1 /*
2 * Copyright (c) 2020, Alliance for Open Media. All rights reserved
3 *
4 * This source code is subject to the terms of the BSD 2 Clause License and
5 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6 * was not distributed with this source code in the LICENSE file, you can
7 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8 * Media Patent License 1.0 was not distributed with this source code in the
9 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10 */
11
12 #include "aom/aomcx.h"
13
14 #include "av1/encoder/bitstream.h"
15 #include "av1/encoder/encodeframe.h"
16 #include "av1/encoder/encoder.h"
17 #include "av1/encoder/encoder_alloc.h"
18 #include "av1/encoder/encodetxb.h"
19 #include "av1/encoder/encoder_utils.h"
20 #include "av1/encoder/grain_test_vectors.h"
21 #include "av1/encoder/mv_prec.h"
22 #include "av1/encoder/rc_utils.h"
23 #include "av1/encoder/rdopt.h"
24 #include "av1/encoder/segmentation.h"
25 #include "av1/encoder/superres_scale.h"
26 #include "av1/encoder/tpl_model.h"
27 #include "av1/encoder/var_based_part.h"
28
29 #if CONFIG_TUNE_VMAF
30 #include "av1/encoder/tune_vmaf.h"
31 #endif
32
33 #define MIN_BOOST_COMBINE_FACTOR 4.0
34 #define MAX_BOOST_COMBINE_FACTOR 12.0
35
36 const int default_tx_type_probs[FRAME_UPDATE_TYPES][TX_SIZES_ALL][TX_TYPES] = {
37 { { 221, 189, 214, 292, 0, 0, 0, 0, 0, 2, 38, 68, 0, 0, 0, 0 },
38 { 262, 203, 216, 239, 0, 0, 0, 0, 0, 1, 37, 66, 0, 0, 0, 0 },
39 { 315, 231, 239, 226, 0, 0, 0, 0, 0, 13, 0, 0, 0, 0, 0, 0 },
40 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
41 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
42 { 222, 188, 214, 287, 0, 0, 0, 0, 0, 2, 50, 61, 0, 0, 0, 0 },
43 { 256, 182, 205, 282, 0, 0, 0, 0, 0, 2, 21, 76, 0, 0, 0, 0 },
44 { 281, 214, 217, 222, 0, 0, 0, 0, 0, 1, 48, 41, 0, 0, 0, 0 },
45 { 263, 194, 225, 225, 0, 0, 0, 0, 0, 2, 15, 100, 0, 0, 0, 0 },
46 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
47 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
48 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
49 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
50 { 170, 192, 242, 293, 0, 0, 0, 0, 0, 1, 68, 58, 0, 0, 0, 0 },
51 { 199, 210, 213, 291, 0, 0, 0, 0, 0, 1, 14, 96, 0, 0, 0, 0 },
52 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
53 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
54 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
55 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 } },
56 { { 106, 69, 107, 278, 9, 15, 20, 45, 49, 23, 23, 88, 36, 74, 25, 57 },
57 { 105, 72, 81, 98, 45, 49, 47, 50, 56, 72, 30, 81, 33, 95, 27, 83 },
58 { 211, 105, 109, 120, 57, 62, 43, 49, 52, 58, 42, 116, 0, 0, 0, 0 },
59 { 1008, 0, 0, 0, 0, 0, 0, 0, 0, 16, 0, 0, 0, 0, 0, 0 },
60 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
61 { 131, 57, 98, 172, 19, 40, 37, 64, 69, 22, 41, 52, 51, 77, 35, 59 },
62 { 176, 83, 93, 202, 22, 24, 28, 47, 50, 16, 12, 93, 26, 76, 17, 59 },
63 { 136, 72, 89, 95, 46, 59, 47, 56, 61, 68, 35, 51, 32, 82, 26, 69 },
64 { 122, 80, 87, 105, 49, 47, 46, 46, 57, 52, 13, 90, 19, 103, 15, 93 },
65 { 1009, 0, 0, 0, 0, 0, 0, 0, 0, 15, 0, 0, 0, 0, 0, 0 },
66 { 1011, 0, 0, 0, 0, 0, 0, 0, 0, 13, 0, 0, 0, 0, 0, 0 },
67 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
68 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
69 { 202, 20, 84, 114, 14, 60, 41, 79, 99, 21, 41, 15, 50, 84, 34, 66 },
70 { 196, 44, 23, 72, 30, 22, 28, 57, 67, 13, 4, 165, 15, 148, 9, 131 },
71 { 882, 0, 0, 0, 0, 0, 0, 0, 0, 142, 0, 0, 0, 0, 0, 0 },
72 { 840, 0, 0, 0, 0, 0, 0, 0, 0, 184, 0, 0, 0, 0, 0, 0 },
73 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
74 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 } },
75 { { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
76 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
77 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
78 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
79 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
80 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
81 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
82 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
83 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
84 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
85 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
86 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
87 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
88 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
89 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
90 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
91 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
92 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
93 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 } },
94 { { 213, 110, 141, 269, 12, 16, 15, 19, 21, 11, 38, 68, 22, 29, 16, 24 },
95 { 216, 119, 128, 143, 38, 41, 26, 30, 31, 30, 42, 70, 23, 36, 19, 32 },
96 { 367, 149, 154, 154, 38, 35, 17, 21, 21, 10, 22, 36, 0, 0, 0, 0 },
97 { 1022, 0, 0, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0 },
98 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
99 { 219, 96, 127, 191, 21, 40, 25, 32, 34, 18, 45, 45, 33, 39, 26, 33 },
100 { 296, 99, 122, 198, 23, 21, 19, 24, 25, 13, 20, 64, 23, 32, 18, 27 },
101 { 275, 128, 142, 143, 35, 48, 23, 30, 29, 18, 42, 36, 18, 23, 14, 20 },
102 { 239, 132, 166, 175, 36, 27, 19, 21, 24, 14, 13, 85, 9, 31, 8, 25 },
103 { 1022, 0, 0, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0 },
104 { 1022, 0, 0, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0 },
105 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
106 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
107 { 309, 25, 79, 59, 25, 80, 34, 53, 61, 25, 49, 23, 43, 64, 36, 59 },
108 { 270, 57, 40, 54, 50, 42, 41, 53, 56, 28, 17, 81, 45, 86, 34, 70 },
109 { 1005, 0, 0, 0, 0, 0, 0, 0, 0, 19, 0, 0, 0, 0, 0, 0 },
110 { 992, 0, 0, 0, 0, 0, 0, 0, 0, 32, 0, 0, 0, 0, 0, 0 },
111 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
112 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 } },
113 { { 133, 63, 55, 83, 57, 87, 58, 72, 68, 16, 24, 35, 29, 105, 25, 114 },
114 { 131, 75, 74, 60, 71, 77, 65, 66, 73, 33, 21, 79, 20, 83, 18, 78 },
115 { 276, 95, 82, 58, 86, 93, 63, 60, 64, 17, 38, 92, 0, 0, 0, 0 },
116 { 1006, 0, 0, 0, 0, 0, 0, 0, 0, 18, 0, 0, 0, 0, 0, 0 },
117 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
118 { 147, 49, 75, 78, 50, 97, 60, 67, 76, 17, 42, 35, 31, 93, 27, 80 },
119 { 157, 49, 58, 75, 61, 52, 56, 67, 69, 12, 15, 79, 24, 119, 11, 120 },
120 { 178, 69, 83, 77, 69, 85, 72, 77, 77, 20, 35, 40, 25, 48, 23, 46 },
121 { 174, 55, 64, 57, 73, 68, 62, 61, 75, 15, 12, 90, 17, 99, 16, 86 },
122 { 1008, 0, 0, 0, 0, 0, 0, 0, 0, 16, 0, 0, 0, 0, 0, 0 },
123 { 1018, 0, 0, 0, 0, 0, 0, 0, 0, 6, 0, 0, 0, 0, 0, 0 },
124 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
125 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
126 { 266, 31, 63, 64, 21, 52, 39, 54, 63, 30, 52, 31, 48, 89, 46, 75 },
127 { 272, 26, 32, 44, 29, 31, 32, 53, 51, 13, 13, 88, 22, 153, 16, 149 },
128 { 923, 0, 0, 0, 0, 0, 0, 0, 0, 101, 0, 0, 0, 0, 0, 0 },
129 { 969, 0, 0, 0, 0, 0, 0, 0, 0, 55, 0, 0, 0, 0, 0, 0 },
130 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
131 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 } },
132 { { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
133 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
134 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
135 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
136 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
137 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
138 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
139 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
140 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
141 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
142 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
143 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
144 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
145 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
146 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
147 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
148 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
149 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 },
150 { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 } },
151 { { 158, 92, 125, 298, 12, 15, 20, 29, 31, 12, 29, 67, 34, 44, 23, 35 },
152 { 147, 94, 103, 123, 45, 48, 38, 41, 46, 48, 37, 78, 33, 63, 27, 53 },
153 { 268, 126, 125, 136, 54, 53, 31, 38, 38, 33, 35, 87, 0, 0, 0, 0 },
154 { 1018, 0, 0, 0, 0, 0, 0, 0, 0, 6, 0, 0, 0, 0, 0, 0 },
155 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
156 { 159, 72, 103, 194, 20, 35, 37, 50, 56, 21, 39, 40, 51, 61, 38, 48 },
157 { 259, 86, 95, 188, 32, 20, 25, 34, 37, 13, 12, 85, 25, 53, 17, 43 },
158 { 189, 99, 113, 123, 45, 59, 37, 46, 48, 44, 39, 41, 31, 47, 26, 37 },
159 { 175, 110, 113, 128, 58, 38, 33, 33, 43, 29, 13, 100, 14, 68, 12, 57 },
160 { 1017, 0, 0, 0, 0, 0, 0, 0, 0, 7, 0, 0, 0, 0, 0, 0 },
161 { 1019, 0, 0, 0, 0, 0, 0, 0, 0, 5, 0, 0, 0, 0, 0, 0 },
162 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
163 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
164 { 208, 22, 84, 101, 21, 59, 44, 70, 90, 25, 59, 13, 64, 67, 49, 48 },
165 { 277, 52, 32, 63, 43, 26, 33, 48, 54, 11, 6, 130, 18, 119, 11, 101 },
166 { 963, 0, 0, 0, 0, 0, 0, 0, 0, 61, 0, 0, 0, 0, 0, 0 },
167 { 979, 0, 0, 0, 0, 0, 0, 0, 0, 45, 0, 0, 0, 0, 0, 0 },
168 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
169 { 1024, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 } }
170 };
171
172 const int default_obmc_probs[FRAME_UPDATE_TYPES][BLOCK_SIZES_ALL] = {
173 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
174 { 0, 0, 0, 106, 90, 90, 97, 67, 59, 70, 28,
175 30, 38, 16, 16, 16, 0, 0, 44, 50, 26, 25 },
176 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
177 { 0, 0, 0, 98, 93, 97, 68, 82, 85, 33, 30,
178 33, 16, 16, 16, 16, 0, 0, 43, 37, 26, 16 },
179 { 0, 0, 0, 91, 80, 76, 78, 55, 49, 24, 16,
180 16, 16, 16, 16, 16, 0, 0, 29, 45, 16, 38 },
181 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
182 { 0, 0, 0, 103, 89, 89, 89, 62, 63, 76, 34,
183 35, 32, 19, 16, 16, 0, 0, 49, 55, 29, 19 }
184 };
185
186 const int default_warped_probs[FRAME_UPDATE_TYPES] = { 64, 64, 64, 64,
187 64, 64, 64 };
188
189 // TODO(yunqing): the default probs can be trained later from better
190 // performance.
