if(!(/*ll|*/l|lt|t|rt|p)){ |
2138 | 1415 if(v){ |
1416 runs[run_index++]= run; | |
1417 run=0; | |
1418 }else{ | |
1419 run++; | |
1420 } | |
1421 } | |
1422 } | |
1423 } | |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1424 max_index= run_index; |
2138 | 1425 runs[run_index++]= run; |
1426 run_index=0; | |
1427 run= runs[run_index++]; | |
1428 | |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1429 put_symbol2(&s->c, b->state[30], max_index, 0); |
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1430 if(run_index <= max_index) |
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1431 put_symbol2(&s->c, b->state[1], run, 3); |
2967 | 1432 |
2138 | 1433 for(y=0; y<h; y++){ |
2435
c89ac0e70c66
10l patch by (matthieu castet <castet.matthieu free fr>)
michael
parents:
2422
diff
changeset
|
1434 if(s->c.bytestream_end - s->c.bytestream < w*40){ |
2422 | 1435 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n"); |
1436 return -1; | |
1437 } | |
2138 | 1438 for(x=0; x<w; x++){ |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1439 int v, p=0; |
2144 | 1440 int /*ll=0, */l=0, lt=0, t=0, rt=0; |
2149 | 1441 v= src[x + y*stride]; |
2138 | 1442 |
1443 if(y){ | |
2149 | 1444 t= src[x + (y-1)*stride]; |
2138 | 1445 if(x){ |
2149 | 1446 lt= src[x - 1 + (y-1)*stride]; |
2138 | 1447 } |
1448 if(x + 1 < w){ | |
2149 | 1449 rt= src[x + 1 + (y-1)*stride]; |
2138 | 1450 } |
1451 } | |
1452 if(x){ | |
2149 | 1453 l= src[x - 1 + y*stride]; |
2144 | 1454 /*if(x > 1){ |
1455 if(orientation==1) ll= src[y + (x-2)*stride]; | |
1456 else ll= src[x - 2 + y*stride]; | |
2138 | 1457 }*/ |
1458 } | |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1459 if(parent){ |
2149 | 1460 int px= x>>1; |
1461 int py= y>>1; | |
2967 | 1462 if(px<b->parent->width && py<b->parent->height) |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1463 p= parent[px + py*2*stride]; |
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1464 } |
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1465 if(/*ll|*/l|lt|t|rt|p){ |
4001 | 1466 int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p)); |
2144 | 1467 |
2335 | 1468 put_rac(&s->c, &b->state[0][context], !!v); |
2138 | 1469 }else{ |
1470 if(!run){ | |
1471 run= runs[run_index++]; | |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1472 |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1473 if(run_index <= max_index) |
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1474 put_symbol2(&s->c, b->state[1], run, 3); |
2138 | 1475 assert(v); |
1476 }else{ | |
1477 run--; | |
1478 assert(!v); | |
1479 } | |
1480 } | |
1481 if(v){ | |
4001 | 1482 int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p)); |
1483 int l2= 2*FFABS(l) + (l<0); | |
1484 int t2= 2*FFABS(t) + (t<0); | |
1485 | |
1486 put_symbol2(&s->c, b->state[context + 2], FFABS(v)-1, context-4); | |
2596 | 1487 put_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l2&0xFF] + 3*quant3bA[t2&0xFF]], v<0); |
2138 | 1488 } |
1489 } | |
1490 } | |
1491 } | |
2422 | 1492 return 0; |
2138 | 1493 } |
1494 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1495 static int encode_subband(SnowContext *s, SubBand *b, IDWTELEM *src, IDWTELEM *parent, int stride, int orientation){ |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1496 // encode_subband_qtree(s, b, src, parent, stride, orientation); |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1497 // encode_subband_z0run(s, b, src, parent, stride, orientation); |
2422 | 1498 return encode_subband_c0run(s, b, src, parent, stride, orientation); |
2155
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1499 // encode_subband_dzr(s, b, src, parent, stride, orientation); |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1500 } |
274a01d80f4a
various subband encoders (all either worse or complicated so they are commented out)
michael
parents:
2152
diff
changeset
|
1501 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1502 static inline void unpack_coeffs(SnowContext *s, SubBand *b, SubBand * parent, int orientation){ |
2138 | 1503 const int w= b->width; |
1504 const int h= b->height; | |
1505 int x,y; | |
2967 | 1506 |
2138 | 1507 if(1){ |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1508 int run, runs; |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1509 x_and_coeff *xc= b->x_coeff; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1510 x_and_coeff *prev_xc= NULL; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1511 x_and_coeff *prev2_xc= xc; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1512 x_and_coeff *parent_xc= parent ? parent->x_coeff : NULL; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1513 x_and_coeff *prev_parent_xc= parent_xc; |
2138 | 1514 |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1515 runs= get_symbol2(&s->c, b->state[30], 0); |
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1516 if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3); |
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1517 else run= INT_MAX; |
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1518 |
2138 | 1519 for(y=0; y<h; y++){ |
2193 | 1520 int v=0; |
1521 int lt=0, t=0, rt=0; | |
1522 | |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1523 if(y && prev_xc->x == 0){ |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1524 rt= prev_xc->coeff; |
2193 | 1525 } |
2138 | 1526 for(x=0; x<w; x++){ |
2193 | 1527 int p=0; |
1528 const int l= v; | |
2967 | 1529 |
2193 | 1530 lt= t; t= rt; |
1531 | |
2194 | 1532 if(y){ |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1533 if(prev_xc->x <= x) |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1534 prev_xc++; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1535 if(prev_xc->x == x + 1) |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1536 rt= prev_xc->coeff; |
2194 | 1537 else |
1538 rt=0; | |
1539 } | |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1540 if(parent_xc){ |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1541 if(x>>1 > parent_xc->x){ |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1542 parent_xc++; |
2192 | 1543 } |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1544 if(x>>1 == parent_xc->x){ |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1545 p= parent_xc->coeff; |
2194 | 1546 } |
2148
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1547 } |
678be5a8f282
use parent sample to predict significance & magnitude
michael
parents:
2146
diff
changeset
|
1548 if(/*ll|*/l|lt|t|rt|p){ |
4001 | 1549 int context= av_log2(/*FFABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1)); |
2144 | 1550 |
2335 | 1551 v=get_rac(&s->c, &b->state[0][context]); |
2605 | 1552 if(v){ |
1553 v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1); | |
1554 v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l&0xFF] + 3*quant3bA[t&0xFF]]); | |
2967 | 1555 |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1556 xc->x=x; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1557 (xc++)->coeff= v; |
2605 | 1558 } |
2138 | 1559 }else{ |
1560 if(!run){ | |
2609
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1561 if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3); |
0f74a379a890
store the number of runs to avoid storing the last run value
michael
parents:
2608
diff
changeset
|
1562 else run= INT_MAX; |
2605 | 1563 v= 2*(get_symbol2(&s->c, b->state[0 + 2], 0-4) + 1); |
1564 v+=get_rac(&s->c, &b->state[0][16 + 1 + 3]); | |
2967 | 1565 |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1566 xc->x=x; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1567 (xc++)->coeff= v; |
2138 | 1568 }else{ |
2606 | 1569 int max_run; |
2138 | 1570 run--; |
1571 v=0; | |
2191 | 1572 |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1573 if(y) max_run= FFMIN(run, prev_xc->x - x - 2); |
2606 | 1574 else max_run= FFMIN(run, w-x-1); |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1575 if(parent_xc) |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1576 max_run= FFMIN(max_run, 2*parent_xc->x - x - 1); |
2606 | 1577 x+= max_run; |
1578 run-= max_run; | |
2138 | 1579 } |
1580 } | |
2192 | 1581 } |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1582 (xc++)->x= w+1; //end marker |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1583 prev_xc= prev2_xc; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1584 prev2_xc= xc; |
2967 | 1585 |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1586 if(parent_xc){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1587 if(y&1){ |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1588 while(parent_xc->x != parent->width+1) |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1589 parent_xc++; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1590 parent_xc++; |
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1591 prev_parent_xc= parent_xc; |
2192 | 1592 }else{ |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1593 parent_xc= prev_parent_xc; |
2138 | 1594 } |
1595 } | |
1596 } | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1597 |
2607
fde7b6fe2aaf
replace complicated pointer dereference + index stuff by pointers in unpack_coeffs()
michael
parents:
2606
diff
changeset
|
1598 (xc++)->x= w+1; //end marker |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1599 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1600 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1601 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1602 static inline void decode_subband_slice_buffered(SnowContext *s, SubBand *b, slice_buffer * sb, int start_y, int h, int save_state[1]){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1603 const int w= b->width; |
2893 | 1604 int y; |
4594 | 1605 const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16); |
2600 | 1606 int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1607 int qadd= (s->qbias*qmul)>>QBIAS_SHIFT; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1608 int new_index = 0; |
2967 | 1609 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1610 START_TIMER |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1611 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1612 if(b->ibuf == s->spatial_idwt_buffer || s->qlog == LOSSLESS_QLOG){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1613 qadd= 0; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1614 qmul= 1<<QEXPSHIFT; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1615 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1616 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1617 /* If we are on the second or later slice, restore our index. */ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1618 if (start_y != 0) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1619 new_index = save_state[0]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1620 |
2967 | 1621 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1622 for(y=start_y; y<h; y++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1623 int x = 0; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1624 int v; |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1625 IDWTELEM * line = slice_buffer_get_line(sb, y * b->stride_line + b->buf_y_offset) + b->buf_x_offset; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
1626 memset(line, 0, b->width*sizeof(IDWTELEM)); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1627 v = b->x_coeff[new_index].coeff; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1628 x = b->x_coeff[new_index++].x; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1629 while(x < w) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1630 { |
2596 | 1631 register int t= ( (v>>1)*qmul + qadd)>>QEXPSHIFT; |
1632 register int u= -(v&1); | |
1633 line[x] = (t^u) - u; | |
1634 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1635 v = b->x_coeff[new_index].coeff; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1636 x = b->x_coeff[new_index++].x; |
2138 | 1637 } |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1638 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1639 if(w > 200 && start_y != 0/*level+1 == s->spatial_decomposition_count*/){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1640 STOP_TIMER("decode_subband") |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1641 } |
2967 | 1642 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1643 /* Save our variables for the next slice. */ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1644 save_state[0] = new_index; |
2967 | 1645 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
1646 return; |
2138 | 1647 } |
1648 | |
4331
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
1649 static void reset_contexts(SnowContext *s){ //FIXME better initial contexts |
2138 | 1650 int plane_index, level, orientation; |
1651 | |
2199 | 1652 for(plane_index=0; plane_index<3; plane_index++){ |
2138 | 1653 for(level=0; level<s->spatial_decomposition_count; level++){ |
1654 for(orientation=level ? 1:0; orientation<4; orientation++){ | |
2335 | 1655 memset(s->plane[plane_index].band[level][orientation].state, MID_STATE, sizeof(s->plane[plane_index].band[level][orientation].state)); |
2138 | 1656 } |
1657 } | |
1658 } | |
2335 | 1659 memset(s->header_state, MID_STATE, sizeof(s->header_state)); |
1660 memset(s->block_state, MID_STATE, sizeof(s->block_state)); | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1661 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1662 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1663 static int alloc_blocks(SnowContext *s){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1664 int w= -((-s->avctx->width )>>LOG2_MB_SIZE); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1665 int h= -((-s->avctx->height)>>LOG2_MB_SIZE); |
2967 | 1666 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1667 s->b_width = w; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1668 s->b_height= h; |
2967 | 1669 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1670 s->block= av_mallocz(w * h * sizeof(BlockNode) << (s->block_max_depth*2)); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1671 return 0; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1672 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1673 |
2335 | 1674 static inline void copy_rac_state(RangeCoder *d, RangeCoder *s){ |
1675 uint8_t *bytestream= d->bytestream; | |
1676 uint8_t *bytestream_start= d->bytestream_start; | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1677 *d= *s; |
2335 | 1678 d->bytestream= bytestream; |
1679 d->bytestream_start= bytestream_start; | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1680 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1681 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1682 //near copy & paste from dsputil, FIXME |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1683 static int pix_sum(uint8_t * pix, int line_size, int w) |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1684 { |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1685 int s, i, j; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1686 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1687 s = 0; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1688 for (i = 0; i < w; i++) { |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1689 for (j = 0; j < w; j++) { |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1690 s += pix[0]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1691 pix ++; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1692 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1693 pix += line_size - w; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1694 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1695 return s; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1696 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1697 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1698 //near copy & paste from dsputil, FIXME |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1699 static int pix_norm1(uint8_t * pix, int line_size, int w) |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1700 { |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1701 int s, i, j; |
4179 | 1702 uint32_t *sq = ff_squareTbl + 256; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1703 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1704 s = 0; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1705 for (i = 0; i < w; i++) { |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1706 for (j = 0; j < w; j ++) { |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1707 s += sq[pix[0]]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1708 pix ++; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1709 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1710 pix += line_size - w; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1711 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1712 return s; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1713 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1714 |
3314 | 1715 static inline void set_blocks(SnowContext *s, int level, int x, int y, int l, int cb, int cr, int mx, int my, int ref, int type){ |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1716 const int w= s->b_width << s->block_max_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1717 const int rem_depth= s->block_max_depth - level; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1718 const int index= (x + y*w) << rem_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1719 const int block_w= 1<<rem_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1720 BlockNode block; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1721 int i,j; |
2967 | 1722 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1723 block.color[0]= l; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1724 block.color[1]= cb; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1725 block.color[2]= cr; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1726 block.mx= mx; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1727 block.my= my; |
3314 | 1728 block.ref= ref; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1729 block.type= type; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1730 block.level= level; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1731 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1732 for(j=0; j<block_w; j++){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1733 for(i=0; i<block_w; i++){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1734 s->block[index + i + j*w]= block; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1735 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1736 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1737 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1738 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1739 static inline void init_ref(MotionEstContext *c, uint8_t *src[3], uint8_t *ref[3], uint8_t *ref2[3], int x, int y, int ref_index){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1740 const int offset[3]= { |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1741 y*c-> stride + x, |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1742 ((y*c->uvstride + x)>>1), |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1743 ((y*c->uvstride + x)>>1), |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1744 }; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1745 int i; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1746 for(i=0; i<3; i++){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1747 c->src[0][i]= src [i]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1748 c->ref[0][i]= ref [i] + offset[i]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1749 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1750 assert(!ref_index); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1751 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1752 |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1753 static inline void pred_mv(SnowContext *s, int *mx, int *my, int ref, |
4408
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
1754 const BlockNode *left, const BlockNode *top, const BlockNode *tr){ |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1755 if(s->ref_frames == 1){ |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1756 *mx = mid_pred(left->mx, top->mx, tr->mx); |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1757 *my = mid_pred(left->my, top->my, tr->my); |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1758 }else{ |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1759 const int *scale = scale_mv_ref[ref]; |
4407
15688bdfe1b5
Brings down the number of snow.c warnings from 33 to 27 by putting parentheses
takis
parents:
4360
diff
changeset
|
1760 *mx = mid_pred((left->mx * scale[left->ref] + 128) >>8, |
15688bdfe1b5
Brings down the number of snow.c warnings from 33 to 27 by putting parentheses
takis
parents:
4360
diff
changeset
|
1761 (top ->mx * scale[top ->ref] + 128) >>8, |
15688bdfe1b5
Brings down the number of snow.c warnings from 33 to 27 by putting parentheses
takis
parents:
4360
diff
changeset
|
1762 (tr ->mx * scale[tr ->ref] + 128) >>8); |
15688bdfe1b5
Brings down the number of snow.c warnings from 33 to 27 by putting parentheses
takis
parents:
4360
diff
changeset
|
1763 *my = mid_pred((left->my * scale[left->ref] + 128) >>8, |
15688bdfe1b5
Brings down the number of snow.c warnings from 33 to 27 by putting parentheses
takis
parents:
4360
diff
changeset
|
1764 (top ->my * scale[top ->ref] + 128) >>8, |
15688bdfe1b5
Brings down the number of snow.c warnings from 33 to 27 by putting parentheses
takis
parents:
4360
diff
changeset
|
1765 (tr ->my * scale[tr ->ref] + 128) >>8); |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1766 } |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1767 } |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1768 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1769 //FIXME copy&paste |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1770 #define P_LEFT P[1] |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1771 #define P_TOP P[2] |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1772 #define P_TOPRIGHT P[3] |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1773 #define P_MEDIAN P[4] |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1774 #define P_MV1 P[9] |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1775 #define FLAG_QPEL 1 //must be 1 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1776 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1777 static int encode_q_branch(SnowContext *s, int level, int x, int y){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1778 uint8_t p_buffer[1024]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1779 uint8_t i_buffer[1024]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1780 uint8_t p_state[sizeof(s->block_state)]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1781 uint8_t i_state[sizeof(s->block_state)]; |
2335 | 1782 RangeCoder pc, ic; |
1783 uint8_t *pbbak= s->c.bytestream; | |
1784 uint8_t *pbbak_start= s->c.bytestream_start; | |
5082 | 1785 int score, score2, iscore, i_len, p_len, block_s, sum, base_bits; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1786 const int w= s->b_width << s->block_max_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1787 const int h= s->b_height << s->block_max_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1788 const int rem_depth= s->block_max_depth - level; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1789 const int index= (x + y*w) << rem_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1790 const int block_w= 1<<(LOG2_MB_SIZE - level); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1791 int trx= (x+1)<<rem_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1792 int try= (y+1)<<rem_depth; |
4408
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
1793 const BlockNode *left = x ? &s->block[index-1] : &null_block; |
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
1794 const BlockNode *top = y ? &s->block[index-w] : &null_block; |
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
1795 const BlockNode *right = trx<w ? &s->block[index+1] : &null_block; |
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
1796 const BlockNode *bottom= try<h ? &s->block[index+w] : &null_block; |
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
1797 const BlockNode *tl = y && x ? &s->block[index-w-1] : left; |
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
1798 const BlockNode *tr = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1799 int pl = left->color[0]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1800 int pcb= left->color[1]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1801 int pcr= left->color[2]; |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1802 int pmx, pmy; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1803 int mx=0, my=0; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
1804 int l,cr,cb; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1805 const int stride= s->current_picture.linesize[0]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1806 const int uvstride= s->current_picture.linesize[1]; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
1807 uint8_t *current_data[3]= { s->input_picture.data[0] + (x + y* stride)*block_w, |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
1808 s->input_picture.data[1] + (x + y*uvstride)*block_w/2, |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
1809 s->input_picture.data[2] + (x + y*uvstride)*block_w/2}; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1810 int P[10][2]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1811 int16_t last_mv[3][2]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1812 int qpel= !!(s->avctx->flags & CODEC_FLAG_QPEL); //unused |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1813 const int shift= 1+qpel; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1814 MotionEstContext *c= &s->m.me; |
3314 | 1815 int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref); |
4001 | 1816 int mx_context= av_log2(2*FFABS(left->mx - top->mx)); |
1817 int my_context= av_log2(2*FFABS(left->my - top->my)); | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1818 int s_context= 2*left->level + 2*top->level + tl->level + tr->level; |
3314 | 1819 int ref, best_ref, ref_score, ref_mx, ref_my; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1820 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1821 assert(sizeof(s->block_state) >= 256); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1822 if(s->keyframe){ |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1823 set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1824 return 0; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1825 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1826 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1827 // clip predictors / edge ? |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1828 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1829 P_LEFT[0]= left->mx; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1830 P_LEFT[1]= left->my; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1831 P_TOP [0]= top->mx; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1832 P_TOP [1]= top->my; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1833 P_TOPRIGHT[0]= tr->mx; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1834 P_TOPRIGHT[1]= tr->my; |
2967 | 1835 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1836 last_mv[0][0]= s->block[index].mx; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1837 last_mv[0][1]= s->block[index].my; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1838 last_mv[1][0]= right->mx; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1839 last_mv[1][1]= right->my; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1840 last_mv[2][0]= bottom->mx; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1841 last_mv[2][1]= bottom->my; |
2967 | 1842 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1843 s->m.mb_stride=2; |
2967 | 1844 s->m.mb_x= |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1845 s->m.mb_y= 0; |
4360
5f887a3c7281
Change the Snow encoder to always use the available MotionEstContext pointer.
takis
parents:
4332
diff
changeset
|
1846 c->skip= 0; |
5f887a3c7281
Change the Snow encoder to always use the available MotionEstContext pointer.
takis
parents:
4332
diff
changeset
|
1847 |
5f887a3c7281
Change the Snow encoder to always use the available MotionEstContext pointer.
takis
parents:
4332
diff
changeset
|
1848 assert(c-> stride == stride); |
5f887a3c7281
Change the Snow encoder to always use the available MotionEstContext pointer.
