/src/aom/av1/common/resize.c
Line | Count | Source (jump to first uncovered line) |
1 | | /* |
2 | | * Copyright (c) 2016, 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 <assert.h> |
13 | | #include <limits.h> |
14 | | #include <math.h> |
15 | | #include <stdbool.h> |
16 | | #include <stdio.h> |
17 | | #include <stdlib.h> |
18 | | #include <string.h> |
19 | | |
20 | | #include "config/aom_config.h" |
21 | | #include "config/av1_rtcd.h" |
22 | | |
23 | | #include "aom_dsp/aom_dsp_common.h" |
24 | | #include "aom_dsp/flow_estimation/corner_detect.h" |
25 | | #include "aom_ports/mem.h" |
26 | | #include "aom_scale/aom_scale.h" |
27 | | #include "av1/common/common.h" |
28 | | #include "av1/common/resize.h" |
29 | | |
30 | | #include "config/aom_dsp_rtcd.h" |
31 | | #include "config/aom_scale_rtcd.h" |
32 | | |
33 | | // Filters for interpolation (0.5-band) - note this also filters integer pels. |
34 | | static const InterpKernel filteredinterp_filters500[(1 << RS_SUBPEL_BITS)] = { |
35 | | { -3, 0, 35, 64, 35, 0, -3, 0 }, { -3, 0, 34, 64, 36, 0, -3, 0 }, |
36 | | { -3, -1, 34, 64, 36, 1, -3, 0 }, { -3, -1, 33, 64, 37, 1, -3, 0 }, |
37 | | { -3, -1, 32, 64, 38, 1, -3, 0 }, { -3, -1, 31, 64, 39, 1, -3, 0 }, |
38 | | { -3, -1, 31, 63, 39, 2, -3, 0 }, { -2, -2, 30, 63, 40, 2, -3, 0 }, |
39 | | { -2, -2, 29, 63, 41, 2, -3, 0 }, { -2, -2, 29, 63, 41, 3, -4, 0 }, |
40 | | { -2, -2, 28, 63, 42, 3, -4, 0 }, { -2, -2, 27, 63, 43, 3, -4, 0 }, |
41 | | { -2, -3, 27, 63, 43, 4, -4, 0 }, { -2, -3, 26, 62, 44, 5, -4, 0 }, |
42 | | { -2, -3, 25, 62, 45, 5, -4, 0 }, { -2, -3, 25, 62, 45, 5, -4, 0 }, |
43 | | { -2, -3, 24, 62, 46, 5, -4, 0 }, { -2, -3, 23, 61, 47, 6, -4, 0 }, |
44 | | { -2, -3, 23, 61, 47, 6, -4, 0 }, { -2, -3, 22, 61, 48, 7, -4, -1 }, |
45 | | { -2, -3, 21, 60, 49, 7, -4, 0 }, { -1, -4, 20, 60, 49, 8, -4, 0 }, |
46 | | { -1, -4, 20, 60, 50, 8, -4, -1 }, { -1, -4, 19, 59, 51, 9, -4, -1 }, |
47 | | { -1, -4, 19, 59, 51, 9, -4, -1 }, { -1, -4, 18, 58, 52, 10, -4, -1 }, |
48 | | { -1, -4, 17, 58, 52, 11, -4, -1 }, { -1, -4, 16, 58, 53, 11, -4, -1 }, |
49 | | { -1, -4, 16, 57, 53, 12, -4, -1 }, { -1, -4, 15, 57, 54, 12, -4, -1 }, |
50 | | { -1, -4, 15, 56, 54, 13, -4, -1 }, { -1, -4, 14, 56, 55, 13, -4, -1 }, |
51 | | { -1, -4, 14, 55, 55, 14, -4, -1 }, { -1, -4, 13, 55, 56, 14, -4, -1 }, |
52 | | { -1, -4, 13, 54, 56, 15, -4, -1 }, { -1, -4, 12, 54, 57, 15, -4, -1 }, |
53 | | { -1, -4, 12, 53, 57, 16, -4, -1 }, { -1, -4, 11, 53, 58, 16, -4, -1 }, |
54 | | { -1, -4, 11, 52, 58, 17, -4, -1 }, { -1, -4, 10, 52, 58, 18, -4, -1 }, |
55 | | { -1, -4, 9, 51, 59, 19, -4, -1 }, { -1, -4, 9, 51, 59, 19, -4, -1 }, |
56 | | { -1, -4, 8, 50, 60, 20, -4, -1 }, { 0, -4, 8, 49, 60, 20, -4, -1 }, |
57 | | { 0, -4, 7, 49, 60, 21, -3, -2 }, { -1, -4, 7, 48, 61, 22, -3, -2 }, |
58 | | { 0, -4, 6, 47, 61, 23, -3, -2 }, { 0, -4, 6, 47, 61, 23, -3, -2 }, |
59 | | { 0, -4, 5, 46, 62, 24, -3, -2 }, { 0, -4, 5, 45, 62, 25, -3, -2 }, |
60 | | { 0, -4, 5, 45, 62, 25, -3, -2 }, { 0, -4, 5, 44, 62, 26, -3, -2 }, |
61 | | { 0, -4, 4, 43, 63, 27, -3, -2 }, { 0, -4, 3, 43, 63, 27, -2, -2 }, |
62 | | { 0, -4, 3, 42, 63, 28, -2, -2 }, { 0, -4, 3, 41, 63, 29, -2, -2 }, |
63 | | { 0, -3, 2, 41, 63, 29, -2, -2 }, { 0, -3, 2, 40, 63, 30, -2, -2 }, |
64 | | { 0, -3, 2, 39, 63, 31, -1, -3 }, { 0, -3, 1, 39, 64, 31, -1, -3 }, |
65 | | { 0, -3, 1, 38, 64, 32, -1, -3 }, { 0, -3, 1, 37, 64, 33, -1, -3 }, |
66 | | { 0, -3, 1, 36, 64, 34, -1, -3 }, { 0, -3, 0, 36, 64, 34, 0, -3 }, |
67 | | }; |
68 | | |
69 | | // Filters for interpolation (0.625-band) - note this also filters integer pels. |
70 | | static const InterpKernel filteredinterp_filters625[(1 << RS_SUBPEL_BITS)] = { |
71 | | { -1, -8, 33, 80, 33, -8, -1, 0 }, { -1, -8, 31, 80, 34, -8, -1, 1 }, |
72 | | { -1, -8, 30, 80, 35, -8, -1, 1 }, { -1, -8, 29, 80, 36, -7, -2, 1 }, |
73 | | { -1, -8, 28, 80, 37, -7, -2, 1 }, { -1, -8, 27, 80, 38, -7, -2, 1 }, |
74 | | { 0, -8, 26, 79, 39, -7, -2, 1 }, { 0, -8, 25, 79, 40, -7, -2, 1 }, |
75 | | { 0, -8, 24, 79, 41, -7, -2, 1 }, { 0, -8, 23, 78, 42, -6, -2, 1 }, |
76 | | { 0, -8, 22, 78, 43, -6, -2, 1 }, { 0, -8, 21, 78, 44, -6, -2, 1 }, |
77 | | { 0, -8, 20, 78, 45, -5, -3, 1 }, { 0, -8, 19, 77, 47, -5, -3, 1 }, |
78 | | { 0, -8, 18, 77, 48, -5, -3, 1 }, { 0, -8, 17, 77, 49, -5, -3, 1 }, |
79 | | { 0, -8, 16, 76, 50, -4, -3, 1 }, { 0, -8, 15, 76, 51, -4, -3, 1 }, |
80 | | { 0, -8, 15, 75, 52, -3, -4, 1 }, { 0, -7, 14, 74, 53, -3, -4, 1 }, |
81 | | { 0, -7, 13, 74, 54, -3, -4, 1 }, { 0, -7, 12, 73, 55, -2, -4, 1 }, |
82 | | { 0, -7, 11, 73, 56, -2, -4, 1 }, { 0, -7, 10, 72, 57, -1, -4, 1 }, |
83 | | { 1, -7, 10, 71, 58, -1, -5, 1 }, { 0, -7, 9, 71, 59, 0, -5, 1 }, |
84 | | { 1, -7, 8, 70, 60, 0, -5, 1 }, { 1, -7, 7, 69, 61, 1, -5, 1 }, |
85 | | { 1, -6, 6, 68, 62, 1, -5, 1 }, { 0, -6, 6, 68, 62, 2, -5, 1 }, |
86 | | { 1, -6, 5, 67, 63, 2, -5, 1 }, { 1, -6, 5, 66, 64, 3, -6, 1 }, |
87 | | { 1, -6, 4, 65, 65, 4, -6, 1 }, { 1, -6, 3, 64, 66, 5, -6, 1 }, |
88 | | { 1, -5, 2, 63, 67, 5, -6, 1 }, { 1, -5, 2, 62, 68, 6, -6, 0 }, |
89 | | { 1, -5, 1, 62, 68, 6, -6, 1 }, { 1, -5, 1, 61, 69, 7, -7, 1 }, |
90 | | { 1, -5, 0, 60, 70, 8, -7, 1 }, { 1, -5, 0, 59, 71, 9, -7, 0 }, |
91 | | { 1, -5, -1, 58, 71, 10, -7, 1 }, { 1, -4, -1, 57, 72, 10, -7, 0 }, |
92 | | { 1, -4, -2, 56, 73, 11, -7, 0 }, { 1, -4, -2, 55, 73, 12, -7, 0 }, |
93 | | { 1, -4, -3, 54, 74, 13, -7, 0 }, { 1, -4, -3, 53, 74, 14, -7, 0 }, |
94 | | { 1, -4, -3, 52, 75, 15, -8, 0 }, { 1, -3, -4, 51, 76, 15, -8, 0 }, |
95 | | { 1, -3, -4, 50, 76, 16, -8, 0 }, { 1, -3, -5, 49, 77, 17, -8, 0 }, |
96 | | { 1, -3, -5, 48, 77, 18, -8, 0 }, { 1, -3, -5, 47, 77, 19, -8, 0 }, |
97 | | { 1, -3, -5, 45, 78, 20, -8, 0 }, { 1, -2, -6, 44, 78, 21, -8, 0 }, |
98 | | { 1, -2, -6, 43, 78, 22, -8, 0 }, { 1, -2, -6, 42, 78, 23, -8, 0 }, |
99 | | { 1, -2, -7, 41, 79, 24, -8, 0 }, { 1, -2, -7, 40, 79, 25, -8, 0 }, |
100 | | { 1, -2, -7, 39, 79, 26, -8, 0 }, { 1, -2, -7, 38, 80, 27, -8, -1 }, |
101 | | { 1, -2, -7, 37, 80, 28, -8, -1 }, { 1, -2, -7, 36, 80, 29, -8, -1 }, |
102 | | { 1, -1, -8, 35, 80, 30, -8, -1 }, { 1, -1, -8, 34, 80, 31, -8, -1 }, |
103 | | }; |
104 | | |
105 | | // Filters for interpolation (0.75-band) - note this also filters integer pels. |
106 | | static const InterpKernel filteredinterp_filters750[(1 << RS_SUBPEL_BITS)] = { |
107 | | { 2, -11, 25, 96, 25, -11, 2, 0 }, { 2, -11, 24, 96, 26, -11, 2, 0 }, |
108 | | { 2, -11, 22, 96, 28, -11, 2, 0 }, { 2, -10, 21, 96, 29, -12, 2, 0 }, |
109 | | { 2, -10, 19, 96, 31, -12, 2, 0 }, { 2, -10, 18, 95, 32, -11, 2, 0 }, |
110 | | { 2, -10, 17, 95, 34, -12, 2, 0 }, { 2, -9, 15, 95, 35, -12, 2, 0 }, |
111 | | { 2, -9, 14, 94, 37, -12, 2, 0 }, { 2, -9, 13, 94, 38, -12, 2, 0 }, |
112 | | { 2, -8, 12, 93, 40, -12, 1, 0 }, { 2, -8, 11, 93, 41, -12, 1, 0 }, |
113 | | { 2, -8, 9, 92, 43, -12, 1, 1 }, { 2, -8, 8, 92, 44, -12, 1, 1 }, |
114 | | { 2, -7, 7, 91, 46, -12, 1, 0 }, { 2, -7, 6, 90, 47, -12, 1, 1 }, |
115 | | { 2, -7, 5, 90, 49, -12, 1, 0 }, { 2, -6, 4, 89, 50, -12, 1, 0 }, |
116 | | { 2, -6, 3, 88, 52, -12, 0, 1 }, { 2, -6, 2, 87, 54, -12, 0, 1 }, |
117 | | { 2, -5, 1, 86, 55, -12, 0, 1 }, { 2, -5, 0, 85, 57, -12, 0, 1 }, |
118 | | { 2, -5, -1, 84, 58, -11, 0, 1 }, { 2, -5, -2, 83, 60, -11, 0, 1 }, |
119 | | { 2, -4, -2, 82, 61, -11, -1, 1 }, { 1, -4, -3, 81, 63, -10, -1, 1 }, |
120 | | { 2, -4, -4, 80, 64, -10, -1, 1 }, { 1, -4, -4, 79, 66, -10, -1, 1 }, |
121 | | { 1, -3, -5, 77, 67, -9, -1, 1 }, { 1, -3, -6, 76, 69, -9, -1, 1 }, |
122 | | { 1, -3, -6, 75, 70, -8, -2, 1 }, { 1, -2, -7, 74, 71, -8, -2, 1 }, |
123 | | { 1, -2, -7, 72, 72, -7, -2, 1 }, { 1, -2, -8, 71, 74, -7, -2, 1 }, |
124 | | { 1, -2, -8, 70, 75, -6, -3, 1 }, { 1, -1, -9, 69, 76, -6, -3, 1 }, |
125 | | { 1, -1, -9, 67, 77, -5, -3, 1 }, { 1, -1, -10, 66, 79, -4, -4, 1 }, |
126 | | { 1, -1, -10, 64, 80, -4, -4, 2 }, { 1, -1, -10, 63, 81, -3, -4, 1 }, |
127 | | { 1, -1, -11, 61, 82, -2, -4, 2 }, { 1, 0, -11, 60, 83, -2, -5, 2 }, |
128 | | { 1, 0, -11, 58, 84, -1, -5, 2 }, { 1, 0, -12, 57, 85, 0, -5, 2 }, |
129 | | { 1, 0, -12, 55, 86, 1, -5, 2 }, { 1, 0, -12, 54, 87, 2, -6, 2 }, |
130 | | { 1, 0, -12, 52, 88, 3, -6, 2 }, { 0, 1, -12, 50, 89, 4, -6, 2 }, |
131 | | { 0, 1, -12, 49, 90, 5, -7, 2 }, { 1, 1, -12, 47, 90, 6, -7, 2 }, |
132 | | { 0, 1, -12, 46, 91, 7, -7, 2 }, { 1, 1, -12, 44, 92, 8, -8, 2 }, |
133 | | { 1, 1, -12, 43, 92, 9, -8, 2 }, { 0, 1, -12, 41, 93, 11, -8, 2 }, |
134 | | { 0, 1, -12, 40, 93, 12, -8, 2 }, { 0, 2, -12, 38, 94, 13, -9, 2 }, |
135 | | { 0, 2, -12, 37, 94, 14, -9, 2 }, { 0, 2, -12, 35, 95, 15, -9, 2 }, |
136 | | { 0, 2, -12, 34, 95, 17, -10, 2 }, { 0, 2, -11, 32, 95, 18, -10, 2 }, |
137 | | { 0, 2, -12, 31, 96, 19, -10, 2 }, { 0, 2, -12, 29, 96, 21, -10, 2 }, |
138 | | { 0, 2, -11, 28, 96, 22, -11, 2 }, { 0, 2, -11, 26, 96, 24, -11, 2 }, |
139 | | }; |
140 | | |
141 | | // Filters for interpolation (0.875-band) - note this also filters integer pels. |
142 | | static const InterpKernel filteredinterp_filters875[(1 << RS_SUBPEL_BITS)] = { |
