/work/svt-av1/Source/Lib/Codec/warped_motion.c
Line | Count | Source |
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 https://www.aomedia.org/license/software-license. 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 https://www.aomedia.org/license/patent-license. |
10 | | */ |
11 | | |
12 | | #include <stdio.h> |
13 | | #include <stdlib.h> |
14 | | #include <assert.h> |
15 | | #include "common_utils.h" |
16 | | #include "aom_dsp_rtcd.h" |
17 | | #include "warped_motion.h" |
18 | | #include "bitstream_unit.h" |
19 | | #include "definitions.h" |
20 | | #include "convolve.h" |
21 | | |
22 | | #define LS_MV_MAX 256 // max mv in 1/8-pel |
23 | | // Use LS_STEP = 8 so that 2 less bits needed for A, bx, by. |
24 | | #define LS_STEP 8 |
25 | | |
26 | | // Assuming LS_MV_MAX is < MAX_SB_SIZE * 8, |
27 | | // the precision needed is: |
28 | | // (MAX_SB_SIZE_LOG2 + 3) [for sx * sx magnitude] + |
29 | | // (MAX_SB_SIZE_LOG2 + 4) [for sx * dx magnitude] + |
30 | | // 1 [for sign] + |
31 | | // LEAST_SQUARES_SAMPLES_MAX_BITS |
32 | | // [for adding up to LEAST_SQUARES_SAMPLES_MAX samples] |
33 | | // The value is 23 |
34 | | #define LS_MAT_RANGE_BITS ((MAX_SB_SIZE_LOG2 + 4) * 2 + LEAST_SQUARES_SAMPLES_MAX_BITS) |
35 | | |
36 | | // Bit-depth reduction from the full-range |
37 | | #define LS_MAT_DOWN_BITS 2 |
38 | | |
39 | | // bits range of A, bx and by after downshifting |
40 | | #define LS_MAT_BITS (LS_MAT_RANGE_BITS - LS_MAT_DOWN_BITS) |
41 | | #define LS_MAT_MIN (-(1 << (LS_MAT_BITS - 1))) |
42 | | #define LS_MAT_MAX ((1 << (LS_MAT_BITS - 1)) - 1) |
43 | | |
44 | | // by setting LS_STEP = 8, the least 2 bits of every elements in A, bx, by are |
45 | | // 0. So, we can reduce LS_MAT_RANGE_BITS(2) bits here. |
46 | | #define LS_SQUARE(a) (((a) * (a) * 4 + (a) * 4 * LS_STEP + LS_STEP * LS_STEP * 2) >> (2 + LS_MAT_DOWN_BITS)) |
47 | | #define LS_PRODUCT1(a, b) (((a) * (b) * 4 + ((a) + (b)) * 2 * LS_STEP + LS_STEP * LS_STEP) >> (2 + LS_MAT_DOWN_BITS)) |
48 | | #define LS_PRODUCT2(a, b) \ |
49 | | (((a) * (b) * 4 + ((a) + (b)) * 2 * LS_STEP + LS_STEP * LS_STEP * 2) >> (2 + LS_MAT_DOWN_BITS)) |
50 | | |
51 | | // For warping, we really use a 6-tap filter, but we do blocks of 8 pixels |
52 | | // at a time. The zoom/rotation/shear in the model are applied to the |
53 | | // "fractional" position of each pixel, which therefore varies within |
54 | | // [-1, 2) * WARPEDPIXEL_PREC_SHIFTS. |
55 | | // We need an extra 2 taps to fit this in, for a total of 8 taps. |
56 | | /* clang-format off */ |
57 | | EB_ALIGN(16) const int16_t svt_aom_warped_filter[WARPEDPIXEL_PREC_SHIFTS * 3 + 1][8] = { |
58 | | #if WARPEDPIXEL_PREC_BITS == 6 |
59 | | // [-1, 0) |
60 | | { 0, 0, 127, 1, 0, 0, 0, 0 }, { 0, - 1, 127, 2, 0, 0, 0, 0 }, |
61 | | { 1, - 3, 127, 4, - 1, 0, 0, 0 }, { 1, - 4, 126, 6, - 2, 1, 0, 0 }, |
62 | | { 1, - 5, 126, 8, - 3, 1, 0, 0 }, { 1, - 6, 125, 11, - 4, 1, 0, 0 }, |
63 | | { 1, - 7, 124, 13, - 4, 1, 0, 0 }, { 2, - 8, 123, 15, - 5, 1, 0, 0 }, |
64 | | { 2, - 9, 122, 18, - 6, 1, 0, 0 }, { 2, -10, 121, 20, - 6, 1, 0, 0 }, |
65 | | { 2, -11, 120, 22, - 7, 2, 0, 0 }, { 2, -12, 119, 25, - 8, 2, 0, 0 }, |
66 | | { 3, -13, 117, 27, - 8, 2, 0, 0 }, { 3, -13, 116, 29, - 9, 2, 0, 0 }, |
67 | | { 3, -14, 114, 32, -10, 3, 0, 0 }, { 3, -15, 113, 35, -10, 2, 0, 0 }, |
68 | | { 3, -15, 111, 37, -11, 3, 0, 0 }, { 3, -16, 109, 40, -11, 3, 0, 0 }, |
69 | | { 3, -16, 108, 42, -12, 3, 0, 0 }, { 4, -17, 106, 45, -13, 3, 0, 0 }, |
70 | | { 4, -17, 104, 47, -13, 3, 0, 0 }, { 4, -17, 102, 50, -14, 3, 0, 0 }, |
71 | | { 4, -17, 100, 52, -14, 3, 0, 0 }, { 4, -18, 98, 55, -15, 4, 0, 0 }, |
72 | | { 4, -18, 96, 58, -15, 3, 0, 0 }, { 4, -18, 94, 60, -16, 4, 0, 0 }, |
73 | | { 4, -18, 91, 63, -16, 4, 0, 0 }, { 4, -18, 89, 65, -16, 4, 0, 0 }, |