takis
parents:
4332
diff
changeset
|
1849 assert(c->uvstride == uvstride); |
2967 | 1850 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1851 c->penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1852 c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1853 c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1854 c->current_mv_penalty= c->mv_penalty[s->m.f_code=1] + MAX_MV; |
2967 | 1855 |
2206 | 1856 c->xmin = - x*block_w - 16+2; |
1857 c->ymin = - y*block_w - 16+2; | |
1858 c->xmax = - (x+1)*block_w + (w<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-2; | |
1859 c->ymax = - (y+1)*block_w + (h<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-2; | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1860 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1861 if(P_LEFT[0] > (c->xmax<<shift)) P_LEFT[0] = (c->xmax<<shift); |
2967 | 1862 if(P_LEFT[1] > (c->ymax<<shift)) P_LEFT[1] = (c->ymax<<shift); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1863 if(P_TOP[0] > (c->xmax<<shift)) P_TOP[0] = (c->xmax<<shift); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1864 if(P_TOP[1] > (c->ymax<<shift)) P_TOP[1] = (c->ymax<<shift); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1865 if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1866 if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift); //due to pmx no clip |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1867 if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1868 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1869 P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1870 P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1871 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1872 if (!y) { |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1873 c->pred_x= P_LEFT[0]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1874 c->pred_y= P_LEFT[1]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1875 } else { |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1876 c->pred_x = P_MEDIAN[0]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1877 c->pred_y = P_MEDIAN[1]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1878 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1879 |
3314 | 1880 score= INT_MAX; |
1881 best_ref= 0; | |
1882 for(ref=0; ref<s->ref_frames; ref++){ | |
1883 init_ref(c, current_data, s->last_picture[ref].data, NULL, block_w*x, block_w*y, 0); | |
1884 | |
1885 ref_score= ff_epzs_motion_search(&s->m, &ref_mx, &ref_my, P, 0, /*ref_index*/ 0, last_mv, | |
1886 (1<<16)>>shift, level-LOG2_MB_SIZE+4, block_w); | |
1887 | |
1888 assert(ref_mx >= c->xmin); | |
1889 assert(ref_mx <= c->xmax); | |
1890 assert(ref_my >= c->ymin); | |
1891 assert(ref_my <= c->ymax); | |
1892 | |
4360
5f887a3c7281
Change the Snow encoder to always use the available MotionEstContext pointer.
takis
parents:
4332
diff
changeset
|
1893 ref_score= c->sub_motion_search(&s->m, &ref_mx, &ref_my, ref_score, 0, 0, level-LOG2_MB_SIZE+4, block_w); |
3314 | 1894 ref_score= ff_get_mb_score(&s->m, ref_mx, ref_my, 0, 0, level-LOG2_MB_SIZE+4, block_w, 0); |
1895 ref_score+= 2*av_log2(2*ref)*c->penalty_factor; | |
1896 if(s->ref_mvs[ref]){ | |
1897 s->ref_mvs[ref][index][0]= ref_mx; | |
1898 s->ref_mvs[ref][index][1]= ref_my; | |
1899 s->ref_scores[ref][index]= ref_score; | |
1900 } | |
1901 if(score > ref_score){ | |
1902 score= ref_score; | |
1903 best_ref= ref; | |
1904 mx= ref_mx; | |
1905 my= ref_my; | |
1906 } | |
1907 } | |
5127 | 1908 //FIXME if mb_cmp != SSE then intra cannot be compared currently and mb_penalty vs. lambda2 |
2967 | 1909 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1910 // subpel search |
5085 | 1911 base_bits= get_rac_count(&s->c) - 8*(s->c.bytestream - s->c.bytestream_start); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1912 pc= s->c; |
2335 | 1913 pc.bytestream_start= |
1914 pc.bytestream= p_buffer; //FIXME end/start? and at the other stoo | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1915 memcpy(p_state, s->block_state, sizeof(s->block_state)); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1916 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1917 if(level!=s->block_max_depth) |
2335 | 1918 put_rac(&pc, &p_state[4 + s_context], 1); |
1919 put_rac(&pc, &p_state[1 + left->type + top->type], 0); | |
3314 | 1920 if(s->ref_frames > 1) |
1921 put_symbol(&pc, &p_state[128 + 1024 + 32*ref_context], best_ref, 0); | |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1922 pred_mv(s, &pmx, &pmy, best_ref, left, top, tr); |
3314 | 1923 put_symbol(&pc, &p_state[128 + 32*(mx_context + 16*!!best_ref)], mx - pmx, 1); |
1924 put_symbol(&pc, &p_state[128 + 32*(my_context + 16*!!best_ref)], my - pmy, 1); | |
2335 | 1925 p_len= pc.bytestream - pc.bytestream_start; |
5082 | 1926 score += (s->lambda2*(get_rac_count(&pc)-base_bits))>>FF_LAMBDA_SHIFT; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1927 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1928 block_s= block_w*block_w; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
1929 sum = pix_sum(current_data[0], stride, block_w); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1930 l= (sum + block_s/2)/block_s; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
1931 iscore = pix_norm1(current_data[0], stride, block_w) - 2*l*sum + l*l*block_s; |
2967 | 1932 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1933 block_s= block_w*block_w>>2; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
1934 sum = pix_sum(current_data[1], uvstride, block_w>>1); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1935 cb= (sum + block_s/2)/block_s; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1936 // iscore += pix_norm1(¤t_mb[1][0], uvstride, block_w>>1) - 2*cb*sum + cb*cb*block_s; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
1937 sum = pix_sum(current_data[2], uvstride, block_w>>1); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1938 cr= (sum + block_s/2)/block_s; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1939 // iscore += pix_norm1(¤t_mb[2][0], uvstride, block_w>>1) - 2*cr*sum + cr*cr*block_s; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1940 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1941 ic= s->c; |
2335 | 1942 ic.bytestream_start= |
1943 ic.bytestream= i_buffer; //FIXME end/start? and at the other stoo | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1944 memcpy(i_state, s->block_state, sizeof(s->block_state)); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1945 if(level!=s->block_max_depth) |
2335 | 1946 put_rac(&ic, &i_state[4 + s_context], 1); |
1947 put_rac(&ic, &i_state[1 + left->type + top->type], 1); | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1948 put_symbol(&ic, &i_state[32], l-pl , 1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1949 put_symbol(&ic, &i_state[64], cb-pcb, 1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1950 put_symbol(&ic, &i_state[96], cr-pcr, 1); |
2335 | 1951 i_len= ic.bytestream - ic.bytestream_start; |
5082 | 1952 iscore += (s->lambda2*(get_rac_count(&ic)-base_bits))>>FF_LAMBDA_SHIFT; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1953 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1954 // assert(score==256*256*256*64-1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1955 assert(iscore < 255*255*256 + s->lambda2*10); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1956 assert(iscore >= 0); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1957 assert(l>=0 && l<=255); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1958 assert(pl>=0 && pl<=255); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1959 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1960 if(level==0){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1961 int varc= iscore >> 8; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1962 int vard= score >> 8; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1963 if (vard <= 64 || vard < varc) |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1964 c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1965 else |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1966 c->scene_change_score+= s->m.qscale; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1967 } |
2967 | 1968 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1969 if(level!=s->block_max_depth){ |
2335 | 1970 put_rac(&s->c, &s->block_state[4 + s_context], 0); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1971 score2 = encode_q_branch(s, level+1, 2*x+0, 2*y+0); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1972 score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+0); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1973 score2+= encode_q_branch(s, level+1, 2*x+0, 2*y+1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1974 score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1975 score2+= s->lambda2>>FF_LAMBDA_SHIFT; //FIXME exact split overhead |
2967 | 1976 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1977 if(score2 < score && score2 < iscore) |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1978 return score2; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1979 } |
2967 | 1980 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1981 if(iscore < score){ |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
1982 pred_mv(s, &pmx, &pmy, 0, left, top, tr); |
2335 | 1983 memcpy(pbbak, i_buffer, i_len); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1984 s->c= ic; |
2335 | 1985 s->c.bytestream_start= pbbak_start; |
1986 s->c.bytestream= pbbak + i_len; | |
3314 | 1987 set_blocks(s, level, x, y, l, cb, cr, pmx, pmy, 0, BLOCK_INTRA); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1988 memcpy(s->block_state, i_state, sizeof(s->block_state)); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1989 return iscore; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1990 }else{ |
2335 | 1991 memcpy(pbbak, p_buffer, p_len); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1992 s->c= pc; |
2335 | 1993 s->c.bytestream_start= pbbak_start; |
1994 s->c.bytestream= pbbak + p_len; | |
3314 | 1995 set_blocks(s, level, x, y, pl, pcb, pcr, mx, my, best_ref, 0); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1996 memcpy(s->block_state, p_state, sizeof(s->block_state)); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1997 return score; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1998 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
1999 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2000 |
4283
d6f83e2f8804
rename always_inline to av_always_inline and move to common.h
mru
parents:
4197
diff
changeset
|
2001 static av_always_inline int same_block(BlockNode *a, BlockNode *b){ |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2002 if((a->type&BLOCK_INTRA) && (b->type&BLOCK_INTRA)){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2003 return !((a->color[0] - b->color[0]) | (a->color[1] - b->color[1]) | (a->color[2] - b->color[2])); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2004 }else{ |
3314 | 2005 return !((a->mx - b->mx) | (a->my - b->my) | (a->ref - b->ref) | ((a->type ^ b->type)&BLOCK_INTRA)); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2006 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2007 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2008 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2009 static void encode_q_branch2(SnowContext *s, int level, int x, int y){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2010 const int w= s->b_width << s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2011 const int rem_depth= s->block_max_depth - level; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2012 const int index= (x + y*w) << rem_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2013 int trx= (x+1)<<rem_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2014 BlockNode *b= &s->block[index]; |
4408
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
2015 const BlockNode *left = x ? &s->block[index-1] : &null_block; |
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
2016 const BlockNode *top = y ? &s->block[index-w] : &null_block; |
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
2017 const BlockNode *tl = y && x ? &s->block[index-w-1] : left; |
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
2018 const BlockNode *tr = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2019 int pl = left->color[0]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2020 int pcb= left->color[1]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2021 int pcr= left->color[2]; |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2022 int pmx, pmy; |
3314 | 2023 int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref); |
4001 | 2024 int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 16*!!b->ref; |
2025 int my_context= av_log2(2*FFABS(left->my - top->my)) + 16*!!b->ref; | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2026 int s_context= 2*left->level + 2*top->level + tl->level + tr->level; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2027 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2028 if(s->keyframe){ |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2029 set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2030 return; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2031 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2032 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2033 if(level!=s->block_max_depth){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2034 if(same_block(b,b+1) && same_block(b,b+w) && same_block(b,b+w+1)){ |
2995
dfc271b90fe6
4mv + iter ME 10l fix (still not bugfree but better)
michael
parents:
2994
diff
changeset
|
2035 put_rac(&s->c, &s->block_state[4 + s_context], 1); |
dfc271b90fe6
4mv + iter ME 10l fix (still not bugfree but better)
michael
parents:
2994
diff
changeset
|
2036 }else{ |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2037 put_rac(&s->c, &s->block_state[4 + s_context], 0); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2038 encode_q_branch2(s, level+1, 2*x+0, 2*y+0); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2039 encode_q_branch2(s, level+1, 2*x+1, 2*y+0); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2040 encode_q_branch2(s, level+1, 2*x+0, 2*y+1); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2041 encode_q_branch2(s, level+1, 2*x+1, 2*y+1); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2042 return; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2043 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2044 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2045 if(b->type & BLOCK_INTRA){ |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2046 pred_mv(s, &pmx, &pmy, 0, left, top, tr); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2047 put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 1); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2048 put_symbol(&s->c, &s->block_state[32], b->color[0]-pl , 1); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2049 put_symbol(&s->c, &s->block_state[64], b->color[1]-pcb, 1); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2050 put_symbol(&s->c, &s->block_state[96], b->color[2]-pcr, 1); |
3314 | 2051 set_blocks(s, level, x, y, b->color[0], b->color[1], b->color[2], pmx, pmy, 0, BLOCK_INTRA); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2052 }else{ |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2053 pred_mv(s, &pmx, &pmy, b->ref, left, top, tr); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2054 put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 0); |
3314 | 2055 if(s->ref_frames > 1) |
2056 put_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], b->ref, 0); | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2057 put_symbol(&s->c, &s->block_state[128 + 32*mx_context], b->mx - pmx, 1); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2058 put_symbol(&s->c, &s->block_state[128 + 32*my_context], b->my - pmy, 1); |
3314 | 2059 set_blocks(s, level, x, y, pl, pcb, pcr, b->mx, b->my, b->ref, 0); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2060 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2061 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2062 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2063 static void decode_q_branch(SnowContext *s, int level, int x, int y){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2064 const int w= s->b_width << s->block_max_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2065 const int rem_depth= s->block_max_depth - level; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2066 const int index= (x + y*w) << rem_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2067 int trx= (x+1)<<rem_depth; |
4408
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
2068 const BlockNode *left = x ? &s->block[index-1] : &null_block; |
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
2069 const BlockNode *top = y ? &s->block[index-w] : &null_block; |
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
2070 const BlockNode *tl = y && x ? &s->block[index-w-1] : left; |
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
2071 const BlockNode *tr = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2072 int s_context= 2*left->level + 2*top->level + tl->level + tr->level; |
2967 | 2073 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2074 if(s->keyframe){ |
3314 | 2075 set_blocks(s, level, x, y, null_block.color[0], null_block.color[1], null_block.color[2], null_block.mx, null_block.my, null_block.ref, BLOCK_INTRA); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2076 return; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2077 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2078 |
2335 | 2079 if(level==s->block_max_depth || get_rac(&s->c, &s->block_state[4 + s_context])){ |
4332 | 2080 int type, mx, my; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2081 int l = left->color[0]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2082 int cb= left->color[1]; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2083 int cr= left->color[2]; |
3314 | 2084 int ref = 0; |
2085 int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref); | |
4001 | 2086 int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 0*av_log2(2*FFABS(tr->mx - top->mx)); |
2087 int my_context= av_log2(2*FFABS(left->my - top->my)) + 0*av_log2(2*FFABS(tr->my - top->my)); | |
2967 | 2088 |
2335 | 2089 type= get_rac(&s->c, &s->block_state[1 + left->type + top->type]) ? BLOCK_INTRA : 0; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2090 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2091 if(type){ |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2092 pred_mv(s, &mx, &my, 0, left, top, tr); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2093 l += get_symbol(&s->c, &s->block_state[32], 1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2094 cb+= get_symbol(&s->c, &s->block_state[64], 1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2095 cr+= get_symbol(&s->c, &s->block_state[96], 1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2096 }else{ |
3314 | 2097 if(s->ref_frames > 1) |
2098 ref= get_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], 0); | |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2099 pred_mv(s, &mx, &my, ref, left, top, tr); |
3314 | 2100 mx+= get_symbol(&s->c, &s->block_state[128 + 32*(mx_context + 16*!!ref)], 1); |
2101 my+= get_symbol(&s->c, &s->block_state[128 + 32*(my_context + 16*!!ref)], 1); | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2102 } |
3314 | 2103 set_blocks(s, level, x, y, l, cb, cr, mx, my, ref, type); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2104 }else{ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2105 decode_q_branch(s, level+1, 2*x+0, 2*y+0); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2106 decode_q_branch(s, level+1, 2*x+1, 2*y+0); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2107 decode_q_branch(s, level+1, 2*x+0, 2*y+1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2108 decode_q_branch(s, level+1, 2*x+1, 2*y+1); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2109 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2110 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2111 |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
2112 static void encode_blocks(SnowContext *s, int search){ |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2113 int x, y; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2114 int w= s->b_width; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2115 int h= s->b_height; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2116 |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
2117 if(s->avctx->me_method == ME_ITER && !s->keyframe && search) |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2118 iterative_me(s); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2119 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2120 for(y=0; y<h; y++){ |
2435
c89ac0e70c66
10l patch by (matthieu castet <castet.matthieu free fr>)
michael
parents:
2422
diff
changeset
|
2121 if(s->c.bytestream_end - s->c.bytestream < w*MB_SIZE*MB_SIZE*3){ //FIXME nicer limit |
2422 | 2122 av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n"); |
2123 return; | |
2124 } | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2125 for(x=0; x<w; x++){ |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