143 | | { 3, -8, 13, 112, 13, -8, 3, 0 }, { 2, -7, 12, 112, 15, -8, 3, -1 }, |
144 | | { 3, -7, 10, 112, 17, -9, 3, -1 }, { 2, -6, 8, 112, 19, -9, 3, -1 }, |
145 | | { 2, -6, 7, 112, 21, -10, 3, -1 }, { 2, -5, 6, 111, 22, -10, 3, -1 }, |
146 | | { 2, -5, 4, 111, 24, -10, 3, -1 }, { 2, -4, 3, 110, 26, -11, 3, -1 }, |
147 | | { 2, -4, 1, 110, 28, -11, 3, -1 }, { 2, -4, 0, 109, 30, -12, 4, -1 }, |
148 | | { 1, -3, -1, 108, 32, -12, 4, -1 }, { 1, -3, -2, 108, 34, -13, 4, -1 }, |
149 | | { 1, -2, -4, 107, 36, -13, 4, -1 }, { 1, -2, -5, 106, 38, -13, 4, -1 }, |
150 | | { 1, -1, -6, 105, 40, -14, 4, -1 }, { 1, -1, -7, 104, 42, -14, 4, -1 }, |
151 | | { 1, -1, -7, 103, 44, -15, 4, -1 }, { 1, 0, -8, 101, 46, -15, 4, -1 }, |
152 | | { 1, 0, -9, 100, 48, -15, 4, -1 }, { 1, 0, -10, 99, 50, -15, 4, -1 }, |
153 | | { 1, 1, -11, 97, 53, -16, 4, -1 }, { 0, 1, -11, 96, 55, -16, 4, -1 }, |
154 | | { 0, 1, -12, 95, 57, -16, 4, -1 }, { 0, 2, -13, 93, 59, -16, 4, -1 }, |
155 | | { 0, 2, -13, 91, 61, -16, 4, -1 }, { 0, 2, -14, 90, 63, -16, 4, -1 }, |
156 | | { 0, 2, -14, 88, 65, -16, 4, -1 }, { 0, 2, -15, 86, 67, -16, 4, 0 }, |
157 | | { 0, 3, -15, 84, 69, -17, 4, 0 }, { 0, 3, -16, 83, 71, -17, 4, 0 }, |
158 | | { 0, 3, -16, 81, 73, -16, 3, 0 }, { 0, 3, -16, 79, 75, -16, 3, 0 }, |
159 | | { 0, 3, -16, 77, 77, -16, 3, 0 }, { 0, 3, -16, 75, 79, -16, 3, 0 }, |
160 | | { 0, 3, -16, 73, 81, -16, 3, 0 }, { 0, 4, -17, 71, 83, -16, 3, 0 }, |
161 | | { 0, 4, -17, 69, 84, -15, 3, 0 }, { 0, 4, -16, 67, 86, -15, 2, 0 }, |
162 | | { -1, 4, -16, 65, 88, -14, 2, 0 }, { -1, 4, -16, 63, 90, -14, 2, 0 }, |
163 | | { -1, 4, -16, 61, 91, -13, 2, 0 }, { -1, 4, -16, 59, 93, -13, 2, 0 }, |
164 | | { -1, 4, -16, 57, 95, -12, 1, 0 }, { -1, 4, -16, 55, 96, -11, 1, 0 }, |
165 | | { -1, 4, -16, 53, 97, -11, 1, 1 }, { -1, 4, -15, 50, 99, -10, 0, 1 }, |
166 | | { -1, 4, -15, 48, 100, -9, 0, 1 }, { -1, 4, -15, 46, 101, -8, 0, 1 }, |
167 | | { -1, 4, -15, 44, 103, -7, -1, 1 }, { -1, 4, -14, 42, 104, -7, -1, 1 }, |
168 | | { -1, 4, -14, 40, 105, -6, -1, 1 }, { -1, 4, -13, 38, 106, -5, -2, 1 }, |
169 | | { -1, 4, -13, 36, 107, -4, -2, 1 }, { -1, 4, -13, 34, 108, -2, -3, 1 }, |
170 | | { -1, 4, -12, 32, 108, -1, -3, 1 }, { -1, 4, -12, 30, 109, 0, -4, 2 }, |
171 | | { -1, 3, -11, 28, 110, 1, -4, 2 }, { -1, 3, -11, 26, 110, 3, -4, 2 }, |
172 | | { -1, 3, -10, 24, 111, 4, -5, 2 }, { -1, 3, -10, 22, 111, 6, -5, 2 }, |
173 | | { -1, 3, -10, 21, 112, 7, -6, 2 }, { -1, 3, -9, 19, 112, 8, -6, 2 }, |
174 | | { -1, 3, -9, 17, 112, 10, -7, 3 }, { -1, 3, -8, 15, 112, 12, -7, 2 }, |
175 | | }; |
176 | | |
177 | | const int16_t av1_resize_filter_normative[( |
178 | | 1 << RS_SUBPEL_BITS)][UPSCALE_NORMATIVE_TAPS] = { |
179 | | #if UPSCALE_NORMATIVE_TAPS == 8 |
180 | | { 0, 0, 0, 128, 0, 0, 0, 0 }, { 0, 0, -1, 128, 2, -1, 0, 0 }, |
181 | | { 0, 1, -3, 127, 4, -2, 1, 0 }, { 0, 1, -4, 127, 6, -3, 1, 0 }, |
182 | | { 0, 2, -6, 126, 8, -3, 1, 0 }, { 0, 2, -7, 125, 11, -4, 1, 0 }, |
183 | | { -1, 2, -8, 125, 13, -5, 2, 0 }, { -1, 3, -9, 124, 15, -6, 2, 0 }, |
184 | | { -1, 3, -10, 123, 18, -6, 2, -1 }, { -1, 3, -11, 122, 20, -7, 3, -1 }, |
185 | | { -1, 4, -12, 121, 22, -8, 3, -1 }, { -1, 4, -13, 120, 25, -9, 3, -1 }, |
186 | | { -1, 4, -14, 118, 28, -9, 3, -1 }, { -1, 4, -15, 117, 30, -10, 4, -1 }, |
187 | | { -1, 5, -16, 116, 32, -11, 4, -1 }, { -1, 5, -16, 114, 35, -12, 4, -1 }, |
188 | | { -1, 5, -17, 112, 38, -12, 4, -1 }, { -1, 5, -18, 111, 40, -13, 5, -1 }, |
189 | | { -1, 5, -18, 109, 43, -14, 5, -1 }, { -1, 6, -19, 107, 45, -14, 5, -1 }, |
190 | | { -1, 6, -19, 105, 48, -15, 5, -1 }, { -1, 6, -19, 103, 51, -16, 5, -1 }, |
191 | | { -1, 6, -20, 101, 53, -16, 6, -1 }, { -1, 6, -20, 99, 56, -17, 6, -1 }, |
192 | | { -1, 6, -20, 97, 58, -17, 6, -1 }, { -1, 6, -20, 95, 61, -18, 6, -1 }, |
193 | | { -2, 7, -20, 93, 64, -18, 6, -2 }, { -2, 7, -20, 91, 66, -19, 6, -1 }, |
194 | | { -2, 7, -20, 88, 69, -19, 6, -1 }, { -2, 7, -20, 86, 71, -19, 6, -1 }, |
195 | | { -2, 7, -20, 84, 74, -20, 7, -2 }, { -2, 7, -20, 81, 76, -20, 7, -1 }, |
196 | | { -2, 7, -20, 79, 79, -20, 7, -2 }, { -1, 7, -20, 76, 81, -20, 7, -2 }, |
197 | | { -2, 7, -20, 74, 84, -20, 7, -2 }, { -1, 6, -19, 71, 86, -20, 7, -2 }, |
198 | | { -1, 6, -19, 69, 88, -20, 7, -2 }, { -1, 6, -19, 66, 91, -20, 7, -2 }, |
199 | | { -2, 6, -18, 64, 93, -20, 7, -2 }, { -1, 6, -18, 61, 95, -20, 6, -1 }, |
200 | | { -1, 6, -17, 58, 97, -20, 6, -1 }, { -1, 6, -17, 56, 99, -20, 6, -1 }, |
201 | | { -1, 6, -16, 53, 101, -20, 6, -1 }, { -1, 5, -16, 51, 103, -19, 6, -1 }, |
202 | | { -1, 5, -15, 48, 105, -19, 6, -1 }, { -1, 5, -14, 45, 107, -19, 6, -1 }, |
203 | | { -1, 5, -14, 43, 109, -18, 5, -1 }, { -1, 5, -13, 40, 111, -18, 5, -1 }, |
204 | | { -1, 4, -12, 38, 112, -17, 5, -1 }, { -1, 4, -12, 35, 114, -16, 5, -1 }, |
205 | | { -1, 4, -11, 32, 116, -16, 5, -1 }, { -1, 4, -10, 30, 117, -15, 4, -1 }, |
206 | | { -1, 3, -9, 28, 118, -14, 4, -1 }, { -1, 3, -9, 25, 120, -13, 4, -1 }, |
207 | | { -1, 3, -8, 22, 121, -12, 4, -1 }, { -1, 3, -7, 20, 122, -11, 3, -1 }, |
208 | | { -1, 2, -6, 18, 123, -10, 3, -1 }, { 0, 2, -6, 15, 124, -9, 3, -1 }, |
209 | | { 0, 2, -5, 13, 125, -8, 2, -1 }, { 0, 1, -4, 11, 125, -7, 2, 0 }, |
210 | | { 0, 1, -3, 8, 126, -6, 2, 0 }, { 0, 1, -3, 6, 127, -4, 1, 0 }, |
211 | | { 0, 1, -2, 4, 127, -3, 1, 0 }, { 0, 0, -1, 2, 128, -1, 0, 0 }, |
212 | | #else |
213 | | #error "Invalid value of UPSCALE_NORMATIVE_TAPS" |
214 | | #endif // UPSCALE_NORMATIVE_TAPS == 8 |
215 | | }; |
216 | | |
217 | | // Filters for interpolation (full-band) - no filtering for integer pixels |
218 | 0 | #define filteredinterp_filters1000 av1_resize_filter_normative |
219 | | |
220 | 0 | static const InterpKernel *choose_interp_filter(int in_length, int out_length) { |
221 | 0 | int out_length16 = out_length * 16; |
222 | 0 | if (out_length16 >= in_length * 16) |
223 | 0 | return filteredinterp_filters1000; |
224 | 0 | else if (out_length16 >= in_length * 13) |
225 | 0 | return filteredinterp_filters875; |
226 | 0 | else if (out_length16 >= in_length * 11) |
227 | 0 | return filteredinterp_filters750; |
228 | 0 | else if (out_length16 >= in_length * 9) |
229 | 0 | return filteredinterp_filters625; |
230 | 0 | else |
231 | 0 | return filteredinterp_filters500; |
232 | 0 | } |
233 | | |
234 | | static void interpolate_core(const uint8_t *const input, int in_length, |
235 | | uint8_t *output, int out_length, |
236 | 0 | const int16_t *interp_filters, int interp_taps) { |
237 | 0 | const int32_t delta = |
238 | 0 | (((uint32_t)in_length << RS_SCALE_SUBPEL_BITS) + out_length / 2) / |
239 | 0 | out_length; |
240 | 0 | const int32_t offset = |
241 | 0 | in_length > out_length |
242 | 0 | ? (((int32_t)(in_length - out_length) << (RS_SCALE_SUBPEL_BITS - 1)) + |
243 | 0 | out_length / 2) / |
244 | 0 | out_length |
245 | 0 | : -(((int32_t)(out_length - in_length) |
246 | 0 | << (RS_SCALE_SUBPEL_BITS - 1)) + |
247 | 0 | out_length / 2) / |
248 | 0 | out_length; |
249 | 0 | uint8_t *optr = output; |
250 | 0 | int x, x1, x2, sum, k, int_pel, sub_pel; |
251 | 0 | int32_t y; |
252 | |
|
253 | 0 | x = 0; |
254 | 0 | y = offset + RS_SCALE_EXTRA_OFF; |
255 | 0 | while ((y >> RS_SCALE_SUBPEL_BITS) < (interp_taps / 2 - 1)) { |
256 | 0 | x++; |
257 | 0 | y += delta; |
258 | 0 | } |
259 | 0 | x1 = x; |
260 | 0 | x = out_length - 1; |
261 | 0 | y = delta * x + offset + RS_SCALE_EXTRA_OFF; |
262 | 0 | while ((y >> RS_SCALE_SUBPEL_BITS) + (int32_t)(interp_taps / 2) >= |
263 | 0 | in_length) { |
264 | 0 | x--; |
265 | 0 | y -= delta; |
266 | 0 | } |
267 | 0 | x2 = x; |
268 | 0 | if (x1 > x2) { |
269 | 0 | for (x = 0, y = offset + RS_SCALE_EXTRA_OFF; x < out_length; |
270 | 0 | ++x, y += delta) { |
271 | 0 | int_pel = y >> RS_SCALE_SUBPEL_BITS; |
272 | 0 | sub_pel = (y >> RS_SCALE_EXTRA_BITS) & RS_SUBPEL_MASK; |
273 | 0 | const int16_t *filter = &interp_filters[sub_pel * interp_taps]; |
274 | 0 | sum = 0; |
275 | 0 | for (k = 0; k < interp_taps; ++k) { |
276 | 0 | const int pk = int_pel - interp_taps / 2 + 1 + k; |
277 | 0 | sum += filter[k] * input[AOMMAX(AOMMIN(pk, in_length - 1), 0)]; |
278 | 0 | } |
279 | 0 | *optr++ = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)); |
280 | 0 | } |
281 | 0 | } else { |
282 | | // Initial part. |
283 | 0 | for (x = 0, y = offset + RS_SCALE_EXTRA_OFF; x < x1; ++x, y += delta) { |
284 | 0 | int_pel = y >> RS_SCALE_SUBPEL_BITS; |
285 | 0 | sub_pel = (y >> RS_SCALE_EXTRA_BITS) & RS_SUBPEL_MASK; |
286 | 0 | const int16_t *filter = &interp_filters[sub_pel * interp_taps]; |
287 | 0 | sum = 0; |
288 | 0 | for (k = 0; k < interp_taps; ++k) |
289 | 0 | sum += filter[k] * input[AOMMAX(int_pel - interp_taps / 2 + 1 + k, 0)]; |
290 | 0 | *optr++ = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)); |
291 | 0 | } |
292 | | // Middle part. |
293 | 0 | for (; x <= x2; ++x, y += delta) { |
294 | 0 | int_pel = y >> RS_SCALE_SUBPEL_BITS; |
295 | 0 | sub_pel = (y >> RS_SCALE_EXTRA_BITS) & RS_SUBPEL_MASK; |
296 | 0 | const int16_t *filter = &interp_filters[sub_pel * interp_taps]; |
297 | 0 | sum = 0; |