74 | | { 4, -18, 87, 68, -17, 4, 0, 0 }, { 4, -18, 85, 70, -17, 4, 0, 0 }, |
75 | | { 4, -18, 82, 73, -17, 4, 0, 0 }, { 4, -18, 80, 75, -17, 4, 0, 0 }, |
76 | | { 4, -18, 78, 78, -18, 4, 0, 0 }, { 4, -17, 75, 80, -18, 4, 0, 0 }, |
77 | | { 4, -17, 73, 82, -18, 4, 0, 0 }, { 4, -17, 70, 85, -18, 4, 0, 0 }, |
78 | | { 4, -17, 68, 87, -18, 4, 0, 0 }, { 4, -16, 65, 89, -18, 4, 0, 0 }, |
79 | | { 4, -16, 63, 91, -18, 4, 0, 0 }, { 4, -16, 60, 94, -18, 4, 0, 0 }, |
80 | | { 3, -15, 58, 96, -18, 4, 0, 0 }, { 4, -15, 55, 98, -18, 4, 0, 0 }, |
81 | | { 3, -14, 52, 100, -17, 4, 0, 0 }, { 3, -14, 50, 102, -17, 4, 0, 0 }, |
82 | | { 3, -13, 47, 104, -17, 4, 0, 0 }, { 3, -13, 45, 106, -17, 4, 0, 0 }, |
83 | | { 3, -12, 42, 108, -16, 3, 0, 0 }, { 3, -11, 40, 109, -16, 3, 0, 0 }, |
84 | | { 3, -11, 37, 111, -15, 3, 0, 0 }, { 2, -10, 35, 113, -15, 3, 0, 0 }, |
85 | | { 3, -10, 32, 114, -14, 3, 0, 0 }, { 2, - 9, 29, 116, -13, 3, 0, 0 }, |
86 | | { 2, - 8, 27, 117, -13, 3, 0, 0 }, { 2, - 8, 25, 119, -12, 2, 0, 0 }, |
87 | | { 2, - 7, 22, 120, -11, 2, 0, 0 }, { 1, - 6, 20, 121, -10, 2, 0, 0 }, |
88 | | { 1, - 6, 18, 122, - 9, 2, 0, 0 }, { 1, - 5, 15, 123, - 8, 2, 0, 0 }, |
89 | | { 1, - 4, 13, 124, - 7, 1, 0, 0 }, { 1, - 4, 11, 125, - 6, 1, 0, 0 }, |
90 | | { 1, - 3, 8, 126, - 5, 1, 0, 0 }, { 1, - 2, 6, 126, - 4, 1, 0, 0 }, |
91 | | { 0, - 1, 4, 127, - 3, 1, 0, 0 }, { 0, 0, 2, 127, - 1, 0, 0, 0 }, |
92 | | // [0, 1) |
93 | | { 0, 0, 0, 127, 1, 0, 0, 0}, { 0, 0, -1, 127, 2, 0, 0, 0}, |
94 | | { 0, 1, -3, 127, 4, -2, 1, 0}, { 0, 1, -5, 127, 6, -2, 1, 0}, |
95 | | { 0, 2, -6, 126, 8, -3, 1, 0}, {-1, 2, -7, 126, 11, -4, 2, -1}, |
96 | | {-1, 3, -8, 125, 13, -5, 2, -1}, {-1, 3, -10, 124, 16, -6, 3, -1}, |
97 | | {-1, 4, -11, 123, 18, -7, 3, -1}, {-1, 4, -12, 122, 20, -7, 3, -1}, |
98 | | {-1, 4, -13, 121, 23, -8, 3, -1}, {-2, 5, -14, 120, 25, -9, 4, -1}, |
99 | | {-1, 5, -15, 119, 27, -10, 4, -1}, {-1, 5, -16, 118, 30, -11, 4, -1}, |
100 | | {-2, 6, -17, 116, 33, -12, 5, -1}, {-2, 6, -17, 114, 35, -12, 5, -1}, |
101 | | {-2, 6, -18, 113, 38, -13, 5, -1}, {-2, 7, -19, 111, 41, -14, 6, -2}, |
102 | | {-2, 7, -19, 110, 43, -15, 6, -2}, {-2, 7, -20, 108, 46, -15, 6, -2}, |
103 | | {-2, 7, -20, 106, 49, -16, 6, -2}, {-2, 7, -21, 104, 51, -16, 7, -2}, |
104 | | {-2, 7, -21, 102, 54, -17, 7, -2}, {-2, 8, -21, 100, 56, -18, 7, -2}, |
105 | | {-2, 8, -22, 98, 59, -18, 7, -2}, {-2, 8, -22, 96, 62, -19, 7, -2}, |
106 | | {-2, 8, -22, 94, 64, -19, 7, -2}, {-2, 8, -22, 91, 67, -20, 8, -2}, |
107 | | {-2, 8, -22, 89, 69, -20, 8, -2}, {-2, 8, -22, 87, 72, -21, 8, -2}, |
108 | | {-2, 8, -21, 84, 74, -21, 8, -2}, {-2, 8, -22, 82, 77, -21, 8, -2}, |
109 | | {-2, 8, -21, 79, 79, -21, 8, -2}, {-2, 8, -21, 77, 82, -22, 8, -2}, |
110 | | {-2, 8, -21, 74, 84, -21, 8, -2}, {-2, 8, -21, 72, 87, -22, 8, -2}, |
111 | | {-2, 8, -20, 69, 89, -22, 8, -2}, {-2, 8, -20, 67, 91, -22, 8, -2}, |
112 | | {-2, 7, -19, 64, 94, -22, 8, -2}, {-2, 7, -19, 62, 96, -22, 8, -2}, |
113 | | {-2, 7, -18, 59, 98, -22, 8, -2}, {-2, 7, -18, 56, 100, -21, 8, -2}, |
114 | | {-2, 7, -17, 54, 102, -21, 7, -2}, {-2, 7, -16, 51, 104, -21, 7, -2}, |
115 | | {-2, 6, -16, 49, 106, -20, 7, -2}, {-2, 6, -15, 46, 108, -20, 7, -2}, |
116 | | {-2, 6, -15, 43, 110, -19, 7, -2}, {-2, 6, -14, 41, 111, -19, 7, -2}, |
117 | | {-1, 5, -13, 38, 113, -18, 6, -2}, {-1, 5, -12, 35, 114, -17, 6, -2}, |
118 | | {-1, 5, -12, 33, 116, -17, 6, -2}, {-1, 4, -11, 30, 118, -16, 5, -1}, |
119 | | {-1, 4, -10, 27, 119, -15, 5, -1}, {-1, 4, -9, 25, 120, -14, 5, -2}, |
120 | | {-1, 3, -8, 23, 121, -13, 4, -1}, {-1, 3, -7, 20, 122, -12, 4, -1}, |
121 | | {-1, 3, -7, 18, 123, -11, 4, -1}, {-1, 3, -6, 16, 124, -10, 3, -1}, |
122 | | {-1, 2, -5, 13, 125, -8, 3, -1}, {-1, 2, -4, 11, 126, -7, 2, -1}, |
123 | | { 0, 1, -3, 8, 126, -6, 2, 0}, { 0, 1, -2, 6, 127, -5, 1, 0}, |