191 const int default_switchable_interp_probs[FRAME_UPDATE_TYPES]
192 [SWITCHABLE_FILTER_CONTEXTS]
193 [SWITCHABLE_FILTERS] = {
194 { { 512, 512, 512 },
195 { 512, 512, 512 },
196 { 512, 512, 512 },
197 { 512, 512, 512 },
198 { 512, 512, 512 },
199 { 512, 512, 512 },
200 { 512, 512, 512 },
201 { 512, 512, 512 },
202 { 512, 512, 512 },
203 { 512, 512, 512 },
204 { 512, 512, 512 },
205 { 512, 512, 512 },
206 { 512, 512, 512 },
207 { 512, 512, 512 },
208 { 512, 512, 512 },
209 { 512, 512, 512 } },
210 { { 512, 512, 512 },
211 { 512, 512, 512 },
212 { 512, 512, 512 },
213 { 512, 512, 512 },
214 { 512, 512, 512 },
215 { 512, 512, 512 },
216 { 512, 512, 512 },
217 { 512, 512, 512 },
218 { 512, 512, 512 },
219 { 512, 512, 512 },
220 { 512, 512, 512 },
221 { 512, 512, 512 },
222 { 512, 512, 512 },
223 { 512, 512, 512 },
224 { 512, 512, 512 },
225 { 512, 512, 512 } },
226 { { 512, 512, 512 },
227 { 512, 512, 512 },
228 { 512, 512, 512 },
229 { 512, 512, 512 },
230 { 512, 512, 512 },
231 { 512, 512, 512 },
232 { 512, 512, 512 },
233 { 512, 512, 512 },
234 { 512, 512, 512 },
235 { 512, 512, 512 },
236 { 512, 512, 512 },
237 { 512, 512, 512 },
238 { 512, 512, 512 },
239 { 512, 512, 512 },
240 { 512, 512, 512 },
241 { 512, 512, 512 } },
242 { { 512, 512, 512 },
243 { 512, 512, 512 },
244 { 512, 512, 512 },
245 { 512, 512, 512 },
246 { 512, 512, 512 },
247 { 512, 512, 512 },
248 { 512, 512, 512 },
249 { 512, 512, 512 },
250 { 512, 512, 512 },
251 { 512, 512, 512 },
252 { 512, 512, 512 },
253 { 512, 512, 512 },
254 { 512, 512, 512 },
255 { 512, 512, 512 },
256 { 512, 512, 512 },
257 { 512, 512, 512 } },
258 { { 512, 512, 512 },
259 { 512, 512, 512 },
260 { 512, 512, 512 },
261 { 512, 512, 512 },
262 { 512, 512, 512 },
263 { 512, 512, 512 },
264 { 512, 512, 512 },
265 { 512, 512, 512 },
266 { 512, 512, 512 },
267 { 512, 512, 512 },
268 { 512, 512, 512 },
269 { 512, 512, 512 },
270 { 512, 512, 512 },
271 { 512, 512, 512 },
272 { 512, 512, 512 },
273 { 512, 512, 512 } },
274 { { 512, 512, 512 },
275 { 512, 512, 512 },
276 { 512, 512, 512 },
277 { 512, 512, 512 },
278 { 512, 512, 512 },
279 { 512, 512, 512 },
280 { 512, 512, 512 },
281 { 512, 512, 512 },
282 { 512, 512, 512 },
283 { 512, 512, 512 },
284 { 512, 512, 512 },
285 { 512, 512, 512 },
286 { 512, 512, 512 },
287 { 512, 512, 512 },
288 { 512, 512, 512 },
289 { 512, 512, 512 } },
290 { { 512, 512, 512 },
291 { 512, 512, 512 },
292 { 512, 512, 512 },
293 { 512, 512, 512 },
294 { 512, 512, 512 },
295 { 512, 512, 512 },
296 { 512, 512, 512 },
297 { 512, 512, 512 },
298 { 512, 512, 512 },
299 { 512, 512, 512 },
300 { 512, 512, 512 },
301 { 512, 512, 512 },
302 { 512, 512, 512 },
303 { 512, 512, 512 },
304 { 512, 512, 512 },
305 { 512, 512, 512 } }
306 };
307
configure_static_seg_features(AV1_COMP * cpi)308 static void configure_static_seg_features(AV1_COMP *cpi) {
309 AV1_COMMON *const cm = &cpi->common;
310 const RATE_CONTROL *const rc = &cpi->rc;
311 struct segmentation *const seg = &cm->seg;
312
313 double avg_q;
314 #if CONFIG_FPMT_TEST
315 avg_q = ((cpi->ppi->gf_group.frame_parallel_level[cpi->gf_frame_index] > 0) &&
316 (cpi->ppi->fpmt_unit_test_cfg == PARALLEL_SIMULATION_ENCODE))
317 ? cpi->ppi->p_rc.temp_avg_q
318 : cpi->ppi->p_rc.avg_q;
319 #else
320 avg_q = cpi->ppi->p_rc.avg_q;
321 #endif
322
323 int high_q = (int)(avg_q > 48.0);
324 int qi_delta;
325
326 // Disable and clear down for KF
327 if (cm->current_frame.frame_type == KEY_FRAME) {
328 // Clear down the global segmentation map
329 memset(cpi->enc_seg.map, 0, cm->mi_params.mi_rows * cm->mi_params.mi_cols);
330 seg->update_map = 0;
331 seg->update_data = 0;
332
333 // Disable segmentation
334 av1_disable_segmentation(seg);
335
336 // Clear down the segment features.
337 av1_clearall_segfeatures(seg);
338 } else if (cpi->refresh_frame.alt_ref_frame) {
339 // If this is an alt ref frame
340 // Clear down the global segmentation map
341 memset(cpi->enc_seg.map, 0, cm->mi_params.mi_rows * cm->mi_params.mi_cols);
342 seg->update_map = 0;
343 seg->update_data = 0;
344
345 // Disable segmentation and individual segment features by default
346 av1_disable_segmentation(seg);
347 av1_clearall_segfeatures(seg);
348
349 // If segmentation was enabled set those features needed for the
350 // arf itself.
351 if (seg->enabled) {
352 seg->update_map = 1;
353 seg->update_data = 1;
354
355 qi_delta = av1_compute_qdelta(rc, avg_q, avg_q * 0.875,
356 cm->seq_params->bit_depth);
357 av1_set_segdata(seg, 1, SEG_LVL_ALT_Q, qi_delta - 2);
358 av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_Y_H, -2);
359 av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_Y_V, -2);
360 av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_U, -2);
361 av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_V, -2);
362
363 av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_Y_H);
364 av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_Y_V);
365 av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_U);
366 av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_V);
367
368 av1_enable_segfeature(seg, 1, SEG_LVL_ALT_Q);
369 }
370 } else if (seg->enabled) {
371 // All other frames if segmentation has been enabled
372
373 // First normal frame in a valid gf or alt ref group
374 if (rc->frames_since_golden == 0) {
375 // Set up segment features for normal frames in an arf group
376 // Disable segmentation and clear down features if alt ref
377 // is not active for this group
378
379 av1_disable_segmentation(seg);
380
381 memset(cpi->enc_seg.map, 0,
382 cm->mi_params.mi_rows * cm->mi_params.mi_cols);
383
384 seg->update_map = 0;
385 seg->update_data = 0;
386
387 av1_clearall_segfeatures(seg);
388 } else if (rc->is_src_frame_alt_ref) {
389 // Special case where we are coding over the top of a previous
390 // alt ref frame.