298 | 0 | for (k = 0; k < interp_taps; ++k) |
299 | 0 | sum += filter[k] * input[int_pel - interp_taps / 2 + 1 + k]; |
300 | 0 | *optr++ = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)); |
301 | 0 | } |
302 | | // End part. |
303 | 0 | for (; x < out_length; ++x, y += delta) { |
304 | 0 | int_pel = y >> RS_SCALE_SUBPEL_BITS; |
305 | 0 | sub_pel = (y >> RS_SCALE_EXTRA_BITS) & RS_SUBPEL_MASK; |
306 | 0 | const int16_t *filter = &interp_filters[sub_pel * interp_taps]; |
307 | 0 | sum = 0; |
308 | 0 | for (k = 0; k < interp_taps; ++k) |
309 | 0 | sum += filter[k] * |
310 | 0 | input[AOMMIN(int_pel - interp_taps / 2 + 1 + k, in_length - 1)]; |
311 | 0 | *optr++ = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)); |
312 | 0 | } |
313 | 0 | } |
314 | 0 | } |
315 | | |
316 | | static void interpolate(const uint8_t *const input, int in_length, |
317 | 0 | uint8_t *output, int out_length) { |
318 | 0 | const InterpKernel *interp_filters = |
319 | 0 | choose_interp_filter(in_length, out_length); |
320 | |
|
321 | 0 | interpolate_core(input, in_length, output, out_length, &interp_filters[0][0], |
322 | 0 | SUBPEL_TAPS); |
323 | 0 | } |
324 | | |
325 | 49.8k | int32_t av1_get_upscale_convolve_step(int in_length, int out_length) { |
326 | 49.8k | return ((in_length << RS_SCALE_SUBPEL_BITS) + out_length / 2) / out_length; |
327 | 49.8k | } |
328 | | |
329 | | static int32_t get_upscale_convolve_x0(int in_length, int out_length, |
330 | 49.8k | int32_t x_step_qn) { |
331 | 49.8k | const int err = out_length * x_step_qn - (in_length << RS_SCALE_SUBPEL_BITS); |
332 | 49.8k | const int32_t x0 = |
333 | 49.8k | (-((out_length - in_length) << (RS_SCALE_SUBPEL_BITS - 1)) + |
334 | 49.8k | out_length / 2) / |
335 | 49.8k | out_length + |
336 | 49.8k | RS_SCALE_EXTRA_OFF - err / 2; |
337 | 49.8k | return (int32_t)((uint32_t)x0 & RS_SCALE_SUBPEL_MASK); |
338 | 49.8k | } |
339 | | |
340 | | void down2_symeven(const uint8_t *const input, int length, uint8_t *output, |
341 | 0 | int start_offset) { |
342 | | // Actual filter len = 2 * filter_len_half. |
343 | 0 | const int16_t *filter = av1_down2_symeven_half_filter; |
344 | 0 | const int filter_len_half = sizeof(av1_down2_symeven_half_filter) / 2; |
345 | 0 | int i, j; |
346 | 0 | uint8_t *optr = output; |
347 | 0 | int l1 = filter_len_half; |
348 | 0 | int l2 = (length - filter_len_half); |
349 | 0 | l1 += (l1 & 1); |
350 | 0 | l2 += (l2 & 1); |
351 | 0 | if (l1 > l2) { |
352 | | // Short input length. |
353 | 0 | for (i = start_offset; i < length; i += 2) { |
354 | 0 | int sum = (1 << (FILTER_BITS - 1)); |
355 | 0 | for (j = 0; j < filter_len_half; ++j) { |
356 | 0 | sum += |
357 | 0 | (input[AOMMAX(i - j, 0)] + input[AOMMIN(i + 1 + j, length - 1)]) * |
358 | 0 | filter[j]; |
359 | 0 | } |
360 | 0 | sum >>= FILTER_BITS; |
361 | 0 | *optr++ = clip_pixel(sum); |
362 | 0 | } |
363 | 0 | } else { |
364 | | // Initial part. |
365 | 0 | for (i = start_offset; i < l1; i += 2) { |
366 | 0 | int sum = (1 << (FILTER_BITS - 1)); |
367 | 0 | for (j = 0; j < filter_len_half; ++j) { |
368 | 0 | sum += (input[AOMMAX(i - j, 0)] + input[i + 1 + j]) * filter[j]; |
369 | 0 | } |
370 | 0 | sum >>= FILTER_BITS; |
371 | 0 | *optr++ = clip_pixel(sum); |
372 | 0 | } |
373 | | // Middle part. |
374 | 0 | for (; i < l2; i += 2) { |
375 | 0 | int sum = (1 << (FILTER_BITS - 1)); |
376 | 0 | for (j = 0; j < filter_len_half; ++j) { |
377 | 0 | sum += (input[i - j] + input[i + 1 + j]) * filter[j]; |
378 | 0 | } |
379 | 0 | sum >>= FILTER_BITS; |
380 | 0 | *optr++ = clip_pixel(sum); |
381 | 0 | } |
382 | | // End part. |
383 | 0 | for (; i < length; i += 2) { |
384 | 0 | int sum = (1 << (FILTER_BITS - 1)); |
385 | 0 | for (j = 0; j < filter_len_half; ++j) { |
386 | 0 | sum += |
387 | 0 | (input[i - j] + input[AOMMIN(i + 1 + j, length - 1)]) * filter[j]; |
388 | 0 | } |
389 | 0 | sum >>= FILTER_BITS; |
390 | 0 | *optr++ = clip_pixel(sum); |
391 | 0 | } |
392 | 0 | } |
393 | 0 | } |
394 | | |
395 | | static void down2_symodd(const uint8_t *const input, int length, |
396 | 0 | uint8_t *output) { |
397 | | // Actual filter len = 2 * filter_len_half - 1. |
398 | 0 | const int16_t *filter = av1_down2_symodd_half_filter; |
399 | 0 | const int filter_len_half = sizeof(av1_down2_symodd_half_filter) / 2; |
400 | 0 | int i, j; |
401 | 0 | uint8_t *optr = output; |
402 | 0 | int l1 = filter_len_half - 1; |
403 | 0 | int l2 = (length - filter_len_half + 1); |
404 | 0 | l1 += (l1 & 1); |
405 | 0 | l2 += (l2 & 1); |
406 | 0 | if (l1 > l2) { |
407 | | // Short input length. |
408 | 0 | for (i = 0; i < length; i += 2) { |
409 | 0 | int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0]; |
410 | 0 | for (j = 1; j < filter_len_half; ++j) { |
411 | 0 | sum += (input[(i - j < 0 ? 0 : i - j)] + |
412 | 0 | input[(i + j >= length ? length - 1 : i + j)]) * |
413 | 0 | filter[j]; |
414 | 0 | } |
415 | 0 | sum >>= FILTER_BITS; |
416 | 0 | *optr++ = clip_pixel(sum); |
417 | 0 | } |
418 | 0 | } else { |
419 | | // Initial part. |
420 | 0 | for (i = 0; i < l1; i += 2) { |
421 | 0 | int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0]; |
422 | 0 | for (j = 1; j < filter_len_half; ++j) { |
423 | 0 | sum += (input[(i - j < 0 ? 0 : i - j)] + input[i + j]) * filter[j]; |
424 | 0 | } |
425 | 0 | sum >>= FILTER_BITS; |
426 | 0 | *optr++ = clip_pixel(sum); |
427 | 0 | } |
428 | | // Middle part. |
429 | 0 | for (; i < l2; i += 2) { |
430 | 0 | int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0]; |
431 | 0 | for (j = 1; j < filter_len_half; ++j) { |
432 | 0 | sum += (input[i - j] + input[i + j]) * filter[j]; |
433 | 0 | } |
434 | 0 | sum >>= FILTER_BITS; |
435 | 0 | *optr++ = clip_pixel(sum); |
436 | 0 | } |
437 | | // End part. |
438 | 0 | for (; i < length; i += 2) { |
439 | 0 | int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0]; |
440 | 0 | for (j = 1; j < filter_len_half; ++j) { |
441 | 0 | sum += (input[i - j] + input[(i + j >= length ? length - 1 : i + j)]) * |
442 | 0 | filter[j]; |
443 | 0 | } |
444 | 0 | sum >>= FILTER_BITS; |
445 | 0 | *optr++ = clip_pixel(sum); |
446 | 0 | } |
447 | 0 | } |
448 | 0 | } |
449 | | |
450 | 0 | static int get_down2_length(int length, int steps) { |
451 | 0 | for (int s = 0; s < steps; ++s) length = (length + 1) >> 1; |
452 | 0 | return length; |
453 | 0 | } |
454 | | |
455 | 0 | static int get_down2_steps(int in_length, int out_length) { |
456 | 0 | int steps = 0; |
457 | 0 | int proj_in_length; |
458 | 0 | while ((proj_in_length = get_down2_length(in_length, 1)) >= out_length) { |
459 | 0 | ++steps; |
460 | 0 | in_length = proj_in_length; |
461 | 0 | if (in_length == 1) { |
462 | | // Special case: we break because any further calls to get_down2_length() |
463 | | // with be with length == 1, which return 1, resulting in an infinite |
464 | | // loop. |
465 | 0 | break; |
466 | 0 | } |
467 | 0 | } |
468 | 0 | return steps; |
469 | 0 | } |
470 | | |
471 | | static void resize_multistep(const uint8_t *const input, int length, |
472 | 0 | uint8_t *output, int olength, uint8_t *otmp) { |
473 | 0 | if (length == olength) { |
474 | 0 | memcpy(output, input, sizeof(output[0]) * length); |
475 | 0 | return; |
476 | 0 | } |
477 | 0 | const int steps = get_down2_steps(length, olength); |
478 | |
|
479 | 0 | if (steps > 0) { |
480 | 0 | uint8_t *out = NULL; |
481 | 0 | int filteredlength = length; |
482 | |
|
483 | 0 | assert(otmp != NULL); |
484 | 0 | uint8_t *otmp2 = otmp + get_down2_length(length, 1); |
485 | 0 | for (int s = 0; s < steps; ++s) { |
486 | 0 | const int proj_filteredlength = get_down2_length(filteredlength, 1); |
487 | 0 | const uint8_t *const in = (s == 0 ? input : out); |
488 | 0 | if (s == steps - 1 && proj_filteredlength == olength) |
489 | 0 | out = output; |
490 | 0 | else |
491 | 0 | out = (s & 1 ? otmp2 : otmp); |
492 | 0 | if (filteredlength & 1) |
493 | 0 | down2_symodd(in, filteredlength, out); |
494 | 0 | else |
495 | 0 | down2_symeven(in, filteredlength, out, 0); |
496 | 0 | filteredlength = proj_filteredlength; |
497 | 0 | } |
498 | 0 | if (filteredlength != olength) { |
499 | 0 | interpolate(out, filteredlength, output, olength); |
500 | 0 | } |
501 | 0 | } else { |
502 | 0 | interpolate(input, length, output, olength); |
503 | 0 | } |
504 | 0 | } |
505 | | |
506 | 0 | static void fill_col_to_arr(uint8_t *img, int stride, int len, uint8_t *arr) { |
507 | 0 | int i; |
508 | 0 | uint8_t *iptr = img; |
509 | 0 | uint8_t *aptr = arr; |
510 | 0 | for (i = 0; i < len; ++i, iptr += stride) { |
511 | 0 | *aptr++ = *iptr; |
512 | 0 | } |
513 | 0 | } |
514 | | |
515 | 0 | static void fill_arr_to_col(uint8_t *img, int stride, int len, uint8_t *arr) { |
516 | 0 | int i; |
517 | 0 | uint8_t *iptr = img; |
518 | 0 | uint8_t *aptr = arr; |
519 | 0 | for (i = 0; i < len; ++i, iptr += stride) { |
520 | 0 | *iptr = *aptr++; |
521 | 0 | } |
522 | 0 | } |
523 | | |
524 | | bool av1_resize_vert_dir_c(uint8_t *intbuf, uint8_t *output, int out_stride, |
525 | 0 | int height, int height2, int width2, int start_col) { |