124 | | { 0, 1, -2, 4, 127, -3, 1, 0}, { 0, 0, 0, 2, 127, -1, 0, 0}, |
125 | | // [1, 2) |
126 | | { 0, 0, 0, 1, 127, 0, 0, 0 }, { 0, 0, 0, - 1, 127, 2, 0, 0 }, |
127 | | { 0, 0, 1, - 3, 127, 4, - 1, 0 }, { 0, 0, 1, - 4, 126, 6, - 2, 1 }, |
128 | | { 0, 0, 1, - 5, 126, 8, - 3, 1 }, { 0, 0, 1, - 6, 125, 11, - 4, 1 }, |
129 | | { 0, 0, 1, - 7, 124, 13, - 4, 1 }, { 0, 0, 2, - 8, 123, 15, - 5, 1 }, |
130 | | { 0, 0, 2, - 9, 122, 18, - 6, 1 }, { 0, 0, 2, -10, 121, 20, - 6, 1 }, |
131 | | { 0, 0, 2, -11, 120, 22, - 7, 2 }, { 0, 0, 2, -12, 119, 25, - 8, 2 }, |
132 | | { 0, 0, 3, -13, 117, 27, - 8, 2 }, { 0, 0, 3, -13, 116, 29, - 9, 2 }, |
133 | | { 0, 0, 3, -14, 114, 32, -10, 3 }, { 0, 0, 3, -15, 113, 35, -10, 2 }, |
134 | | { 0, 0, 3, -15, 111, 37, -11, 3 }, { 0, 0, 3, -16, 109, 40, -11, 3 }, |
135 | | { 0, 0, 3, -16, 108, 42, -12, 3 }, { 0, 0, 4, -17, 106, 45, -13, 3 }, |
136 | | { 0, 0, 4, -17, 104, 47, -13, 3 }, { 0, 0, 4, -17, 102, 50, -14, 3 }, |
137 | | { 0, 0, 4, -17, 100, 52, -14, 3 }, { 0, 0, 4, -18, 98, 55, -15, 4 }, |
138 | | { 0, 0, 4, -18, 96, 58, -15, 3 }, { 0, 0, 4, -18, 94, 60, -16, 4 }, |
139 | | { 0, 0, 4, -18, 91, 63, -16, 4 }, { 0, 0, 4, -18, 89, 65, -16, 4 }, |
140 | | { 0, 0, 4, -18, 87, 68, -17, 4 }, { 0, 0, 4, -18, 85, 70, -17, 4 }, |
141 | | { 0, 0, 4, -18, 82, 73, -17, 4 }, { 0, 0, 4, -18, 80, 75, -17, 4 }, |
142 | | { 0, 0, 4, -18, 78, 78, -18, 4 }, { 0, 0, 4, -17, 75, 80, -18, 4 }, |
143 | | { 0, 0, 4, -17, 73, 82, -18, 4 }, { 0, 0, 4, -17, 70, 85, -18, 4 }, |
144 | | { 0, 0, 4, -17, 68, 87, -18, 4 }, { 0, 0, 4, -16, 65, 89, -18, 4 }, |
145 | | { 0, 0, 4, -16, 63, 91, -18, 4 }, { 0, 0, 4, -16, 60, 94, -18, 4 }, |
146 | | { 0, 0, 3, -15, 58, 96, -18, 4 }, { 0, 0, 4, -15, 55, 98, -18, 4 }, |
147 | | { 0, 0, 3, -14, 52, 100, -17, 4 }, { 0, 0, 3, -14, 50, 102, -17, 4 }, |
148 | | { 0, 0, 3, -13, 47, 104, -17, 4 }, { 0, 0, 3, -13, 45, 106, -17, 4 }, |
149 | | { 0, 0, 3, -12, 42, 108, -16, 3 }, { 0, 0, 3, -11, 40, 109, -16, 3 }, |
150 | | { 0, 0, 3, -11, 37, 111, -15, 3 }, { 0, 0, 2, -10, 35, 113, -15, 3 }, |
151 | | { 0, 0, 3, -10, 32, 114, -14, 3 }, { 0, 0, 2, - 9, 29, 116, -13, 3 }, |
152 | | { 0, 0, 2, - 8, 27, 117, -13, 3 }, { 0, 0, 2, - 8, 25, 119, -12, 2 }, |
153 | | { 0, 0, 2, - 7, 22, 120, -11, 2 }, { 0, 0, 1, - 6, 20, 121, -10, 2 }, |
154 | | { 0, 0, 1, - 6, 18, 122, - 9, 2 }, { 0, 0, 1, - 5, 15, 123, - 8, 2 }, |
155 | | { 0, 0, 1, - 4, 13, 124, - 7, 1 }, { 0, 0, 1, - 4, 11, 125, - 6, 1 }, |
156 | | { 0, 0, 1, - 3, 8, 126, - 5, 1 }, { 0, 0, 1, - 2, 6, 126, - 4, 1 }, |
157 | | { 0, 0, 0, - 1, 4, 127, - 3, 1 }, { 0, 0, 0, 0, 2, 127, - 1, 0 }, |
158 | | // dummy (replicate row index 191) |
159 | | { 0, 0, 0, 0, 2, 127, - 1, 0 }, |
160 | | #elif WARPEDPIXEL_PREC_BITS == 5 |
161 | | // [-1, 0) |
162 | | {0, 0, 127, 1, 0, 0, 0, 0}, {1, -3, 127, 4, -1, 0, 0, 0}, |
163 | | {1, -5, 126, 8, -3, 1, 0, 0}, {1, -7, 124, 13, -4, 1, 0, 0}, |
164 | | {2, -9, 122, 18, -6, 1, 0, 0}, {2, -11, 120, 22, -7, 2, 0, 0}, |
165 | | {3, -13, 117, 27, -8, 2, 0, 0}, {3, -14, 114, 32, -10, 3, 0, 0}, |
166 | | {3, -15, 111, 37, -11, 3, 0, 0}, {3, -16, 108, 42, -12, 3, 0, 0}, |
167 | | {4, -17, 104, 47, -13, 3, 0, 0}, {4, -17, 100, 52, -14, 3, 0, 0}, |
168 | | {4, -18, 96, 58, -15, 3, 0, 0}, {4, -18, 91, 63, -16, 4, 0, 0}, |
169 | | {4, -18, 87, 68, -17, 4, 0, 0}, {4, -18, 82, 73, -17, 4, 0, 0}, |
170 | | {4, -18, 78, 78, -18, 4, 0, 0}, {4, -17, 73, 82, -18, 4, 0, 0}, |
171 | | {4, -17, 68, 87, -18, 4, 0, 0}, {4, -16, 63, 91, -18, 4, 0, 0}, |
172 | | {3, -15, 58, 96, -18, 4, 0, 0}, {3, -14, 52, 100, -17, 4, 0, 0}, |
173 | | {3, -13, 47, 104, -17, 4, 0, 0}, {3, -12, 42, 108, -16, 3, 0, 0}, |
174 | | {3, -11, 37, 111, -15, 3, 0, 0}, {3, -10, 32, 114, -14, 3, 0, 0}, |
175 | | {2, -8, 27, 117, -13, 3, 0, 0}, {2, -7, 22, 120, -11, 2, 0, 0}, |