2126 if(s->avctx->me_method == ME_ITER || !search) |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2127 encode_q_branch2(s, 0, x, y); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2128 else |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2129 encode_q_branch (s, 0, x, y); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2130 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2131 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2132 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2133 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2134 static void decode_blocks(SnowContext *s){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2135 int x, y; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2136 int w= s->b_width; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2137 int h= s->b_height; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2138 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2139 for(y=0; y<h; y++){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2140 for(x=0; x<w; x++){ |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2141 decode_q_branch(s, 0, x, y); |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2142 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2143 } |
2138 | 2144 } |
2145 | |
5254 | 2146 static void mc_block(uint8_t *dst, const uint8_t *src, uint8_t *tmp, int stride, int b_w, int b_h, int dx, int dy){ |
2138 | 2147 int x, y; |
2221 | 2148 START_TIMER |
5634
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2149 for(y=0; y < b_h+HTAPS-1; y++){ |
2138 | 2150 for(x=0; x < b_w; x++){ |
5634
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2151 int a_2=src[x + HTAPS/2-5]; |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2152 int a_1=src[x + HTAPS/2-4]; |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2153 int a0= src[x + HTAPS/2-3]; |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2154 int a1= src[x + HTAPS/2-2]; |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2155 int a2= src[x + HTAPS/2-1]; |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2156 int a3= src[x + HTAPS/2+0]; |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2157 int a4= src[x + HTAPS/2+1]; |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2158 int a5= src[x + HTAPS/2+2]; |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2159 int a6= src[x + HTAPS/2+3]; |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2160 int a7= src[x + HTAPS/2+4]; |
2138 | 2161 // int am= 9*(a1+a2) - (a0+a3); |
5637 | 2162 #if HTAPS==6 |
2138 | 2163 int am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5); |
5637 | 2164 #else |
2165 int am= 21*(a2+a3) - 7*(a1+a4) + 3*(a0+a5) - (a_1+a6); | |
2166 #endif | |
2138 | 2167 // int am= 18*(a2+a3) - 2*(a1+a4); |
2168 // int aL= (-7*a0 + 105*a1 + 35*a2 - 5*a3)>>3; | |
2169 // int aR= (-7*a3 + 105*a2 + 35*a1 - 5*a0)>>3; | |
2170 | |
2171 // if(b_w==16) am= 8*(a1+a2); | |
2172 | |
2610
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2173 if(dx<8) am = (32*a2*( 8-dx) + am* dx + 128)>>8; |
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2174 else am = ( am*(16-dx) + 32*a3*(dx-8) + 128)>>8; |
2967 | 2175 |
2610
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2176 /* FIXME Try increasing tmp buffer to 16 bits and not clipping here. Should give marginally better results. - Robert*/ |
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2177 if(am&(~255)) am= ~(am>>31); |
2967 | 2178 |
2610
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2179 tmp[x] = am; |
2138 | 2180 |
2181 /* if (dx< 4) tmp[x + y*stride]= (16*a1*( 4-dx) + aL* dx + 32)>>6; | |
2182 else if(dx< 8) tmp[x + y*stride]= ( aL*( 8-dx) + am*(dx- 4) + 32)>>6; | |
2183 else if(dx<12) tmp[x + y*stride]= ( am*(12-dx) + aR*(dx- 8) + 32)>>6; | |
2184 else tmp[x + y*stride]= ( aR*(16-dx) + 16*a2*(dx-12) + 32)>>6;*/ | |
2185 } | |
2221 | 2186 tmp += stride; |
2187 src += stride; | |
2138 | 2188 } |
5634
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2189 tmp -= (b_h+HTAPS-1)*stride; |
2967 | 2190 |
2138 | 2191 for(y=0; y < b_h; y++){ |
2192 for(x=0; x < b_w; x++){ | |
5634
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2193 int a_2=tmp[x + (HTAPS/2-5)*stride]; |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2194 int a_1=tmp[x + (HTAPS/2-4)*stride]; |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2195 int a0= tmp[x + (HTAPS/2-3)*stride]; |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2196 int a1= tmp[x + (HTAPS/2-2)*stride]; |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2197 int a2= tmp[x + (HTAPS/2-1)*stride]; |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2198 int a3= tmp[x + (HTAPS/2+0)*stride]; |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2199 int a4= tmp[x + (HTAPS/2+1)*stride]; |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2200 int a5= tmp[x + (HTAPS/2+2)*stride]; |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2201 int a6= tmp[x + (HTAPS/2+3)*stride]; |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2202 int a7= tmp[x + (HTAPS/2+4)*stride]; |
5637 | 2203 #if HTAPS==6 |
2138 | 2204 int am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5); |
5637 | 2205 #else |
2206 int am= 21*(a2+a3) - 7*(a1+a4) + 3*(a0+a5) - (a_1+a6); | |
2207 #endif | |
2138 | 2208 // int am= 18*(a2+a3) - 2*(a1+a4); |
2209 /* int aL= (-7*a0 + 105*a1 + 35*a2 - 5*a3)>>3; | |
2210 int aR= (-7*a3 + 105*a2 + 35*a1 - 5*a0)>>3;*/ | |
2967 | 2211 |
2138 | 2212 // if(b_w==16) am= 8*(a1+a2); |
2213 | |
2610
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2214 if(dy<8) am = (32*a2*( 8-dy) + am* dy + 128)>>8; |
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2215 else am = ( am*(16-dy) + 32*a3*(dy-8) + 128)>>8; |
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2216 |
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2217 if(am&(~255)) am= ~(am>>31); |
2967 | 2218 |
2610
f794026f4551
fix overflow in mc_block() patch by (Yartrebo )yartrebo earthlink net
michael
parents:
2609
diff
changeset
|
2219 dst[x] = am; |
2138 | 2220 /* if (dy< 4) tmp[x + y*stride]= (16*a1*( 4-dy) + aL* dy + 32)>>6; |
2221 else if(dy< 8) tmp[x + y*stride]= ( aL*( 8-dy) + am*(dy- 4) + 32)>>6; | |
2222 else if(dy<12) tmp[x + y*stride]= ( am*(12-dy) + aR*(dy- 8) + 32)>>6; | |
2223 else tmp[x + y*stride]= ( aR*(16-dy) + 16*a2*(dy-12) + 32)>>6;*/ | |
2224 } | |
2221 | 2225 dst += stride; |
2226 tmp += stride; | |
2138 | 2227 } |
2221 | 2228 STOP_TIMER("mc_block") |
2138 | 2229 } |
2230 | |
2231 #define mca(dx,dy,b_w)\ | |
5254 | 2232 static void mc_block_hpel ## dx ## dy ## b_w(uint8_t *dst, const uint8_t *src, int stride, int h){\ |
5634
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2233 uint8_t tmp[stride*(b_w+HTAPS-1)];\ |
2138 | 2234 assert(h==b_w);\ |
5634
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2235 mc_block(dst, src-(HTAPS/2-1)-(HTAPS/2-1)*stride, tmp, stride, b_w, b_w, dx, dy);\ |
2138 | 2236 } |
2237 | |
2238 mca( 0, 0,16) | |
2239 mca( 8, 0,16) | |
2240 mca( 0, 8,16) | |
2241 mca( 8, 8,16) | |
2224
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
2242 mca( 0, 0,8) |
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
2243 mca( 8, 0,8) |
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
2244 mca( 0, 8,8) |
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
2245 mca( 8, 8,8) |
2138 | 2246 |
3314 | 2247 static void pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, int stride, int sx, int sy, int b_w, int b_h, BlockNode *block, int plane_index, int w, int h){ |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2248 if(block->type & BLOCK_INTRA){ |
2206 | 2249 int x, y; |
3018 | 2250 const int color = block->color[plane_index]; |
2251 const int color4= color*0x01010101; | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2252 if(b_w==32){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2253 for(y=0; y < b_h; y++){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2254 *(uint32_t*)&dst[0 + y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2255 *(uint32_t*)&dst[4 + y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2256 *(uint32_t*)&dst[8 + y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2257 *(uint32_t*)&dst[12+ y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2258 *(uint32_t*)&dst[16+ y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2259 *(uint32_t*)&dst[20+ y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2260 *(uint32_t*)&dst[24+ y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2261 *(uint32_t*)&dst[28+ y*stride]= color4; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2262 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2263 }else if(b_w==16){ |
3018 | 2264 for(y=0; y < b_h; y++){ |
2265 *(uint32_t*)&dst[0 + y*stride]= color4; | |
2266 *(uint32_t*)&dst[4 + y*stride]= color4; | |
2267 *(uint32_t*)&dst[8 + y*stride]= color4; | |
2268 *(uint32_t*)&dst[12+ y*stride]= color4; | |
2269 } | |
2270 }else if(b_w==8){ | |
2271 for(y=0; y < b_h; y++){ | |
2272 *(uint32_t*)&dst[0 + y*stride]= color4; | |
2273 *(uint32_t*)&dst[4 + y*stride]= color4; | |
2274 } | |
2275 }else if(b_w==4){ | |
2276 for(y=0; y < b_h; y++){ | |
2277 *(uint32_t*)&dst[0 + y*stride]= color4; | |
2278 } | |
2279 }else{ | |
2280 for(y=0; y < b_h; y++){ | |
2281 for(x=0; x < b_w; x++){ | |
2282 dst[x + y*stride]= color; | |
2283 } | |
2138 | 2284 } |
2285 } | |
2286 }else{ | |
3314 | 2287 uint8_t *src= s->last_picture[block->ref].data[plane_index]; |
2206 | 2288 const int scale= plane_index ? s->mv_scale : 2*s->mv_scale; |
2289 int mx= block->mx*scale; | |
2290 int my= block->my*scale; | |
2223 | 2291 const int dx= mx&15; |
2292 const int dy= my&15; | |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3018
diff
changeset
|
2293 const int tab_index= 3 - (b_w>>2) + (b_w>>4); |
5634
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2294 sx += (mx>>4) - (HTAPS/2-1); |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2295 sy += (my>>4) - (HTAPS/2-1); |
2206 | 2296 src += sx + sy*stride; |
5634
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2297 if( (unsigned)sx >= w - b_w - (HTAPS-2) |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2298 || (unsigned)sy >= h - b_h - (HTAPS-2)){ |
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2299 ff_emulated_edge_mc(tmp + MB_SIZE, src, stride, b_w+HTAPS-1, b_h+HTAPS-1, sx, sy, w, h); |
2206 | 2300 src= tmp + MB_SIZE; |
2138 | 2301 } |
3189 | 2302 // assert(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h); |
2303 // assert(!(b_w&(b_w-1))); | |
3018 | 2304 assert(b_w>1 && b_h>1); |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2305 assert(tab_index>=0 && tab_index<4 || b_w==32); |
5634
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2306 if((dx&3) || (dy&3) || !(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h) || (b_w&(b_w-1)) || HTAPS != 6) |
2223 | 2307 mc_block(dst, src, tmp, stride, b_w, b_h, dx, dy); |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2308 else if(b_w==32){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2309 int y; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2310 for(y=0; y<b_h; y+=16){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2311 s->dsp.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + y*stride, src + 2 + (y+2)*stride,stride); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2312 s->dsp.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + 16 + y*stride, src + 18 + (y+2)*stride,stride); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2313 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2314 }else if(b_w==b_h) |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3018
diff
changeset
|
2315 s->dsp.put_h264_qpel_pixels_tab[tab_index ][dy+(dx>>2)](dst,src + 2 + 2*stride,stride); |
3018 | 2316 else if(b_w==2*b_h){ |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3018
diff
changeset
|
2317 s->dsp.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst ,src + 2 + 2*stride,stride); |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3018
diff
changeset
|
2318 s->dsp.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst+b_h,src + 2 + b_h + 2*stride,stride); |
3018 | 2319 }else{ |
2320 assert(2*b_w==b_h); | |
3020
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3018
diff
changeset
|
2321 s->dsp.put_h264_qpel_pixels_tab[tab_index ][dy+(dx>>2)](dst ,src + 2 + 2*stride ,stride); |
c75fb0747e74
use h264 MC functions for 2xX Xx2 blocks in snow too
michael
parents:
3018
diff
changeset
|
2322 s->dsp.put_h264_qpel_pixels_tab[tab_index ][dy+(dx>>2)](dst+b_w*stride,src + 2 + 2*stride+b_w*stride,stride); |
3018 | 2323 } |
2138 | 2324 } |
2325 } | |
2326 | |
4436
d3e389536b0a
Add the const specifier as needed to reduce the number of warnings.
takis
parents:
4409
diff
changeset
|
2327 void ff_snow_inner_add_yblock(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2328 int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){ |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2329 int y, x; |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
2330 IDWTELEM * dst; |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2331 for(y=0; y<b_h; y++){ |
5409 | 2332 //FIXME ugly misuse of obmc_stride |
4436
d3e389536b0a
Add the const specifier as needed to reduce the number of warnings.
takis
parents:
4409
diff
changeset
|
2333 const uint8_t *obmc1= obmc + y*obmc_stride; |
d3e389536b0a
Add the const specifier as needed to reduce the number of warnings.
takis
parents:
4409
diff
changeset
|
2334 const uint8_t *obmc2= obmc1+ (obmc_stride>>1); |
d3e389536b0a
Add the const specifier as needed to reduce the number of warnings.
takis
parents:
4409
diff
changeset
|
2335 const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1); |
d3e389536b0a
Add the const specifier as needed to reduce the number of warnings.
takis
parents:
4409
diff
changeset
|
2336 const uint8_t *obmc4= obmc3+ (obmc_stride>>1); |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2337 dst = slice_buffer_get_line(sb, src_y + y); |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2338 for(x=0; x<b_w; x++){ |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2339 int v= obmc1[x] * block[3][x + y*src_stride] |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2340 +obmc2[x] * block[2][x + y*src_stride] |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2341 +obmc3[x] * block[1][x + y*src_stride] |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2342 +obmc4[x] * block[0][x + y*src_stride]; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2343 |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2344 v <<= 8 - LOG2_OBMC_MAX; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2345 if(FRAC_BITS != 8){ |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2346 v >>= 8 - FRAC_BITS; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2347 } |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2348 if(add){ |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2349 v += dst[x + src_x]; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2350 v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2351 if(v&(~255)) v= ~(v>>31); |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2352 dst8[x + y*src_stride] = v; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2353 }else{ |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2354 dst[x + src_x] -= v; |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2355 } |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2356 } |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2357 } |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2358 } |
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
2359 |
2206 | 2360 //FIXME name clenup (b_w, block_w, b_width stuff) |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
2361 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){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2362 const int b_width = s->b_width << s->block_max_depth; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2363 const int b_height= s->b_height << s->block_max_depth; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2364 const int b_stride= b_width; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2365 BlockNode *lt= &s->block[b_x + b_y*b_stride]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2366 BlockNode *rt= lt+1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2367 BlockNode *lb= lt+b_stride; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2368 BlockNode *rb= lb+1; |
2967 | 2369 uint8_t *block[4]; |
2842 | 2370 int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride; |
2371 uint8_t tmp[src_stride*7*MB_SIZE]; //FIXME align | |
2372 uint8_t *ptmp; | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2373 int x,y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2374 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2375 if(b_x<0){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2376 lt= rt; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2377 lb= rb; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2378 }else if(b_x + 1 >= b_width){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2379 rt= lt; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2380 rb= lb; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2381 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2382 if(b_y<0){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2383 lt= lb; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2384 rt= rb; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2385 }else if(b_y + 1 >= b_height){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2386 lb= lt; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2387 rb= rt; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2388 } |
2967 | 2389 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2390 if(src_x<0){ //FIXME merge with prev & always round internal width upto *16 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2391 obmc -= src_x; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2392 b_w += src_x; |
3662
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2393 if(!sliced && !offset_dst) |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2394 dst -= src_x; |
2206 | 2395 src_x=0; |
2396 }else if(src_x + b_w > w){ | |
2397 b_w = w - src_x; | |
2398 } | |
2399 if(src_y<0){ | |
2400 obmc -= src_y*obmc_stride; | |
2401 b_h += src_y; | |
3662
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2402 if(!sliced && !offset_dst) |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2403 dst -= src_y*dst_stride; |
2206 | 2404 src_y=0; |
2405 }else if(src_y + b_h> h){ | |
2406 b_h = h - src_y; | |
2407 } | |
2967 | 2408 |
2206 | 2409 if(b_w<=0 || b_h<=0) return; |
2410 | |
2842 | 2411 assert(src_stride > 2*MB_SIZE + 5); |
3662
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2412 if(!sliced && offset_dst) |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2413 dst += src_x + src_y*dst_stride; |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2414 dst8+= src_x + src_y*src_stride; |
2206 | 2415 // src += src_x + src_y*src_stride; |
2416 | |
2842 | 2417 ptmp= tmp + 3*tmp_step; |
2418 block[0]= ptmp; | |
2419 ptmp+=tmp_step; | |
3314 | 2420 pred_block(s, block[0], tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h); |
2206 | 2421 |
2422 if(same_block(lt, rt)){ | |
2423 block[1]= block[0]; | |
2424 }else{ | |
2842 | 2425 block[1]= ptmp; |
2426 ptmp+=tmp_step; | |
3314 | 2427 pred_block(s, block[1], tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h); |
2206 | 2428 } |
2967 | 2429 |
2206 | 2430 if(same_block(lt, lb)){ |
2431 block[2]= block[0]; | |
2432 }else if(same_block(rt, lb)){ | |
2433 block[2]= block[1]; | |
2434 }else{ | |
2842 | 2435 block[2]= ptmp; |
2436 ptmp+=tmp_step; | |
3314 | 2437 pred_block(s, block[2], tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h); |
2206 | 2438 } |
2439 | |
2440 if(same_block(lt, rb) ){ | |
2441 block[3]= block[0]; | |
2442 }else if(same_block(rt, rb)){ | |
2443 block[3]= block[1]; | |
2444 }else if(same_block(lb, rb)){ | |
2445 block[3]= block[2]; | |
2446 }else{ | |
2842 | 2447 block[3]= ptmp; |
3314 | 2448 pred_block(s, block[3], tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h); |
2206 | 2449 } |
2450 #if 0 | |
2451 for(y=0; y<b_h; y++){ | |
2452 for(x=0; x<b_w; x++){ | |
2453 int v= obmc [x + y*obmc_stride] * block[3][x + y*src_stride] * (256/OBMC_MAX); | |
2454 if(add) dst[x + y*dst_stride] += v; | |
2455 else dst[x + y*dst_stride] -= v; | |
2456 } | |
2457 } | |
2458 for(y=0; y<b_h; y++){ | |
2459 uint8_t *obmc2= obmc + (obmc_stride>>1); | |
2460 for(x=0; x<b_w; x++){ | |
2461 int v= obmc2[x + y*obmc_stride] * block[2][x + y*src_stride] * (256/OBMC_MAX); | |
2462 if(add) dst[x + y*dst_stride] += v; | |
2463 else dst[x + y*dst_stride] -= v; | |
2464 } | |
2465 } | |
2466 for(y=0; y<b_h; y++){ | |
2467 uint8_t *obmc3= obmc + obmc_stride*(obmc_stride>>1); | |
2468 for(x=0; x<b_w; x++){ | |
2469 int v= obmc3[x + y*obmc_stride] * block[1][x + y*src_stride] * (256/OBMC_MAX); | |
2470 if(add) dst[x + y*dst_stride] += v; | |
2471 else dst[x + y*dst_stride] -= v; | |
2472 } | |
2473 } | |
2474 for(y=0; y<b_h; y++){ | |
2475 uint8_t *obmc3= obmc + obmc_stride*(obmc_stride>>1); | |
2476 uint8_t *obmc4= obmc3+ (obmc_stride>>1); | |
2477 for(x=0; x<b_w; x++){ | |
2478 int v= obmc4[x + y*obmc_stride] * block[0][x + y*src_stride] * (256/OBMC_MAX); | |
2479 if(add) dst[x + y*dst_stride] += v; | |
2480 else dst[x + y*dst_stride] -= v; | |
2481 } | |
2482 } | |
2483 #else | |
3662
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2484 if(sliced){ |
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2485 START_TIMER |
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2486 |
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2487 s->dsp.inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8); |
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2488 STOP_TIMER("inner_add_yblock") |
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2489 }else |
2206 | 2490 for(y=0; y<b_h; y++){ |
5409 | 2491 //FIXME ugly misuse of obmc_stride |
4436
d3e389536b0a
Add the const specifier as needed to reduce the number of warnings.
takis
parents:
4409
diff
changeset
|
2492 const uint8_t *obmc1= obmc + y*obmc_stride; |
d3e389536b0a
Add the const specifier as needed to reduce the number of warnings.
takis
parents:
4409
diff
changeset
|
2493 const uint8_t *obmc2= obmc1+ (obmc_stride>>1); |
d3e389536b0a
Add the const specifier as needed to reduce the number of warnings.
takis
parents:
4409
diff
changeset
|
2494 const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1); |
d3e389536b0a
Add the const specifier as needed to reduce the number of warnings.