526 | 0 | bool mem_status = true; |
527 | 0 | uint8_t *arrbuf = (uint8_t *)aom_malloc(sizeof(*arrbuf) * height); |
528 | 0 | uint8_t *arrbuf2 = (uint8_t *)aom_malloc(sizeof(*arrbuf2) * height2); |
529 | 0 | if (arrbuf == NULL || arrbuf2 == NULL) { |
530 | 0 | mem_status = false; |
531 | 0 | goto Error; |
532 | 0 | } |
533 | | |
534 | 0 | for (int i = start_col; i < width2; ++i) { |
535 | 0 | fill_col_to_arr(intbuf + i, width2, height, arrbuf); |
536 | 0 | down2_symeven(arrbuf, height, arrbuf2, 0); |
537 | 0 | fill_arr_to_col(output + i, out_stride, height2, arrbuf2); |
538 | 0 | } |
539 | |
|
540 | 0 | Error: |
541 | 0 | aom_free(arrbuf); |
542 | 0 | aom_free(arrbuf2); |
543 | 0 | return mem_status; |
544 | 0 | } |
545 | | |
546 | | void av1_resize_horz_dir_c(const uint8_t *const input, int in_stride, |
547 | | uint8_t *intbuf, int height, int filtered_length, |
548 | 0 | int width2) { |
549 | 0 | for (int i = 0; i < height; ++i) |
550 | 0 | down2_symeven(input + in_stride * i, filtered_length, intbuf + width2 * i, |
551 | 0 | 0); |
552 | 0 | } |
553 | | |
554 | | bool av1_resize_plane_to_half(const uint8_t *const input, int height, int width, |
555 | | int in_stride, uint8_t *output, int height2, |
556 | 0 | int width2, int out_stride) { |
557 | 0 | uint8_t *intbuf = (uint8_t *)aom_malloc(sizeof(*intbuf) * width2 * height); |
558 | 0 | if (intbuf == NULL) { |
559 | 0 | return false; |
560 | 0 | } |
561 | | |
562 | | // Resize in the horizontal direction |
563 | 0 | av1_resize_horz_dir(input, in_stride, intbuf, height, width, width2); |
564 | | // Resize in the vertical direction |
565 | 0 | bool mem_status = av1_resize_vert_dir(intbuf, output, out_stride, height, |
566 | 0 | height2, width2, 0 /*start_col*/); |
567 | 0 | aom_free(intbuf); |
568 | 0 | return mem_status; |
569 | 0 | } |
570 | | |
571 | | // Check if both the output width and height are half of input width and |
572 | | // height respectively. |
573 | 0 | bool should_resize_by_half(int height, int width, int height2, int width2) { |
574 | 0 | const bool is_width_by_2 = get_down2_length(width, 1) == width2; |
575 | 0 | const bool is_height_by_2 = get_down2_length(height, 1) == height2; |
576 | 0 | return (is_width_by_2 && is_height_by_2); |
577 | 0 | } |
578 | | |
579 | | bool av1_resize_plane(const uint8_t *input, int height, int width, |
580 | | int in_stride, uint8_t *output, int height2, int width2, |
581 | 0 | int out_stride) { |
582 | 0 | int i; |
583 | 0 | bool mem_status = true; |
584 | 0 | uint8_t *intbuf = (uint8_t *)aom_malloc(sizeof(uint8_t) * width2 * height); |
585 | 0 | uint8_t *tmpbuf = |
586 | 0 | (uint8_t *)aom_malloc(sizeof(uint8_t) * AOMMAX(width, height)); |
587 | 0 | uint8_t *arrbuf = (uint8_t *)aom_malloc(sizeof(uint8_t) * height); |
588 | 0 | uint8_t *arrbuf2 = (uint8_t *)aom_malloc(sizeof(uint8_t) * height2); |
589 | 0 | if (intbuf == NULL || tmpbuf == NULL || arrbuf == NULL || arrbuf2 == NULL) { |
590 | 0 | mem_status = false; |
591 | 0 | goto Error; |
592 | 0 | } |
593 | 0 | assert(width > 0); |
594 | 0 | assert(height > 0); |
595 | 0 | assert(width2 > 0); |
596 | 0 | assert(height2 > 0); |
597 | 0 | for (i = 0; i < height; ++i) |
598 | 0 | resize_multistep(input + in_stride * i, width, intbuf + width2 * i, width2, |
599 | 0 | tmpbuf); |
600 | 0 | for (i = 0; i < width2; ++i) { |
601 | 0 | fill_col_to_arr(intbuf + i, width2, height, arrbuf); |
602 | 0 | resize_multistep(arrbuf, height, arrbuf2, height2, tmpbuf); |
603 | 0 | fill_arr_to_col(output + i, out_stride, height2, arrbuf2); |
604 | 0 | } |
605 | |
|
606 | 0 | Error: |
607 | 0 | aom_free(intbuf); |
608 | 0 | aom_free(tmpbuf); |
609 | 0 | aom_free(arrbuf); |
610 | 0 | aom_free(arrbuf2); |
611 | 0 | return mem_status; |
612 | 0 | } |
613 | | |
614 | | static bool upscale_normative_rect(const uint8_t *const input, int height, |
615 | | int width, int in_stride, uint8_t *output, |
616 | | int height2, int width2, int out_stride, |
617 | | int x_step_qn, int x0_qn, int pad_left, |
618 | 32.2k | int pad_right) { |
619 | 32.2k | assert(width > 0); |
620 | 32.2k | assert(height > 0); |
621 | 32.2k | assert(width2 > 0); |
622 | 32.2k | assert(height2 > 0); |
623 | 32.2k | assert(height2 == height); |
624 | | |
625 | | // Extend the left/right pixels of the tile column if needed |
626 | | // (either because we can't sample from other tiles, or because we're at |
627 | | // a frame edge). |
628 | | // Save the overwritten pixels into tmp_left and tmp_right. |
629 | | // Note: Because we pass input-1 to av1_convolve_horiz_rs, we need one extra |
630 | | // column of border pixels compared to what we'd naively think. |
631 | 32.2k | const int border_cols = UPSCALE_NORMATIVE_TAPS / 2 + 1; |
632 | 32.2k | uint8_t *tmp_left = |
633 | 32.2k | NULL; // Silence spurious "may be used uninitialized" warnings |
634 | 32.2k | uint8_t *tmp_right = NULL; |
635 | 32.2k | uint8_t *const in_tl = (uint8_t *)(input - border_cols); // Cast off 'const' |
636 | 32.2k | uint8_t *const in_tr = (uint8_t *)(input + width); |
637 | 32.2k | if (pad_left) { |
638 | 30.9k | tmp_left = (uint8_t *)aom_malloc(sizeof(*tmp_left) * border_cols * height); |
639 | 30.9k | if (!tmp_left) return false; |
640 | 869k | for (int i = 0; i < height; i++) { |
641 | 838k | memcpy(tmp_left + i * border_cols, in_tl + i * in_stride, border_cols); |
642 | 838k | memset(in_tl + i * in_stride, input[i * in_stride], border_cols); |
643 | 838k | } |
644 | 30.9k | } |
645 | 32.2k | if (pad_right) { |
646 | 30.9k | tmp_right = |
647 | 30.9k | (uint8_t *)aom_malloc(sizeof(*tmp_right) * border_cols * height); |
648 | 30.9k | if (!tmp_right) { |
649 | 0 | aom_free(tmp_left); |
650 | 0 | return false; |
651 | 0 | } |
652 | 869k | for (int i = 0; i < height; i++) { |
653 | 838k | memcpy(tmp_right + i * border_cols, in_tr + i * in_stride, border_cols); |
654 | 838k | memset(in_tr + i * in_stride, input[i * in_stride + width - 1], |
655 | 838k | border_cols); |
656 | 838k | } |
657 | 30.9k | } |
658 | | |
659 | 32.2k | av1_convolve_horiz_rs(input - 1, in_stride, output, out_stride, width2, |
660 | 32.2k | height2, &av1_resize_filter_normative[0][0], x0_qn, |
661 | 32.2k | x_step_qn); |
662 | | |
663 | | // Restore the left/right border pixels |
664 | 32.2k | if (pad_left) { |
665 | 869k | for (int i = 0; i < height; i++) { |
666 | 838k | memcpy(in_tl + i * in_stride, tmp_left + i * border_cols, border_cols); |
667 | 838k | } |
668 | 30.9k | aom_free(tmp_left); |
669 | 30.9k | } |
670 | 32.2k | if (pad_right) { |
671 | 869k | for (int i = 0; i < height; i++) { |
672 | 838k | memcpy(in_tr + i * in_stride, tmp_right + i * border_cols, border_cols); |
673 | 838k | } |
674 | 30.9k | aom_free(tmp_right); |
675 | 30.9k | } |
676 | 32.2k | return true; |
677 | 32.2k | } |
678 | | |
679 | | #if CONFIG_AV1_HIGHBITDEPTH |
680 | | static void highbd_interpolate_core(const uint16_t *const input, int in_length, |
681 | | uint16_t *output, int out_length, int bd, |
682 | | const int16_t *interp_filters, |
683 | 0 | int interp_taps) { |
684 | 0 | const int32_t delta = |
685 | 0 | (((uint32_t)in_length << RS_SCALE_SUBPEL_BITS) + out_length / 2) / |
686 | 0 | out_length; |
687 | 0 | const int32_t offset = |
688 | 0 | in_length > out_length |
689 | 0 | ? (((int32_t)(in_length - out_length) << (RS_SCALE_SUBPEL_BITS - 1)) + |
690 | 0 | out_length / 2) / |
691 | 0 | out_length |
692 | 0 | : -(((int32_t)(out_length - in_length) |
693 | 0 | << (RS_SCALE_SUBPEL_BITS - 1)) + |
694 | 0 | out_length / 2) / |
695 | 0 | out_length; |
696 | 0 | uint16_t *optr = output; |
697 | 0 | int x, x1, x2, sum, k, int_pel, sub_pel; |
698 | 0 | int32_t y; |
699 | |
|
700 | 0 | x = 0; |
701 | 0 | y = offset + RS_SCALE_EXTRA_OFF; |
702 | 0 | while ((y >> RS_SCALE_SUBPEL_BITS) < (interp_taps / 2 - 1)) { |
703 | 0 | x++; |
704 | 0 | y += delta; |
705 | 0 | } |
706 | 0 | x1 = x; |
707 | 0 | x = out_length - 1; |
708 | 0 | y = delta * x + offset + RS_SCALE_EXTRA_OFF; |
709 | 0 | while ((y >> RS_SCALE_SUBPEL_BITS) + (int32_t)(interp_taps / 2) >= |
710 | 0 | in_length) { |
711 | 0 | x--; |
712 | 0 | y -= delta; |
713 | 0 | } |
714 | 0 | x2 = x; |
715 | 0 | if (x1 > x2) { |
716 | 0 | for (x = 0, y = offset + RS_SCALE_EXTRA_OFF; x < out_length; |
717 | 0 | ++x, y += delta) { |
718 | 0 | int_pel = y >> RS_SCALE_SUBPEL_BITS; |
719 | 0 | sub_pel = (y >> RS_SCALE_EXTRA_BITS) & RS_SUBPEL_MASK; |
720 | 0 | const int16_t *filter = &interp_filters[sub_pel * interp_taps]; |
721 | 0 | sum = 0; |
722 | 0 | for (k = 0; k < interp_taps; ++k) { |
723 | 0 | const int pk = int_pel - interp_taps / 2 + 1 + k; |
724 | 0 | sum += filter[k] * input[AOMMAX(AOMMIN(pk, in_length - 1), 0)]; |
725 | 0 | } |
726 | 0 | *optr++ = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd); |
727 | 0 | } |
728 | 0 | } else { |
729 | | // Initial part. |
730 | 0 | for (x = 0, y = offset + RS_SCALE_EXTRA_OFF; x < x1; ++x, y += delta) { |
731 | 0 | int_pel = y >> RS_SCALE_SUBPEL_BITS; |
732 | 0 | sub_pel = (y >> RS_SCALE_EXTRA_BITS) & RS_SUBPEL_MASK; |