176 | | {1, -6, 18, 122, -9, 2, 0, 0}, {1, -4, 13, 124, -7, 1, 0, 0}, |
177 | | {1, -3, 8, 126, -5, 1, 0, 0}, {0, -1, 4, 127, -3, 1, 0, 0}, |
178 | | // [0, 1) |
179 | | { 0, 0, 0, 127, 1, 0, 0, 0}, { 0, 1, -3, 127, 4, -2, 1, 0}, |
180 | | { 0, 2, -6, 126, 8, -3, 1, 0}, {-1, 3, -8, 125, 13, -5, 2, -1}, |
181 | | {-1, 4, -11, 123, 18, -7, 3, -1}, {-1, 4, -13, 121, 23, -8, 3, -1}, |
182 | | {-1, 5, -15, 119, 27, -10, 4, -1}, {-2, 6, -17, 116, 33, -12, 5, -1}, |
183 | | {-2, 6, -18, 113, 38, -13, 5, -1}, {-2, 7, -19, 110, 43, -15, 6, -2}, |
184 | | {-2, 7, -20, 106, 49, -16, 6, -2}, {-2, 7, -21, 102, 54, -17, 7, -2}, |
185 | | {-2, 8, -22, 98, 59, -18, 7, -2}, {-2, 8, -22, 94, 64, -19, 7, -2}, |
186 | | {-2, 8, -22, 89, 69, -20, 8, -2}, {-2, 8, -21, 84, 74, -21, 8, -2}, |
187 | | {-2, 8, -21, 79, 79, -21, 8, -2}, {-2, 8, -21, 74, 84, -21, 8, -2}, |
188 | | {-2, 8, -20, 69, 89, -22, 8, -2}, {-2, 7, -19, 64, 94, -22, 8, -2}, |
189 | | {-2, 7, -18, 59, 98, -22, 8, -2}, {-2, 7, -17, 54, 102, -21, 7, -2}, |
190 | | {-2, 6, -16, 49, 106, -20, 7, -2}, {-2, 6, -15, 43, 110, -19, 7, -2}, |
191 | | {-1, 5, -13, 38, 113, -18, 6, -2}, {-1, 5, -12, 33, 116, -17, 6, -2}, |
192 | | {-1, 4, -10, 27, 119, -15, 5, -1}, {-1, 3, -8, 23, 121, -13, 4, -1}, |
193 | | {-1, 3, -7, 18, 123, -11, 4, -1}, {-1, 2, -5, 13, 125, -8, 3, -1}, |
194 | | { 0, 1, -3, 8, 126, -6, 2, 0}, { 0, 1, -2, 4, 127, -3, 1, 0}, |
195 | | // [1, 2) |
196 | | {0, 0, 0, 1, 127, 0, 0, 0}, {0, 0, 1, -3, 127, 4, -1, 0}, |
197 | | {0, 0, 1, -5, 126, 8, -3, 1}, {0, 0, 1, -7, 124, 13, -4, 1}, |
198 | | {0, 0, 2, -9, 122, 18, -6, 1}, {0, 0, 2, -11, 120, 22, -7, 2}, |
199 | | {0, 0, 3, -13, 117, 27, -8, 2}, {0, 0, 3, -14, 114, 32, -10, 3}, |
200 | | {0, 0, 3, -15, 111, 37, -11, 3}, {0, 0, 3, -16, 108, 42, -12, 3}, |
201 | | {0, 0, 4, -17, 104, 47, -13, 3}, {0, 0, 4, -17, 100, 52, -14, 3}, |
202 | | {0, 0, 4, -18, 96, 58, -15, 3}, {0, 0, 4, -18, 91, 63, -16, 4}, |
203 | | {0, 0, 4, -18, 87, 68, -17, 4}, {0, 0, 4, -18, 82, 73, -17, 4}, |
204 | | {0, 0, 4, -18, 78, 78, -18, 4}, {0, 0, 4, -17, 73, 82, -18, 4}, |
205 | | {0, 0, 4, -17, 68, 87, -18, 4}, {0, 0, 4, -16, 63, 91, -18, 4}, |
206 | | {0, 0, 3, -15, 58, 96, -18, 4}, {0, 0, 3, -14, 52, 100, -17, 4}, |
207 | | {0, 0, 3, -13, 47, 104, -17, 4}, {0, 0, 3, -12, 42, 108, -16, 3}, |
208 | | {0, 0, 3, -11, 37, 111, -15, 3}, {0, 0, 3, -10, 32, 114, -14, 3}, |
209 | | {0, 0, 2, -8, 27, 117, -13, 3}, {0, 0, 2, -7, 22, 120, -11, 2}, |
210 | | {0, 0, 1, -6, 18, 122, -9, 2}, {0, 0, 1, -4, 13, 124, -7, 1}, |
211 | | {0, 0, 1, -3, 8, 126, -5, 1}, {0, 0, 0, -1, 4, 127, -3, 1}, |
212 | | // dummy (replicate row index 95) |
213 | | {0, 0, 0, -1, 4, 127, -3, 1}, |
214 | | #endif // WARPEDPIXEL_PREC_BITS == 6 |
215 | | }; |
216 | | /* clang-format on */ |
217 | | |
218 | | #define USE_LIMITED_PREC_MULT 0 |
219 | | |
220 | | #if USE_LIMITED_PREC_MULT |
221 | | |
222 | | #define MUL_PREC_BITS 16 |
223 | | |
224 | | static uint16_t resolve_multiplier_64(uint64_t D, int16_t* shift) { |
225 | | int msb = 0; |
226 | | uint16_t mult = 0; |
227 | | *shift = 0; |
228 | | if (D != 0) { |
229 | | msb = (int16_t)((D >> 32) ? get_msb((unsigned int)(D >> 32)) + 32 : get_msb((unsigned int)D)); |
230 | | if (msb >= MUL_PREC_BITS) { |
231 | | mult = (uint16_t)ROUND_POWER_OF_TWO_64(D, msb + 1 - MUL_PREC_BITS); |
232 | | *shift = msb + 1 - MUL_PREC_BITS; |
233 | | } else { |
234 | | mult = (uint16_t)D; |
235 | | *shift = 0; |
236 | | } |
237 | | } |
238 | | return mult; |
239 | | } |
240 | | |
241 | | static int32_t get_mult_shift_ndiag(int64_t p_x, int16_t i_det, int shift) { |
242 | | int32_t ret; |
243 | | int16_t mshift; |
244 | | uint16_t Mul = resolve_multiplier_64(llabs(p_x), &mshift); |
245 | | int32_t v = (int32_t)Mul * (int32_t)i_det * (p_x < 0 ? -1 : 1); |