takis
parents:
4409
diff
changeset
|
2495 const uint8_t *obmc4= obmc3+ (obmc_stride>>1); |
2206 | 2496 for(x=0; x<b_w; x++){ |
2497 int v= obmc1[x] * block[3][x + y*src_stride] | |
2498 +obmc2[x] * block[2][x + y*src_stride] | |
2499 +obmc3[x] * block[1][x + y*src_stride] | |
2500 +obmc4[x] * block[0][x + y*src_stride]; | |
2967 | 2501 |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2502 v <<= 8 - LOG2_OBMC_MAX; |
2246 | 2503 if(FRAC_BITS != 8){ |
2504 v >>= 8 - FRAC_BITS; | |
2505 } | |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2506 if(add){ |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2507 v += dst[x + y*dst_stride]; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2508 v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2509 if(v&(~255)) v= ~(v>>31); |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2510 dst8[x + y*src_stride] = v; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2511 }else{ |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2512 dst[x + y*dst_stride] -= v; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2513 } |
2206 | 2514 } |
2515 } | |
2516 #endif | |
2517 } | |
2518 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
2519 static av_always_inline void predict_slice_buffered(SnowContext *s, slice_buffer * sb, IDWTELEM * old_buffer, int plane_index, int add, int mb_y){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2520 Plane *p= &s->plane[plane_index]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2521 const int mb_w= s->b_width << s->block_max_depth; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2522 const int mb_h= s->b_height << s->block_max_depth; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2523 int x, y, mb_x; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2524 int block_size = MB_SIZE >> s->block_max_depth; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2525 int block_w = plane_index ? block_size/2 : block_size; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2526 const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2527 int obmc_stride= plane_index ? block_size : 2*block_size; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2528 int ref_stride= s->current_picture.linesize[plane_index]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2529 uint8_t *dst8= s->current_picture.data[plane_index]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2530 int w= p->width; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2531 int h= p->height; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2532 START_TIMER |
2967 | 2533 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2534 if(s->keyframe || (s->avctx->debug&512)){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2535 if(mb_y==mb_h) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2536 return; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2537 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2538 if(add){ |
2604
b7e6c3d31c65
Snow segfault bug in revision 1.43 patch by (Yartrebo /yartrebo earthlink net/)
michael
parents:
2602
diff
changeset
|
2539 for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++) |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2540 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2541 // DWTELEM * line = slice_buffer_get_line(sb, y); |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
2542 IDWTELEM * line = sb->line[y]; |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2543 for(x=0; x<w; x++) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2544 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2545 // int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1)); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2546 int v= line[x] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1)); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2547 v >>= FRAC_BITS; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2548 if(v&(~255)) v= ~(v>>31); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2549 dst8[x + y*ref_stride]= v; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2550 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2551 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2552 }else{ |
2604
b7e6c3d31c65
Snow segfault bug in revision 1.43 patch by (Yartrebo /yartrebo earthlink net/)
michael
parents:
2602
diff
changeset
|
2553 for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++) |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2554 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2555 // DWTELEM * line = slice_buffer_get_line(sb, y); |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
2556 IDWTELEM * line = sb->line[y]; |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2557 for(x=0; x<w; x++) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2558 { |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2559 line[x] -= 128 << FRAC_BITS; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2560 // buf[x + y*w]-= 128<<FRAC_BITS; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2561 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2562 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2563 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2564 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2565 return; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2566 } |
2967 | 2567 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2568 for(mb_x=0; mb_x<=mb_w; mb_x++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2569 START_TIMER |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2570 |
3662
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2571 add_yblock(s, 1, sb, old_buffer, dst8, obmc, |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2572 block_w*mb_x - block_w/2, |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2573 block_w*mb_y - block_w/2, |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2574 block_w, block_w, |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2575 w, h, |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2576 w, ref_stride, obmc_stride, |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2577 mb_x - 1, mb_y - 1, |
3662
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2578 add, 0, plane_index); |
2967 | 2579 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2580 STOP_TIMER("add_yblock") |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2581 } |
2967 | 2582 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2583 STOP_TIMER("predict_slice") |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2584 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
2585 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
2586 static av_always_inline void predict_slice(SnowContext *s, IDWTELEM *buf, int plane_index, int add, int mb_y){ |
2138 | 2587 Plane *p= &s->plane[plane_index]; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2588 const int mb_w= s->b_width << s->block_max_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2589 const int mb_h= s->b_height << s->block_max_depth; |
2562 | 2590 int x, y, mb_x; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2591 int block_size = MB_SIZE >> s->block_max_depth; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
2592 int block_w = plane_index ? block_size/2 : block_size; |
2206 | 2593 const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth]; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2594 const int obmc_stride= plane_index ? block_size : 2*block_size; |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2595 int ref_stride= s->current_picture.linesize[plane_index]; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2596 uint8_t *dst8= s->current_picture.data[plane_index]; |
2138 | 2597 int w= p->width; |
2598 int h= p->height; | |
2197 | 2599 START_TIMER |
2967 | 2600 |
2206 | 2601 if(s->keyframe || (s->avctx->debug&512)){ |
2562 | 2602 if(mb_y==mb_h) |
2603 return; | |
2604 | |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2605 if(add){ |
2604
b7e6c3d31c65
Snow segfault bug in revision 1.43 patch by (Yartrebo /yartrebo earthlink net/)
michael
parents:
2602
diff
changeset
|
2606 for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){ |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2607 for(x=0; x<w; x++){ |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2608 int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1)); |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2609 v >>= FRAC_BITS; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2610 if(v&(~255)) v= ~(v>>31); |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2611 dst8[x + y*ref_stride]= v; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2612 } |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2613 } |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2614 }else{ |
2604
b7e6c3d31c65
Snow segfault bug in revision 1.43 patch by (Yartrebo /yartrebo earthlink net/)
michael
parents:
2602
diff
changeset
|
2615 for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){ |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2616 for(x=0; x<w; x++){ |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2617 buf[x + y*w]-= 128<<FRAC_BITS; |
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
2618 } |
2206 | 2619 } |
2138 | 2620 } |
2206 | 2621 |
2622 return; | |
2138 | 2623 } |
2967 | 2624 |
2206 | 2625 for(mb_x=0; mb_x<=mb_w; mb_x++){ |
2197 | 2626 START_TIMER |
2206 | 2627 |
3662
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2628 add_yblock(s, 0, NULL, buf, dst8, obmc, |
2206 | 2629 block_w*mb_x - block_w/2, |
2138 | 2630 block_w*mb_y - block_w/2, |
2206 | 2631 block_w, block_w, |
2138 | 2632 w, h, |
2206 | 2633 w, ref_stride, obmc_stride, |
2634 mb_x - 1, mb_y - 1, | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2635 add, 1, plane_index); |
2967 | 2636 |
2206 | 2637 STOP_TIMER("add_yblock") |
2138 | 2638 } |
2967 | 2639 |
2562 | 2640 STOP_TIMER("predict_slice") |
2641 } | |
2642 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
2643 static av_always_inline void predict_plane(SnowContext *s, IDWTELEM *buf, int plane_index, int add){ |
2562 | 2644 const int mb_h= s->b_height << s->block_max_depth; |
2645 int mb_y; | |
2646 for(mb_y=0; mb_y<=mb_h; mb_y++) | |
2647 predict_slice(s, buf, plane_index, add, mb_y); | |
2138 | 2648 } |
2649 | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2650 static int get_dc(SnowContext *s, int mb_x, int mb_y, int plane_index){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2651 int i, x2, y2; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2652 Plane *p= &s->plane[plane_index]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2653 const int block_size = MB_SIZE >> s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2654 const int block_w = plane_index ? block_size/2 : block_size; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2655 const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2656 const int obmc_stride= plane_index ? block_size : 2*block_size; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2657 const int ref_stride= s->current_picture.linesize[plane_index]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2658 uint8_t *src= s-> input_picture.data[plane_index]; |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
2659 IDWTELEM *dst= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4; //FIXME change to unsigned |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2660 const int b_stride = s->b_width << s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2661 const int w= p->width; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2662 const int h= p->height; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2663 int index= mb_x + mb_y*b_stride; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2664 BlockNode *b= &s->block[index]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2665 BlockNode backup= *b; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2666 int ab=0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2667 int aa=0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2668 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2669 b->type|= BLOCK_INTRA; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2670 b->color[plane_index]= 0; |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
2671 memset(dst, 0, obmc_stride*obmc_stride*sizeof(IDWTELEM)); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2672 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2673 for(i=0; i<4; i++){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2674 int mb_x2= mb_x + (i &1) - 1; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2675 int mb_y2= mb_y + (i>>1) - 1; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2676 int x= block_w*mb_x2 + block_w/2; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2677 int y= block_w*mb_y2 + block_w/2; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2678 |
3662
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2679 add_yblock(s, 0, NULL, dst + ((i&1)+(i>>1)*obmc_stride)*block_w, NULL, obmc, |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2680 x, y, block_w, block_w, w, h, obmc_stride, ref_stride, obmc_stride, mb_x2, mb_y2, 0, 0, plane_index); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2681 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2682 for(y2= FFMAX(y, 0); y2<FFMIN(h, y+block_w); y2++){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2683 for(x2= FFMAX(x, 0); x2<FFMIN(w, x+block_w); x2++){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2684 int index= x2-(block_w*mb_x - block_w/2) + (y2-(block_w*mb_y - block_w/2))*obmc_stride; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2685 int obmc_v= obmc[index]; |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2686 int d; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2687 if(y<0) obmc_v += obmc[index + block_w*obmc_stride]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2688 if(x<0) obmc_v += obmc[index + block_w]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2689 if(y+block_w>h) obmc_v += obmc[index - block_w*obmc_stride]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2690 if(x+block_w>w) obmc_v += obmc[index - block_w]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2691 //FIXME precalc this or simplify it somehow else |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2692 |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2693 d = -dst[index] + (1<<(FRAC_BITS-1)); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2694 dst[index] = d; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2695 ab += (src[x2 + y2*ref_stride] - (d>>FRAC_BITS)) * obmc_v; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2696 aa += obmc_v * obmc_v; //FIXME precalclate this |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2697 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2698 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2699 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2700 *b= backup; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2701 |
5127 | 2702 return av_clip(((ab<<LOG2_OBMC_MAX) + aa/2)/aa, 0, 255); //FIXME we should not need clipping |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2703 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2704 |
3051 | 2705 static inline int get_block_bits(SnowContext *s, int x, int y, int w){ |
2706 const int b_stride = s->b_width << s->block_max_depth; | |
2707 const int b_height = s->b_height<< s->block_max_depth; | |
2708 int index= x + y*b_stride; | |
4408
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
2709 const BlockNode *b = &s->block[index]; |
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
2710 const BlockNode *left = x ? &s->block[index-1] : &null_block; |
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
2711 const BlockNode *top = y ? &s->block[index-b_stride] : &null_block; |
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
2712 const BlockNode *tl = y && x ? &s->block[index-b_stride-1] : left; |
c3be13eac2d6
Bring down the number of snow.c warnings from 27 to 17 by using the const
takis
parents:
4407
diff
changeset
|
2713 const BlockNode *tr = y && x+w<b_stride ? &s->block[index-b_stride+w] : tl; |
3051 | 2714 int dmx, dmy; |
4001 | 2715 // int mx_context= av_log2(2*FFABS(left->mx - top->mx)); |
2716 // int my_context= av_log2(2*FFABS(left->my - top->my)); | |
3051 | 2717 |
2718 if(x<0 || x>=b_stride || y>=b_height) | |
2719 return 0; | |
2720 /* | |
2721 1 0 0 | |
2722 01X 1-2 1 | |
2723 001XX 3-6 2-3 | |
2724 0001XXX 7-14 4-7 | |
2725 00001XXXX 15-30 8-15 | |
2726 */ | |
2727 //FIXME try accurate rate | |
2728 //FIXME intra and inter predictors if surrounding blocks arent the same type | |
2729 if(b->type & BLOCK_INTRA){ | |
4001 | 2730 return 3+2*( av_log2(2*FFABS(left->color[0] - b->color[0])) |
2731 + av_log2(2*FFABS(left->color[1] - b->color[1])) | |
2732 + av_log2(2*FFABS(left->color[2] - b->color[2]))); | |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2733 }else{ |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2734 pred_mv(s, &dmx, &dmy, b->ref, left, top, tr); |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2735 dmx-= b->mx; |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2736 dmy-= b->my; |
4001 | 2737 return 2*(1 + av_log2(2*FFABS(dmx)) //FIXME kill the 2* can be merged in lambda |
2738 + av_log2(2*FFABS(dmy)) | |
3314 | 2739 + av_log2(2*b->ref)); |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
2740 } |
3051 | 2741 } |
2742 | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2743 static int get_block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index, const uint8_t *obmc_edged){ |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2744 Plane *p= &s->plane[plane_index]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2745 const int block_size = MB_SIZE >> s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2746 const int block_w = plane_index ? block_size/2 : block_size; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2747 const int obmc_stride= plane_index ? block_size : 2*block_size; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2748 const int ref_stride= s->current_picture.linesize[plane_index]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2749 uint8_t *dst= s->current_picture.data[plane_index]; |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2750 uint8_t *src= s-> input_picture.data[plane_index]; |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
2751 IDWTELEM *pred= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4; |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2752 uint8_t cur[ref_stride*2*MB_SIZE]; //FIXME alignment |
5634
9960732c7d7b
Replace constants by HTAPS (half pel interpolation filter taps) where appropriate.
michael
parents:
5633
diff
changeset
|
2753 uint8_t tmp[ref_stride*(2*MB_SIZE+HTAPS-1)]; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2754 const int b_stride = s->b_width << s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2755 const int b_height = s->b_height<< s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2756 const int w= p->width; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2757 const int h= p->height; |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2758 int distortion; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2759 int rate= 0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2760 const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp); |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2761 int sx= block_w*mb_x - block_w/2; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2762 int sy= block_w*mb_y - block_w/2; |
3206 | 2763 int x0= FFMAX(0,-sx); |
2764 int y0= FFMAX(0,-sy); | |
2765 int x1= FFMIN(block_w*2, w-sx); | |
2766 int y1= FFMIN(block_w*2, h-sy); | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2767 int i,x,y; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2768 |
3314 | 2769 pred_block(s, cur, tmp, ref_stride, sx, sy, block_w*2, block_w*2, &s->block[mb_x + mb_y*b_stride], plane_index, w, h); |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2770 |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2771 for(y=y0; y<y1; y++){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2772 const uint8_t *obmc1= obmc_edged + y*obmc_stride; |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
2773 const IDWTELEM *pred1 = pred + y*obmc_stride; |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2774 uint8_t *cur1 = cur + y*ref_stride; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2775 uint8_t *dst1 = dst + sx + (sy+y)*ref_stride; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2776 for(x=x0; x<x1; x++){ |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
2777 #if FRAC_BITS >= LOG2_OBMC_MAX |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2778 int v = (cur1[x] * obmc1[x]) << (FRAC_BITS - LOG2_OBMC_MAX); |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
2779 #else |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
2780 int v = (cur1[x] * obmc1[x] + (1<<(LOG2_OBMC_MAX - FRAC_BITS-1))) >> (LOG2_OBMC_MAX - FRAC_BITS); |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
2781 #endif |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2782 v = (v + pred1[x]) >> FRAC_BITS; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2783 if(v&(~255)) v= ~(v>>31); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2784 dst1[x] = v; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2785 } |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2786 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2787 |
3206 | 2788 /* copy the regions where obmc[] = (uint8_t)256 */ |
2789 if(LOG2_OBMC_MAX == 8 | |
2790 && (mb_x == 0 || mb_x == b_stride-1) | |
2791 && (mb_y == 0 || mb_y == b_height-1)){ | |
2792 if(mb_x == 0) | |
2793 x1 = block_w; | |
2794 else | |
2795 x0 = block_w; | |
2796 if(mb_y == 0) | |
2797 y1 = block_w; | |
2798 else | |
2799 y0 = block_w; | |
2800 for(y=y0; y<y1; y++) | |
2801 memcpy(dst + sx+x0 + (sy+y)*ref_stride, cur + x0 + y*ref_stride, x1-x0); | |
2802 } | |
2803 | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2804 if(block_w==16){ |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
2805 /* FIXME rearrange dsputil to fit 32x32 cmp functions */ |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
2806 /* FIXME check alignment of the cmp wavelet vs the encoding wavelet */ |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
2807 /* FIXME cmps overlap but don't cover the wavelet's whole support, |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
2808 * so improving the score of one block is not strictly guaranteed to |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
2809 * improve the score of the whole frame, so iterative motion est |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
2810 * doesn't always converge. */ |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
2811 if(s->avctx->me_cmp == FF_CMP_W97) |
4197 | 2812 distortion = w97_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32); |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
2813 else if(s->avctx->me_cmp == FF_CMP_W53) |
4197 | 2814 distortion = w53_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32); |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
2815 else{ |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
2816 distortion = 0; |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
2817 for(i=0; i<4; i++){ |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
2818 int off = sx+16*(i&1) + (sy+16*(i>>1))*ref_stride; |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
2819 distortion += s->dsp.me_cmp[0](&s->m, src + off, dst + off, ref_stride, 16); |
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
2820 } |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2821 } |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2822 }else{ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2823 assert(block_w==8); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2824 distortion = s->dsp.me_cmp[0](&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, block_w*2); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2825 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2826 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2827 if(plane_index==0){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2828 for(i=0; i<4; i++){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2829 /* ..RRr |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2830 * .RXx. |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2831 * rxx.. |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2832 */ |
3051 | 2833 rate += get_block_bits(s, mb_x + (i&1) - (i>>1), mb_y + (i>>1), 1); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2834 } |
3057 | 2835 if(mb_x == b_stride-2) |
2836 rate += get_block_bits(s, mb_x + 1, mb_y + 1, 1); | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2837 } |
3051 | 2838 return distortion + rate*penalty_factor; |
2839 } | |
2840 | |
2841 static int get_4block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index){ | |
2842 int i, y2; | |
2843 Plane *p= &s->plane[plane_index]; | |
2844 const int block_size = MB_SIZE >> s->block_max_depth; | |
2845 const int block_w = plane_index ? block_size/2 : block_size; | |
2846 const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth]; | |
2847 const int obmc_stride= plane_index ? block_size : 2*block_size; | |
2848 const int ref_stride= s->current_picture.linesize[plane_index]; | |
2849 uint8_t *dst= s->current_picture.data[plane_index]; | |
2850 uint8_t *src= s-> input_picture.data[plane_index]; | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
2851 static const IDWTELEM zero_dst[4096]; //FIXME |
3051 | 2852 const int b_stride = s->b_width << s->block_max_depth; |
2853 const int w= p->width; | |
2854 const int h= p->height; | |
2855 int distortion= 0; | |
2856 int rate= 0; | |
2857 const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp); | |
2858 | |
2859 for(i=0; i<9; i++){ | |
2860 int mb_x2= mb_x + (i%3) - 1; | |
2861 int mb_y2= mb_y + (i/3) - 1; | |
2862 int x= block_w*mb_x2 + block_w/2; | |
2863 int y= block_w*mb_y2 + block_w/2; | |
2864 | |
3662
fc714e9a5419
snow cosmetics: merge the sliced and non-sliced versions of add_yblock
lorenm
parents:
3661
diff
changeset
|
2865 add_yblock(s, 0, NULL, zero_dst, dst, obmc, |
3051 | 2866 x, y, block_w, block_w, w, h, /*dst_stride*/0, ref_stride, obmc_stride, mb_x2, mb_y2, 1, 1, plane_index); |
2867 | |
2868 //FIXME find a cleaner/simpler way to skip the outside stuff | |
2869 for(y2= y; y2<0; y2++) | |
2870 memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w); | |
2871 for(y2= h; y2<y+block_w; y2++) | |
2872 memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w); | |
2873 if(x<0){ | |
2874 for(y2= y; y2<y+block_w; y2++) | |
2875 memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, -x); | |
2876 } | |
2877 if(x+block_w > w){ | |
2878 for(y2= y; y2<y+block_w; y2++) | |
2879 memcpy(dst + w + y2*ref_stride, src + w + y2*ref_stride, x+block_w - w); | |
2880 } | |
2881 | |
2882 assert(block_w== 8 || block_w==16); | |
2883 distortion += s->dsp.me_cmp[block_w==8](&s->m, src + x + y*ref_stride, dst + x + y*ref_stride, ref_stride, block_w); | |
2884 } | |
2885 | |
2886 if(plane_index==0){ | |
2887 BlockNode *b= &s->block[mb_x+mb_y*b_stride]; | |
2888 int merged= same_block(b,b+1) && same_block(b,b+b_stride) && same_block(b,b+b_stride+1); | |
2889 | |
2890 /* ..RRRr | |
2891 * .RXXx. | |
2892 * .RXXx. | |
2893 * rxxx. | |
2894 */ | |
2895 if(merged) | |
2896 rate = get_block_bits(s, mb_x, mb_y, 2); | |
2897 for(i=merged?4:0; i<9; i++){ | |
2898 static const int dxy[9][2] = {{0,0},{1,0},{0,1},{1,1},{2,0},{2,1},{-1,2},{0,2},{1,2}}; | |
2899 rate += get_block_bits(s, mb_x + dxy[i][0], mb_y + dxy[i][1], 1); | |
2900 } | |
2901 } | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2902 return distortion + rate*penalty_factor; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2903 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2904 |
4283
d6f83e2f8804
rename always_inline to av_always_inline and move to common.h
mru
parents:
4197
diff
changeset
|
2905 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){ |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2906 const int b_stride= s->b_width << s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2907 BlockNode *block= &s->block[mb_x + mb_y * b_stride]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2908 BlockNode backup= *block; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2909 int rd, index, value; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2910 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2911 assert(mb_x>=0 && mb_y>=0); |
2994 | 2912 assert(mb_x<b_stride); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2913 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2914 if(intra){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2915 block->color[0] = p[0]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2916 block->color[1] = p[1]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2917 block->color[2] = p[2]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2918 block->type |= BLOCK_INTRA; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2919 }else{ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2920 index= (p[0] + 31*p[1]) & (ME_CACHE_SIZE-1); |
3314 | 2921 value= s->me_cache_generation + (p[0]>>10) + (p[1]<<6) + (block->ref<<12); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2922 if(s->me_cache[index] == value) |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2923 return 0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2924 s->me_cache[index]= value; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2925 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2926 block->mx= p[0]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2927 block->my= p[1]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2928 block->type &= ~BLOCK_INTRA; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2929 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2930 |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
2931 rd= get_block_rd(s, mb_x, mb_y, 0, obmc_edged); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2932 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2933 //FIXME chroma |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2934 if(rd < *best_rd){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2935 *best_rd= rd; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2936 return 1; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2937 }else{ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2938 *block= backup; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2939 return 0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2940 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2941 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2942 |
2999
bc83e7a080a5
this fixes compilation with gcc 2.95 which seems to not like array casting... special case inline of check_block which makes the array and passes it
mmu_man
parents:
2998
diff
changeset
|
2943 /* special case for int[2] args we discard afterward, fixes compilation prob with gcc 2.95 */ |
4283
d6f83e2f8804
rename always_inline to av_always_inline and move to common.h
mru
parents:
4197
diff
changeset
|
2944 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){ |
2999
bc83e7a080a5
this fixes compilation with gcc 2.95 which seems to not like array casting... special case inline of check_block which makes the array and passes it
mmu_man
parents:
2998
diff
changeset
|
2945 int p[2] = {p0, p1}; |
3197 | 2946 return check_block(s, mb_x, mb_y, p, 0, obmc_edged, best_rd); |
2999
bc83e7a080a5
this fixes compilation with gcc 2.95 which seems to not like array casting... special case inline of check_block which makes the array and passes it
mmu_man
parents:
2998
diff
changeset
|
2947 } |
bc83e7a080a5
this fixes compilation with gcc 2.95 which seems to not like array casting... special case inline of check_block which makes the array and passes it
mmu_man
parents:
2998
diff
changeset
|
2948 |
4283
d6f83e2f8804
rename always_inline to av_always_inline and move to common.h
mru
parents:
4197
diff
changeset
|
2949 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){ |
3051 | 2950 const int b_stride= s->b_width << s->block_max_depth; |
2951 BlockNode *block= &s->block[mb_x + mb_y * b_stride]; | |
2952 BlockNode backup[4]= {block[0], block[1], block[b_stride], block[b_stride+1]}; | |
2953 int rd, index, value; | |
2954 | |
2955 assert(mb_x>=0 && mb_y>=0); | |
2956 assert(mb_x<b_stride); | |
2957 assert(((mb_x|mb_y)&1) == 0); | |
2958 | |
2959 index= (p0 + 31*p1) & (ME_CACHE_SIZE-1); | |
3314 | 2960 value= s->me_cache_generation + (p0>>10) + (p1<<6) + (block->ref<<12); |
3051 | 2961 if(s->me_cache[index] == value) |
2962 return 0; | |
2963 s->me_cache[index]= value; | |
2964 | |
2965 block->mx= p0; | |
2966 block->my= p1; | |
3314 | 2967 block->ref= ref; |
3051 | 2968 block->type &= ~BLOCK_INTRA; |
2969 block[1]= block[b_stride]= block[b_stride+1]= *block; | |
2970 | |
2971 rd= get_4block_rd(s, mb_x, mb_y, 0); | |
2972 | |
2973 //FIXME chroma | |
2974 if(rd < *best_rd){ | |
2975 *best_rd= rd; | |
2976 return 1; | |
2977 }else{ | |
2978 block[0]= backup[0]; | |
2979 block[1]= backup[1]; | |
2980 block[b_stride]= backup[2]; | |
2981 block[b_stride+1]= backup[3]; | |
2982 return 0; | |
2983 } | |
2984 } | |
2985 | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2986 static void iterative_me(SnowContext *s){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2987 int pass, mb_x, mb_y; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2988 const int b_width = s->b_width << s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2989 const int b_height= s->b_height << s->block_max_depth; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2990 const int b_stride= b_width; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2991 int color[3]; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
2992 |
3194
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
2993 { |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
2994 RangeCoder r = s->c; |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
2995 uint8_t state[sizeof(s->block_state)]; |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
2996 memcpy(state, s->block_state, sizeof(s->block_state)); |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
2997 for(mb_y= 0; mb_y<s->b_height; mb_y++) |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
2998 for(mb_x= 0; mb_x<s->b_width; mb_x++) |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
2999 encode_q_branch(s, 0, mb_x, mb_y); |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
3000 s->c = r; |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
3001 memcpy(s->block_state, state, sizeof(s->block_state)); |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
3002 } |
c30e9bcbb716
seed iterative_me with mvs from conventional search.