391 // Segment coding disabled for compred testing
392
393 // Enable ref frame features for segment 0 as well
394 av1_enable_segfeature(seg, 0, SEG_LVL_REF_FRAME);
395 av1_enable_segfeature(seg, 1, SEG_LVL_REF_FRAME);
396
397 // All mbs should use ALTREF_FRAME
398 av1_clear_segdata(seg, 0, SEG_LVL_REF_FRAME);
399 av1_set_segdata(seg, 0, SEG_LVL_REF_FRAME, ALTREF_FRAME);
400 av1_clear_segdata(seg, 1, SEG_LVL_REF_FRAME);
401 av1_set_segdata(seg, 1, SEG_LVL_REF_FRAME, ALTREF_FRAME);
402
403 // Skip all MBs if high Q (0,0 mv and skip coeffs)
404 if (high_q) {
405 av1_enable_segfeature(seg, 0, SEG_LVL_SKIP);
406 av1_enable_segfeature(seg, 1, SEG_LVL_SKIP);
407 }
408 // Enable data update
409 seg->update_data = 1;
410 } else {
411 // All other frames.
412
413 // No updates.. leave things as they are.
414 seg->update_map = 0;
415 seg->update_data = 0;
416 }
417 }
418 }
419
av1_apply_active_map(AV1_COMP * cpi)420 void av1_apply_active_map(AV1_COMP *cpi) {
421 struct segmentation *const seg = &cpi->common.seg;
422 unsigned char *const seg_map = cpi->enc_seg.map;
423 const unsigned char *const active_map = cpi->active_map.map;
424
425 assert(AM_SEGMENT_ID_ACTIVE == CR_SEGMENT_ID_BASE);
426
427 // Disable the active_maps on intra_only frames or if the
428 // input map for the current frame has no inactive blocks.
429 if (frame_is_intra_only(&cpi->common) ||
430 cpi->rc.percent_blocks_inactive == 0) {
431 cpi->active_map.enabled = 0;
432 cpi->active_map.update = 1;
433 }
434
435 if (cpi->active_map.update) {
436 if (cpi->active_map.enabled) {
437 const int num_mis =
438 cpi->common.mi_params.mi_rows * cpi->common.mi_params.mi_cols;
439 memcpy(seg_map, active_map, sizeof(active_map[0]) * num_mis);
440 av1_enable_segmentation(seg);
441 av1_enable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_SKIP);
442 av1_enable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_Y_H);
443 av1_enable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_Y_V);
444 av1_enable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_U);
445 av1_enable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_V);
446
447 av1_set_segdata(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_Y_H,
448 -MAX_LOOP_FILTER);
449 av1_set_segdata(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_Y_V,
450 -MAX_LOOP_FILTER);
451 av1_set_segdata(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_U,
452 -MAX_LOOP_FILTER);
453 av1_set_segdata(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_V,
454 -MAX_LOOP_FILTER);
455 } else {
456 av1_disable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_SKIP);
457 av1_disable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_Y_H);
458 av1_disable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_Y_V);
459 av1_disable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_U);
460 av1_disable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_V);
461 if (seg->enabled) {
462 seg->update_data = 1;
463 seg->update_map = 1;
464 }
465 }
466 cpi->active_map.update = 0;
467 }
468 }
469
470 #if !CONFIG_REALTIME_ONLY
process_tpl_stats_frame(AV1_COMP * cpi)471 static void process_tpl_stats_frame(AV1_COMP *cpi) {
472 const GF_GROUP *const gf_group = &cpi->ppi->gf_group;
473 AV1_COMMON *const cm = &cpi->common;
474
475 assert(IMPLIES(gf_group->size > 0, cpi->gf_frame_index < gf_group->size));
476
477 const int tpl_idx = cpi->gf_frame_index;
478 TplParams *const tpl_data = &cpi->ppi->tpl_data;
479 TplDepFrame *tpl_frame = &tpl_data->tpl_frame[tpl_idx];
480 TplDepStats *tpl_stats = tpl_frame->tpl_stats_ptr;
481
482 if (tpl_frame->is_valid) {
483 int tpl_stride = tpl_frame->stride;
484 double intra_cost_base = 0;
485 double mc_dep_cost_base = 0;
486 double cbcmp_base = 1;
487 const int step = 1 << tpl_data->tpl_stats_block_mis_log2;
488 const int row_step = step;
489 const int col_step_sr =
490 coded_to_superres_mi(step, cm->superres_scale_denominator);
491 const int mi_cols_sr = av1_pixels_to_mi(cm->superres_upscaled_width);
492
493 for (int row = 0; row < cm->mi_params.mi_rows; row += row_step) {
494 for (int col = 0; col < mi_cols_sr; col += col_step_sr) {
495 TplDepStats *this_stats = &tpl_stats[av1_tpl_ptr_pos(
496 row, col, tpl_stride, tpl_data->tpl_stats_block_mis_log2)];
497 double cbcmp = (double)(this_stats->srcrf_dist);
498 int64_t mc_dep_delta =
499 RDCOST(tpl_frame->base_rdmult, this_stats->mc_dep_rate,
500 this_stats->mc_dep_dist);
501 double dist_scaled = (double)(this_stats->recrf_dist << RDDIV_BITS);
502 intra_cost_base += log(dist_scaled) * cbcmp;
503 mc_dep_cost_base += log(dist_scaled + mc_dep_delta) * cbcmp;
504 cbcmp_base += cbcmp;
505 }
506 }
507
508 if (mc_dep_cost_base == 0) {
509 tpl_frame->is_valid = 0;
510 } else {
511 cpi->rd.r0 = exp((intra_cost_base - mc_dep_cost_base) / cbcmp_base);
512 if (is_frame_tpl_eligible(gf_group, cpi->gf_frame_index)) {
513 if (cpi->ppi->lap_enabled) {
514 double min_boost_factor = sqrt(cpi->ppi->p_rc.baseline_gf_interval);
515 const int gfu_boost = get_gfu_boost_from_r0_lap(
516 min_boost_factor, MAX_GFUBOOST_FACTOR, cpi->rd.r0,
517 cpi->ppi->p_rc.num_stats_required_for_gfu_boost);
518 // printf("old boost %d new boost %d\n", cpi->rc.gfu_boost,
519 // gfu_boost);
520 cpi->ppi->p_rc.gfu_boost = combine_prior_with_tpl_boost(
521 min_boost_factor, MAX_BOOST_COMBINE_FACTOR,
522 cpi->ppi->p_rc.gfu_boost, gfu_boost,
523 cpi->ppi->p_rc.num_stats_used_for_gfu_boost);
524 } else {
525 // TPL may only look at a subset of frame in the gf group when the
526 // speed feature 'reduce_num_frames' is on, which affects the r0
527 // calcuation. Thus, to compensate for TPL not using all frames a
528 // factor to adjust r0 is used.
529 const int gfu_boost =
530 (int)(200.0 * cpi->ppi->tpl_data.r0_adjust_factor / cpi->rd.r0);
531 cpi->ppi->p_rc.gfu_boost = combine_prior_with_tpl_boost(
532 MIN_BOOST_COMBINE_FACTOR, MAX_BOOST_COMBINE_FACTOR,
533 cpi->ppi->p_rc.gfu_boost, gfu_boost, cpi->rc.frames_to_key);
534 }
535 }
536 }
537 }
538 }
539 #endif // !CONFIG_REALTIME_ONLY
540
av1_set_size_dependent_vars(AV1_COMP * cpi,int * q,int * bottom_index,int * top_index)541 void av1_set_size_dependent_vars(AV1_COMP *cpi, int *q, int *bottom_index,
542 int *top_index) {
543 AV1_COMMON *const cm = &cpi->common;
544
545 // Setup variables that depend on the dimensions of the frame.
546 av1_set_speed_features_framesize_dependent(cpi, cpi->speed);
547
548 #if !CONFIG_REALTIME_ONLY
549 GF_GROUP *gf_group = &cpi->ppi->gf_group;
550 if (cpi->oxcf.algo_cfg.enable_tpl_model &&
551 av1_tpl_stats_ready(&cpi->ppi->tpl_data, cpi->gf_frame_index)) {
552 process_tpl_stats_frame(cpi);
553 av1_tpl_rdmult_setup(cpi);
554 }
555 #endif
556
557 // Decide q and q bounds.