246 | | shift -= mshift; |
247 | | if (shift > 0) { |
248 | | return (int32_t)clamp(ROUND_POWER_OF_TWO_SIGNED(v, shift), |
249 | | -WARPEDMODEL_NONDIAGAFFINE_CLAMP + 1, |
250 | | WARPEDMODEL_NONDIAGAFFINE_CLAMP - 1); |
251 | | } else { |
252 | | return (int32_t)clamp( |
253 | | v * (1 << (-shift)), -WARPEDMODEL_NONDIAGAFFINE_CLAMP + 1, WARPEDMODEL_NONDIAGAFFINE_CLAMP - 1); |
254 | | } |
255 | | return ret; |
256 | | } |
257 | | |
258 | | static int32_t get_mult_shift_diag(int64_t p_x, int16_t i_det, int shift) { |
259 | | int16_t mshift; |
260 | | uint16_t Mul = resolve_multiplier_64(llabs(p_x), &mshift); |
261 | | int32_t v = (int32_t)Mul * (int32_t)i_det * (p_x < 0 ? -1 : 1); |
262 | | shift -= mshift; |
263 | | if (shift > 0) { |
264 | | return (int32_t)clamp(ROUND_POWER_OF_TWO_SIGNED(v, shift), |
265 | | (1 << WARPEDMODEL_PREC_BITS) - WARPEDMODEL_NONDIAGAFFINE_CLAMP + 1, |
266 | | (1 << WARPEDMODEL_PREC_BITS) + WARPEDMODEL_NONDIAGAFFINE_CLAMP - 1); |
267 | | } else { |
268 | | return (int32_t)clamp(v * (1 << (-shift)), |
269 | | (1 << WARPEDMODEL_PREC_BITS) - WARPEDMODEL_NONDIAGAFFINE_CLAMP + 1, |
270 | | (1 << WARPEDMODEL_PREC_BITS) + WARPEDMODEL_NONDIAGAFFINE_CLAMP - 1); |
271 | | } |
272 | | } |
273 | | |
274 | | #else |
275 | | |
276 | | static int32_t get_mult_shift_ndiag(int64_t p_x, int16_t i_det, int shift) { |
277 | | int64_t v = p_x * (int64_t)i_det; |
278 | | return (int32_t)clamp64(ROUND_POWER_OF_TWO_SIGNED_64(v, shift), |
279 | | -WARPEDMODEL_NONDIAGAFFINE_CLAMP + 1, |
280 | | WARPEDMODEL_NONDIAGAFFINE_CLAMP - 1); |
281 | | } |
282 | | |
283 | | static int32_t get_mult_shift_diag(int64_t p_x, int16_t i_det, int shift) { |
284 | | int64_t v = p_x * (int64_t)i_det; |
285 | | return (int32_t)clamp64(ROUND_POWER_OF_TWO_SIGNED_64(v, shift), |
286 | | (1 << WARPEDMODEL_PREC_BITS) - WARPEDMODEL_NONDIAGAFFINE_CLAMP + 1, |
287 | | (1 << WARPEDMODEL_PREC_BITS) + WARPEDMODEL_NONDIAGAFFINE_CLAMP - 1); |
288 | | } |
289 | | #endif // USE_LIMITED_PREC_MULT |
290 | | |
291 | | #define DIV_LUT_PREC_BITS 14 |
292 | | #define DIV_LUT_BITS 8 |
293 | | #define DIV_LUT_NUM (1 << DIV_LUT_BITS) |
294 | | |
295 | | static const uint16_t div_lut[DIV_LUT_NUM + 1] = { |
296 | | 16384, 16320, 16257, 16194, 16132, 16070, 16009, 15948, 15888, 15828, 15768, 15709, 15650, 15592, 15534, 15477, |
297 | | 15420, 15364, 15308, 15252, 15197, 15142, 15087, 15033, 14980, 14926, 14873, 14821, 14769, 14717, 14665, 14614, |
298 | | 14564, 14513, 14463, 14413, 14364, 14315, 14266, 14218, 14170, 14122, 14075, 14028, 13981, 13935, 13888, 13843, |
299 | | 13797, 13752, 13707, 13662, 13618, 13574, 13530, 13487, 13443, 13400, 13358, 13315, 13273, 13231, 13190, 13148, |
300 | | 13107, 13066, 13026, 12985, 12945, 12906, 12866, 12827, 12788, 12749, 12710, 12672, 12633, 12596, 12558, 12520, |
301 | | 12483, 12446, 12409, 12373, 12336, 12300, 12264, 12228, 12193, 12157, 12122, 12087, 12053, 12018, 11984, 11950, |
302 | | 11916, 11882, 11848, 11815, 11782, 11749, 11716, 11683, 11651, 11619, 11586, 11555, 11523, 11491, 11460, 11429, |
303 | | 11398, 11367, 11336, 11305, 11275, 11245, 11215, 11185, 11155, 11125, 11096, 11067, 11038, 11009, 10980, 10951, |
304 | | 10923, 10894, 10866, 10838, 10810, 10782, 10755, 10727, 10700, 10673, 10645, 10618, 10592, 10565, 10538, 10512, |
305 | | 10486, 10460, 10434, 10408, 10382, 10356, 10331, 10305, 10280, 10255, 10230, 10205, 10180, 10156, 10131, 10107, |
306 | | 10082, 10058, 10034, 10010, 9986, 9963, 9939, 9916, 9892, 9869, 9846, 9823, 9800, 9777, 9754, 9732, |
307 | | 9709, 9687, 9664, 9642, 9620, 9598, 9576, 9554, 9533, 9511, 9489, 9468, 9447, 9425, 9404, 9383, |
308 | | 9362, 9341, 9321, 9300, 9279, 9259, 9239, 9218, 9198, 9178, 9158, 9138, 9118, 9098, 9079, 9059, |