lorenm
parents:
3193
diff
changeset
|
3003 |
3323
87c54a3f8d19
Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
lorenm
parents:
3322
diff
changeset
|
3004 for(pass=0; pass<25; pass++){ |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3005 int change= 0; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3006 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3007 for(mb_y= 0; mb_y<b_height; mb_y++){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3008 for(mb_x= 0; mb_x<b_width; mb_x++){ |
3314 | 3009 int dia_change, i, j, ref; |
3010 int best_rd= INT_MAX, ref_rd; | |
3011 BlockNode backup, ref_b; | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3012 const int index= mb_x + mb_y * b_stride; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3013 BlockNode *block= &s->block[index]; |
3324 | 3014 BlockNode *tb = mb_y ? &s->block[index-b_stride ] : NULL; |
3015 BlockNode *lb = mb_x ? &s->block[index -1] : NULL; | |
3016 BlockNode *rb = mb_x+1<b_width ? &s->block[index +1] : NULL; | |
3017 BlockNode *bb = mb_y+1<b_height ? &s->block[index+b_stride ] : NULL; | |
3018 BlockNode *tlb= mb_x && mb_y ? &s->block[index-b_stride-1] : NULL; | |
3019 BlockNode *trb= mb_x+1<b_width && mb_y ? &s->block[index-b_stride+1] : NULL; | |
3020 BlockNode *blb= mb_x && mb_y+1<b_height ? &s->block[index+b_stride-1] : NULL; | |
3021 BlockNode *brb= mb_x+1<b_width && mb_y+1<b_height ? &s->block[index+b_stride+1] : NULL; | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3022 const int b_w= (MB_SIZE >> s->block_max_depth); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3023 uint8_t obmc_edged[b_w*2][b_w*2]; |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3024 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3025 if(pass && (block->type & BLOCK_OPT)) |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3026 continue; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3027 block->type |= BLOCK_OPT; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3028 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3029 backup= *block; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3030 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3031 if(!s->me_cache_generation) |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3032 memset(s->me_cache, 0, sizeof(s->me_cache)); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3033 s->me_cache_generation += 1<<22; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3034 |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3035 //FIXME precalc |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3036 { |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3037 int x, y; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3038 memcpy(obmc_edged, obmc_tab[s->block_max_depth], b_w*b_w*4); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3039 if(mb_x==0) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3040 for(y=0; y<b_w*2; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3041 memset(obmc_edged[y], obmc_edged[y][0] + obmc_edged[y][b_w-1], b_w); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3042 if(mb_x==b_stride-1) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3043 for(y=0; y<b_w*2; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3044 memset(obmc_edged[y]+b_w, obmc_edged[y][b_w] + obmc_edged[y][b_w*2-1], b_w); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3045 if(mb_y==0){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3046 for(x=0; x<b_w*2; x++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3047 obmc_edged[0][x] += obmc_edged[b_w-1][x]; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3048 for(y=1; y<b_w; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3049 memcpy(obmc_edged[y], obmc_edged[0], b_w*2); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3050 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3051 if(mb_y==b_height-1){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3052 for(x=0; x<b_w*2; x++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3053 obmc_edged[b_w*2-1][x] += obmc_edged[b_w][x]; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3054 for(y=b_w; y<b_w*2-1; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3055 memcpy(obmc_edged[y], obmc_edged[b_w*2-1], b_w*2); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3056 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3057 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3058 |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3059 //skip stuff outside the picture |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3060 if(mb_x==0 || mb_y==0 || mb_x==b_width-1 || mb_y==b_height-1) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3061 { |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3062 uint8_t *src= s-> input_picture.data[0]; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3063 uint8_t *dst= s->current_picture.data[0]; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3064 const int stride= s->current_picture.linesize[0]; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3065 const int block_w= MB_SIZE >> s->block_max_depth; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3066 const int sx= block_w*mb_x - block_w/2; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3067 const int sy= block_w*mb_y - block_w/2; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3068 const int w= s->plane[0].width; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3069 const int h= s->plane[0].height; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3070 int y; |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3071 |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3072 for(y=sy; y<0; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3073 memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3074 for(y=h; y<sy+block_w*2; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3075 memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3076 if(sx<0){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3077 for(y=sy; y<sy+block_w*2; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3078 memcpy(dst + sx + y*stride, src + sx + y*stride, -sx); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3079 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3080 if(sx+block_w*2 > w){ |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3081 for(y=sy; y<sy+block_w*2; y++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3082 memcpy(dst + w + y*stride, src + w + y*stride, sx+block_w*2 - w); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3083 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3084 } |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3085 |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3086 // intra(black) = neighbors' contribution to the current block |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3087 for(i=0; i<3; i++) |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3088 color[i]= get_dc(s, mb_x, mb_y, i); |
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3089 |
5127 | 3090 // get previous score (cannot be cached due to OBMC) |
3057 | 3091 if(pass > 0 && (block->type&BLOCK_INTRA)){ |
3092 int color0[3]= {block->color[0], block->color[1], block->color[2]}; | |
3093 check_block(s, mb_x, mb_y, color0, 1, *obmc_edged, &best_rd); | |
3094 }else | |
3197 | 3095 check_block_inter(s, mb_x, mb_y, block->mx, block->my, *obmc_edged, &best_rd); |
3096 | |
3314 | 3097 ref_b= *block; |
3098 ref_rd= best_rd; | |
3099 for(ref=0; ref < s->ref_frames; ref++){ | |
3100 int16_t (*mvr)[2]= &s->ref_mvs[ref][index]; | |
3101 if(s->ref_scores[ref][index] > s->ref_scores[ref_b.ref][index]*3/2) //FIXME tune threshold | |
3102 continue; | |
3103 block->ref= ref; | |
3104 best_rd= INT_MAX; | |
3105 | |
3106 check_block_inter(s, mb_x, mb_y, mvr[0][0], mvr[0][1], *obmc_edged, &best_rd); | |
3107 check_block_inter(s, mb_x, mb_y, 0, 0, *obmc_edged, &best_rd); | |
3324 | 3108 if(tb) |
3314 | 3109 check_block_inter(s, mb_x, mb_y, mvr[-b_stride][0], mvr[-b_stride][1], *obmc_edged, &best_rd); |
3324 | 3110 if(lb) |
3314 | 3111 check_block_inter(s, mb_x, mb_y, mvr[-1][0], mvr[-1][1], *obmc_edged, &best_rd); |
3324 | 3112 if(rb) |
3314 | 3113 check_block_inter(s, mb_x, mb_y, mvr[1][0], mvr[1][1], *obmc_edged, &best_rd); |
3324 | 3114 if(bb) |
3314 | 3115 check_block_inter(s, mb_x, mb_y, mvr[b_stride][0], mvr[b_stride][1], *obmc_edged, &best_rd); |
3116 | |
3117 /* fullpel ME */ | |
3118 //FIXME avoid subpel interpol / round to nearest integer | |
3119 do{ | |
3120 dia_change=0; | |
3121 for(i=0; i<FFMAX(s->avctx->dia_size, 1); i++){ | |
3122 for(j=0; j<i; j++){ | |
3123 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my+(4*j), *obmc_edged, &best_rd); | |
3124 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my-(4*j), *obmc_edged, &best_rd); | |
3125 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my-(4*j), *obmc_edged, &best_rd); | |
3126 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my+(4*j), *obmc_edged, &best_rd); | |
3127 } | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3128 } |
3314 | 3129 }while(dia_change); |
3130 /* subpel ME */ | |
3131 do{ | |
3132 static const int square[8][2]= {{+1, 0},{-1, 0},{ 0,+1},{ 0,-1},{+1,+1},{-1,-1},{+1,-1},{-1,+1},}; | |
3133 dia_change=0; | |
3134 for(i=0; i<8; i++) | |
3135 dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+square[i][0], block->my+square[i][1], *obmc_edged, &best_rd); | |
3136 }while(dia_change); | |
3137 //FIXME or try the standard 2 pass qpel or similar | |
3138 | |
3139 mvr[0][0]= block->mx; | |
3140 mvr[0][1]= block->my; | |
3141 if(ref_rd > best_rd){ | |
3142 ref_rd= best_rd; | |
3143 ref_b= *block; | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3144 } |
3314 | 3145 } |
3146 best_rd= ref_rd; | |
3147 *block= ref_b; | |
2998 | 3148 #if 1 |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3149 check_block(s, mb_x, mb_y, color, 1, *obmc_edged, &best_rd); |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3150 //FIXME RD style color selection |
2998 | 3151 #endif |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3152 if(!same_block(block, &backup)){ |
3324 | 3153 if(tb ) tb ->type &= ~BLOCK_OPT; |
3154 if(lb ) lb ->type &= ~BLOCK_OPT; | |
3155 if(rb ) rb ->type &= ~BLOCK_OPT; | |
3156 if(bb ) bb ->type &= ~BLOCK_OPT; | |
3157 if(tlb) tlb->type &= ~BLOCK_OPT; | |
3158 if(trb) trb->type &= ~BLOCK_OPT; | |
3159 if(blb) blb->type &= ~BLOCK_OPT; | |
3160 if(brb) brb->type &= ~BLOCK_OPT; | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3161 change ++; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3162 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3163 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3164 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3165 av_log(NULL, AV_LOG_ERROR, "pass:%d changed:%d\n", pass, change); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3166 if(!change) |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3167 break; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3168 } |
3051 | 3169 |
3170 if(s->block_max_depth == 1){ | |
3171 int change= 0; | |
3172 for(mb_y= 0; mb_y<b_height; mb_y+=2){ | |
3173 for(mb_x= 0; mb_x<b_width; mb_x+=2){ | |
3324 | 3174 int i; |
3051 | 3175 int best_rd, init_rd; |
3176 const int index= mb_x + mb_y * b_stride; | |
3177 BlockNode *b[4]; | |
3178 | |
3179 b[0]= &s->block[index]; | |
3180 b[1]= b[0]+1; | |
3181 b[2]= b[0]+b_stride; | |
3182 b[3]= b[2]+1; | |
3183 if(same_block(b[0], b[1]) && | |
3184 same_block(b[0], b[2]) && | |
3185 same_block(b[0], b[3])) | |
3186 continue; | |
3187 | |
3188 if(!s->me_cache_generation) | |
3189 memset(s->me_cache, 0, sizeof(s->me_cache)); | |
3190 s->me_cache_generation += 1<<22; | |
3191 | |
3192 init_rd= best_rd= get_4block_rd(s, mb_x, mb_y, 0); | |
3193 | |
3314 | 3194 //FIXME more multiref search? |
3051 | 3195 check_4block_inter(s, mb_x, mb_y, |
3196 (b[0]->mx + b[1]->mx + b[2]->mx + b[3]->mx + 2) >> 2, | |
3314 | 3197 (b[0]->my + b[1]->my + b[2]->my + b[3]->my + 2) >> 2, 0, &best_rd); |
3051 | 3198 |
3199 for(i=0; i<4; i++) | |
3200 if(!(b[i]->type&BLOCK_INTRA)) | |
3314 | 3201 check_4block_inter(s, mb_x, mb_y, b[i]->mx, b[i]->my, b[i]->ref, &best_rd); |
3051 | 3202 |
3203 if(init_rd != best_rd) | |
3204 change++; | |
3205 } | |
3206 } | |
3207 av_log(NULL, AV_LOG_ERROR, "pass:4mv changed:%d\n", change*4); | |
3208 } | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3209 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3210 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3211 static void quantize(SnowContext *s, SubBand *b, IDWTELEM *dst, DWTELEM *src, int stride, int bias){ |
2138 | 3212 const int level= b->level; |
3213 const int w= b->width; | |
3214 const int h= b->height; | |
4594 | 3215 const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16); |
5575 | 3216 const int qmul= qexp[qlog&(QROOT-1)]<<((qlog>>QSHIFT) + ENCODER_EXTRA_BITS); |
2150 | 3217 int x,y, thres1, thres2; |
2893 | 3218 // START_TIMER |
2138 | 3219 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3220 if(s->qlog == LOSSLESS_QLOG){ |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3221 for(y=0; y<h; y++) |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3222 for(x=0; x<w; x++) |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3223 dst[x + y*stride]= src[x + y*stride]; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3224 return; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3225 } |
2967 | 3226 |
2138 | 3227 bias= bias ? 0 : (3*qmul)>>3; |
2150 | 3228 thres1= ((qmul - bias)>>QEXPSHIFT) - 1; |
3229 thres2= 2*thres1; | |
2967 | 3230 |
2138 | 3231 if(!bias){ |
3232 for(y=0; y<h; y++){ | |
3233 for(x=0; x<w; x++){ | |
2150 | 3234 int i= src[x + y*stride]; |
2967 | 3235 |
2150 | 3236 if((unsigned)(i+thres1) > thres2){ |
3237 if(i>=0){ | |
3238 i<<= QEXPSHIFT; | |
3239 i/= qmul; //FIXME optimize | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3240 dst[x + y*stride]= i; |
2150 | 3241 }else{ |
3242 i= -i; | |
3243 i<<= QEXPSHIFT; | |
3244 i/= qmul; //FIXME optimize | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3245 dst[x + y*stride]= -i; |
2150 | 3246 } |
3247 }else | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3248 dst[x + y*stride]= 0; |
2138 | 3249 } |
3250 } | |
3251 }else{ | |
3252 for(y=0; y<h; y++){ | |
3253 for(x=0; x<w; x++){ | |
2967 | 3254 int i= src[x + y*stride]; |
3255 | |
2150 | 3256 if((unsigned)(i+thres1) > thres2){ |
3257 if(i>=0){ | |
3258 i<<= QEXPSHIFT; | |
3259 i= (i + bias) / qmul; //FIXME optimize | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3260 dst[x + y*stride]= i; |
2150 | 3261 }else{ |
3262 i= -i; | |
3263 i<<= QEXPSHIFT; | |
3264 i= (i + bias) / qmul; //FIXME optimize | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3265 dst[x + y*stride]= -i; |
2150 | 3266 } |
3267 }else | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3268 dst[x + y*stride]= 0; |
2138 | 3269 } |
3270 } | |
3271 } | |
2150 | 3272 if(level+1 == s->spatial_decomposition_count){ |
3273 // STOP_TIMER("quantize") | |
3274 } | |
2138 | 3275 } |
3276 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3277 static void dequantize_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, IDWTELEM *src, int stride, int start_y, int end_y){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3278 const int w= b->width; |
4594 | 3279 const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16); |
2600 | 3280 const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3281 const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3282 int x,y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3283 START_TIMER |
2967 | 3284 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3285 if(s->qlog == LOSSLESS_QLOG) return; |
2967 | 3286 |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
3287 for(y=start_y; y<end_y; y++){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3288 // DWTELEM * line = slice_buffer_get_line_from_address(sb, src + (y * stride)); |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3289 IDWTELEM * line = slice_buffer_get_line(sb, (y * b->stride_line) + b->buf_y_offset) + b->buf_x_offset; |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3290 for(x=0; x<w; x++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3291 int i= line[x]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3292 if(i<0){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3293 line[x]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3294 }else if(i>0){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3295 line[x]= (( i*qmul + qadd)>>(QEXPSHIFT)); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3296 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3297 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3298 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3299 if(w > 200 /*level+1 == s->spatial_decomposition_count*/){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3300 STOP_TIMER("dquant") |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3301 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3302 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3303 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3304 static void dequantize(SnowContext *s, SubBand *b, IDWTELEM *src, int stride){ |
2138 | 3305 const int w= b->width; |
3306 const int h= b->height; | |
4594 | 3307 const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16); |
2600 | 3308 const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); |
2138 | 3309 const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT; |
3310 int x,y; | |
2195 | 3311 START_TIMER |
2967 | 3312 |
2161 | 3313 if(s->qlog == LOSSLESS_QLOG) return; |
2967 | 3314 |
2138 | 3315 for(y=0; y<h; y++){ |
3316 for(x=0; x<w; x++){ | |
3317 int i= src[x + y*stride]; | |
3318 if(i<0){ | |
3319 src[x + y*stride]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias | |
3320 }else if(i>0){ | |
3321 src[x + y*stride]= (( i*qmul + qadd)>>(QEXPSHIFT)); | |
3322 } | |
3323 } | |
3324 } | |
2195 | 3325 if(w > 200 /*level+1 == s->spatial_decomposition_count*/){ |
3326 STOP_TIMER("dquant") | |
3327 } | |
2138 | 3328 } |
3329 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3330 static void decorrelate(SnowContext *s, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median){ |
2138 | 3331 const int w= b->width; |
3332 const int h= b->height; | |
3333 int x,y; | |
2967 | 3334 |
2138 | 3335 for(y=h-1; y>=0; y--){ |
3336 for(x=w-1; x>=0; x--){ | |
3337 int i= x + y*stride; | |
2967 | 3338 |
2138 | 3339 if(x){ |
3340 if(use_median){ | |
3341 if(y && x+1<w) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]); | |
3342 else src[i] -= src[i - 1]; | |
3343 }else{ | |
3344 if(y) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]); | |
3345 else src[i] -= src[i - 1]; | |
3346 } | |
3347 }else{ | |
3348 if(y) src[i] -= src[i - stride]; | |
3349 } | |
3350 } | |
3351 } | |
3352 } | |
3353 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3354 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){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3355 const int w= b->width; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3356 int x,y; |
2967 | 3357 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3358 // START_TIMER |
2967 | 3359 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3360 IDWTELEM * line=0; // silence silly "could be used without having been initialized" warning |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3361 IDWTELEM * prev; |
2967 | 3362 |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
3363 if (start_y != 0) |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
3364 line = slice_buffer_get_line(sb, ((start_y - 1) * b->stride_line) + b->buf_y_offset) + b->buf_x_offset; |
2967 | 3365 |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
3366 for(y=start_y; y<end_y; y++){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3367 prev = line; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3368 // line = slice_buffer_get_line_from_address(sb, src + (y * stride)); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3369 line = slice_buffer_get_line(sb, (y * b->stride_line) + b->buf_y_offset) + b->buf_x_offset; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3370 for(x=0; x<w; x++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3371 if(x){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3372 if(use_median){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3373 if(y && x+1<w) line[x] += mid_pred(line[x - 1], prev[x], prev[x + 1]); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3374 else line[x] += line[x - 1]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3375 }else{ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3376 if(y) line[x] += mid_pred(line[x - 1], prev[x], line[x - 1] + prev[x] - prev[x - 1]); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3377 else line[x] += line[x - 1]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3378 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3379 }else{ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3380 if(y) line[x] += prev[x]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3381 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3382 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3383 } |
2967 | 3384 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3385 // STOP_TIMER("correlate") |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3386 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3387 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3388 static void correlate(SnowContext *s, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median){ |
2138 | 3389 const int w= b->width; |
3390 const int h= b->height; | |
3391 int x,y; | |
2967 | 3392 |
2138 | 3393 for(y=0; y<h; y++){ |
3394 for(x=0; x<w; x++){ | |
3395 int i= x + y*stride; | |
2967 | 3396 |
2138 | 3397 if(x){ |
3398 if(use_median){ | |
3399 if(y && x+1<w) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]); | |
3400 else src[i] += src[i - 1]; | |
3401 }else{ | |
3402 if(y) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]); | |
3403 else src[i] += src[i - 1]; | |
3404 } | |
3405 }else{ | |
3406 if(y) src[i] += src[i - stride]; | |
3407 } | |
3408 } | |
3409 } | |
3410 } | |
3411 | |
3412 static void encode_header(SnowContext *s){ | |
3413 int plane_index, level, orientation; | |
2967 | 3414 uint8_t kstate[32]; |
3415 | |
3416 memset(kstate, MID_STATE, sizeof(kstate)); | |
2138 | 3417 |
2335 | 3418 put_rac(&s->c, kstate, s->keyframe); |
4331
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3419 if(s->keyframe || s->always_reset){ |
2199 | 3420 reset_contexts(s); |
4331
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3421 s->last_spatial_decomposition_type= |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3422 s->last_qlog= |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3423 s->last_qbias= |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3424 s->last_mv_scale= |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3425 s->last_block_max_depth= 0; |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3426 } |
2138 | 3427 if(s->keyframe){ |
3428 put_symbol(&s->c, s->header_state, s->version, 0); | |
2335 | 3429 put_rac(&s->c, s->header_state, s->always_reset); |
2138 | 3430 put_symbol(&s->c, s->header_state, s->temporal_decomposition_type, 0); |
3431 put_symbol(&s->c, s->header_state, s->temporal_decomposition_count, 0); | |
3432 put_symbol(&s->c, s->header_state, s->spatial_decomposition_count, 0); | |
3433 put_symbol(&s->c, s->header_state, s->colorspace_type, 0); | |
3434 put_symbol(&s->c, s->header_state, s->chroma_h_shift, 0); | |
3435 put_symbol(&s->c, s->header_state, s->chroma_v_shift, 0); | |
2335 | 3436 put_rac(&s->c, s->header_state, s->spatial_scalability); |
3437 // put_rac(&s->c, s->header_state, s->rate_scalability); | |
3314 | 3438 put_symbol(&s->c, s->header_state, s->max_ref_frames-1, 0); |
2138 | 3439 |
3440 for(plane_index=0; plane_index<2; plane_index++){ | |
3441 for(level=0; level<s->spatial_decomposition_count; level++){ | |
3442 for(orientation=level ? 1:0; orientation<4; orientation++){ | |
3443 if(orientation==2) continue; | |
3444 put_symbol(&s->c, s->header_state, s->plane[plane_index].band[level][orientation].qlog, 1); | |
3445 } | |
3446 } | |
3447 } | |
3448 } | |
4331
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3449 put_symbol(&s->c, s->header_state, s->spatial_decomposition_type - s->last_spatial_decomposition_type, 1); |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3450 put_symbol(&s->c, s->header_state, s->qlog - s->last_qlog , 1); |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3451 put_symbol(&s->c, s->header_state, s->mv_scale - s->last_mv_scale, 1); |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3452 put_symbol(&s->c, s->header_state, s->qbias - s->last_qbias , 1); |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3453 put_symbol(&s->c, s->header_state, s->block_max_depth - s->last_block_max_depth, 1); |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3454 |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3455 s->last_spatial_decomposition_type= s->spatial_decomposition_type; |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3456 s->last_qlog = s->qlog; |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3457 s->last_qbias = s->qbias; |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3458 s->last_mv_scale = s->mv_scale; |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3459 s->last_block_max_depth = s->block_max_depth; |
2138 | 3460 } |
3461 | |
3462 static int decode_header(SnowContext *s){ | |
3463 int plane_index, level, orientation; | |
2335 | 3464 uint8_t kstate[32]; |
2138 | 3465 |
2967 | 3466 memset(kstate, MID_STATE, sizeof(kstate)); |
2335 | 3467 |
3468 s->keyframe= get_rac(&s->c, kstate); | |
4331
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3469 if(s->keyframe || s->always_reset){ |
2199 | 3470 reset_contexts(s); |
4331
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3471 s->spatial_decomposition_type= |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3472 s->qlog= |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3473 s->qbias= |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3474 s->mv_scale= |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3475 s->block_max_depth= 0; |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3476 } |
2138 | 3477 if(s->keyframe){ |
3478 s->version= get_symbol(&s->c, s->header_state, 0); | |
3479 if(s->version>0){ | |
3480 av_log(s->avctx, AV_LOG_ERROR, "version %d not supported", s->version); | |
3481 return -1; | |
3482 } | |
2335 | 3483 s->always_reset= get_rac(&s->c, s->header_state); |
2138 | 3484 s->temporal_decomposition_type= get_symbol(&s->c, s->header_state, 0); |
3485 s->temporal_decomposition_count= get_symbol(&s->c, s->header_state, 0); | |
3486 s->spatial_decomposition_count= get_symbol(&s->c, s->header_state, 0); | |
3487 s->colorspace_type= get_symbol(&s->c, s->header_state, 0); | |
3488 s->chroma_h_shift= get_symbol(&s->c, s->header_state, 0); | |
3489 s->chroma_v_shift= get_symbol(&s->c, s->header_state, 0); | |
2335 | 3490 s->spatial_scalability= get_rac(&s->c, s->header_state); |
3491 // s->rate_scalability= get_rac(&s->c, s->header_state); | |
3314 | 3492 s->max_ref_frames= get_symbol(&s->c, s->header_state, 0)+1; |
2138 | 3493 |
3494 for(plane_index=0; plane_index<3; plane_index++){ | |
3495 for(level=0; level<s->spatial_decomposition_count; level++){ | |
3496 for(orientation=level ? 1:0; orientation<4; orientation++){ | |
3497 int q; | |
3498 if (plane_index==2) q= s->plane[1].band[level][orientation].qlog; | |
3499 else if(orientation==2) q= s->plane[plane_index].band[level][1].qlog; | |
3500 else q= get_symbol(&s->c, s->header_state, 1); | |
3501 s->plane[plane_index].band[level][orientation].qlog= q; | |
3502 } | |
3503 } | |
3504 } | |
3505 } | |
2967 | 3506 |
4331
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3507 s->spatial_decomposition_type+= get_symbol(&s->c, s->header_state, 1); |
5588
effa59ca89b3
we only have 2 wavelets, the 3rd was just for experimentation ...