733 | 0 | const int16_t *filter = &interp_filters[sub_pel * interp_taps]; |
734 | 0 | sum = 0; |
735 | 0 | for (k = 0; k < interp_taps; ++k) |
736 | 0 | sum += filter[k] * input[AOMMAX(int_pel - interp_taps / 2 + 1 + k, 0)]; |
737 | 0 | *optr++ = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd); |
738 | 0 | } |
739 | | // Middle part. |
740 | 0 | for (; x <= x2; ++x, y += delta) { |
741 | 0 | int_pel = y >> RS_SCALE_SUBPEL_BITS; |
742 | 0 | sub_pel = (y >> RS_SCALE_EXTRA_BITS) & RS_SUBPEL_MASK; |
743 | 0 | const int16_t *filter = &interp_filters[sub_pel * interp_taps]; |
744 | 0 | sum = 0; |
745 | 0 | for (k = 0; k < interp_taps; ++k) |
746 | 0 | sum += filter[k] * input[int_pel - interp_taps / 2 + 1 + k]; |
747 | 0 | *optr++ = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd); |
748 | 0 | } |
749 | | // End part. |
750 | 0 | for (; x < out_length; ++x, y += delta) { |
751 | 0 | int_pel = y >> RS_SCALE_SUBPEL_BITS; |
752 | 0 | sub_pel = (y >> RS_SCALE_EXTRA_BITS) & RS_SUBPEL_MASK; |
753 | 0 | const int16_t *filter = &interp_filters[sub_pel * interp_taps]; |
754 | 0 | sum = 0; |
755 | 0 | for (k = 0; k < interp_taps; ++k) |
756 | 0 | sum += filter[k] * |
757 | 0 | input[AOMMIN(int_pel - interp_taps / 2 + 1 + k, in_length - 1)]; |
758 | 0 | *optr++ = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd); |
759 | 0 | } |
760 | 0 | } |
761 | 0 | } |
762 | | |
763 | | static void highbd_interpolate(const uint16_t *const input, int in_length, |
764 | 0 | uint16_t *output, int out_length, int bd) { |
765 | 0 | const InterpKernel *interp_filters = |
766 | 0 | choose_interp_filter(in_length, out_length); |
767 | |
|
768 | 0 | highbd_interpolate_core(input, in_length, output, out_length, bd, |
769 | 0 | &interp_filters[0][0], SUBPEL_TAPS); |
770 | 0 | } |
771 | | |
772 | | static void highbd_down2_symeven(const uint16_t *const input, int length, |
773 | 0 | uint16_t *output, int bd) { |
774 | | // Actual filter len = 2 * filter_len_half. |
775 | 0 | static const int16_t *filter = av1_down2_symeven_half_filter; |
776 | 0 | const int filter_len_half = sizeof(av1_down2_symeven_half_filter) / 2; |
777 | 0 | int i, j; |
778 | 0 | uint16_t *optr = output; |
779 | 0 | int l1 = filter_len_half; |
780 | 0 | int l2 = (length - filter_len_half); |
781 | 0 | l1 += (l1 & 1); |
782 | 0 | l2 += (l2 & 1); |
783 | 0 | if (l1 > l2) { |
784 | | // Short input length. |
785 | 0 | for (i = 0; i < length; i += 2) { |
786 | 0 | int sum = (1 << (FILTER_BITS - 1)); |
787 | 0 | for (j = 0; j < filter_len_half; ++j) { |
788 | 0 | sum += |
789 | 0 | (input[AOMMAX(0, i - j)] + input[AOMMIN(i + 1 + j, length - 1)]) * |
790 | 0 | filter[j]; |
791 | 0 | } |
792 | 0 | sum >>= FILTER_BITS; |
793 | 0 | *optr++ = clip_pixel_highbd(sum, bd); |
794 | 0 | } |
795 | 0 | } else { |
796 | | // Initial part. |
797 | 0 | for (i = 0; i < l1; i += 2) { |
798 | 0 | int sum = (1 << (FILTER_BITS - 1)); |
799 | 0 | for (j = 0; j < filter_len_half; ++j) { |
800 | 0 | sum += (input[AOMMAX(0, i - j)] + input[i + 1 + j]) * filter[j]; |
801 | 0 | } |
802 | 0 | sum >>= FILTER_BITS; |
803 | 0 | *optr++ = clip_pixel_highbd(sum, bd); |
804 | 0 | } |
805 | | // Middle part. |
806 | 0 | for (; i < l2; i += 2) { |
807 | 0 | int sum = (1 << (FILTER_BITS - 1)); |
808 | 0 | for (j = 0; j < filter_len_half; ++j) { |
809 | 0 | sum += (input[i - j] + input[i + 1 + j]) * filter[j]; |
810 | 0 | } |
811 | 0 | sum >>= FILTER_BITS; |
812 | 0 | *optr++ = clip_pixel_highbd(sum, bd); |
813 | 0 | } |
814 | | // End part. |
815 | 0 | for (; i < length; i += 2) { |
816 | 0 | int sum = (1 << (FILTER_BITS - 1)); |
817 | 0 | for (j = 0; j < filter_len_half; ++j) { |
818 | 0 | sum += |
819 | 0 | (input[i - j] + input[AOMMIN(i + 1 + j, length - 1)]) * filter[j]; |
820 | 0 | } |
821 | 0 | sum >>= FILTER_BITS; |
822 | 0 | *optr++ = clip_pixel_highbd(sum, bd); |
823 | 0 | } |
824 | 0 | } |
825 | 0 | } |
826 | | |
827 | | static void highbd_down2_symodd(const uint16_t *const input, int length, |
828 | 0 | uint16_t *output, int bd) { |
829 | | // Actual filter len = 2 * filter_len_half - 1. |
830 | 0 | static const int16_t *filter = av1_down2_symodd_half_filter; |
831 | 0 | const int filter_len_half = sizeof(av1_down2_symodd_half_filter) / 2; |
832 | 0 | int i, j; |
833 | 0 | uint16_t *optr = output; |
834 | 0 | int l1 = filter_len_half - 1; |
835 | 0 | int l2 = (length - filter_len_half + 1); |
836 | 0 | l1 += (l1 & 1); |
837 | 0 | l2 += (l2 & 1); |
838 | 0 | if (l1 > l2) { |
839 | | // Short input length. |
840 | 0 | for (i = 0; i < length; i += 2) { |
841 | 0 | int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0]; |
842 | 0 | for (j = 1; j < filter_len_half; ++j) { |
843 | 0 | sum += (input[AOMMAX(i - j, 0)] + input[AOMMIN(i + j, length - 1)]) * |
844 | 0 | filter[j]; |
845 | 0 | } |
846 | 0 | sum >>= FILTER_BITS; |
847 | 0 | *optr++ = clip_pixel_highbd(sum, bd); |
848 | 0 | } |
849 | 0 | } else { |
850 | | // Initial part. |
851 | 0 | for (i = 0; i < l1; i += 2) { |
852 | 0 | int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0]; |
853 | 0 | for (j = 1; j < filter_len_half; ++j) { |
854 | 0 | sum += (input[AOMMAX(i - j, 0)] + input[i + j]) * filter[j]; |
855 | 0 | } |
856 | 0 | sum >>= FILTER_BITS; |
857 | 0 | *optr++ = clip_pixel_highbd(sum, bd); |
858 | 0 | } |
859 | | // Middle part. |
860 | 0 | for (; i < l2; i += 2) { |
861 | 0 | int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0]; |
862 | 0 | for (j = 1; j < filter_len_half; ++j) { |
863 | 0 | sum += (input[i - j] + input[i + j]) * filter[j]; |
864 | 0 | } |
865 | 0 | sum >>= FILTER_BITS; |
866 | 0 | *optr++ = clip_pixel_highbd(sum, bd); |
867 | 0 | } |
868 | | // End part. |
869 | 0 | for (; i < length; i += 2) { |
870 | 0 | int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0]; |
871 | 0 | for (j = 1; j < filter_len_half; ++j) { |
872 | 0 | sum += (input[i - j] + input[AOMMIN(i + j, length - 1)]) * filter[j]; |
873 | 0 | } |
874 | 0 | sum >>= FILTER_BITS; |
875 | 0 | *optr++ = clip_pixel_highbd(sum, bd); |
876 | 0 | } |
877 | 0 | } |
878 | 0 | } |
879 | | |
880 | | static void highbd_resize_multistep(const uint16_t *const input, int length, |
881 | | uint16_t *output, int olength, |
882 | 0 | uint16_t *otmp, int bd) { |
883 | 0 | if (length == olength) { |
884 | 0 | memcpy(output, input, sizeof(output[0]) * length); |
885 | 0 | return; |
886 | 0 | } |
887 | 0 | const int steps = get_down2_steps(length, olength); |
888 | |
|
889 | 0 | if (steps > 0) { |
890 | 0 | uint16_t *out = NULL; |
891 | 0 | int filteredlength = length; |
892 | |
|
893 | 0 | assert(otmp != NULL); |
894 | 0 | uint16_t *otmp2 = otmp + get_down2_length(length, 1); |
895 | 0 | for (int s = 0; s < steps; ++s) { |
896 | 0 | const int proj_filteredlength = get_down2_length(filteredlength, 1); |
897 | 0 | const uint16_t *const in = (s == 0 ? input : out); |
898 | 0 | if (s == steps - 1 && proj_filteredlength == olength) |
899 | 0 | out = output; |
900 | 0 | else |
901 | 0 | out = (s & 1 ? otmp2 : otmp); |
902 | 0 | if (filteredlength & 1) |
903 | 0 | highbd_down2_symodd(in, filteredlength, out, bd); |
904 | 0 | else |
905 | 0 | highbd_down2_symeven(in, filteredlength, out, bd); |
906 | 0 | filteredlength = proj_filteredlength; |
907 | 0 | } |
908 | 0 | if (filteredlength != olength) { |
909 | 0 | highbd_interpolate(out, filteredlength, output, olength, bd); |
910 | 0 | } |
911 | 0 | } else { |
912 | 0 | highbd_interpolate(input, length, output, olength, bd); |
913 | 0 | } |
914 | 0 | } |
915 | | |
916 | | static void highbd_fill_col_to_arr(uint16_t *img, int stride, int len, |
917 | 0 | uint16_t *arr) { |
918 | 0 | int i; |
919 | 0 | uint16_t *iptr = img; |
920 | 0 | uint16_t *aptr = arr; |
921 | 0 | for (i = 0; i < len; ++i, iptr += stride) { |
922 | 0 | *aptr++ = *iptr; |
923 | 0 | } |
924 | 0 | } |
925 | | |
926 | | static void highbd_fill_arr_to_col(uint16_t *img, int stride, int len, |
927 | 0 | uint16_t *arr) { |
928 | 0 | int i; |
929 | 0 | uint16_t *iptr = img; |
930 | 0 | uint16_t *aptr = arr; |
931 | 0 | for (i = 0; i < len; ++i, iptr += stride) { |
932 | 0 | *iptr = *aptr++; |
933 | 0 | } |
934 | 0 | } |
935 | | |
936 | | void av1_highbd_resize_plane(const uint8_t *input, int height, int width, |
937 | | int in_stride, uint8_t *output, int height2, |
938 | 0 | int width2, int out_stride, int bd) { |
939 | 0 | int i; |
940 | 0 | uint16_t *intbuf = (uint16_t *)aom_malloc(sizeof(uint16_t) * width2 * height); |
941 | 0 | uint16_t *tmpbuf = |
942 | 0 | (uint16_t *)aom_malloc(sizeof(uint16_t) * AOMMAX(width, height)); |
943 | 0 | uint16_t *arrbuf = (uint16_t *)aom_malloc(sizeof(uint16_t) * height); |
944 | 0 | uint16_t *arrbuf2 = (uint16_t *)aom_malloc(sizeof(uint16_t) * height2); |
945 | 0 | if (intbuf == NULL || tmpbuf == NULL || arrbuf == NULL || arrbuf2 == NULL) |
946 | 0 | goto Error; |
947 | 0 | for (i = 0; i < height; ++i) { |
948 | 0 | highbd_resize_multistep(CONVERT_TO_SHORTPTR(input + in_stride * i), width, |
949 | 0 | intbuf + width2 * i, width2, tmpbuf, bd); |
950 | 0 | } |
951 | 0 | for (i = 0; i < width2; ++i) { |