309 | | 9039, 9020, 9001, 8981, 8962, 8943, 8924, 8905, 8886, 8867, 8849, 8830, 8812, 8793, 8775, 8756, |
310 | | 8738, 8720, 8702, 8684, 8666, 8648, 8630, 8613, 8595, 8577, 8560, 8542, 8525, 8508, 8490, 8473, |
311 | | 8456, 8439, 8422, 8405, 8389, 8372, 8355, 8339, 8322, 8306, 8289, 8273, 8257, 8240, 8224, 8208, |
312 | | 8192, |
313 | | }; |
314 | | |
315 | | // Decomposes a divisor D such that 1/D = y/2^shift, where y is returned |
316 | | // at precision of DIV_LUT_PREC_BITS along with the shift. |
317 | | static int16_t resolve_divisor_64(uint64_t D, int16_t* shift) { |
318 | | int64_t f; |
319 | | *shift = (int16_t)((D >> 32) ? get_msb((unsigned int)(D >> 32)) + 32 : get_msb((unsigned int)D)); |
320 | | // e is obtained from D after resetting the most significant 1 bit. |
321 | | const int64_t e = D - ((uint64_t)1 << *shift); |
322 | | // Get the most significant DIV_LUT_BITS (8) bits of e into f |
323 | | if (*shift > DIV_LUT_BITS) { |
324 | | f = ROUND_POWER_OF_TWO_64(e, *shift - DIV_LUT_BITS); |
325 | | } else { |
326 | | f = e << (DIV_LUT_BITS - *shift); |
327 | | } |
328 | | assert(f <= DIV_LUT_NUM); |
329 | | *shift += DIV_LUT_PREC_BITS; |
330 | | // Use f as lookup into the precomputed table of multipliers |
331 | | return div_lut[f]; |
332 | | } |
333 | | |
334 | | static int16_t resolve_divisor_32(uint32_t D, int16_t* shift) { |
335 | | int32_t f; |
336 | | *shift = get_msb(D); |
337 | | // e is obtained from D after resetting the most significant 1 bit. |
338 | | const int32_t e = D - ((uint32_t)1 << *shift); |
339 | | // Get the most significant DIV_LUT_BITS (8) bits of e into f |
340 | | if (*shift > DIV_LUT_BITS) { |
341 | | f = ROUND_POWER_OF_TWO(e, *shift - DIV_LUT_BITS); |
342 | | } else { |
343 | | f = e << (DIV_LUT_BITS - *shift); |
344 | | } |
345 | | assert(f <= DIV_LUT_NUM); |
346 | | *shift += DIV_LUT_PREC_BITS; |
347 | | // Use f as lookup into the precomputed table of multipliers |
348 | | return div_lut[f]; |
349 | | } |
350 | | |
351 | | static int is_affine_valid(const WarpedMotionParams* const wm) { |
352 | | const int32_t* mat = wm->wmmat; |
353 | | return (mat[2] > 0); |
354 | | } |
355 | | |
356 | | static int is_affine_shear_allowed(int16_t alpha, int16_t beta, int16_t gamma, int16_t delta) { |
357 | | if ((4 * abs(alpha) + 7 * abs(beta) >= (1 << WARPEDMODEL_PREC_BITS)) || |
358 | | (4 * abs(gamma) + 4 * abs(delta) >= (1 << WARPEDMODEL_PREC_BITS))) { |
359 | | return 0; |
360 | | } else { |
361 | | return 1; |
362 | | } |
363 | | } |
364 | | |
365 | | static int find_affine_int(int np, const int* pts1, const int* pts2, BlockSize bsize, Mv mv, WarpedMotionParams* wm, |
366 | | int mi_row, int mi_col) { |
367 | | int32_t A[2][2] = {{0, 0}, {0, 0}}; |
368 | | int32_t bx[2] = {0, 0}; |
369 | | int32_t by[2] = {0, 0}; |
370 | | int i; |
371 | | |
372 | | const int bw = block_size_wide[bsize]; |
373 | | const int bh = block_size_high[bsize]; |
374 | | const int rsuy = (AOMMAX(bh, MI_SIZE) / 2 - 1); |
375 | | const int rsux = (AOMMAX(bw, MI_SIZE) / 2 - 1); |
376 | | const int suy = rsuy * 8; |
377 | | const int sux = rsux * 8; |
378 | | const int duy = suy + mv.y; |
379 | | const int dux = sux + mv.x; |
380 | | const int isuy = (mi_row * MI_SIZE + rsuy); |
381 | | const int isux = (mi_col * MI_SIZE + rsux); |
382 | | |
383 | | // Assume the center pixel of the block has exactly the same motion vector |
384 | | // as transmitted for the block. First shift the origin of the source |
385 | | // points to the block center, and the origin of the destination points to |
386 | | // the block center added to the motion vector transmitted. |
387 | | // Let (xi, yi) denote the source points and (xi', yi') denote destination |
388 | | // points after origin shfifting, for i = 0, 1, 2, .... n-1. |
389 | | // Then if P = [x0, y0, |