michael
parents:
5587
diff
changeset
|
3508 if(s->spatial_decomposition_type > 1){ |
2138 | 3509 av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_type %d not supported", s->spatial_decomposition_type); |
3510 return -1; | |
3511 } | |
2967 | 3512 |
4331
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3513 s->qlog += get_symbol(&s->c, s->header_state, 1); |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3514 s->mv_scale += get_symbol(&s->c, s->header_state, 1); |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3515 s->qbias += get_symbol(&s->c, s->header_state, 1); |
e571dfe677be
store a few values in the header as difference to the last
michael
parents:
4283
diff
changeset
|
3516 s->block_max_depth+= get_symbol(&s->c, s->header_state, 1); |
3303
68721b62a528
sanity checks, some might have been exploitable ...
michael
parents:
3206
diff
changeset
|
3517 if(s->block_max_depth > 1 || s->block_max_depth < 0){ |
2952 | 3518 av_log(s->avctx, AV_LOG_ERROR, "block_max_depth= %d is too large", s->block_max_depth); |
3519 s->block_max_depth= 0; | |
3520 return -1; | |
3521 } | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3522 |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3523 return 0; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3524 } |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3525 |
3075 | 3526 static void init_qexp(void){ |
2600 | 3527 int i; |
3528 double v=128; | |
3529 | |
3530 for(i=0; i<QROOT; i++){ | |
3531 qexp[i]= lrintf(v); | |
2967 | 3532 v *= pow(2, 1.0 / QROOT); |
2600 | 3533 } |
3534 } | |
3535 | |
2138 | 3536 static int common_init(AVCodecContext *avctx){ |
3537 SnowContext *s = avctx->priv_data; | |
3538 int width, height; | |
3539 int level, orientation, plane_index, dec; | |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
3540 int i, j; |
2138 | 3541 |
3542 s->avctx= avctx; | |
2967 | 3543 |
2138 | 3544 dsputil_init(&s->dsp, avctx); |
3545 | |
3546 #define mcf(dx,dy)\ | |
3547 s->dsp.put_qpel_pixels_tab [0][dy+dx/4]=\ | |
3548 s->dsp.put_no_rnd_qpel_pixels_tab[0][dy+dx/4]=\ | |
2224
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
3549 s->dsp.put_h264_qpel_pixels_tab[0][dy+dx/4];\ |
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
3550 s->dsp.put_qpel_pixels_tab [1][dy+dx/4]=\ |
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
3551 s->dsp.put_no_rnd_qpel_pixels_tab[1][dy+dx/4]=\ |
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
3552 s->dsp.put_h264_qpel_pixels_tab[1][dy+dx/4]; |
2138 | 3553 |
3554 mcf( 0, 0) | |
3555 mcf( 4, 0) | |
3556 mcf( 8, 0) | |
3557 mcf(12, 0) | |
3558 mcf( 0, 4) | |
3559 mcf( 4, 4) | |
3560 mcf( 8, 4) | |
3561 mcf(12, 4) | |
3562 mcf( 0, 8) | |
3563 mcf( 4, 8) | |
3564 mcf( 8, 8) | |
3565 mcf(12, 8) | |
3566 mcf( 0,12) | |
3567 mcf( 4,12) | |
3568 mcf( 8,12) | |
3569 mcf(12,12) | |
3570 | |
3571 #define mcfh(dx,dy)\ | |
3572 s->dsp.put_pixels_tab [0][dy/4+dx/8]=\ | |
3573 s->dsp.put_no_rnd_pixels_tab[0][dy/4+dx/8]=\ | |
2224
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
3574 mc_block_hpel ## dx ## dy ## 16;\ |
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
3575 s->dsp.put_pixels_tab [1][dy/4+dx/8]=\ |
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
3576 s->dsp.put_no_rnd_pixels_tab[1][dy/4+dx/8]=\ |
11d54cb7ac4e
100l (forgot to set the 8x8 mc functions, so ME did use the mpeg ones ...)
michael
parents:
2223
diff
changeset
|
3577 mc_block_hpel ## dx ## dy ## 8; |
2138 | 3578 |
3579 mcfh(0, 0) | |
3580 mcfh(8, 0) | |
3581 mcfh(0, 8) | |
3582 mcfh(8, 8) | |
2600 | 3583 |
3584 if(!qexp[0]) | |
3585 init_qexp(); | |
3586 | |
2138 | 3587 dec= s->spatial_decomposition_count= 5; |
3588 s->spatial_decomposition_type= avctx->prediction_method; //FIXME add decorrelator type r transform_type | |
2967 | 3589 |
2138 | 3590 s->chroma_h_shift= 1; //FIXME XXX |
3591 s->chroma_v_shift= 1; | |
2967 | 3592 |
2138 | 3593 // dec += FFMAX(s->chroma_h_shift, s->chroma_v_shift); |
2967 | 3594 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3595 width= s->avctx->width; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3596 height= s->avctx->height; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3597 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3598 s->spatial_idwt_buffer= av_mallocz(width*height*sizeof(IDWTELEM)); |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3599 s->spatial_dwt_buffer= av_mallocz(width*height*sizeof(DWTELEM)); //FIXME this doesnt belong here |
2967 | 3600 |
2138 | 3601 s->mv_scale= (s->avctx->flags & CODEC_FLAG_QPEL) ? 2 : 4; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3602 s->block_max_depth= (s->avctx->flags & CODEC_FLAG_4MV) ? 1 : 0; |
2967 | 3603 |
3604 for(plane_index=0; plane_index<3; plane_index++){ | |
2138 | 3605 int w= s->avctx->width; |
3606 int h= s->avctx->height; | |
3607 | |
3608 if(plane_index){ | |
3609 w>>= s->chroma_h_shift; | |
3610 h>>= s->chroma_v_shift; | |
3611 } | |
3612 s->plane[plane_index].width = w; | |
3613 s->plane[plane_index].height= h; | |
2160 | 3614 //av_log(NULL, AV_LOG_DEBUG, "%d %d\n", w, h); |
2138 | 3615 for(level=s->spatial_decomposition_count-1; level>=0; level--){ |
3616 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
3617 SubBand *b= &s->plane[plane_index].band[level][orientation]; | |
2967 | 3618 |
2138 | 3619 b->buf= s->spatial_dwt_buffer; |
3620 b->level= level; | |
3621 b->stride= s->plane[plane_index].width << (s->spatial_decomposition_count - level); | |
3622 b->width = (w + !(orientation&1))>>1; | |
3623 b->height= (h + !(orientation>1))>>1; | |
2967 | 3624 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3625 b->stride_line = 1 << (s->spatial_decomposition_count - level); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3626 b->buf_x_offset = 0; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3627 b->buf_y_offset = 0; |
2967 | 3628 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3629 if(orientation&1){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3630 b->buf += (w+1)>>1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3631 b->buf_x_offset = (w+1)>>1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3632 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3633 if(orientation>1){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3634 b->buf += b->stride>>1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3635 b->buf_y_offset = b->stride_line >> 1; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3636 } |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3637 b->ibuf= s->spatial_idwt_buffer + (b->buf - s->spatial_dwt_buffer); |
2967 | 3638 |
2138 | 3639 if(level) |
3640 b->parent= &s->plane[plane_index].band[level-1][orientation]; | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
3641 b->x_coeff=av_mallocz(((b->width+1) * b->height+1)*sizeof(x_and_coeff)); |
2138 | 3642 } |
3643 w= (w+1)>>1; | |
3644 h= (h+1)>>1; | |
3645 } | |
3646 } | |
2967 | 3647 |
3325
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
3648 for(i=0; i<MAX_REF_FRAMES; i++) |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
3649 for(j=0; j<MAX_REF_FRAMES; j++) |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
3650 scale_mv_ref[i][j] = 256*(i+1)/(j+1); |
c2a017de6bea
Snow: scale predicted mv based on which reference frame the neighbors used.
lorenm
parents:
3324
diff
changeset
|
3651 |
2138 | 3652 reset_contexts(s); |
2967 | 3653 /* |
2138 | 3654 width= s->width= avctx->width; |
3655 height= s->height= avctx->height; | |
2967 | 3656 |
2138 | 3657 assert(width && height); |
3658 */ | |
3659 s->avctx->get_buffer(s->avctx, &s->mconly_picture); | |
2967 | 3660 |
2138 | 3661 return 0; |
3662 } | |
3663 | |
3322
0b4f548dfb44
Snow: 10l. 1pass ratecontrol failed to set snow's internal quant.
lorenm
parents:
3314
diff
changeset
|
3664 static int qscale2qlog(int qscale){ |
0b4f548dfb44
Snow: 10l. 1pass ratecontrol failed to set snow's internal quant.
lorenm
parents:
3314
diff
changeset
|
3665 return rint(QROOT*log(qscale / (float)FF_QP2LAMBDA)/log(2)) |
0b4f548dfb44
Snow: 10l. 1pass ratecontrol failed to set snow's internal quant.
lorenm
parents:
3314
diff
changeset
|
3666 + 61*QROOT/8; //<64 >60 |
0b4f548dfb44
Snow: 10l. 1pass ratecontrol failed to set snow's internal quant.
lorenm
parents:
3314
diff
changeset
|
3667 } |
0b4f548dfb44
Snow: 10l. 1pass ratecontrol failed to set snow's internal quant.
lorenm
parents:
3314
diff
changeset
|
3668 |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
3669 static int ratecontrol_1pass(SnowContext *s, AVFrame *pict) |
3313 | 3670 { |
3671 /* estimate the frame's complexity as a sum of weighted dwt coefs. | |
3672 * FIXME we know exact mv bits at this point, | |
3673 * but ratecontrol isn't set up to include them. */ | |
3674 uint32_t coef_sum= 0; | |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
3675 int level, orientation, delta_qlog; |
3313 | 3676 |
3677 for(level=0; level<s->spatial_decomposition_count; level++){ | |
3678 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
3679 SubBand *b= &s->plane[0].band[level][orientation]; | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3680 IDWTELEM *buf= b->ibuf; |
3313 | 3681 const int w= b->width; |
3682 const int h= b->height; | |
3683 const int stride= b->stride; | |
4594 | 3684 const int qlog= av_clip(2*QROOT + b->qlog, 0, QROOT*16); |
3313 | 3685 const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); |
3686 const int qdiv= (1<<16)/qmul; | |
3687 int x, y; | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3688 //FIXME this is ugly |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3689 for(y=0; y<h; y++) |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3690 for(x=0; x<w; x++) |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3691 buf[x+y*stride]= b->buf[x+y*stride]; |
3313 | 3692 if(orientation==0) |
3693 decorrelate(s, b, buf, stride, 1, 0); | |
3694 for(y=0; y<h; y++) | |
3695 for(x=0; x<w; x++) | |
3696 coef_sum+= abs(buf[x+y*stride]) * qdiv >> 16; | |
3697 } | |
3698 } | |
3699 | |
3700 /* ugly, ratecontrol just takes a sqrt again */ | |
3701 coef_sum = (uint64_t)coef_sum * coef_sum >> 16; | |
3702 assert(coef_sum < INT_MAX); | |
3703 | |
3704 if(pict->pict_type == I_TYPE){ | |
3705 s->m.current_picture.mb_var_sum= coef_sum; | |
3706 s->m.current_picture.mc_mb_var_sum= 0; | |
3707 }else{ | |
3708 s->m.current_picture.mc_mb_var_sum= coef_sum; | |
3709 s->m.current_picture.mb_var_sum= 0; | |
3710 } | |
3711 | |
3712 pict->quality= ff_rate_estimate_qscale(&s->m, 1); | |
3766 | 3713 if (pict->quality < 0) |
4011
5bce97c30a69
-1 is a valid return value in ratecontrol_1pass() -> 100l for takis
michael
parents:
4001
diff
changeset
|
3714 return INT_MIN; |
3313 | 3715 s->lambda= pict->quality * 3/2; |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
3716 delta_qlog= qscale2qlog(pict->quality) - s->qlog; |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
3717 s->qlog+= delta_qlog; |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
3718 return delta_qlog; |
3313 | 3719 } |
2138 | 3720 |
3721 static void calculate_vissual_weight(SnowContext *s, Plane *p){ | |
3722 int width = p->width; | |
3723 int height= p->height; | |
2198 | 3724 int level, orientation, x, y; |
2138 | 3725 |
3726 for(level=0; level<s->spatial_decomposition_count; level++){ | |
3727 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
3728 SubBand *b= &p->band[level][orientation]; | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3729 IDWTELEM *ibuf= b->ibuf; |
2138 | 3730 int64_t error=0; |
2967 | 3731 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3732 memset(s->spatial_idwt_buffer, 0, sizeof(*s->spatial_idwt_buffer)*width*height); |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3733 ibuf[b->width/2 + b->height/2*b->stride]= 256*16; |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3734 ff_spatial_idwt(s->spatial_idwt_buffer, width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count); |
2138 | 3735 for(y=0; y<height; y++){ |
3736 for(x=0; x<width; x++){ | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
3737 int64_t d= s->spatial_idwt_buffer[x + y*width]*16; |
2138 | 3738 error += d*d; |
3739 } | |
3740 } | |
3741 | |
3742 b->qlog= (int)(log(352256.0/sqrt(error)) / log(pow(2.0, 1.0/QROOT))+0.5); | |
2164 | 3743 // av_log(NULL, AV_LOG_DEBUG, "%d %d %d\n", level, orientation, b->qlog/*, sqrt(error)*/); |
2138 | 3744 } |
3745 } | |
3746 } | |
3747 | |
3748 static int encode_init(AVCodecContext *avctx) | |
3749 { | |
3750 SnowContext *s = avctx->priv_data; | |
2198 | 3751 int plane_index; |
2138 | 3752 |
2658
d1609cfeb1d0
#defines for strict_std_compliance and split between inofficial extensions and non standarized things
michael
parents:
2635
diff
changeset
|
3753 if(avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL){ |
2628
511e3afc43e1
Ministry of English Composition, reporting for duty (and the word is "skipped", not "skiped"; "skiped" would rhyme with "hyped")
melanson
parents:
2610
diff
changeset
|
3754 av_log(avctx, AV_LOG_ERROR, "this codec is under development, files encoded with it may not be decodable with future versions!!!\n" |
2658
d1609cfeb1d0
#defines for strict_std_compliance and split between inofficial extensions and non standarized things
michael
parents:
2635
diff
changeset
|
3755 "use vstrict=-2 / -strict -2 to use it anyway\n"); |
2151 | 3756 return -1; |
3757 } | |
2967 | 3758 |
3327
955096780e7c
Snow: don't try to encode lossless with 9/7 wavelet, because it isn't lossless.
lorenm
parents:
3326
diff
changeset
|
3759 if(avctx->prediction_method == DWT_97 |
955096780e7c
Snow: don't try to encode lossless with 9/7 wavelet, because it isn't lossless.
lorenm
parents:
3326
diff
changeset
|
3760 && (avctx->flags & CODEC_FLAG_QSCALE) |
955096780e7c
Snow: don't try to encode lossless with 9/7 wavelet, because it isn't lossless.
lorenm
parents:
3326
diff
changeset
|
3761 && avctx->global_quality == 0){ |
955096780e7c
Snow: don't try to encode lossless with 9/7 wavelet, because it isn't lossless.
lorenm
parents:
3326
diff
changeset
|
3762 av_log(avctx, AV_LOG_ERROR, "the 9/7 wavelet is incompatible with lossless mode\n"); |
955096780e7c
Snow: don't try to encode lossless with 9/7 wavelet, because it isn't lossless.
lorenm
parents:
3326
diff
changeset
|
3763 return -1; |
955096780e7c
Snow: don't try to encode lossless with 9/7 wavelet, because it isn't lossless.
lorenm
parents:
3326
diff
changeset
|
3764 } |
955096780e7c
Snow: don't try to encode lossless with 9/7 wavelet, because it isn't lossless.
lorenm
parents:
3326
diff
changeset
|
3765 |
2138 | 3766 common_init(avctx); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3767 alloc_blocks(s); |
2967 | 3768 |
2138 | 3769 s->version=0; |
2967 | 3770 |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3771 s->m.avctx = avctx; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3772 s->m.flags = avctx->flags; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3773 s->m.bit_rate= avctx->bit_rate; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3774 |
2138 | 3775 s->m.me.scratchpad= av_mallocz((avctx->width+64)*2*16*2*sizeof(uint8_t)); |
3776 s->m.me.map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t)); | |
3777 s->m.me.score_map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t)); | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
3778 s->m.obmc_scratchpad= av_mallocz(MB_SIZE*MB_SIZE*12*sizeof(uint32_t)); |
2138 | 3779 h263_encode_init(&s->m); //mv_penalty |
3780 | |
3314 | 3781 s->max_ref_frames = FFMAX(FFMIN(avctx->refs, MAX_REF_FRAMES), 1); |
3782 | |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3783 if(avctx->flags&CODEC_FLAG_PASS1){ |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3784 if(!avctx->stats_out) |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3785 avctx->stats_out = av_mallocz(256); |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3786 } |
3322
0b4f548dfb44
Snow: 10l. 1pass ratecontrol failed to set snow's internal quant.