558 *q = av1_rc_pick_q_and_bounds(cpi, cm->width, cm->height, cpi->gf_frame_index,
559 bottom_index, top_index);
560
561 #if !CONFIG_REALTIME_ONLY
562 if (cpi->oxcf.rc_cfg.mode == AOM_Q &&
563 cpi->ppi->tpl_data.tpl_frame[cpi->gf_frame_index].is_valid &&
564 !is_lossless_requested(&cpi->oxcf.rc_cfg)) {
565 const RateControlCfg *const rc_cfg = &cpi->oxcf.rc_cfg;
566 const int tpl_q = av1_tpl_get_q_index(
567 &cpi->ppi->tpl_data, cpi->gf_frame_index, cpi->rc.active_worst_quality,
568 cm->seq_params->bit_depth);
569 *q = clamp(tpl_q, rc_cfg->best_allowed_q, rc_cfg->worst_allowed_q);
570 *top_index = *bottom_index = *q;
571 if (gf_group->update_type[cpi->gf_frame_index] == ARF_UPDATE)
572 cpi->ppi->p_rc.arf_q = *q;
573 }
574
575 if (cpi->oxcf.q_cfg.use_fixed_qp_offsets && cpi->oxcf.rc_cfg.mode == AOM_Q) {
576 if (is_frame_tpl_eligible(gf_group, cpi->gf_frame_index)) {
577 const double qratio_grad =
578 cpi->ppi->p_rc.baseline_gf_interval > 20 ? 0.2 : 0.3;
579 const double qstep_ratio =
580 0.2 +
581 (1.0 - (double)cpi->rc.active_worst_quality / MAXQ) * qratio_grad;
582 *q = av1_get_q_index_from_qstep_ratio(
583 cpi->rc.active_worst_quality, qstep_ratio, cm->seq_params->bit_depth);
584 *top_index = *bottom_index = *q;
585 if (gf_group->update_type[cpi->gf_frame_index] == ARF_UPDATE ||
586 gf_group->update_type[cpi->gf_frame_index] == KF_UPDATE ||
587 gf_group->update_type[cpi->gf_frame_index] == GF_UPDATE)
588 cpi->ppi->p_rc.arf_q = *q;
589 } else if (gf_group->layer_depth[cpi->gf_frame_index] <
590 gf_group->max_layer_depth) {
591 int this_height = gf_group->layer_depth[cpi->gf_frame_index];
592 int arf_q = cpi->ppi->p_rc.arf_q;
593 while (this_height > 1) {
594 arf_q = (arf_q + cpi->oxcf.rc_cfg.cq_level + 1) / 2;
595 --this_height;
596 }
597 *top_index = *bottom_index = *q = arf_q;
598 }
599 }
600 #endif
601
602 // Configure experimental use of segmentation for enhanced coding of
603 // static regions if indicated.
604 // Only allowed in the second pass of a two pass encode, as it requires
605 // lagged coding, and if the relevant speed feature flag is set.
606 if (is_stat_consumption_stage_twopass(cpi) &&
607 cpi->sf.hl_sf.static_segmentation)
608 configure_static_seg_features(cpi);
609 }
610
reset_film_grain_chroma_params(aom_film_grain_t * pars)611 static void reset_film_grain_chroma_params(aom_film_grain_t *pars) {
612 pars->num_cr_points = 0;
613 pars->cr_mult = 0;
614 pars->cr_luma_mult = 0;
615 memset(pars->scaling_points_cr, 0, sizeof(pars->scaling_points_cr));
616 memset(pars->ar_coeffs_cr, 0, sizeof(pars->ar_coeffs_cr));
617 pars->num_cb_points = 0;
618 pars->cb_mult = 0;
619 pars->cb_luma_mult = 0;
620 pars->chroma_scaling_from_luma = 0;
621 memset(pars->scaling_points_cb, 0, sizeof(pars->scaling_points_cb));
622 memset(pars->ar_coeffs_cb, 0, sizeof(pars->ar_coeffs_cb));
623 }
624
av1_update_film_grain_parameters_seq(struct AV1_PRIMARY * ppi,const AV1EncoderConfig * oxcf)625 void av1_update_film_grain_parameters_seq(struct AV1_PRIMARY *ppi,
626 const AV1EncoderConfig *oxcf) {
627 SequenceHeader *const seq_params = &ppi->seq_params;
628 const TuneCfg *const tune_cfg = &oxcf->tune_cfg;
629
630 if (tune_cfg->film_grain_test_vector || tune_cfg->film_grain_table_filename ||
631 tune_cfg->content == AOM_CONTENT_FILM) {
632 seq_params->film_grain_params_present = 1;
633 } else {
634 #if CONFIG_DENOISE
635 seq_params->film_grain_params_present = (oxcf->noise_level > 0);
636 #else
637 seq_params->film_grain_params_present = 0;
638 #endif
639 }
640 }
641
av1_update_film_grain_parameters(struct AV1_COMP * cpi,const AV1EncoderConfig * oxcf)642 void av1_update_film_grain_parameters(struct AV1_COMP *cpi,
643 const AV1EncoderConfig *oxcf) {
644 AV1_COMMON *const cm = &cpi->common;
645 const TuneCfg *const tune_cfg = &oxcf->tune_cfg;
646
647 if (cpi->film_grain_table) {
648 aom_film_grain_table_free(cpi->film_grain_table);
649 aom_free(cpi->film_grain_table);
650 cpi->film_grain_table = NULL;
651 }
652
653 if (tune_cfg->film_grain_test_vector) {
654 if (cm->current_frame.frame_type == KEY_FRAME) {
655 memcpy(&cm->film_grain_params,
656 film_grain_test_vectors + tune_cfg->film_grain_test_vector - 1,
657 sizeof(cm->film_grain_params));
658 if (oxcf->tool_cfg.enable_monochrome)
659 reset_film_grain_chroma_params(&cm->film_grain_params);
660 cm->film_grain_params.bit_depth = cm->seq_params->bit_depth;
661 if (cm->seq_params->color_range == AOM_CR_FULL_RANGE) {
662 cm->film_grain_params.clip_to_restricted_range = 0;
663 }
664 }
665 } else if (tune_cfg->film_grain_table_filename) {
666 CHECK_MEM_ERROR(cm, cpi->film_grain_table,
667 aom_calloc(1, sizeof(*cpi->film_grain_table)));
668
669 aom_film_grain_table_read(cpi->film_grain_table,
670 tune_cfg->film_grain_table_filename, cm->error);
671 } else if (tune_cfg->content == AOM_CONTENT_FILM) {
672 cm->film_grain_params.bit_depth = cm->seq_params->bit_depth;
673 if (oxcf->tool_cfg.enable_monochrome)
674 reset_film_grain_chroma_params(&cm->film_grain_params);
675 if (cm->seq_params->color_range == AOM_CR_FULL_RANGE)
676 cm->film_grain_params.clip_to_restricted_range = 0;
677 } else {
678 memset(&cm->film_grain_params, 0, sizeof(cm->film_grain_params));
679 }
680 }
681
av1_scale_references(AV1_COMP * cpi,const InterpFilter filter,const int phase,const int use_optimized_scaler)682 void av1_scale_references(AV1_COMP *cpi, const InterpFilter filter,
683 const int phase, const int use_optimized_scaler) {
684 AV1_COMMON *cm = &cpi->common;
685 const int num_planes = av1_num_planes(cm);
686 MV_REFERENCE_FRAME ref_frame;
687
688 for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
689 // Need to convert from AOM_REFFRAME to index into ref_mask (subtract 1).
690 if (cpi->ref_frame_flags & av1_ref_frame_flag_list[ref_frame]) {
691 BufferPool *const pool = cm->buffer_pool;
692 const YV12_BUFFER_CONFIG *const ref =
693 get_ref_frame_yv12_buf(cm, ref_frame);
694
695 if (ref == NULL) {
696 cpi->scaled_ref_buf[ref_frame - 1] = NULL;
697 continue;
698 }
699
700 // For RTC-SVC: if force_zero_mode_spatial_ref is enabled, check if the
701 // motion search can be skipped for the references: last, golden, altref.
702 // If so, we can skip scaling that reference.
703 if (cpi->ppi->use_svc && cpi->svc.force_zero_mode_spatial_ref &&
704 cpi->ppi->rtc_ref.set_ref_frame_config) {
705 if (ref_frame == LAST_FRAME && cpi->svc.skip_mvsearch_last) continue;
706 if (ref_frame == GOLDEN_FRAME && cpi->svc.skip_mvsearch_gf) continue;
707 if (ref_frame == ALTREF_FRAME && cpi->svc.skip_mvsearch_altref)
708 continue;
709 }
710 // For RTC with superres on: golden reference only needs to be scaled
711 // if it was refreshed in previous frame.
712 if (is_one_pass_rt_params(cpi) &&
713 cpi->oxcf.superres_cfg.enable_superres && ref_frame == GOLDEN_FRAME &&
714 cpi->rc.frame_num_last_gf_refresh <
715 (int)cm->current_frame.frame_number - 1) {
716 continue;
717 }
718
719 if (ref->y_crop_width != cm->width || ref->y_crop_height != cm->height) {
720 // Replace the reference buffer with a copy having a thicker border,
721 // if the reference buffer is higher resolution than the current
722 // frame, and the border is thin.
723 if ((ref->y_crop_width > cm->width ||
724 ref->y_crop_height > cm->height) &&
725 ref->border < AOM_BORDER_IN_PIXELS) {
726 RefCntBuffer *ref_fb = get_ref_frame_buf(cm, ref_frame);
727 if (aom_yv12_realloc_with_new_border(
728 &ref_fb->buf, AOM_BORDER_IN_PIXELS,
729 cm->features.byte_alignment, cpi->alloc_pyramid,
730 num_planes) != 0) {
731 aom_internal_error(cm->error, AOM_CODEC_MEM_ERROR,
732 "Failed to allocate frame buffer");
733 }
734 }
735 int force_scaling = 0;
736 RefCntBuffer *new_fb = cpi->scaled_ref_buf[ref_frame - 1];
737 if (new_fb == NULL) {
738 const int new_fb_idx = get_free_fb(cm);
739 if (new_fb_idx == INVALID_IDX) {
740 aom_internal_error(cm->error, AOM_CODEC_MEM_ERROR,
741 "Unable to find free frame buffer");
742 }
743 force_scaling = 1;
744 new_fb = &pool->frame_bufs[new_fb_idx];
745 }
746
747 if (force_scaling || new_fb->buf.y_crop_width != cm->width ||
748 new_fb->buf.y_crop_height != cm->height) {
749 if (aom_realloc_frame_buffer(
750 &new_fb->buf, cm->width, cm->height,
751 cm->seq_params->subsampling_x, cm->seq_params->subsampling_y,
752 cm->seq_params->use_highbitdepth, AOM_BORDER_IN_PIXELS,
753 cm->features.byte_alignment, NULL, NULL, NULL, false, 0)) {
754 if (force_scaling) {
755 // Release the reference acquired in the get_free_fb() call above.