390 | | // x1, y1 |
391 | | // x2, y1, |
392 | | // .... |
393 | | // ] |
394 | | // q = [x0', x1', x2', ... ]' |
395 | | // r = [y0', y1', y2', ... ]' |
396 | | // the least squares problems that need to be solved are: |
397 | | // [h1, h2]' = inv(P'P)P'q and |
398 | | // [h3, h4]' = inv(P'P)P'r |
399 | | // where the affine transformation is given by: |
400 | | // x' = h1.x + h2.y |
401 | | // y' = h3.x + h4.y |
402 | | // |
403 | | // The loop below computes: A = P'P, bx = P'q, by = P'r |
404 | | // We need to just compute inv(A).bx and inv(A).by for the solutions. |
405 | | // Contribution from neighbor block |
406 | | for (i = 0; i < np; i++) { |
407 | | const int dx = pts2[i * 2] - dux; |
408 | | const int dy = pts2[i * 2 + 1] - duy; |
409 | | const int sx = pts1[i * 2] - sux; |
410 | | const int sy = pts1[i * 2 + 1] - suy; |
411 | | if (abs(sx - dx) < LS_MV_MAX && abs(sy - dy) < LS_MV_MAX) { |
412 | | A[0][0] += LS_SQUARE(sx); |
413 | | A[0][1] += LS_PRODUCT1(sx, sy); |
414 | | A[1][1] += LS_SQUARE(sy); |
415 | | bx[0] += LS_PRODUCT2(sx, dx); |
416 | | bx[1] += LS_PRODUCT1(sy, dx); |
417 | | by[0] += LS_PRODUCT1(sx, dy); |
418 | | by[1] += LS_PRODUCT2(sy, dy); |
419 | | } |
420 | | } |
421 | | |
422 | | // Just for debugging, and can be removed later. |
423 | | assert(A[0][0] >= LS_MAT_MIN && A[0][0] <= LS_MAT_MAX); |
424 | | assert(A[0][1] >= LS_MAT_MIN && A[0][1] <= LS_MAT_MAX); |
425 | | assert(A[1][1] >= LS_MAT_MIN && A[1][1] <= LS_MAT_MAX); |
426 | | assert(bx[0] >= LS_MAT_MIN && bx[0] <= LS_MAT_MAX); |
427 | | assert(bx[1] >= LS_MAT_MIN && bx[1] <= LS_MAT_MAX); |
428 | | assert(by[0] >= LS_MAT_MIN && by[0] <= LS_MAT_MAX); |
429 | | assert(by[1] >= LS_MAT_MIN && by[1] <= LS_MAT_MAX); |
430 | | |
431 | | int64_t det; |
432 | | int16_t i_det, shift; |
433 | | |
434 | | // Compute Determinant of A |
435 | | det = (int64_t)A[0][0] * A[1][1] - (int64_t)A[0][1] * A[0][1]; |
436 | | if (det == 0) { |
437 | | return 1; |
438 | | } |
439 | | i_det = resolve_divisor_64(llabs(det), &shift) * (det < 0 ? -1 : 1); |
440 | | shift -= WARPEDMODEL_PREC_BITS; |
441 | | if (shift < 0) { |
442 | | i_det <<= (-shift); |
443 | | shift = 0; |
444 | | } |
445 | | |
446 | | int64_t p_x[2], p_y[2]; |
447 | | |
448 | | // These divided by the det, are the least squares solutions |
449 | | p_x[0] = (int64_t)A[1][1] * bx[0] - (int64_t)A[0][1] * bx[1]; |
450 | | p_x[1] = -(int64_t)A[0][1] * bx[0] + (int64_t)A[0][0] * bx[1]; |
451 | | p_y[0] = (int64_t)A[1][1] * by[0] - (int64_t)A[0][1] * by[1]; |
452 | | p_y[1] = -(int64_t)A[0][1] * by[0] + (int64_t)A[0][0] * by[1]; |
453 | | |
454 | | wm->wmmat[2] = get_mult_shift_diag(p_x[0], i_det, shift); |
455 | | wm->wmmat[3] = get_mult_shift_ndiag(p_x[1], i_det, shift); |
456 | | wm->wmmat[4] = get_mult_shift_ndiag(p_y[0], i_det, shift); |
457 | | wm->wmmat[5] = get_mult_shift_diag(p_y[1], i_det, shift); |
458 | | |
459 | | // Note: In the vx, vy expressions below, the max value of each of the |
460 | | // 2nd and 3rd terms are (2^16 - 1) * (2^13 - 1). That leaves enough room |
461 | | // for the first term so that the overall sum in the worst case fits |
462 | | // within 32 bits overall. |
463 | | int32_t vx = mv.x * (1 << (WARPEDMODEL_PREC_BITS - 3)) - |
464 | | (isux * (wm->wmmat[2] - (1 << WARPEDMODEL_PREC_BITS)) + isuy * wm->wmmat[3]); |
465 | | int32_t vy = mv.y * (1 << (WARPEDMODEL_PREC_BITS - 3)) - |
466 | | (isux * wm->wmmat[4] + isuy * (wm->wmmat[5] - (1 << WARPEDMODEL_PREC_BITS))); |
467 | | wm->wmmat[0] = clamp(vx, -WARPEDMODEL_TRANS_CLAMP, WARPEDMODEL_TRANS_CLAMP - 1); |
468 | | wm->wmmat[1] = clamp(vy, -WARPEDMODEL_TRANS_CLAMP, WARPEDMODEL_TRANS_CLAMP - 1); |
469 | | |
470 | | return 0; |
471 | | } |
472 | | |
473 | | bool svt_find_projection(int np, int* pts1, int* pts2, BlockSize bsize, Mv mv, WarpedMotionParams* wm_params, |