lorenm
parents:
3314
diff
changeset
|
3787 if((avctx->flags&CODEC_FLAG_PASS2) || !(avctx->flags&CODEC_FLAG_QSCALE)){ |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3788 if(ff_rate_control_init(&s->m) < 0) |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3789 return -1; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3790 } |
3313 | 3791 s->pass1_rc= !(avctx->flags & (CODEC_FLAG_QSCALE|CODEC_FLAG_PASS2)); |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3792 |
2138 | 3793 for(plane_index=0; plane_index<3; plane_index++){ |
3794 calculate_vissual_weight(s, &s->plane[plane_index]); | |
3795 } | |
2967 | 3796 |
3797 | |
2138 | 3798 avctx->coded_frame= &s->current_picture; |
3799 switch(avctx->pix_fmt){ | |
3800 // case PIX_FMT_YUV444P: | |
3801 // case PIX_FMT_YUV422P: | |
3802 case PIX_FMT_YUV420P: | |
3803 case PIX_FMT_GRAY8: | |
3804 // case PIX_FMT_YUV411P: | |
3805 // case PIX_FMT_YUV410P: | |
3806 s->colorspace_type= 0; | |
3807 break; | |
4494
ce643a22f049
Replace deprecated PIX_FMT names by the newer variants.
diego
parents:
4436
diff
changeset
|
3808 /* case PIX_FMT_RGB32: |
2138 | 3809 s->colorspace= 1; |
3810 break;*/ | |
3811 default: | |
3812 av_log(avctx, AV_LOG_ERROR, "format not supported\n"); | |
3813 return -1; | |
3814 } | |
3815 // avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_h_shift, &s->chroma_v_shift); | |
3816 s->chroma_h_shift= 1; | |
3817 s->chroma_v_shift= 1; | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3818 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3819 ff_set_cmp(&s->dsp, s->dsp.me_cmp, s->avctx->me_cmp); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3820 ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, s->avctx->me_sub_cmp); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3821 |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3822 s->avctx->get_buffer(s->avctx, &s->input_picture); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3823 |
3314 | 3824 if(s->avctx->me_method == ME_ITER){ |
3825 int i; | |
3826 int size= s->b_width * s->b_height << 2*s->block_max_depth; | |
3827 for(i=0; i<s->max_ref_frames; i++){ | |
3828 s->ref_mvs[i]= av_mallocz(size*sizeof(int16_t[2])); | |
3829 s->ref_scores[i]= av_mallocz(size*sizeof(uint32_t)); | |
3830 } | |
3831 } | |
3832 | |
2138 | 3833 return 0; |
3834 } | |
3835 | |
5633
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3836 static void halfpel_interpol(SnowContext *s, uint8_t *halfpel[4][4], AVFrame *frame){ |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3837 int p,x,y; |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3838 |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3839 assert(!(s->avctx->flags & CODEC_FLAG_EMU_EDGE)); |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3840 |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3841 for(p=0; p<3; p++){ |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3842 int is_chroma= !!p; |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3843 int w= s->avctx->width >>is_chroma; |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3844 int h= s->avctx->height >>is_chroma; |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3845 int ls= frame->linesize[p]; |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3846 uint8_t *src= frame->data[p]; |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3847 |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3848 halfpel[1][p]= (uint8_t*)av_malloc(ls * (h+2*EDGE_WIDTH)) + EDGE_WIDTH*(1+ls); |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3849 halfpel[2][p]= (uint8_t*)av_malloc(ls * (h+2*EDGE_WIDTH)) + EDGE_WIDTH*(1+ls); |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3850 halfpel[3][p]= (uint8_t*)av_malloc(ls * (h+2*EDGE_WIDTH)) + EDGE_WIDTH*(1+ls); |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3851 |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3852 halfpel[0][p]= src; |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3853 for(y=0; y<h; y++){ |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3854 for(x=0; x<w; x++){ |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3855 int i= y*ls + x; |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3856 |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3857 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; |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3858 } |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3859 } |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3860 for(y=0; y<h; y++){ |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3861 for(x=0; x<w; x++){ |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3862 int i= y*ls + x; |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3863 |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3864 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; |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3865 } |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3866 } |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3867 src= halfpel[1][p]; |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3868 for(y=0; y<h; y++){ |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3869 for(x=0; x<w; x++){ |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3870 int i= y*ls + x; |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3871 |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3872 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; |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3873 } |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3874 } |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3875 |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3876 //FIXME border! |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3877 } |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3878 } |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3879 |
2138 | 3880 static int frame_start(SnowContext *s){ |
3881 AVFrame tmp; | |
2187 | 3882 int w= s->avctx->width; //FIXME round up to x16 ? |
3883 int h= s->avctx->height; | |
2138 | 3884 |
2187 | 3885 if(s->current_picture.data[0]){ |
3886 draw_edges(s->current_picture.data[0], s->current_picture.linesize[0], w , h , EDGE_WIDTH ); | |
3887 draw_edges(s->current_picture.data[1], s->current_picture.linesize[1], w>>1, h>>1, EDGE_WIDTH/2); | |
3888 draw_edges(s->current_picture.data[2], s->current_picture.linesize[2], w>>1, h>>1, EDGE_WIDTH/2); | |
3889 } | |
3890 | |
3314 | 3891 tmp= s->last_picture[s->max_ref_frames-1]; |
3892 memmove(s->last_picture+1, s->last_picture, (s->max_ref_frames-1)*sizeof(AVFrame)); | |
5633
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3893 memmove(s->halfpel_plane+1, s->halfpel_plane, (s->max_ref_frames-1)*sizeof(void*)*4*4); |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3894 #ifdef USE_HALFPEL_PLANE |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3895 if(s->current_picture.data[0]) |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3896 halfpel_interpol(s, s->halfpel_plane[0], &s->current_picture); |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
3897 #endif |
3314 | 3898 s->last_picture[0]= s->current_picture; |
2138 | 3899 s->current_picture= tmp; |
2967 | 3900 |
3314 | 3901 if(s->keyframe){ |
3902 s->ref_frames= 0; | |
3903 }else{ | |
3904 int i; | |
3905 for(i=0; i<s->max_ref_frames && s->last_picture[i].data[0]; i++) | |
3906 if(i && s->last_picture[i-1].key_frame) | |
3907 break; | |
3908 s->ref_frames= i; | |
3909 } | |
3910 | |
2138 | 3911 s->current_picture.reference= 1; |
3912 if(s->avctx->get_buffer(s->avctx, &s->current_picture) < 0){ | |
3913 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n"); | |
3914 return -1; | |
3915 } | |
2967 | 3916 |
3314 | 3917 s->current_picture.key_frame= s->keyframe; |
3918 | |
2138 | 3919 return 0; |
3920 } | |
3921 | |
3922 static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size, void *data){ | |
3923 SnowContext *s = avctx->priv_data; | |
2335 | 3924 RangeCoder * const c= &s->c; |
2138 | 3925 AVFrame *pict = data; |
3926 const int width= s->avctx->width; | |
3927 const int height= s->avctx->height; | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3928 int level, orientation, plane_index, i, y; |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
3929 uint8_t rc_header_bak[sizeof(s->header_state)]; |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
3930 uint8_t rc_block_bak[sizeof(s->block_state)]; |
2138 | 3931 |
2335 | 3932 ff_init_range_encoder(c, buf, buf_size); |
3933 ff_build_rac_states(c, 0.05*(1LL<<32), 256-8); | |
2967 | 3934 |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3935 for(i=0; i<3; i++){ |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3936 int shift= !!i; |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3937 for(y=0; y<(height>>shift); y++) |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3938 memcpy(&s->input_picture.data[i][y * s->input_picture.linesize[i]], |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3939 &pict->data[i][y * pict->linesize[i]], |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3940 width>>shift); |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3941 } |
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
3942 s->new_picture = *pict; |
2138 | 3943 |
3313 | 3944 s->m.picture_number= avctx->frame_number; |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3945 if(avctx->flags&CODEC_FLAG_PASS2){ |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3946 s->m.pict_type = |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3947 pict->pict_type= s->m.rc_context.entry[avctx->frame_number].new_pict_type; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3948 s->keyframe= pict->pict_type==FF_I_TYPE; |
3766 | 3949 if(!(avctx->flags&CODEC_FLAG_QSCALE)) { |
3193 | 3950 pict->quality= ff_rate_estimate_qscale(&s->m, 0); |
3766 | 3951 if (pict->quality < 0) |
3952 return -1; | |
3953 } | |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3954 }else{ |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3955 s->keyframe= avctx->gop_size==0 || avctx->frame_number % avctx->gop_size == 0; |
3313 | 3956 s->m.pict_type= |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3957 pict->pict_type= s->keyframe ? FF_I_TYPE : FF_P_TYPE; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3958 } |
2967 | 3959 |
3313 | 3960 if(s->pass1_rc && avctx->frame_number == 0) |
3961 pict->quality= 2*FF_QP2LAMBDA; | |
2161 | 3962 if(pict->quality){ |
3322
0b4f548dfb44
Snow: 10l. 1pass ratecontrol failed to set snow's internal quant.
lorenm
parents:
3314
diff
changeset
|
3963 s->qlog= qscale2qlog(pict->quality); |
3313 | 3964 s->lambda = pict->quality * 3/2; |
3965 } | |
3966 if(s->qlog < 0 || (!pict->quality && (avctx->flags & CODEC_FLAG_QSCALE))){ | |
2161 | 3967 s->qlog= LOSSLESS_QLOG; |
3313 | 3968 s->lambda = 0; |
3969 }//else keep previous frame's qlog until after motion est | |
2138 | 3970 |
3971 frame_start(s); | |
3972 | |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
3973 s->m.current_picture_ptr= &s->m.current_picture; |
2138 | 3974 if(pict->pict_type == P_TYPE){ |
3975 int block_width = (width +15)>>4; | |
3976 int block_height= (height+15)>>4; | |
3977 int stride= s->current_picture.linesize[0]; | |
2967 | 3978 |
2138 | 3979 assert(s->current_picture.data[0]); |
3314 | 3980 assert(s->last_picture[0].data[0]); |
2967 | 3981 |
2138 | 3982 s->m.avctx= s->avctx; |
3983 s->m.current_picture.data[0]= s->current_picture.data[0]; | |
3314 | 3984 s->m. last_picture.data[0]= s->last_picture[0].data[0]; |
2138 | 3985 s->m. new_picture.data[0]= s-> input_picture.data[0]; |
3986 s->m. last_picture_ptr= &s->m. last_picture; | |
3987 s->m.linesize= | |
3988 s->m. last_picture.linesize[0]= | |
3989 s->m. new_picture.linesize[0]= | |
3990 s->m.current_picture.linesize[0]= stride; | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
3991 s->m.uvlinesize= s->current_picture.linesize[1]; |
2138 | 3992 s->m.width = width; |
3993 s->m.height= height; | |
3994 s->m.mb_width = block_width; | |
3995 s->m.mb_height= block_height; | |
3996 s->m.mb_stride= s->m.mb_width+1; | |
3997 s->m.b8_stride= 2*s->m.mb_width+1; | |
3998 s->m.f_code=1; | |
3999 s->m.pict_type= pict->pict_type; | |
4000 s->m.me_method= s->avctx->me_method; | |
4001 s->m.me.scene_change_score=0; | |
4002 s->m.flags= s->avctx->flags; | |
4003 s->m.quarter_sample= (s->avctx->flags & CODEC_FLAG_QPEL)!=0; | |
4004 s->m.out_format= FMT_H263; | |
4005 s->m.unrestricted_mv= 1; | |
4006 | |
3313 | 4007 s->m.lambda = s->lambda; |
2138 | 4008 s->m.qscale= (s->m.lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7); |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4009 s->lambda2= s->m.lambda2= (s->m.lambda*s->m.lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4010 |
2138 | 4011 s->m.dsp= s->dsp; //move |
4012 ff_init_me(&s->m); | |
2993
cb0e26759cca
iterative overlapped block based motion estimation for snow
michael
parents:
2979
diff
changeset
|
4013 s->dsp= s->m.dsp; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4014 } |
2967 | 4015 |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4016 if(s->pass1_rc){ |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4017 memcpy(rc_header_bak, s->header_state, sizeof(s->header_state)); |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4018 memcpy(rc_block_bak, s->block_state, sizeof(s->block_state)); |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4019 } |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4020 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4021 redo_frame: |
2967 | 4022 |
3313 | 4023 s->m.pict_type = pict->pict_type; |
2138 | 4024 s->qbias= pict->pict_type == P_TYPE ? 2 : 0; |
4025 | |
4026 encode_header(s); | |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4027 s->m.misc_bits = 8*(s->c.bytestream - s->c.bytestream_start); |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4028 encode_blocks(s, 1); |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4029 s->m.mv_bits = 8*(s->c.bytestream - s->c.bytestream_start) - s->m.misc_bits; |
2967 | 4030 |
2138 | 4031 for(plane_index=0; plane_index<3; plane_index++){ |
4032 Plane *p= &s->plane[plane_index]; | |
4033 int w= p->width; | |
4034 int h= p->height; | |
4035 int x, y; | |
2198 | 4036 // int bits= put_bits_count(&s->c.pb); |
2138 | 4037 |
3338
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4038 if(!(avctx->flags2 & CODEC_FLAG2_MEMC_ONLY)){ |
2138 | 4039 //FIXME optimize |
4040 if(pict->data[plane_index]) //FIXME gray hack | |
4041 for(y=0; y<h; y++){ | |
4042 for(x=0; x<w; x++){ | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4043 s->spatial_idwt_buffer[y*w + x]= pict->data[plane_index][y*pict->linesize[plane_index] + x]<<FRAC_BITS; |
2138 | 4044 } |
4045 } | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4046 predict_plane(s, s->spatial_idwt_buffer, plane_index, 0); |
2967 | 4047 |
4048 if( plane_index==0 | |
4049 && pict->pict_type == P_TYPE | |
3313 | 4050 && !(avctx->flags&CODEC_FLAG_PASS2) |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4051 && s->m.me.scene_change_score > s->avctx->scenechange_threshold){ |
2335 | 4052 ff_init_range_encoder(c, buf, buf_size); |
4053 ff_build_rac_states(c, 0.05*(1LL<<32), 256-8); | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4054 pict->pict_type= FF_I_TYPE; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4055 s->keyframe=1; |
3314 | 4056 s->current_picture.key_frame=1; |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4057 goto redo_frame; |
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4058 } |
2967 | 4059 |
2161 | 4060 if(s->qlog == LOSSLESS_QLOG){ |
4061 for(y=0; y<h; y++){ | |
4062 for(x=0; x<w; x++){ | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4063 s->spatial_dwt_buffer[y*w + x]= (s->spatial_idwt_buffer[y*w + x] + (1<<(FRAC_BITS-1))-1)>>FRAC_BITS; |
2161 | 4064 } |
4065 } | |
5575 | 4066 }else{ |
4067 for(y=0; y<h; y++){ | |
4068 for(x=0; x<w; x++){ | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4069 s->spatial_dwt_buffer[y*w + x]=s->spatial_idwt_buffer[y*w + x]<<ENCODER_EXTRA_BITS; |
5575 | 4070 } |
4071 } | |
2161 | 4072 } |
2967 | 4073 |
2164 | 4074 ff_spatial_dwt(s->spatial_dwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count); |
2161 | 4075 |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4076 if(s->pass1_rc && plane_index==0){ |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4077 int delta_qlog = ratecontrol_1pass(s, pict); |
4011
5bce97c30a69
-1 is a valid return value in ratecontrol_1pass() -> 100l for takis
michael
parents:
4001
diff
changeset
|
4078 if (delta_qlog <= INT_MIN) |
3766 | 4079 return -1; |
3661
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4080 if(delta_qlog){ |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4081 //reordering qlog in the bitstream would eliminate this reset |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4082 ff_init_range_encoder(c, buf, buf_size); |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4083 memcpy(s->header_state, rc_header_bak, sizeof(s->header_state)); |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4084 memcpy(s->block_state, rc_block_bak, sizeof(s->block_state)); |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4085 encode_header(s); |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4086 encode_blocks(s, 0); |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4087 } |
b4425339894b
fix snow 1pass rc: the qlog used and the one written to the bitstream weren't always the same.
lorenm
parents:
3556
diff
changeset
|
4088 } |
3313 | 4089 |
2138 | 4090 for(level=0; level<s->spatial_decomposition_count; level++){ |
4091 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
4092 SubBand *b= &p->band[level][orientation]; | |
2967 | 4093 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4094 quantize(s, b, b->ibuf, b->buf, b->stride, s->qbias); |
2138 | 4095 if(orientation==0) |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4096 decorrelate(s, b, b->ibuf, b->stride, pict->pict_type == P_TYPE, 0); |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4097 encode_subband(s, b, b->ibuf, b->parent ? b->parent->ibuf : NULL, b->stride, orientation); |
2138 | 4098 assert(b->parent==NULL || b->parent->stride == b->stride*2); |
4099 if(orientation==0) | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4100 correlate(s, b, b->ibuf, b->stride, 1, 0); |
2138 | 4101 } |
4102 } | |
4103 // av_log(NULL, AV_LOG_DEBUG, "plane:%d bits:%d\n", plane_index, put_bits_count(&s->c.pb) - bits); | |
4104 | |
4105 for(level=0; level<s->spatial_decomposition_count; level++){ | |
4106 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
4107 SubBand *b= &p->band[level][orientation]; | |
4108 | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4109 dequantize(s, b, b->ibuf, b->stride); |
2138 | 4110 } |
4111 } | |
2161 | 4112 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4113 ff_spatial_idwt(s->spatial_idwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count); |
2161 | 4114 if(s->qlog == LOSSLESS_QLOG){ |
4115 for(y=0; y<h; y++){ | |
4116 for(x=0; x<w; x++){ | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4117 s->spatial_idwt_buffer[y*w + x]<<=FRAC_BITS; |
2161 | 4118 } |
4119 } | |
4120 } | |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
4121 {START_TIMER |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4122 predict_plane(s, s->spatial_idwt_buffer, plane_index, 1); |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
4123 STOP_TIMER("pred-conv")} |
3338
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4124 }else{ |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4125 //ME/MC only |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4126 if(pict->pict_type == I_TYPE){ |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4127 for(y=0; y<h; y++){ |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4128 for(x=0; x<w; x++){ |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4129 s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x]= |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4130 pict->data[plane_index][y*pict->linesize[plane_index] + x]; |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4131 } |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4132 } |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4133 }else{ |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4134 memset(s->spatial_idwt_buffer, 0, sizeof(IDWTELEM)*w*h); |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4135 predict_plane(s, s->spatial_idwt_buffer, plane_index, 1); |
3338
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4136 } |
937f14bb0f23
support doing motion estimation and compensation without any residual transform or coding
michael
parents:
3327
diff
changeset
|
4137 } |
2138 | 4138 if(s->avctx->flags&CODEC_FLAG_PSNR){ |
4139 int64_t error= 0; | |
2967 | 4140 |
2138 | 4141 if(pict->data[plane_index]) //FIXME gray hack |
4142 for(y=0; y<h; y++){ | |
4143 for(x=0; x<w; x++){ | |
2161 | 4144 int d= s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x] - pict->data[plane_index][y*pict->linesize[plane_index] + x]; |
2138 | 4145 error += d*d; |
4146 } | |
4147 } | |
4148 s->avctx->error[plane_index] += error; | |
2232 | 4149 s->current_picture.error[plane_index] = error; |
2138 | 4150 } |
4151 } | |
4152 | |
5633
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
4153 if(s->last_picture[s->max_ref_frames-1].data[0]){ |
3314 | 4154 avctx->release_buffer(avctx, &s->last_picture[s->max_ref_frames-1]); |
5633
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
4155 for(i=0; i<9; i++) |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
4156 if(s->halfpel_plane[s->max_ref_frames-1][1+i/3][i%3]) |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
4157 av_free(s->halfpel_plane[s->max_ref_frames-1][1+i/3][i%3] - EDGE_WIDTH*(1+s->current_picture.linesize[i%3])); |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
4158 } |
2138 | 4159 |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4160 s->current_picture.coded_picture_number = avctx->frame_number; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4161 s->current_picture.pict_type = pict->pict_type; |
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4162 s->current_picture.quality = pict->quality; |
3313 | 4163 s->m.frame_bits = 8*(s->c.bytestream - s->c.bytestream_start); |
4164 s->m.p_tex_bits = s->m.frame_bits - s->m.misc_bits - s->m.mv_bits; | |
4165 s->m.current_picture.display_picture_number = | |
4166 s->m.current_picture.coded_picture_number = avctx->frame_number; | |
4167 s->m.current_picture.quality = pict->quality; | |
4168 s->m.total_bits += 8*(s->c.bytestream - s->c.bytestream_start); | |
4169 if(s->pass1_rc) | |
3766 | 4170 if (ff_rate_estimate_qscale(&s->m, 0) < 0) |
4171 return -1; | |
3313 | 4172 if(avctx->flags&CODEC_FLAG_PASS1) |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4173 ff_write_pass1_stats(&s->m); |
3313 | 4174 s->m.last_pict_type = s->m.pict_type; |
4123
5c86acb39889
outputs bit spent on various encoding functions (motion vectors, overhead, etc)
gpoirier
parents:
4122
diff
changeset
|
4175 avctx->frame_bits = s->m.frame_bits; |
5c86acb39889
outputs bit spent on various encoding functions (motion vectors, overhead, etc)
gpoirier
parents:
4122
diff
changeset
|
4176 avctx->mv_bits = s->m.mv_bits; |
5c86acb39889
outputs bit spent on various encoding functions (motion vectors, overhead, etc)
gpoirier
parents:
4122
diff
changeset
|
4177 avctx->misc_bits = s->m.misc_bits; |
5c86acb39889
outputs bit spent on various encoding functions (motion vectors, overhead, etc)
gpoirier
parents:
4122
diff
changeset
|
4178 avctx->p_tex_bits = s->m.p_tex_bits; |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4179 |
2138 | 4180 emms_c(); |
2967 | 4181 |
2335 | 4182 return ff_rac_terminate(c); |
2138 | 4183 } |
4184 | |
4185 static void common_end(SnowContext *s){ | |
3314 | 4186 int plane_index, level, orientation, i; |
2192 | 4187 |
2138 | 4188 av_freep(&s->spatial_dwt_buffer); |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4189 av_freep(&s->spatial_idwt_buffer); |
2138 | 4190 |
2967 | 4191 av_freep(&s->m.me.scratchpad); |
2138 | 4192 av_freep(&s->m.me.map); |
4193 av_freep(&s->m.me.score_map); | |
3033
e8599ab02b38
faster iterative_me: avoid duplicate mc of neighboring blocks.