756 --new_fb->ref_count;
757 }
758 aom_internal_error(cm->error, AOM_CODEC_MEM_ERROR,
759 "Failed to allocate frame buffer");
760 }
761 bool has_optimized_scaler = av1_has_optimized_scaler(
762 ref->y_crop_width, ref->y_crop_height, new_fb->buf.y_crop_width,
763 new_fb->buf.y_crop_height);
764 if (num_planes > 1) {
765 has_optimized_scaler =
766 has_optimized_scaler &&
767 av1_has_optimized_scaler(
768 ref->uv_crop_width, ref->uv_crop_height,
769 new_fb->buf.uv_crop_width, new_fb->buf.uv_crop_height);
770 }
771 #if CONFIG_AV1_HIGHBITDEPTH
772 if (use_optimized_scaler && has_optimized_scaler &&
773 cm->seq_params->bit_depth == AOM_BITS_8) {
774 av1_resize_and_extend_frame(ref, &new_fb->buf, filter, phase,
775 num_planes);
776 } else if (!av1_resize_and_extend_frame_nonnormative(
777 ref, &new_fb->buf, (int)cm->seq_params->bit_depth,
778 num_planes)) {
779 aom_internal_error(cm->error, AOM_CODEC_MEM_ERROR,
780 "Failed to allocate buffer during resize");
781 }
782 #else
783 if (use_optimized_scaler && has_optimized_scaler) {
784 av1_resize_and_extend_frame(ref, &new_fb->buf, filter, phase,
785 num_planes);
786 } else if (!av1_resize_and_extend_frame_nonnormative(
787 ref, &new_fb->buf, (int)cm->seq_params->bit_depth,
788 num_planes)) {
789 aom_internal_error(cm->error, AOM_CODEC_MEM_ERROR,
790 "Failed to allocate buffer during resize");
791 }
792 #endif
793 cpi->scaled_ref_buf[ref_frame - 1] = new_fb;
794 alloc_frame_mvs(cm, new_fb);
795 }
796 } else {
797 RefCntBuffer *buf = get_ref_frame_buf(cm, ref_frame);
798 buf->buf.y_crop_width = ref->y_crop_width;
799 buf->buf.y_crop_height = ref->y_crop_height;
800 cpi->scaled_ref_buf[ref_frame - 1] = buf;
801 ++buf->ref_count;
802 }
803 } else {
804 if (!has_no_stats_stage(cpi)) cpi->scaled_ref_buf[ref_frame - 1] = NULL;
805 }
806 }
807 }
808
av1_select_sb_size(const AV1EncoderConfig * const oxcf,int width,int height,int number_spatial_layers)809 BLOCK_SIZE av1_select_sb_size(const AV1EncoderConfig *const oxcf, int width,
810 int height, int number_spatial_layers) {
811 if (oxcf->tool_cfg.superblock_size == AOM_SUPERBLOCK_SIZE_64X64) {
812 return BLOCK_64X64;
813 }
814 if (oxcf->tool_cfg.superblock_size == AOM_SUPERBLOCK_SIZE_128X128) {
815 return BLOCK_128X128;
816 }
817 #if CONFIG_TFLITE
818 if (oxcf->q_cfg.deltaq_mode == DELTA_Q_USER_RATING_BASED) return BLOCK_64X64;
819 #endif
820 // Force 64x64 superblock size to increase resolution in perceptual
821 // AQ mode.
822 if (oxcf->mode == ALLINTRA &&
823 (oxcf->q_cfg.deltaq_mode == DELTA_Q_PERCEPTUAL_AI ||
824 oxcf->q_cfg.deltaq_mode == DELTA_Q_USER_RATING_BASED)) {
825 return BLOCK_64X64;
826 }
827 assert(oxcf->tool_cfg.superblock_size == AOM_SUPERBLOCK_SIZE_DYNAMIC);
828
829 if (number_spatial_layers > 1 ||
830 oxcf->resize_cfg.resize_mode != RESIZE_NONE) {
831 // Use the configured size (top resolution) for spatial layers or
832 // on resize.
833 return AOMMIN(oxcf->frm_dim_cfg.width, oxcf->frm_dim_cfg.height) > 720
834 ? BLOCK_128X128
835 : BLOCK_64X64;
836 } else if (oxcf->mode == REALTIME) {
837 if (oxcf->tune_cfg.content == AOM_CONTENT_SCREEN) {
838 const TileConfig *const tile_cfg = &oxcf->tile_cfg;
839 const int num_tiles =
840 (1 << tile_cfg->tile_columns) * (1 << tile_cfg->tile_rows);
841 // For multi-thread encode: if the number of (128x128) superblocks
842 // per tile is low use 64X64 superblock.
843 if (oxcf->row_mt == 1 && oxcf->max_threads >= 4 &&
844 oxcf->max_threads >= num_tiles && AOMMIN(width, height) > 720 &&
845 (width * height) / (128 * 128 * num_tiles) <= 38)
846 return BLOCK_64X64;
847 else
848 return AOMMIN(width, height) >= 720 ? BLOCK_128X128 : BLOCK_64X64;
849 } else {
850 return AOMMIN(width, height) > 720 ? BLOCK_128X128 : BLOCK_64X64;
851 }
852 }
853
854 // TODO(any): Possibly could improve this with a heuristic.
855 // When superres / resize is on, 'cm->width / height' can change between
856 // calls, so we don't apply this heuristic there.
857 // Things break if superblock size changes between the first pass and second
858 // pass encoding, which is why this heuristic is not configured as a
859 // speed-feature.
860 if (oxcf->superres_cfg.superres_mode == AOM_SUPERRES_NONE &&
861 oxcf->resize_cfg.resize_mode == RESIZE_NONE) {
862 int is_480p_or_lesser = AOMMIN(width, height) <= 480;
863 if (oxcf->speed >= 1 && is_480p_or_lesser) return BLOCK_64X64;
864
865 // For 1080p and lower resolutions, choose SB size adaptively based on
866 // resolution and speed level for multi-thread encode.
867 int is_1080p_or_lesser = AOMMIN(width, height) <= 1080;
868 if (!is_480p_or_lesser && is_1080p_or_lesser && oxcf->mode == GOOD &&
869 oxcf->row_mt == 1 && oxcf->max_threads > 1 && oxcf->speed >= 5)
870 return BLOCK_64X64;
871
872 // For allintra encode, since the maximum partition size is set to 32X32 for
873 // speed>=6, superblock size is set to 64X64 instead of 128X128. This
874 // improves the multithread performance due to reduction in top right delay
875 // and thread sync wastage. Currently, this setting is selectively enabled
876 // only for speed>=9 and resolutions less than 4k since cost update
877 // frequency is set to INTERNAL_COST_UPD_OFF in these cases.
878 const int is_4k_or_larger = AOMMIN(width, height) >= 2160;
879 if (oxcf->mode == ALLINTRA && oxcf->speed >= 9 && !is_4k_or_larger)
880 return BLOCK_64X64;
881 }
882 return BLOCK_128X128;
883 }
884
av1_setup_frame(AV1_COMP * cpi)885 void av1_setup_frame(AV1_COMP *cpi) {
886 AV1_COMMON *const cm = &cpi->common;
887 // Set up entropy context depending on frame type. The decoder mandates
888 // the use of the default context, index 0, for keyframes and inter
889 // frames where the error_resilient_mode or intra_only flag is set. For
890 // other inter-frames the encoder currently uses only two contexts;
891 // context 1 for ALTREF frames and context 0 for the others.