952 | 0 | highbd_fill_col_to_arr(intbuf + i, width2, height, arrbuf); |
953 | 0 | highbd_resize_multistep(arrbuf, height, arrbuf2, height2, tmpbuf, bd); |
954 | 0 | highbd_fill_arr_to_col(CONVERT_TO_SHORTPTR(output + i), out_stride, height2, |
955 | 0 | arrbuf2); |
956 | 0 | } |
957 | |
|
958 | 0 | Error: |
959 | 0 | aom_free(intbuf); |
960 | 0 | aom_free(tmpbuf); |
961 | 0 | aom_free(arrbuf); |
962 | 0 | aom_free(arrbuf2); |
963 | 0 | } |
964 | | |
965 | | static bool highbd_upscale_normative_rect(const uint8_t *const input, |
966 | | int height, int width, int in_stride, |
967 | | uint8_t *output, int height2, |
968 | | int width2, int out_stride, |
969 | | int x_step_qn, int x0_qn, |
970 | 20.5k | int pad_left, int pad_right, int bd) { |
971 | 20.5k | assert(width > 0); |
972 | 20.5k | assert(height > 0); |
973 | 20.5k | assert(width2 > 0); |
974 | 20.5k | assert(height2 > 0); |
975 | 20.5k | assert(height2 == height); |
976 | | |
977 | | // Extend the left/right pixels of the tile column if needed |
978 | | // (either because we can't sample from other tiles, or because we're at |
979 | | // a frame edge). |
980 | | // Save the overwritten pixels into tmp_left and tmp_right. |
981 | | // Note: Because we pass input-1 to av1_convolve_horiz_rs, we need one extra |
982 | | // column of border pixels compared to what we'd naively think. |
983 | 20.5k | const int border_cols = UPSCALE_NORMATIVE_TAPS / 2 + 1; |
984 | 20.5k | const int border_size = border_cols * sizeof(uint16_t); |
985 | 20.5k | uint16_t *tmp_left = |
986 | 20.5k | NULL; // Silence spurious "may be used uninitialized" warnings |
987 | 20.5k | uint16_t *tmp_right = NULL; |
988 | 20.5k | uint16_t *const input16 = CONVERT_TO_SHORTPTR(input); |
989 | 20.5k | uint16_t *const in_tl = input16 - border_cols; |
990 | 20.5k | uint16_t *const in_tr = input16 + width; |
991 | 20.5k | if (pad_left) { |
992 | 18.9k | tmp_left = (uint16_t *)aom_malloc(sizeof(*tmp_left) * border_cols * height); |
993 | 18.9k | if (!tmp_left) return false; |
994 | 623k | for (int i = 0; i < height; i++) { |
995 | 604k | memcpy(tmp_left + i * border_cols, in_tl + i * in_stride, border_size); |
996 | 604k | aom_memset16(in_tl + i * in_stride, input16[i * in_stride], border_cols); |
997 | 604k | } |
998 | 18.9k | } |
999 | 20.5k | if (pad_right) { |
1000 | 18.9k | tmp_right = |
1001 | 18.9k | (uint16_t *)aom_malloc(sizeof(*tmp_right) * border_cols * height); |
1002 | 18.9k | if (!tmp_right) { |
1003 | 0 | aom_free(tmp_left); |
1004 | 0 | return false; |
1005 | 0 | } |
1006 | 623k | for (int i = 0; i < height; i++) { |
1007 | 604k | memcpy(tmp_right + i * border_cols, in_tr + i * in_stride, border_size); |
1008 | 604k | aom_memset16(in_tr + i * in_stride, input16[i * in_stride + width - 1], |
1009 | 604k | border_cols); |
1010 | 604k | } |
1011 | 18.9k | } |
1012 | | |
1013 | 20.5k | av1_highbd_convolve_horiz_rs(CONVERT_TO_SHORTPTR(input - 1), in_stride, |
1014 | 20.5k | CONVERT_TO_SHORTPTR(output), out_stride, width2, |
1015 | 20.5k | height2, &av1_resize_filter_normative[0][0], |
1016 | 20.5k | x0_qn, x_step_qn, bd); |
1017 | | |
1018 | | // Restore the left/right border pixels |
1019 | 20.5k | if (pad_left) { |
1020 | 623k | for (int i = 0; i < height; i++) { |
1021 | 604k | memcpy(in_tl + i * in_stride, tmp_left + i * border_cols, border_size); |
1022 | 604k | } |
1023 | 18.9k | aom_free(tmp_left); |
1024 | 18.9k | } |
1025 | 20.5k | if (pad_right) { |
1026 | 623k | for (int i = 0; i < height; i++) { |
1027 | 604k | memcpy(in_tr + i * in_stride, tmp_right + i * border_cols, border_size); |
1028 | 604k | } |
1029 | 18.9k | aom_free(tmp_right); |
1030 | 18.9k | } |
1031 | 20.5k | return true; |
1032 | 20.5k | } |
1033 | | #endif // CONFIG_AV1_HIGHBITDEPTH |
1034 | | |
1035 | | void av1_resize_frame420(const uint8_t *y, int y_stride, const uint8_t *u, |
1036 | | const uint8_t *v, int uv_stride, int height, int width, |
1037 | | uint8_t *oy, int oy_stride, uint8_t *ou, uint8_t *ov, |
1038 | 0 | int ouv_stride, int oheight, int owidth) { |
1039 | 0 | if (!av1_resize_plane(y, height, width, y_stride, oy, oheight, owidth, |
1040 | 0 | oy_stride)) |
1041 | 0 | abort(); |
1042 | 0 | if (!av1_resize_plane(u, height / 2, width / 2, uv_stride, ou, oheight / 2, |
1043 | 0 | owidth / 2, ouv_stride)) |
1044 | 0 | abort(); |
1045 | 0 | if (!av1_resize_plane(v, height / 2, width / 2, uv_stride, ov, oheight / 2, |
1046 | 0 | owidth / 2, ouv_stride)) |
1047 | 0 | abort(); |
1048 | 0 | } |
1049 | | |
1050 | | bool av1_resize_frame422(const uint8_t *y, int y_stride, const uint8_t *u, |
1051 | | const uint8_t *v, int uv_stride, int height, int width, |
1052 | | uint8_t *oy, int oy_stride, uint8_t *ou, uint8_t *ov, |
1053 | 0 | int ouv_stride, int oheight, int owidth) { |
1054 | 0 | if (!av1_resize_plane(y, height, width, y_stride, oy, oheight, owidth, |
1055 | 0 | oy_stride)) |
1056 | 0 | return false; |
1057 | 0 | if (!av1_resize_plane(u, height, width / 2, uv_stride, ou, oheight, |
1058 | 0 | owidth / 2, ouv_stride)) |
1059 | 0 | return false; |
1060 | 0 | if (!av1_resize_plane(v, height, width / 2, uv_stride, ov, oheight, |
1061 | 0 | owidth / 2, ouv_stride)) |
1062 | 0 | return false; |
1063 | 0 | return true; |
1064 | 0 | } |
1065 | | |
1066 | | bool av1_resize_frame444(const uint8_t *y, int y_stride, const uint8_t *u, |
1067 | | const uint8_t *v, int uv_stride, int height, int width, |
1068 | | uint8_t *oy, int oy_stride, uint8_t *ou, uint8_t *ov, |
1069 | 0 | int ouv_stride, int oheight, int owidth) { |
1070 | 0 | if (!av1_resize_plane(y, height, width, y_stride, oy, oheight, owidth, |
1071 | 0 | oy_stride)) |
1072 | 0 | return false; |
1073 | 0 | if (!av1_resize_plane(u, height, width, uv_stride, ou, oheight, owidth, |
1074 | 0 | ouv_stride)) |
1075 | 0 | return false; |
1076 | 0 | if (!av1_resize_plane(v, height, width, uv_stride, ov, oheight, owidth, |
1077 | 0 | ouv_stride)) |
1078 | 0 | return false; |
1079 | 0 | return true; |
1080 | 0 | } |
1081 | | |
1082 | | #if CONFIG_AV1_HIGHBITDEPTH |
1083 | | void av1_highbd_resize_frame420(const uint8_t *y, int y_stride, |
1084 | | const uint8_t *u, const uint8_t *v, |
1085 | | int uv_stride, int height, int width, |
1086 | | uint8_t *oy, int oy_stride, uint8_t *ou, |
1087 | | uint8_t *ov, int ouv_stride, int oheight, |
1088 | 0 | int owidth, int bd) { |
1089 | 0 | av1_highbd_resize_plane(y, height, width, y_stride, oy, oheight, owidth, |
1090 | 0 | oy_stride, bd); |
1091 | 0 | av1_highbd_resize_plane(u, height / 2, width / 2, uv_stride, ou, oheight / 2, |
1092 | 0 | owidth / 2, ouv_stride, bd); |
1093 | 0 | av1_highbd_resize_plane(v, height / 2, width / 2, uv_stride, ov, oheight / 2, |
1094 | 0 | owidth / 2, ouv_stride, bd); |
1095 | 0 | } |
1096 | | |
1097 | | void av1_highbd_resize_frame422(const uint8_t *y, int y_stride, |
1098 | | const uint8_t *u, const uint8_t *v, |
1099 | | int uv_stride, int height, int width, |
1100 | | uint8_t *oy, int oy_stride, uint8_t *ou, |
1101 | | uint8_t *ov, int ouv_stride, int oheight, |
1102 | 0 | int owidth, int bd) { |
1103 | 0 | av1_highbd_resize_plane(y, height, width, y_stride, oy, oheight, owidth, |
1104 | 0 | oy_stride, bd); |
1105 | 0 | av1_highbd_resize_plane(u, height, width / 2, uv_stride, ou, oheight, |
1106 | 0 | owidth / 2, ouv_stride, bd); |
1107 | 0 | av1_highbd_resize_plane(v, height, width / 2, uv_stride, ov, oheight, |
1108 | 0 | owidth / 2, ouv_stride, bd); |
1109 | 0 | } |
1110 | | |
1111 | | void av1_highbd_resize_frame444(const uint8_t *y, int y_stride, |
1112 | | const uint8_t *u, const uint8_t *v, |
1113 | | int uv_stride, int height, int width, |
1114 | | uint8_t *oy, int oy_stride, uint8_t *ou, |
1115 | | uint8_t *ov, int ouv_stride, int oheight, |
1116 | 0 | int owidth, int bd) { |
1117 | 0 | av1_highbd_resize_plane(y, height, width, y_stride, oy, oheight, owidth, |
1118 | 0 | oy_stride, bd); |
1119 | 0 | av1_highbd_resize_plane(u, height, width, uv_stride, ou, oheight, owidth, |
1120 | 0 | ouv_stride, bd); |
1121 | 0 | av1_highbd_resize_plane(v, height, width, uv_stride, ov, oheight, owidth, |
1122 | 0 | ouv_stride, bd); |
1123 | 0 | } |
1124 | | #endif // CONFIG_AV1_HIGHBITDEPTH |
1125 | | |
1126 | | void av1_resize_and_extend_frame_c(const YV12_BUFFER_CONFIG *src, |
1127 | | YV12_BUFFER_CONFIG *dst, |
1128 | | const InterpFilter filter, |
1129 | | const int phase_scaler, |
1130 | 0 | const int num_planes) { |
1131 | 0 | assert(filter == BILINEAR || filter == EIGHTTAP_SMOOTH || |
1132 | 0 | filter == EIGHTTAP_REGULAR); |
1133 | 0 | const InterpKernel *const kernel = |
1134 | 0 | (const InterpKernel *)av1_interp_filter_params_list[filter].filter_ptr; |
1135 | |
|
1136 | 0 | for (int i = 0; i < AOMMIN(num_planes, MAX_MB_PLANE); ++i) { |
1137 | 0 | const int is_uv = i > 0; |
1138 | 0 | const int src_w = src->crop_widths[is_uv]; |
1139 | 0 | const int src_h = src->crop_heights[is_uv]; |
1140 | 0 | const uint8_t *src_buffer = src->buffers[i]; |
1141 | 0 | const int src_stride = src->strides[is_uv]; |
1142 | 0 | const int dst_w = dst->crop_widths[is_uv]; |