lorenm
parents:
3020
diff
changeset
|
4194 av_freep(&s->m.obmc_scratchpad); |
2967 | 4195 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4196 av_freep(&s->block); |
2192 | 4197 |
3314 | 4198 for(i=0; i<MAX_REF_FRAMES; i++){ |
4199 av_freep(&s->ref_mvs[i]); | |
4200 av_freep(&s->ref_scores[i]); | |
4201 if(s->last_picture[i].data[0]) | |
4202 s->avctx->release_buffer(s->avctx, &s->last_picture[i]); | |
4203 } | |
4204 | |
2967 | 4205 for(plane_index=0; plane_index<3; plane_index++){ |
2192 | 4206 for(level=s->spatial_decomposition_count-1; level>=0; level--){ |
4207 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
4208 SubBand *b= &s->plane[plane_index].band[level][orientation]; | |
2967 | 4209 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4210 av_freep(&b->x_coeff); |
2192 | 4211 } |
4212 } | |
4213 } | |
2138 | 4214 } |
4215 | |
4216 static int encode_end(AVCodecContext *avctx) | |
4217 { | |
4218 SnowContext *s = avctx->priv_data; | |
4219 | |
4220 common_end(s); | |
2608
4fb7fa34050b
allow 2pass ratecontrol. also fixes psnr displayed by mencoder.
lorenm
parents:
2607
diff
changeset
|
4221 av_free(avctx->stats_out); |
2138 | 4222 |
4223 return 0; | |
4224 } | |
4225 | |
4226 static int decode_init(AVCodecContext *avctx) | |
4227 { | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4228 SnowContext *s = avctx->priv_data; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4229 int block_size; |
2967 | 4230 |
2635 | 4231 avctx->pix_fmt= PIX_FMT_YUV420P; |
2138 | 4232 |
4233 common_init(avctx); | |
2967 | 4234 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4235 block_size = MB_SIZE >> s->block_max_depth; |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4236 slice_buffer_init(&s->sb, s->plane[0].height, (block_size) + (s->spatial_decomposition_count * (s->spatial_decomposition_count + 3)) + 1, s->plane[0].width, s->spatial_idwt_buffer); |
2967 | 4237 |
2138 | 4238 return 0; |
4239 } | |
4240 | |
4241 static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, uint8_t *buf, int buf_size){ | |
4242 SnowContext *s = avctx->priv_data; | |
2335 | 4243 RangeCoder * const c= &s->c; |
2138 | 4244 int bytes_read; |
4245 AVFrame *picture = data; | |
5633
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
4246 int level, orientation, plane_index, i; |
2138 | 4247 |
2335 | 4248 ff_init_range_decoder(c, buf, buf_size); |
4249 ff_build_rac_states(c, 0.05*(1LL<<32), 256-8); | |
2138 | 4250 |
4251 s->current_picture.pict_type= FF_I_TYPE; //FIXME I vs. P | |
4252 decode_header(s); | |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4253 if(!s->block) alloc_blocks(s); |
2138 | 4254 |
4255 frame_start(s); | |
4256 //keyframe flag dupliaction mess FIXME | |
4257 if(avctx->debug&FF_DEBUG_PICT_INFO) | |
4258 av_log(avctx, AV_LOG_ERROR, "keyframe:%d qlog:%d\n", s->keyframe, s->qlog); | |
2967 | 4259 |
2189
70b27300a496
quad tree based motion compensation (currently only 16x16 & 8x8 OBMC blocks, but can be extended to other block sizes easily)
michael
parents:
2187
diff
changeset
|
4260 decode_blocks(s); |
2138 | 4261 |
4262 for(plane_index=0; plane_index<3; plane_index++){ | |
4263 Plane *p= &s->plane[plane_index]; | |
4264 int w= p->width; | |
4265 int h= p->height; | |
4266 int x, y; | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4267 int decode_state[MAX_DECOMPOSITIONS][4][1]; /* Stored state info for unpack_coeffs. 1 variable per instance. */ |
2967 | 4268 |
2138 | 4269 if(s->avctx->debug&2048){ |
4270 memset(s->spatial_dwt_buffer, 0, sizeof(DWTELEM)*w*h); | |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4271 predict_plane(s, s->spatial_idwt_buffer, plane_index, 1); |
2138 | 4272 |
4273 for(y=0; y<h; y++){ | |
4274 for(x=0; x<w; x++){ | |
2249
2b1a5e1fd449
merge predict_plane() with DWTELEM->8bit conversation (21% faster)
michael
parents:
2246
diff
changeset
|
4275 int v= s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x]; |
2138 | 4276 s->mconly_picture.data[plane_index][y*s->mconly_picture.linesize[plane_index] + x]= v; |
4277 } | |
4278 } | |
4279 } | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4280 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4281 { START_TIMER |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4282 for(level=0; level<s->spatial_decomposition_count; level++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4283 for(orientation=level ? 1 : 0; orientation<4; orientation++){ |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4284 SubBand *b= &p->band[level][orientation]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4285 unpack_coeffs(s, b, b->parent, orientation); |
2138 | 4286 } |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4287 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4288 STOP_TIMER("unpack coeffs"); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4289 } |
2138 | 4290 |
2562 | 4291 {START_TIMER |
4292 const int mb_h= s->b_height << s->block_max_depth; | |
4293 const int block_size = MB_SIZE >> s->block_max_depth; | |
4294 const int block_w = plane_index ? block_size/2 : block_size; | |
4295 int mb_y; | |
4296 dwt_compose_t cs[MAX_DECOMPOSITIONS]; | |
4297 int yd=0, yq=0; | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4298 int y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4299 int end_y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4300 |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4301 ff_spatial_idwt_buffered_init(cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count); |
2562 | 4302 for(mb_y=0; mb_y<=mb_h; mb_y++){ |
2967 | 4303 |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4304 int slice_starty = block_w*mb_y; |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4305 int slice_h = block_w*(mb_y+1); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4306 if (!(s->keyframe || s->avctx->debug&512)){ |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4307 slice_starty = FFMAX(0, slice_starty - (block_w >> 1)); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4308 slice_h -= (block_w >> 1); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4309 } |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4310 |
2967 | 4311 { |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4312 START_TIMER |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4313 for(level=0; level<s->spatial_decomposition_count; level++){ |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4314 for(orientation=level ? 1 : 0; orientation<4; orientation++){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4315 SubBand *b= &p->band[level][orientation]; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4316 int start_y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4317 int end_y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4318 int our_mb_start = mb_y; |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4319 int our_mb_end = (mb_y + 1); |
3012 | 4320 const int extra= 3; |
4321 start_y = (mb_y ? ((block_w * our_mb_start) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra: 0); | |
4322 end_y = (((block_w * our_mb_end) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra); | |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4323 if (!(s->keyframe || s->avctx->debug&512)){ |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4324 start_y = FFMAX(0, start_y - (block_w >> (1+s->spatial_decomposition_count - level))); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4325 end_y = FFMAX(0, end_y - (block_w >> (1+s->spatial_decomposition_count - level))); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4326 } |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4327 start_y = FFMIN(b->height, start_y); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4328 end_y = FFMIN(b->height, end_y); |
2967 | 4329 |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4330 if (start_y != end_y){ |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4331 if (orientation == 0){ |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4332 SubBand * correlate_band = &p->band[0][0]; |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4333 int correlate_end_y = FFMIN(b->height, end_y + 1); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4334 int correlate_start_y = FFMIN(b->height, (start_y ? start_y + 1 : 0)); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4335 decode_subband_slice_buffered(s, correlate_band, &s->sb, correlate_start_y, correlate_end_y, decode_state[0][0]); |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4336 correlate_slice_buffered(s, &s->sb, correlate_band, correlate_band->ibuf, correlate_band->stride, 1, 0, correlate_start_y, correlate_end_y); |
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4337 dequantize_slice_buffered(s, &s->sb, correlate_band, correlate_band->ibuf, correlate_band->stride, start_y, end_y); |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4338 } |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4339 else |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4340 decode_subband_slice_buffered(s, b, &s->sb, start_y, end_y, decode_state[level][orientation]); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4341 } |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4342 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4343 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4344 STOP_TIMER("decode_subband_slice"); |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4345 } |
2967 | 4346 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4347 { START_TIMER |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4348 for(; yd<slice_h; yd+=4){ |
3198
6b9f0c4fbdbe
First part of a series of speed-enchancing patches.
gpoirier
parents:
3197
diff
changeset
|
4349 ff_spatial_idwt_buffered_slice(&s->dsp, cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count, yd); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4350 } |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4351 STOP_TIMER("idwt slice");} |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4352 |
2967 | 4353 |
2161 | 4354 if(s->qlog == LOSSLESS_QLOG){ |
2562 | 4355 for(; yq<slice_h && yq<h; yq++){ |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4356 IDWTELEM * line = slice_buffer_get_line(&s->sb, yq); |
2161 | 4357 for(x=0; x<w; x++){ |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4358 line[x] <<= FRAC_BITS; |
2161 | 4359 } |
4360 } | |
4361 } | |
2562 | 4362 |
5587
3ae03eacbe9f
use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
michael
parents:
5575
diff
changeset
|
4363 predict_slice_buffered(s, &s->sb, s->spatial_idwt_buffer, plane_index, 1, mb_y); |
2967 | 4364 |
2634
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4365 y = FFMIN(p->height, slice_starty); |
337217ecbb3e
tighter snow slicing patch by (Yartrebo || yartrebo earthlink net)
michael
parents:
2628
diff
changeset
|
4366 end_y = FFMIN(p->height, slice_h); |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4367 while(y < end_y) |
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4368 slice_buffer_release(&s->sb, y++); |
2562 | 4369 } |
2967 | 4370 |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4371 slice_buffer_flush(&s->sb); |
2967 | 4372 |
2562 | 4373 STOP_TIMER("idwt + predict_slices")} |
2138 | 4374 } |
2967 | 4375 |
2138 | 4376 emms_c(); |
4377 | |
5633
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
4378 if(s->last_picture[s->max_ref_frames-1].data[0]){ |
3314 | 4379 avctx->release_buffer(avctx, &s->last_picture[s->max_ref_frames-1]); |
5633
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
4380 for(i=0; i<9; i++) |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
4381 if(s->halfpel_plane[s->max_ref_frames-1][1+i/3][i%3]) |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
4382 av_free(s->halfpel_plane[s->max_ref_frames-1][1+i/3][i%3] - EDGE_WIDTH*(1+s->current_picture.linesize[i%3])); |
873ea64637d9
code to do halfpel interpolation per frame (unfinished and under ifdef but it
michael
parents:
5627
diff
changeset
|
4383 } |
2138 | 4384 |
2967 | 4385 if(!(s->avctx->debug&2048)) |
2138 | 4386 *picture= s->current_picture; |
4387 else | |
4388 *picture= s->mconly_picture; | |
2967 | 4389 |
2138 | 4390 *data_size = sizeof(AVFrame); |
2967 | 4391 |
2335 | 4392 bytes_read= c->bytestream - c->bytestream_start; |
4393 if(bytes_read ==0) av_log(s->avctx, AV_LOG_ERROR, "error at end of frame\n"); //FIXME | |
2138 | 4394 |
4395 return bytes_read; | |
4396 } | |
4397 | |
4398 static int decode_end(AVCodecContext *avctx) | |
4399 { | |
4400 SnowContext *s = avctx->priv_data; | |
4401 | |
2589
a5a62827f195
Snow Slicing patch by (Yartrebo) yartrebo earthlink net
michael
parents:
2562
diff
changeset
|
4402 slice_buffer_destroy(&s->sb); |
2967 | 4403 |
2138 | 4404 common_end(s); |
4405 | |
4406 return 0; | |
4407 } | |
4408 | |
4409 AVCodec snow_decoder = { | |
4410 "snow", | |
4411 CODEC_TYPE_VIDEO, | |
4412 CODEC_ID_SNOW, | |
4413 sizeof(SnowContext), | |
4414 decode_init, | |
4415 NULL, | |
4416 decode_end, | |
4417 decode_frame, | |
4418 0 /*CODEC_CAP_DR1*/ /*| CODEC_CAP_DRAW_HORIZ_BAND*/, | |
4419 NULL | |
4420 }; | |
4421 | |
5224 | 4422 #ifdef CONFIG_SNOW_ENCODER |
2138 | 4423 AVCodec snow_encoder = { |
4424 "snow", | |
4425 CODEC_TYPE_VIDEO, | |
4426 CODEC_ID_SNOW, | |
4427 sizeof(SnowContext), | |
4428 encode_init, | |
4429 encode_frame, | |
4430 encode_end, | |
4431 }; | |
2408
a6e4da1c28ee
Disable encoders patch by (Gianluigi Tiesi <mplayer netfarm it>)
michael
parents:
2368
diff
changeset
|
4432 #endif |
2138 | 4433 |
4434 | |
4435 #if 0 | |
4436 #undef malloc | |
4437 #undef free | |
4438 #undef printf | |
5380
389366aa3458
Fix the self tests which are contained in some codecs and are using random().
takis
parents:
5254
diff
changeset
|
4439 #undef random |
2138 | 4440 |
4441 int main(){ | |
4442 int width=256; | |
4443 int height=256; | |
4444 int buffer[2][width*height]; | |
4445 SnowContext s; | |
4446 int i; | |
4447 s.spatial_decomposition_count=6; | |
4448 s.spatial_decomposition_type=1; | |
2967 | 4449 |
2138 | 4450 printf("testing 5/3 DWT\n"); |
4451 for(i=0; i<width*height; i++) | |
4452 buffer[0][i]= buffer[1][i]= random()%54321 - 12345; | |
2967 | 4453 |
2951 | 4454 ff_spatial_dwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); |
4455 ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); | |
2967 | 4456 |
2138 | 4457 for(i=0; i<width*height; i++) |
4458 if(buffer[0][i]!= buffer[1][i]) printf("fsck: %d %d %d\n",i, buffer[0][i], buffer[1][i]); | |
4459 | |
4460 printf("testing 9/7 DWT\n"); | |
4461 s.spatial_decomposition_type=0; | |
4462 for(i=0; i<width*height; i++) | |
4463 buffer[0][i]= buffer[1][i]= random()%54321 - 12345; | |
2967 | 4464 |
2951 | 4465 ff_spatial_dwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); |
4466 ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); | |
2967 | 4467 |
2138 | 4468 for(i=0; i<width*height; i++) |
4001 | 4469 if(FFABS(buffer[0][i] - buffer[1][i])>20) printf("fsck: %d %d %d\n",i, buffer[0][i], buffer[1][i]); |
2967 | 4470 |
2951 | 4471 #if 0 |
2138 | 4472 printf("testing AC coder\n"); |
4473 memset(s.header_state, 0, sizeof(s.header_state)); | |
2335 | 4474 ff_init_range_encoder(&s.c, buffer[0], 256*256); |
2138 | 4475 ff_init_cabac_states(&s.c, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64); |
2967 | 4476 |
2138 | 4477 for(i=-256; i<256; i++){ |
4478 START_TIMER | |
4001 | 4479 put_symbol(&s.c, s.header_state, i*i*i/3*FFABS(i), 1); |
2138 | 4480 STOP_TIMER("put_symbol") |
4481 } | |
2335 | 4482 ff_rac_terminate(&s.c); |
2138 | 4483 |
4484 memset(s.header_state, 0, sizeof(s.header_state)); | |
2335 | 4485 ff_init_range_decoder(&s.c, buffer[0], 256*256); |
2138 | 4486 ff_init_cabac_states(&s.c, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64); |
2967 | 4487 |
2138 | 4488 for(i=-256; i<256; i++){ |
4489 int j; | |
4490 START_TIMER | |
4491 j= get_symbol(&s.c, s.header_state, 1); | |
4492 STOP_TIMER("get_symbol") | |
4001 | 4493 if(j!=i*i*i/3*FFABS(i)) printf("fsck: %d != %d\n", i, j); |
2138 | 4494 } |
2951 | 4495 #endif |
2138 | 4496 { |
4497 int level, orientation, x, y; | |
4498 int64_t errors[8][4]; | |
4499 int64_t g=0; | |
4500 | |
4501 memset(errors, 0, sizeof(errors)); | |
4502 s.spatial_decomposition_count=3; | |
4503 s.spatial_decomposition_type=0; | |
4504 for(level=0; level<s.spatial_decomposition_count; level++){ | |
4505 for(orientation=level ? 1 : 0; orientation<4; orientation++){ | |
4506 int w= width >> (s.spatial_decomposition_count-level); | |
4507 int h= height >> (s.spatial_decomposition_count-level); | |
4508 int stride= width << (s.spatial_decomposition_count-level); | |
4509 DWTELEM *buf= buffer[0]; | |
4510 int64_t error=0; | |
4511 | |
4512 if(orientation&1) buf+=w; | |
4513 if(orientation>1) buf+=stride>>1; | |
2967 | 4514 |
2138 | 4515 memset(buffer[0], 0, sizeof(int)*width*height); |
4516 buf[w/2 + h/2*stride]= 256*256; | |
2951 | 4517 ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); |
2138 | 4518 for(y=0; y<height; y++){ |
4519 for(x=0; x<width; x++){ | |
4520 int64_t d= buffer[0][x + y*width]; | |
4521 error += d*d; | |
4122
daae66c03857
Replace most of the %lld and %llx by their (cleaner) PRI*64 counterparts.
diego
parents:
4011
diff
changeset
|
4522 if(FFABS(width/2-x)<9 && FFABS(height/2-y)<9 && level==2) printf("%8"PRId64" ", d); |
2138 | 4523 } |
4001 | 4524 if(FFABS(height/2-y)<9 && level==2) printf("\n"); |
2138 | 4525 } |
4526 error= (int)(sqrt(error)+0.5); | |
4527 errors[level][orientation]= error; | |
4528 if(g) g=ff_gcd(g, error); | |
4529 else g= error; | |
4530 } | |
4531 } | |
4532 printf("static int const visual_weight[][4]={\n"); | |
4533 for(level=0; level<s.spatial_decomposition_count; level++){ | |
4534 printf(" {"); | |
4535 for(orientation=0; orientation<4; orientation++){ | |
4122
daae66c03857
Replace most of the %lld and %llx by their (cleaner) PRI*64 counterparts.
diego
parents:
4011
diff
changeset
|
4536 printf("%8"PRId64",", errors[level][orientation]/g); |
2138 | 4537 } |
4538 printf("},\n"); | |
4539 } | |
4540 printf("};\n"); | |
4541 { | |
4542 int level=2; | |
4543 int orientation=3; | |
4544 int w= width >> (s.spatial_decomposition_count-level); | |
4545 int h= height >> (s.spatial_decomposition_count-level); | |
4546 int stride= width << (s.spatial_decomposition_count-level); | |
4547 DWTELEM *buf= buffer[0]; | |
4548 int64_t error=0; | |
4549 | |
4550 buf+=w; | |
4551 buf+=stride>>1; | |
2967 | 4552 |
2138 | 4553 memset(buffer[0], 0, sizeof(int)*width*height); |
4554 #if 1 | |
4555 for(y=0; y<height; y++){ | |
4556 for(x=0; x<width; x++){ | |
4557 int tab[4]={0,2,3,1}; | |
4558 buffer[0][x+width*y]= 256*256*tab[(x&1) + 2*(y&1)]; | |
4559 } | |
4560 } | |
2951 | 4561 ff_spatial_dwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); |
2138 | 4562 #else |
4563 for(y=0; y<h; y++){ | |
4564 for(x=0; x<w; x++){ | |
4565 buf[x + y*stride ]=169; | |
4566 buf[x + y*stride-w]=64; | |
4567 } | |
4568 } | |
2951 | 4569 ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); |
2138 | 4570 #endif |
4571 for(y=0; y<height; y++){ | |
4572 for(x=0; x<width; x++){ | |
4573 int64_t d= buffer[0][x + y*width]; | |
4574 error += d*d; | |
4122
daae66c03857
Replace most of the %lld and %llx by their (cleaner) PRI*64 counterparts.
diego
parents:
4011
diff
changeset
|
4575 if(FFABS(width/2-x)<9 && FFABS(height/2-y)<9) printf("%8"PRId64" ", d); |
2138 | 4576 } |
4001 | 4577 if(FFABS(height/2-y)<9) printf("\n"); |
2138 | 4578 } |
4579 } | |
4580 | |
4581 } | |
4582 return 0; | |
4583 } | |
4584 #endif | |
4585 |