892
893 if (frame_is_intra_only(cm) || cm->features.error_resilient_mode ||
894 cpi->ext_flags.use_primary_ref_none) {
895 av1_setup_past_independence(cm);
896 }
897
898 if ((cm->current_frame.frame_type == KEY_FRAME && cm->show_frame) ||
899 frame_is_sframe(cm)) {
900 if (!cpi->ppi->seq_params_locked) {
901 set_sb_size(cm->seq_params,
902 av1_select_sb_size(&cpi->oxcf, cm->width, cm->height,
903 cpi->ppi->number_spatial_layers));
904 }
905 } else {
906 const RefCntBuffer *const primary_ref_buf = get_primary_ref_frame_buf(cm);
907 if (primary_ref_buf == NULL) {
908 av1_setup_past_independence(cm);
909 cm->seg.update_map = 1;
910 cm->seg.update_data = 1;
911 } else {
912 *cm->fc = primary_ref_buf->frame_context;
913 }
914 }
915
916 av1_zero(cm->cur_frame->interp_filter_selected);
917 cm->prev_frame = get_primary_ref_frame_buf(cm);
918 cpi->vaq_refresh = 0;
919 }
920
921 #if !CONFIG_REALTIME_ONLY
get_interp_filter_selected(const AV1_COMMON * const cm,MV_REFERENCE_FRAME ref,InterpFilter ifilter)922 static int get_interp_filter_selected(const AV1_COMMON *const cm,
923 MV_REFERENCE_FRAME ref,
924 InterpFilter ifilter) {
925 const RefCntBuffer *const buf = get_ref_frame_buf(cm, ref);
926 if (buf == NULL) return 0;
927 return buf->interp_filter_selected[ifilter];
928 }
929
av1_setup_interp_filter_search_mask(AV1_COMP * cpi)930 uint16_t av1_setup_interp_filter_search_mask(AV1_COMP *cpi) {
931 const AV1_COMMON *const cm = &cpi->common;
932 int ref_total[REF_FRAMES] = { 0 };
933 uint16_t mask = ALLOW_ALL_INTERP_FILT_MASK;
934
935 if (cpi->last_frame_type == KEY_FRAME || cpi->refresh_frame.alt_ref_frame)
936 return mask;
937
938 for (MV_REFERENCE_FRAME ref = LAST_FRAME; ref <= ALTREF_FRAME; ++ref) {
939 for (InterpFilter ifilter = EIGHTTAP_REGULAR; ifilter <= MULTITAP_SHARP;
940 ++ifilter) {
941 ref_total[ref] += get_interp_filter_selected(cm, ref, ifilter);
942 }
943 }
944 int ref_total_total = (ref_total[LAST2_FRAME] + ref_total[LAST3_FRAME] +
945 ref_total[GOLDEN_FRAME] + ref_total[BWDREF_FRAME] +
946 ref_total[ALTREF2_FRAME] + ref_total[ALTREF_FRAME]);
947
948 for (InterpFilter ifilter = EIGHTTAP_REGULAR; ifilter <= MULTITAP_SHARP;
949 ++ifilter) {
950 int last_score = get_interp_filter_selected(cm, LAST_FRAME, ifilter) * 30;
951 if (ref_total[LAST_FRAME] && last_score <= ref_total[LAST_FRAME]) {
952 int filter_score =
953 get_interp_filter_selected(cm, LAST2_FRAME, ifilter) * 20 +
954 get_interp_filter_selected(cm, LAST3_FRAME, ifilter) * 20 +
955 get_interp_filter_selected(cm, GOLDEN_FRAME, ifilter) * 20 +
956 get_interp_filter_selected(cm, BWDREF_FRAME, ifilter) * 10 +
957 get_interp_filter_selected(cm, ALTREF2_FRAME, ifilter) * 10 +
958 get_interp_filter_selected(cm, ALTREF_FRAME, ifilter) * 10;
959 if (filter_score < ref_total_total) {
960 DUAL_FILTER_TYPE filt_type = ifilter + SWITCHABLE_FILTERS * ifilter;
961 reset_interp_filter_allowed_mask(&mask, filt_type);
962 }
963 }
964 }
965 return mask;
966 }
967
968 #define STRICT_PSNR_DIFF_THRESH 0.9
969 // Encode key frame with/without screen content tools to determine whether
970 // screen content tools should be enabled for this key frame group or not.
971 // The first encoding is without screen content tools.
972 // The second encoding is with screen content tools.
973 // We compare the psnr and frame size to make the decision.
screen_content_tools_determination(AV1_COMP * cpi,const int allow_screen_content_tools_orig_decision,const int allow_intrabc_orig_decision,const int use_screen_content_tools_orig_decision,const int is_screen_content_type_orig_decision,const int pass,int * projected_size_pass,PSNR_STATS * psnr)974 static void screen_content_tools_determination(
975 AV1_COMP *cpi, const int allow_screen_content_tools_orig_decision,
976 const int allow_intrabc_orig_decision,
977 const int use_screen_content_tools_orig_decision,
978 const int is_screen_content_type_orig_decision, const int pass,
979 int *projected_size_pass, PSNR_STATS *psnr) {
980 AV1_COMMON *const cm = &cpi->common;
981 FeatureFlags *const features = &cm->features;
982
983 #if CONFIG_FPMT_TEST
984 projected_size_pass[pass] =
985 ((cpi->ppi->gf_group.frame_parallel_level[cpi->gf_frame_index] > 0) &&
986 (cpi->ppi->fpmt_unit_test_cfg == PARALLEL_SIMULATION_ENCODE))
987 ? cpi->ppi->p_rc.temp_projected_frame_size
988 : cpi->rc.projected_frame_size;
989 #else
990 projected_size_pass[pass] = cpi->rc.projected_frame_size;
991 #endif
992
993 #if CONFIG_AV1_HIGHBITDEPTH
994 const uint32_t in_bit_depth = cpi->oxcf.input_cfg.input_bit_depth;
995 const uint32_t bit_depth = cpi->td.mb.e_mbd.bd;
996 aom_calc_highbd_psnr(cpi->source, &cpi->common.cur_frame->buf, &psnr[pass],
997 bit_depth, in_bit_depth);
998 #else
999 aom_calc_psnr(cpi->source, &cpi->common.cur_frame->buf, &psnr[pass]);
1000 #endif
1001 if (pass != 1) return;
1002
1003 const double psnr_diff = psnr[1].psnr[0] - psnr[0].psnr[0];
1004 // Calculate % of palette mode to be chosen in a frame from mode decision.
1005 const double palette_ratio =
1006 (double)cpi->palette_pixel_num / (double)(cm->height * cm->width);
1007 const int psnr_diff_is_large = (psnr_diff > STRICT_PSNR_DIFF_THRESH);
1008 const int ratio_is_large =
1009 ((palette_ratio >= 0.0001) && ((psnr_diff / palette_ratio) > 4));
1010 const int is_sc_encoding_much_better = (psnr_diff_is_large || ratio_is_large);
1011 if (is_sc_encoding_much_better) {
1012 // Use screen content tools, if we get coding gain.
1013 features->allow_screen_content_tools = 1;
1014 features->allow_intrabc = cpi->intrabc_used;
1015 cpi->use_screen_content_tools = 1;
1016 cpi->is_screen_content_type = 1;
1017 } else {
1018 // Use original screen content decision.
1019 features->allow_screen_content_tools =
1020 allow_screen_content_tools_orig_decision;
1021 features->allow_intrabc = allow_intrabc_orig_decision;
1022 cpi->use_screen_content_tools = use_screen_content_tools_orig_decision;
1023 cpi->is_screen_content_type = is_screen_content_type_orig_decision;
1024 }
1025 }
1026
1027 // Set some encoding parameters to make the encoding process fast.
1028 // A fixed block partition size, and a large q is used.
set_encoding_params_for_screen_content(AV1_COMP * cpi,const int pass)1029 static void set_encoding_params_for_screen_content(AV1_COMP *cpi,
1030 const int pass) {
1031 AV1_COMMON *const cm = &cpi->common;
1032 if (pass == 0) {
1033 // In the first pass, encode without screen content tools.
1034 // Use a high q, and a fixed block size for fast encoding.
1035 cm->features.allow_screen_content_tools = 0;
1036 cm->features.allow_intrabc = 0;
1037 cpi->use_screen_content_tools = 0;
1038 cpi->sf.part_sf.partition_search_type = FIXED_PARTITION;
1039 cpi->sf.part_sf.fixed_partition_size = BLOCK_32X32;
1040 return;
1041 }
1042 assert(pass == 1);
1043 // In the second pass, encode with screen content tools.
1044 // Use a high q, and a fixed block size for fast encoding.
1045 cm->features.allow_screen_content_tools = 1;
1046 // TODO(chengchen): turn intrabc on could lead to data race issue.
1047 // cm->allow_intrabc = 1;
1048 cpi->use_screen_content_tools = 1;
1049 cpi->sf.part_sf.partition_search_type = FIXED_PARTITION;
1050 cpi->sf.part_sf.fixed_partition_size = BLOCK_32X32;
1051 }
1052
1053 // Determines whether to use screen content tools for the key frame group.
1054 // This function modifies "cm->features.allow_screen_content_tools",
1055 // "cm->features.allow_intrabc" and "cpi->use_screen_content_tools".
av1_determine_sc_tools_with_encoding(AV1_COMP * cpi,const int q_orig)1056 void av1_determine_sc_tools_with_encoding(AV1_COMP *cpi, const int q_orig) {
1057 AV1_COMMON *const cm = &cpi->common;
1058 const AV1EncoderConfig *const oxcf = &cpi->oxcf;
1059 const QuantizationCfg *const q_cfg = &oxcf->q_cfg;
1060 // Variables to help determine if we should allow screen content tools.
1061 int projected_size_pass[3] = { 0 };
1062 PSNR_STATS psnr[3];
1063 const int is_key_frame = cm->current_frame.frame_type == KEY_FRAME;
1064 const int allow_screen_content_tools_orig_decision =
1065 cm->features.allow_screen_content_tools;
1066 const int allow_intrabc_orig_decision = cm->features.allow_intrabc;
1067 const int use_screen_content_tools_orig_decision =
1068 cpi->use_screen_content_tools;
1069 const int is_screen_content_type_orig_decision = cpi->is_screen_content_type;
1070 // Turn off the encoding trial for forward key frame and superres.
1071 if (cpi->sf.rt_sf.use_nonrd_pick_mode || oxcf->kf_cfg.fwd_kf_enabled ||
1072 cpi->superres_mode != AOM_SUPERRES_NONE || oxcf->mode == REALTIME ||
1073 use_screen_content_tools_orig_decision || !is_key_frame) {
1074 return;
1075 }
1076
1077 // TODO(chengchen): multiple encoding for the lossless mode is time consuming.
1078 // Find a better way to determine whether screen content tools should be used
1079 // for lossless coding.
1080 // Use a high q and a fixed partition to do quick encoding.
1081 const int q_for_screen_content_quick_run =
1082 is_lossless_requested(&oxcf->rc_cfg) ? q_orig : AOMMAX(q_orig, 244);
1083 const int partition_search_type_orig = cpi->sf.part_sf.partition_search_type;
1084 const BLOCK_SIZE fixed_partition_block_size_orig =
1085 cpi->sf.part_sf.fixed_partition_size;
1086
1087 // Setup necessary params for encoding, including frame source, etc.