1143 | 0 | const int dst_h = dst->crop_heights[is_uv]; |
1144 | 0 | uint8_t *dst_buffer = dst->buffers[i]; |
1145 | 0 | const int dst_stride = dst->strides[is_uv]; |
1146 | 0 | for (int y = 0; y < dst_h; y += 16) { |
1147 | 0 | const int y_q4 = |
1148 | 0 | src_h == dst_h ? 0 : y * 16 * src_h / dst_h + phase_scaler; |
1149 | 0 | for (int x = 0; x < dst_w; x += 16) { |
1150 | 0 | const int x_q4 = |
1151 | 0 | src_w == dst_w ? 0 : x * 16 * src_w / dst_w + phase_scaler; |
1152 | 0 | const uint8_t *src_ptr = |
1153 | 0 | src_buffer + y * src_h / dst_h * src_stride + x * src_w / dst_w; |
1154 | 0 | uint8_t *dst_ptr = dst_buffer + y * dst_stride + x; |
1155 | | |
1156 | | // Width and height of the actual working area. |
1157 | 0 | const int work_w = AOMMIN(16, dst_w - x); |
1158 | 0 | const int work_h = AOMMIN(16, dst_h - y); |
1159 | | // SIMD versions of aom_scaled_2d() have some trouble handling |
1160 | | // nonstandard sizes, so fall back on the C version to handle borders. |
1161 | 0 | if (work_w != 16 || work_h != 16) { |
1162 | 0 | aom_scaled_2d_c(src_ptr, src_stride, dst_ptr, dst_stride, kernel, |
1163 | 0 | x_q4 & 0xf, 16 * src_w / dst_w, y_q4 & 0xf, |
1164 | 0 | 16 * src_h / dst_h, work_w, work_h); |
1165 | 0 | } else { |
1166 | 0 | aom_scaled_2d(src_ptr, src_stride, dst_ptr, dst_stride, kernel, |
1167 | 0 | x_q4 & 0xf, 16 * src_w / dst_w, y_q4 & 0xf, |
1168 | 0 | 16 * src_h / dst_h, 16, 16); |
1169 | 0 | } |
1170 | 0 | } |
1171 | 0 | } |
1172 | 0 | } |
1173 | 0 | aom_extend_frame_borders(dst, num_planes); |
1174 | 0 | } |
1175 | | |
1176 | | bool av1_resize_and_extend_frame_nonnormative(const YV12_BUFFER_CONFIG *src, |
1177 | | YV12_BUFFER_CONFIG *dst, int bd, |
1178 | 0 | int num_planes) { |
1179 | | // TODO(dkovalev): replace YV12_BUFFER_CONFIG with aom_image_t |
1180 | | |
1181 | | // We use AOMMIN(num_planes, MAX_MB_PLANE) instead of num_planes to quiet |
1182 | | // the static analysis warnings. |
1183 | 0 | for (int i = 0; i < AOMMIN(num_planes, MAX_MB_PLANE); ++i) { |
1184 | 0 | const int is_uv = i > 0; |
1185 | 0 | #if CONFIG_AV1_HIGHBITDEPTH |
1186 | 0 | if (src->flags & YV12_FLAG_HIGHBITDEPTH) { |
1187 | 0 | av1_highbd_resize_plane(src->buffers[i], src->crop_heights[is_uv], |
1188 | 0 | src->crop_widths[is_uv], src->strides[is_uv], |
1189 | 0 | dst->buffers[i], dst->crop_heights[is_uv], |
1190 | 0 | dst->crop_widths[is_uv], dst->strides[is_uv], bd); |
1191 | 0 | } else if (!av1_resize_plane(src->buffers[i], src->crop_heights[is_uv], |
1192 | 0 | src->crop_widths[is_uv], src->strides[is_uv], |
1193 | 0 | dst->buffers[i], dst->crop_heights[is_uv], |
1194 | 0 | dst->crop_widths[is_uv], |
1195 | 0 | dst->strides[is_uv])) { |
1196 | 0 | return false; |
1197 | 0 | } |
1198 | | #else |
1199 | | (void)bd; |
1200 | | if (!av1_resize_plane(src->buffers[i], src->crop_heights[is_uv], |
1201 | | src->crop_widths[is_uv], src->strides[is_uv], |
1202 | | dst->buffers[i], dst->crop_heights[is_uv], |
1203 | | dst->crop_widths[is_uv], dst->strides[is_uv])) |
1204 | | return false; |
1205 | | #endif |
1206 | 0 | } |
1207 | 0 | aom_extend_frame_borders(dst, num_planes); |
1208 | 0 | return true; |
1209 | 0 | } |
1210 | | |
1211 | | void av1_upscale_normative_rows(const AV1_COMMON *cm, const uint8_t *src, |
1212 | | int src_stride, uint8_t *dst, int dst_stride, |
1213 | 49.8k | int plane, int rows) { |
1214 | 49.8k | const int is_uv = (plane > 0); |
1215 | 49.8k | const int ss_x = is_uv && cm->seq_params->subsampling_x; |
1216 | 49.8k | const int downscaled_plane_width = ROUND_POWER_OF_TWO(cm->width, ss_x); |
1217 | 49.8k | const int upscaled_plane_width = |
1218 | 49.8k | ROUND_POWER_OF_TWO(cm->superres_upscaled_width, ss_x); |
1219 | 49.8k | const int superres_denom = cm->superres_scale_denominator; |
1220 | | |
1221 | 49.8k | TileInfo tile_col; |
1222 | 49.8k | const int32_t x_step_qn = av1_get_upscale_convolve_step( |
1223 | 49.8k | downscaled_plane_width, upscaled_plane_width); |
1224 | 49.8k | int32_t x0_qn = get_upscale_convolve_x0(downscaled_plane_width, |
1225 | 49.8k | upscaled_plane_width, x_step_qn); |
1226 | | |
1227 | 102k | for (int j = 0; j < cm->tiles.cols; j++) { |
1228 | 52.8k | av1_tile_set_col(&tile_col, cm, j); |
1229 | | // Determine the limits of this tile column in both the source |
1230 | | // and destination images. |
1231 | | // Note: The actual location which we start sampling from is |
1232 | | // (downscaled_x0 - 1 + (x0_qn/2^14)), and this quantity increases |
1233 | | // by exactly dst_width * (x_step_qn/2^14) pixels each iteration. |
1234 | 52.8k | const int downscaled_x0 = tile_col.mi_col_start << (MI_SIZE_LOG2 - ss_x); |
1235 | 52.8k | const int downscaled_x1 = tile_col.mi_col_end << (MI_SIZE_LOG2 - ss_x); |
1236 | 52.8k | const int src_width = downscaled_x1 - downscaled_x0; |
1237 | | |
1238 | 52.8k | const int upscaled_x0 = (downscaled_x0 * superres_denom) / SCALE_NUMERATOR; |
1239 | 52.8k | int upscaled_x1; |
1240 | 52.8k | if (j == cm->tiles.cols - 1) { |
1241 | | // Note that we can't just use AOMMIN here - due to rounding, |
1242 | | // (downscaled_x1 * superres_denom) / SCALE_NUMERATOR may be less than |
1243 | | // upscaled_plane_width. |
1244 | 49.8k | upscaled_x1 = upscaled_plane_width; |
1245 | 49.8k | } else { |
1246 | 2.94k | upscaled_x1 = (downscaled_x1 * superres_denom) / SCALE_NUMERATOR; |
1247 | 2.94k | } |
1248 | | |
1249 | 52.8k | const uint8_t *const src_ptr = src + downscaled_x0; |
1250 | 52.8k | uint8_t *const dst_ptr = dst + upscaled_x0; |
1251 | 52.8k | const int dst_width = upscaled_x1 - upscaled_x0; |
1252 | | |
1253 | 52.8k | const int pad_left = (j == 0); |
1254 | 52.8k | const int pad_right = (j == cm->tiles.cols - 1); |
1255 | | |
1256 | 52.8k | bool success; |
1257 | 52.8k | #if CONFIG_AV1_HIGHBITDEPTH |
1258 | 52.8k | if (cm->seq_params->use_highbitdepth) |
1259 | 20.5k | success = highbd_upscale_normative_rect( |
1260 | 20.5k | src_ptr, rows, src_width, src_stride, dst_ptr, rows, dst_width, |
1261 | 20.5k | dst_stride, x_step_qn, x0_qn, pad_left, pad_right, |
1262 | 20.5k | cm->seq_params->bit_depth); |
1263 | 32.2k | else |
1264 | 32.2k | success = upscale_normative_rect(src_ptr, rows, src_width, src_stride, |
1265 | 32.2k | dst_ptr, rows, dst_width, dst_stride, |
1266 | 32.2k | x_step_qn, x0_qn, pad_left, pad_right); |
1267 | | #else |
1268 | | success = upscale_normative_rect(src_ptr, rows, src_width, src_stride, |
1269 | | dst_ptr, rows, dst_width, dst_stride, |
1270 | | x_step_qn, x0_qn, pad_left, pad_right); |
1271 | | #endif |
1272 | 52.8k | if (!success) { |
1273 | 0 | aom_internal_error(cm->error, AOM_CODEC_MEM_ERROR, |
1274 | 0 | "Error upscaling frame"); |
1275 | 0 | } |
1276 | | // Update the fractional pixel offset to prepare for the next tile column. |
1277 | 52.8k | x0_qn += (dst_width * x_step_qn) - (src_width << RS_SCALE_SUBPEL_BITS); |
1278 | 52.8k | } |
1279 | 49.8k | } |
1280 | | |
1281 | | void av1_upscale_normative_and_extend_frame(const AV1_COMMON *cm, |
1282 | | const YV12_BUFFER_CONFIG *src, |
1283 | 10.3k | YV12_BUFFER_CONFIG *dst) { |
1284 | 10.3k | const int num_planes = av1_num_planes(cm); |
1285 | 38.6k | for (int i = 0; i < num_planes; ++i) { |
1286 | 28.3k | const int is_uv = (i > 0); |
1287 | 28.3k | av1_upscale_normative_rows(cm, src->buffers[i], src->strides[is_uv], |
1288 | 28.3k | dst->buffers[i], dst->strides[is_uv], i, |
1289 | 28.3k | src->crop_heights[is_uv]); |
1290 | 28.3k | } |
1291 | | |
1292 | 10.3k | aom_extend_frame_borders(dst, num_planes); |
1293 | 10.3k | } |
1294 | | |
1295 | | YV12_BUFFER_CONFIG *av1_realloc_and_scale_if_required( |
1296 | | AV1_COMMON *cm, YV12_BUFFER_CONFIG *unscaled, YV12_BUFFER_CONFIG *scaled, |
1297 | | const InterpFilter filter, const int phase, const bool use_optimized_scaler, |
1298 | 0 | const bool for_psnr, const int border_in_pixels, const bool alloc_pyramid) { |
1299 | | // If scaling is performed for the sole purpose of calculating PSNR, then our |
1300 | | // target dimensions are superres upscaled width/height. Otherwise our target |
1301 | | // dimensions are coded width/height. |
1302 | 0 | const int scaled_width = for_psnr ? cm->superres_upscaled_width : cm->width; |
1303 | 0 | const int scaled_height = |
1304 | 0 | for_psnr ? cm->superres_upscaled_height : cm->height; |
1305 | 0 | const bool scaling_required = (scaled_width != unscaled->y_crop_width) || |
1306 | 0 | (scaled_height != unscaled->y_crop_height); |
1307 | |
|
1308 | 0 | if (scaling_required) { |
1309 | 0 | const int num_planes = av1_num_planes(cm); |
1310 | 0 | const SequenceHeader *seq_params = cm->seq_params; |
1311 | | |
1312 | | // Reallocate the frame buffer based on the target dimensions when scaling |
1313 | | // is required. |
1314 | 0 | if (aom_realloc_frame_buffer( |
1315 | 0 | scaled, scaled_width, scaled_height, seq_params->subsampling_x, |
1316 | 0 | seq_params->subsampling_y, seq_params->use_highbitdepth, |
1317 | 0 | border_in_pixels, cm->features.byte_alignment, NULL, NULL, NULL, |
1318 | 0 | alloc_pyramid, 0)) |