1088
1089 cpi->source = av1_realloc_and_scale_if_required(
1090 cm, cpi->unscaled_source, &cpi->scaled_source, cm->features.interp_filter,
1091 0, false, false, cpi->oxcf.border_in_pixels, cpi->alloc_pyramid);
1092 if (cpi->unscaled_last_source != NULL) {
1093 cpi->last_source = av1_realloc_and_scale_if_required(
1094 cm, cpi->unscaled_last_source, &cpi->scaled_last_source,
1095 cm->features.interp_filter, 0, false, false, cpi->oxcf.border_in_pixels,
1096 cpi->alloc_pyramid);
1097 }
1098
1099 av1_setup_frame(cpi);
1100
1101 if (cm->seg.enabled) {
1102 if (!cm->seg.update_data && cm->prev_frame) {
1103 segfeatures_copy(&cm->seg, &cm->prev_frame->seg);
1104 cm->seg.enabled = cm->prev_frame->seg.enabled;
1105 } else {
1106 av1_calculate_segdata(&cm->seg);
1107 }
1108 } else {
1109 memset(&cm->seg, 0, sizeof(cm->seg));
1110 }
1111 segfeatures_copy(&cm->cur_frame->seg, &cm->seg);
1112 cm->cur_frame->seg.enabled = cm->seg.enabled;
1113
1114 // The two encoding passes aim to help determine whether to use screen
1115 // content tools, with a high q and fixed partition.
1116 for (int pass = 0; pass < 2; ++pass) {
1117 set_encoding_params_for_screen_content(cpi, pass);
1118 av1_set_quantizer(cm, q_cfg->qm_minlevel, q_cfg->qm_maxlevel,
1119 q_for_screen_content_quick_run,
1120 q_cfg->enable_chroma_deltaq, q_cfg->enable_hdr_deltaq);
1121 av1_set_speed_features_qindex_dependent(cpi, oxcf->speed);
1122 av1_init_quantizer(&cpi->enc_quant_dequant_params, &cm->quant_params,
1123 cm->seq_params->bit_depth);
1124
1125 av1_set_variance_partition_thresholds(cpi, q_for_screen_content_quick_run,
1126 0);
1127 // transform / motion compensation build reconstruction frame
1128 av1_encode_frame(cpi);
1129 // Screen content decision
1130 screen_content_tools_determination(
1131 cpi, allow_screen_content_tools_orig_decision,
1132 allow_intrabc_orig_decision, use_screen_content_tools_orig_decision,
1133 is_screen_content_type_orig_decision, pass, projected_size_pass, psnr);
1134 }
1135
1136 // Set partition speed feature back.
1137 cpi->sf.part_sf.partition_search_type = partition_search_type_orig;
1138 cpi->sf.part_sf.fixed_partition_size = fixed_partition_block_size_orig;
1139
1140 // Free token related info if screen content coding tools are not enabled.
1141 if (!cm->features.allow_screen_content_tools)
1142 free_token_info(&cpi->token_info);
1143 }
1144 #endif // CONFIG_REALTIME_ONLY
1145
fix_interp_filter(InterpFilter * const interp_filter,const FRAME_COUNTS * const counts)1146 static void fix_interp_filter(InterpFilter *const interp_filter,
1147 const FRAME_COUNTS *const counts) {
1148 if (*interp_filter == SWITCHABLE) {
1149 // Check to see if only one of the filters is actually used
1150 int count[SWITCHABLE_FILTERS] = { 0 };
1151 int num_filters_used = 0;
1152 for (int i = 0; i < SWITCHABLE_FILTERS; ++i) {
1153 for (int j = 0; j < SWITCHABLE_FILTER_CONTEXTS; ++j)
1154 count[i] += counts->switchable_interp[j][i];
1155 num_filters_used += (count[i] > 0);
1156 }
1157 if (num_filters_used == 1) {
1158 // Only one filter is used. So set the filter at frame level
1159 for (int i = 0; i < SWITCHABLE_FILTERS; ++i) {
1160 if (count[i]) {
1161 *interp_filter = i;
1162 break;
1163 }
1164 }
1165 }
1166 }
1167 }
1168
av1_finalize_encoded_frame(AV1_COMP * const cpi)1169 void av1_finalize_encoded_frame(AV1_COMP *const cpi) {
1170 AV1_COMMON *const cm = &cpi->common;
1171 CurrentFrame *const current_frame = &cm->current_frame;
1172
1173 if (!cm->seq_params->reduced_still_picture_hdr &&
1174 encode_show_existing_frame(cm)) {
1175 RefCntBuffer *const frame_to_show =
1176 cm->ref_frame_map[cpi->existing_fb_idx_to_show];
1177
1178 if (frame_to_show == NULL) {
1179 aom_internal_error(cm->error, AOM_CODEC_UNSUP_BITSTREAM,
1180 "Buffer does not contain a reconstructed frame");
1181 }
1182 assert(frame_to_show->ref_count > 0);
1183 assign_frame_buffer_p(&cm->cur_frame, frame_to_show);
1184 }
1185
1186 if (!encode_show_existing_frame(cm) &&
1187 cm->seq_params->film_grain_params_present &&
1188 (cm->show_frame || cm->showable_frame)) {
1189 // Copy the current frame's film grain params to the its corresponding
1190 // RefCntBuffer slot.
1191 cm->cur_frame->film_grain_params = cm->film_grain_params;
1192
1193 // We must update the parameters if this is not an INTER_FRAME
1194 if (current_frame->frame_type != INTER_FRAME)
1195 cm->cur_frame->film_grain_params.update_parameters = 1;
1196
1197 // Iterate the random seed for the next frame.
1198 cm->film_grain_params.random_seed += 3381;
1199 if (cm->film_grain_params.random_seed == 0)
1200 cm->film_grain_params.random_seed = 7391;
1201 }
1202
1203 // Initialise all tiles' contexts from the global frame context
1204 for (int tile_col = 0; tile_col < cm->tiles.cols; tile_col++) {
1205 for (int tile_row = 0; tile_row < cm->tiles.rows; tile_row++) {
1206 const int tile_idx = tile_row * cm->tiles.cols + tile_col;
1207 cpi->tile_data[tile_idx].tctx = *cm->fc;
1208 }
1209 }
1210
1211 if (!frame_is_intra_only(cm))
1212 fix_interp_filter(&cm->features.interp_filter, cpi->td.counts);
1213 }
1214
av1_is_integer_mv(const YV12_BUFFER_CONFIG * cur_picture,const YV12_BUFFER_CONFIG * last_picture,ForceIntegerMVInfo * const force_intpel_info)1215 int av1_is_integer_mv(const YV12_BUFFER_CONFIG *cur_picture,
1216 const YV12_BUFFER_CONFIG *last_picture,
1217 ForceIntegerMVInfo *const force_intpel_info) {
1218 // check use hash ME
1219 int k;
1220
1221 const int block_size = FORCE_INT_MV_DECISION_BLOCK_SIZE;
1222 const double threshold_current = 0.8;
1223 const double threshold_average = 0.95;
1224 const int max_history_size = 32;
1225 int T = 0; // total block
1226 int C = 0; // match with collocated block
1227 int S = 0; // smooth region but not match with collocated block
1228
1229 const int pic_width = cur_picture->y_width;
1230 const int pic_height = cur_picture->y_height;
1231 for (int i = 0; i + block_size <= pic_height; i += block_size) {
1232 for (int j = 0; j + block_size <= pic_width; j += block_size) {
1233 const int x_pos = j;
1234 const int y_pos = i;
1235 int match = 1;
1236 T++;
1237
1238 // check whether collocated block match with current
1239 uint8_t *p_cur = cur_picture->y_buffer;
1240 uint8_t *p_ref = last_picture->y_buffer;
1241 int stride_cur = cur_picture->y_stride;
1242 int stride_ref = last_picture->y_stride;
1243 p_cur += (y_pos * stride_cur + x_pos);
1244 p_ref += (y_pos * stride_ref + x_pos);
1245
1246 if (cur_picture->flags & YV12_FLAG_HIGHBITDEPTH) {
1247 uint16_t *p16_cur = CONVERT_TO_SHORTPTR(p_cur);
1248 uint16_t *p16_ref = CONVERT_TO_SHORTPTR(p_ref);
1249 for (int tmpY = 0; tmpY < block_size && match; tmpY++) {
1250 for (int tmpX = 0; tmpX < block_size && match; tmpX++) {
1251 if (p16_cur[tmpX] != p16_ref[tmpX]) {
1252 match = 0;
1253 }
1254 }
1255 p16_cur += stride_cur;
1256 p16_ref += stride_ref;
1257 }
1258 } else {
1259 for (int tmpY = 0; tmpY < block_size && match; tmpY++) {
1260 for (int tmpX = 0; tmpX < block_size && match; tmpX++) {
1261 if (p_cur[tmpX] != p_ref[tmpX]) {
1262 match = 0;
1263 }
1264 }
1265 p_cur += stride_cur;
1266 p_ref += stride_ref;
1267 }
1268 }
1269
1270 if (match) {
1271 C++;
1272 continue;
1273 }
1274
1275 if (av1_hash_is_horizontal_perfect(cur_picture, block_size, x_pos,
1276 y_pos) ||
1277 av1_hash_is_vertical_perfect(cur_picture, block_size, x_pos, y_pos)) {
1278 S++;
1279 continue;
1280 }
1281 }
1282 }
1283
1284 assert(T > 0);
1285 double cs_rate = ((double)(C + S)) / ((double)(T));
1286
1287 force_intpel_info->cs_rate_array[force_intpel_info->rate_index] = cs_rate;
1288
1289 force_intpel_info->rate_index =
1290 (force_intpel_info->rate_index + 1) % max_history_size;
1291 force_intpel_info->rate_size++;
1292 force_intpel_info->rate_size =
1293 AOMMIN(force_intpel_info->rate_size, max_history_size);
1294
1295 if (cs_rate < threshold_current) {
1296 return 0;
1297 }
1298
1299 if (C == T) {
1300 return 1;
1301 }
1302
1303 double cs_average = 0.0;
1304
1305 for (k = 0; k < force_intpel_info->rate_size; k++) {
1306 cs_average += force_intpel_info->cs_rate_array[k];
1307 }
1308 cs_average /= force_intpel_info->rate_size;
1309
1310 if (cs_average < threshold_average) {
1311 return 0;
1312 }
1313
1314 if ((T - C - S) < 0) {
1315 return 1;
1316 }
1317
1318 if (cs_average > 1.01) {
1319 return 1;
1320 }
1321
1322 return 0;
1323 }
1324
av1_set_mb_ssim_rdmult_scaling(AV1_COMP * cpi)1325 void av1_set_mb_ssim_rdmult_scaling(AV1_COMP *cpi) {
1326 const CommonModeInfoParams *const mi_params = &cpi->common.mi_params;
1327 const MACROBLOCKD *const xd = &cpi->td.mb.e_mbd;
1328 uint8_t *y_buffer = cpi->source->y_buffer;
1329 const int y_stride = cpi->source->y_stride;
1330 const int block_size = BLOCK_16X16;
1331
1332 const int num_mi_w = mi_size_wide[block_size];
1333 const int num_mi_h = mi_size_high[block_size];
1334 const int num_cols = (mi_params->mi_cols + num_mi_w - 1) / num_mi_w;
1335 const int num_rows = (mi_params->mi_rows + num_mi_h - 1) / num_mi_h;
1336 double log_sum = 0.0;
1337
1338 // Loop through each 16x16 block.
1339 for (int row = 0; row < num_rows; ++row) {
1340 for (int col = 0; col < num_cols; ++col) {
1341 double var = 0.0, num_of_var = 0.0;
1342 const int index = row * num_cols + col;
1343
1344 // Loop through each 8x8 block.
1345 for (int mi_row = row * num_mi_h;
1346 mi_row < mi_params->mi_rows && mi_row < (row + 1) * num_mi_h;
1347 mi_row += 2) {
1348 for (int mi_col = col * num_mi_w;
1349 mi_col < mi_params->mi_cols && mi_col < (col + 1) * num_mi_w;
1350 mi_col += 2) {
1351 struct buf_2d buf;
1352 const int row_offset_y = mi_row << 2;
1353 const int col_offset_y = mi_col << 2;
1354
1355 buf.buf = y_buffer + row_offset_y * y_stride + col_offset_y;
1356 buf.stride = y_stride;
1357
1358 var += av1_get_perpixel_variance_facade(cpi, xd, &buf, BLOCK_8X8,
1359 AOM_PLANE_Y);
1360 num_of_var += 1.0;
1361 }
1362 }
1363 var = var / num_of_var;
1364
1365 // Curve fitting with an exponential model on all 16x16 blocks from the
1366 // midres dataset.
1367 var = 67.035434 * (1 - exp(-0.0021489 * var)) + 17.492222;
1368
1369 // As per the above computation, var will be in the range of
1370 // [17.492222, 84.527656], assuming the data type is of infinite
1371 // precision. The following assert conservatively checks if var is in the
1372 // range of [17.0, 85.0] to avoid any issues due to the precision of the
1373 // relevant data type.
1374 assert(var > 17.0 && var < 85.0);
1375 cpi->ssim_rdmult_scaling_factors[index] = var;
1376 log_sum += log(var);
1377 }
1378 }
1379
1380 // As log_sum holds the geometric mean, it will be in the range
1381 // [17.492222, 84.527656]. Hence, in the below loop, the value of
1382 // cpi->ssim_rdmult_scaling_factors[index] would be in the range
1383 // [0.2069, 4.8323].
1384 log_sum = exp(log_sum / (double)(num_rows * num_cols));
1385
1386 for (int row = 0; row < num_rows; ++row) {
1387 for (int col = 0; col < num_cols; ++col) {
1388 const int index = row * num_cols + col;
1389 cpi->ssim_rdmult_scaling_factors[index] /= log_sum;
1390 }
1391 }
1392 }
1393
1394 // Coding context that only needs to be saved when recode loop includes
1395 // filtering (deblocking, CDEF, superres post-encode upscale and/or loop
1396 // restoraton).
save_extra_coding_context(AV1_COMP * cpi)1397 static void save_extra_coding_context(AV1_COMP *cpi) {
1398 CODING_CONTEXT *const cc = &cpi->coding_context;
1399 AV1_COMMON *cm = &cpi->common;
1400
1401 cc->lf = cm->lf;
1402 cc->cdef_info = cm->cdef_info;
1403 cc->rc = cpi->rc;
1404 cc->mv_stats = cpi->ppi->mv_stats;
1405 }
1406
av1_save_all_coding_context(AV1_COMP * cpi)1407 void av1_save_all_coding_context(AV1_COMP *cpi) {
1408 save_extra_coding_context(cpi);
1409 if (!frame_is_intra_only(&cpi->common)) release_scaled_references(cpi);
1410 }
1411
1412 #if DUMP_RECON_FRAMES == 1
1413
1414 // NOTE(zoeliu): For debug - Output the filtered reconstructed video.
av1_dump_filtered_recon_frames(AV1_COMP * cpi)1415 void av1_dump_filtered_recon_frames(AV1_COMP *cpi) {
1416 AV1_COMMON *const cm = &cpi->common;
1417 const CurrentFrame *const current_frame = &cm->current_frame;
1418 const YV12_BUFFER_CONFIG *recon_buf = &cm->cur_frame->buf;
1419
1420 if (recon_buf == NULL) {
1421 printf("Frame %d is not ready.\n", current_frame->frame_number);
1422 return;
1423 }
1424
1425 static const int flag_list[REF_FRAMES] = { 0,
1426 AOM_LAST_FLAG,
1427 AOM_LAST2_FLAG,
1428 AOM_LAST3_FLAG,
1429 AOM_GOLD_FLAG,
1430 AOM_BWD_FLAG,
1431 AOM_ALT2_FLAG,
1432 AOM_ALT_FLAG };
1433 printf(
1434 "\n***Frame=%d (frame_offset=%d, show_frame=%d, "
1435 "show_existing_frame=%d) "
1436 "[LAST LAST2 LAST3 GOLDEN BWD ALT2 ALT]=[",
1437 current_frame->frame_number, current_frame->order_hint, cm->show_frame,
1438 cm->show_existing_frame);
1439 for (int ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
1440 const RefCntBuffer *const buf = get_ref_frame_buf(cm, ref_frame);
1441 const int ref_offset = buf != NULL ? (int)buf->order_hint : -1;
1442 printf(" %d(%c)", ref_offset,
1443 (cpi->ref_frame_flags & flag_list[ref_frame]) ? 'Y' : 'N');
1444 }
1445 printf(" ]\n");
1446
1447 if (!cm->show_frame) {
1448 printf("Frame %d is a no show frame, so no image dump.\n",
1449 current_frame->frame_number);
1450 return;
1451 }
1452
1453 int h;
1454 char file_name[256] = "/tmp/enc_filtered_recon.yuv";
1455 FILE *f_recon = NULL;
1456
1457 if (current_frame->frame_number == 0) {
1458 if ((f_recon = fopen(file_name, "wb")) == NULL) {
1459 printf("Unable to open file %s to write.\n", file_name);
1460 return;
1461 }
1462 } else {
1463 if ((f_recon = fopen(file_name, "ab")) == NULL) {
1464 printf("Unable to open file %s to append.\n", file_name);
1465 return;
1466 }
1467 }
1468 printf(
1469 "\nFrame=%5d, encode_update_type[%5d]=%1d, frame_offset=%d, "
1470 "show_frame=%d, show_existing_frame=%d, source_alt_ref_active=%d, "
1471 "refresh_alt_ref_frame=%d, "
1472 "y_stride=%4d, uv_stride=%4d, cm->width=%4d, cm->height=%4d\n\n",
1473 current_frame->frame_number, cpi->gf_frame_index,
1474 cpi->ppi->gf_group.update_type[cpi->gf_frame_index],
1475 current_frame->order_hint, cm->show_frame, cm->show_existing_frame,
1476 cpi->rc.source_alt_ref_active, cpi->refresh_frame.alt_ref_frame,
1477 recon_buf->y_stride, recon_buf->uv_stride, cm->width, cm->height);
1478 #if 0
1479 int ref_frame;
1480 printf("get_ref_frame_map_idx: [");
1481 for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame)
1482 printf(" %d", get_ref_frame_map_idx(cm, ref_frame));
1483 printf(" ]\n");
1484 #endif // 0
1485
1486 // --- Y ---
1487 for (h = 0; h < cm->height; ++h) {
1488 fwrite(&recon_buf->y_buffer[h * recon_buf->y_stride], 1, cm->width,
1489 f_recon);
1490 }
1491 // --- U ---
1492 for (h = 0; h < (cm->height >> 1); ++h) {
1493 fwrite(&recon_buf->u_buffer[h * recon_buf->uv_stride], 1, (cm->width >> 1),
1494 f_recon);
1495 }
1496 // --- V ---
1497 for (h = 0; h < (cm->height >> 1); ++h) {
1498 fwrite(&recon_buf->v_buffer[h * recon_buf->uv_stride], 1, (cm->width >> 1),
1499 f_recon);
1500 }
1501
1502 fclose(f_recon);
1503 }
1504 #endif // DUMP_RECON_FRAMES
1505