1319 | 0 | aom_internal_error(cm->error, AOM_CODEC_MEM_ERROR, |
1320 | 0 | "Failed to allocate scaled buffer"); |
1321 | |
|
1322 | 0 | bool has_optimized_scaler = av1_has_optimized_scaler( |
1323 | 0 | unscaled->y_crop_width, unscaled->y_crop_height, scaled_width, |
1324 | 0 | scaled_height); |
1325 | 0 | if (num_planes > 1) { |
1326 | 0 | has_optimized_scaler = has_optimized_scaler && |
1327 | 0 | av1_has_optimized_scaler(unscaled->uv_crop_width, |
1328 | 0 | unscaled->uv_crop_height, |
1329 | 0 | scaled->uv_crop_width, |
1330 | 0 | scaled->uv_crop_height); |
1331 | 0 | } |
1332 | |
|
1333 | 0 | #if CONFIG_AV1_HIGHBITDEPTH |
1334 | 0 | if (use_optimized_scaler && has_optimized_scaler && |
1335 | 0 | cm->seq_params->bit_depth == AOM_BITS_8) { |
1336 | 0 | av1_resize_and_extend_frame(unscaled, scaled, filter, phase, num_planes); |
1337 | 0 | } else { |
1338 | 0 | if (!av1_resize_and_extend_frame_nonnormative( |
1339 | 0 | unscaled, scaled, (int)cm->seq_params->bit_depth, num_planes)) |
1340 | 0 | aom_internal_error(cm->error, AOM_CODEC_MEM_ERROR, |
1341 | 0 | "Failed to allocate buffers during resize"); |
1342 | 0 | } |
1343 | | #else |
1344 | | if (use_optimized_scaler && has_optimized_scaler) { |
1345 | | av1_resize_and_extend_frame(unscaled, scaled, filter, phase, num_planes); |
1346 | | } else { |
1347 | | if (!av1_resize_and_extend_frame_nonnormative( |
1348 | | unscaled, scaled, (int)cm->seq_params->bit_depth, num_planes)) |
1349 | | aom_internal_error(cm->error, AOM_CODEC_MEM_ERROR, |
1350 | | "Failed to allocate buffers during resize"); |
1351 | | } |
1352 | | #endif |
1353 | 0 | return scaled; |
1354 | 0 | } |
1355 | 0 | return unscaled; |
1356 | 0 | } |
1357 | | |
1358 | | // Calculates the scaled dimension given the original dimension and the scale |
1359 | | // denominator. |
1360 | 22.0k | static void calculate_scaled_size_helper(int *dim, int denom) { |
1361 | 22.0k | if (denom != SCALE_NUMERATOR) { |
1362 | | // We need to ensure the constraint in "Appendix A" of the spec: |
1363 | | // * FrameWidth is greater than or equal to 16 |
1364 | | // * FrameHeight is greater than or equal to 16 |
1365 | | // For this, we clamp the downscaled dimension to at least 16. One |
1366 | | // exception: if original dimension itself was < 16, then we keep the |
1367 | | // downscaled dimension to be same as the original, to ensure that resizing |
1368 | | // is valid. |
1369 | 22.0k | const int min_dim = AOMMIN(16, *dim); |
1370 | | // Use this version if we need *dim to be even |
1371 | | // *width = (*width * SCALE_NUMERATOR + denom) / (2 * denom); |
1372 | | // *width <<= 1; |
1373 | 22.0k | *dim = (*dim * SCALE_NUMERATOR + denom / 2) / (denom); |
1374 | 22.0k | *dim = AOMMAX(*dim, min_dim); |
1375 | 22.0k | } |
1376 | 22.0k | } |
1377 | | |
1378 | 0 | void av1_calculate_scaled_size(int *width, int *height, int resize_denom) { |
1379 | 0 | calculate_scaled_size_helper(width, resize_denom); |
1380 | 0 | calculate_scaled_size_helper(height, resize_denom); |
1381 | 0 | } |
1382 | | |
1383 | | void av1_calculate_scaled_superres_size(int *width, int *height, |
1384 | 22.0k | int superres_denom) { |
1385 | 22.0k | (void)height; |
1386 | 22.0k | calculate_scaled_size_helper(width, superres_denom); |
1387 | 22.0k | } |
1388 | | |
1389 | 0 | void av1_calculate_unscaled_superres_size(int *width, int *height, int denom) { |
1390 | 0 | if (denom != SCALE_NUMERATOR) { |
1391 | | // Note: av1_calculate_scaled_superres_size() rounds *up* after division |
1392 | | // when the resulting dimensions are odd. So here, we round *down*. |
1393 | 0 | *width = *width * denom / SCALE_NUMERATOR; |
1394 | 0 | (void)height; |
1395 | 0 | } |
1396 | 0 | } |
1397 | | |
1398 | | // Copy only the config data from 'src' to 'dst'. |
1399 | | static void copy_buffer_config(const YV12_BUFFER_CONFIG *const src, |
1400 | 0 | YV12_BUFFER_CONFIG *const dst) { |
1401 | 0 | dst->bit_depth = src->bit_depth; |
1402 | 0 | dst->color_primaries = src->color_primaries; |
1403 | 0 | dst->transfer_characteristics = src->transfer_characteristics; |
1404 | 0 | dst->matrix_coefficients = src->matrix_coefficients; |
1405 | 0 | dst->monochrome = src->monochrome; |
1406 | 0 | dst->chroma_sample_position = src->chroma_sample_position; |
1407 | 0 | dst->color_range = src->color_range; |
1408 | 0 | } |
1409 | | |
1410 | | // TODO(afergs): Look for in-place upscaling |
1411 | | // TODO(afergs): aom_ vs av1_ functions? Which can I use? |
1412 | | // Upscale decoded image. |
1413 | | void av1_superres_upscale(AV1_COMMON *cm, BufferPool *const pool, |
1414 | 10.3k | bool alloc_pyramid) { |
1415 | 10.3k | const int num_planes = av1_num_planes(cm); |
1416 | 10.3k | if (!av1_superres_scaled(cm)) return; |
1417 | 10.3k | const SequenceHeader *const seq_params = cm->seq_params; |
1418 | 10.3k | const int byte_alignment = cm->features.byte_alignment; |
1419 | | |
1420 | 10.3k | YV12_BUFFER_CONFIG copy_buffer; |
1421 | 10.3k | memset(©_buffer, 0, sizeof(copy_buffer)); |
1422 | | |
1423 | 10.3k | YV12_BUFFER_CONFIG *const frame_to_show = &cm->cur_frame->buf; |
1424 | | |
1425 | 10.3k | const int aligned_width = ALIGN_POWER_OF_TWO(cm->width, 3); |
1426 | 10.3k | if (aom_alloc_frame_buffer( |
1427 | 10.3k | ©_buffer, aligned_width, cm->height, seq_params->subsampling_x, |
1428 | 10.3k | seq_params->subsampling_y, seq_params->use_highbitdepth, |
1429 | 10.3k | AOM_BORDER_IN_PIXELS, byte_alignment, false, 0)) |
1430 | 0 | aom_internal_error(cm->error, AOM_CODEC_MEM_ERROR, |
1431 | 0 | "Failed to allocate copy buffer for superres upscaling"); |
1432 | | |
1433 | | // Copy function assumes the frames are the same size. |
1434 | | // Note that it does not copy YV12_BUFFER_CONFIG config data. |
1435 | 10.3k | aom_yv12_copy_frame(frame_to_show, ©_buffer, num_planes); |
1436 | | |
1437 | 10.3k | assert(copy_buffer.y_crop_width == aligned_width); |
1438 | 10.3k | assert(copy_buffer.y_crop_height == cm->height); |
1439 | | |
1440 | | // Realloc the current frame buffer at a higher resolution in place. |
1441 | 10.3k | if (pool != NULL) { |
1442 | | // Use callbacks if on the decoder. |
1443 | 10.3k | aom_codec_frame_buffer_t *fb = &cm->cur_frame->raw_frame_buffer; |
1444 | 10.3k | aom_release_frame_buffer_cb_fn_t release_fb_cb = pool->release_fb_cb; |
1445 | 10.3k | aom_get_frame_buffer_cb_fn_t cb = pool->get_fb_cb; |
1446 | 10.3k | void *cb_priv = pool->cb_priv; |
1447 | | |
1448 | 10.3k | lock_buffer_pool(pool); |
1449 | | // Realloc with callback does not release the frame buffer - release first. |
1450 | 10.3k | if (release_fb_cb(cb_priv, fb)) { |
1451 | 0 | unlock_buffer_pool(pool); |
1452 | 0 | aom_internal_error( |
1453 | 0 | cm->error, AOM_CODEC_MEM_ERROR, |
1454 | 0 | "Failed to free current frame buffer before superres upscaling"); |
1455 | 0 | } |
1456 | | // aom_realloc_frame_buffer() leaves config data for frame_to_show intact |
1457 | 10.3k | if (aom_realloc_frame_buffer( |
1458 | 10.3k | frame_to_show, cm->superres_upscaled_width, |
1459 | 10.3k | cm->superres_upscaled_height, seq_params->subsampling_x, |
1460 | 10.3k | seq_params->subsampling_y, seq_params->use_highbitdepth, |
1461 | 10.3k | AOM_BORDER_IN_PIXELS, byte_alignment, fb, cb, cb_priv, |
1462 | 10.3k | alloc_pyramid, 0)) { |
1463 | 0 | unlock_buffer_pool(pool); |
1464 | 0 | aom_internal_error( |
1465 | 0 | cm->error, AOM_CODEC_MEM_ERROR, |
1466 | 0 | "Failed to allocate current frame buffer for superres upscaling"); |
1467 | 0 | } |
1468 | 10.3k | unlock_buffer_pool(pool); |
1469 | 10.3k | } else { |
1470 | | // Make a copy of the config data for frame_to_show in copy_buffer |
1471 | 0 | copy_buffer_config(frame_to_show, ©_buffer); |
1472 | | |
1473 | | // Don't use callbacks on the encoder. |
1474 | | // aom_alloc_frame_buffer() clears the config data for frame_to_show |
1475 | 0 | if (aom_alloc_frame_buffer( |
1476 | 0 | frame_to_show, cm->superres_upscaled_width, |
1477 | 0 | cm->superres_upscaled_height, seq_params->subsampling_x, |
1478 | 0 | seq_params->subsampling_y, seq_params->use_highbitdepth, |
1479 | 0 | AOM_BORDER_IN_PIXELS, byte_alignment, alloc_pyramid, 0)) |
1480 | 0 | aom_internal_error( |
1481 | 0 | cm->error, AOM_CODEC_MEM_ERROR, |
1482 | 0 | "Failed to reallocate current frame buffer for superres upscaling"); |
1483 | | |
1484 | | // Restore config data back to frame_to_show |
1485 | 0 | copy_buffer_config(©_buffer, frame_to_show); |
1486 | 0 | } |
1487 | | // TODO(afergs): verify frame_to_show is correct after realloc |
1488 | | // encoder: |
1489 | | // decoder: |
1490 | | |
1491 | 10.3k | assert(frame_to_show->y_crop_width == cm->superres_upscaled_width); |
1492 | 10.3k | assert(frame_to_show->y_crop_height == cm->superres_upscaled_height); |
1493 | | |
1494 | | // Scale up and back into frame_to_show. |
1495 | 10.3k | assert(frame_to_show->y_crop_width != cm->width); |
1496 | 10.3k | av1_upscale_normative_and_extend_frame(cm, ©_buffer, frame_to_show); |
1497 | | |
1498 | | // Free the copy buffer |
1499 | 10.3k | aom_free_frame_buffer(©_buffer); |
1500 | 10.3k | } |