474 | 0 | int mi_row, int mi_col) { |
475 | 0 | if (find_affine_int(np, pts1, pts2, bsize, mv, wm_params, mi_row, mi_col)) { |
476 | 0 | return 1; |
477 | 0 | } |
478 | | |
479 | | // check compatibility with the fast warp filter |
480 | 0 | if (!svt_get_shear_params(wm_params)) { |
481 | 0 | return 1; |
482 | 0 | } |
483 | | |
484 | 0 | return 0; |
485 | 0 | } |
486 | | |
487 | | /* The warp filter for ROTZOOM and AFFINE models works as follows: |
488 | | * Split the input into 8x8 blocks |
489 | | * For each block, project the point (4, 4) within the block, to get the |
490 | | overall block position. Split into integer and fractional coordinates, |
491 | | maintaining full WARPEDMODEL precision |
492 | | * Filter horizontally: Generate 15 rows of 8 pixels each. Each pixel gets a |
493 | | variable horizontal offset. This means that, while the rows of the |
494 | | intermediate buffer align with the rows of the *reference* image, the |
495 | | columns align with the columns of the *destination* image. |
496 | | * Filter vertically: Generate the output block (up to 8x8 pixels, but if the |
497 | | destination is too small we crop the output at this stage). Each pixel has |
498 | | a variable vertical offset, so that the resulting rows are aligned with |
499 | | the rows of the destination image. |
500 | | |
501 | | To accomplish these alignments, we factor the warp matrix as a |
502 | | product of two shear / asymmetric zoom matrices: |
503 | | / a b \ = / 1 0 \ * / 1+alpha beta \ |
504 | | \ c d / \ gamma 1+delta / \ 0 1 / |
505 | | where a, b, c, d are wmmat[2], wmmat[3], wmmat[4], wmmat[5] respectively. |
506 | | The horizontal shear (with alpha and beta) is applied first, |
507 | | then the vertical shear (with gamma and delta) is applied second. |
508 | | |
509 | | The only limitation is that, to fit this in a fixed 8-tap filter size, |
510 | | the fractional pixel offsets must be at most +-1. Since the horizontal filter |
511 | | generates 15 rows of 8 columns, and the initial point we project is at (4, 4) |
512 | | within the block, the parameters must satisfy |
513 | | 4 * |alpha| + 7 * |beta| <= 1 and 4 * |gamma| + 4 * |delta| <= 1 |
514 | | for this filter to be applicable. |
515 | | |
516 | | Note: This function assumes that the caller has done all of the relevant |
517 | | checks, ie. that we have a ROTZOOM or AFFINE model, that wm[4] and wm[5] |
518 | | are set appropriately (if using a ROTZOOM model), and that alpha, beta, |
519 | | gamma, delta are all in range. |
520 | | |
521 | | */ |
522 | | /* A note on hardware implementation: |
523 | | The warp filter is intended to be implementable using the same hardware as |
524 | | the high-precision svt_aom_convolve filters from the loop-restoration and |
525 | | convolve-round experiments. |
526 | | |
527 | | For a single filter stage, considering all of the coefficient sets for the |
528 | | warp filter and the regular convolution filter, an input in the range |
529 | | [0, 2^k - 1] is mapped into the range [-56 * (2^k - 1), 184 * (2^k - 1)] |
530 | | before rounding. |
531 | | |
532 | | Allowing for some changes to the filter coefficient sets, call the range |
533 | | [-64 * 2^k, 192 * 2^k]. Then, if we initialize the accumulator to 64 * 2^k, |
534 | | we can replace this by the range [0, 256 * 2^k], which can be stored in an |
535 | | unsigned value with 8 + k bits. |
536 | | |
537 | | This allows the derivation of the appropriate bit widths and offsets for |
538 | | the various intermediate values: If |
539 | | |
540 | | F := FILTER_BITS = 7 (or else the above ranges need adjusting) |
541 | | So a *single* filter stage maps a k-bit input to a (k + F + 1)-bit |
542 | | intermediate value. |
543 | | H := ROUND0_BITS |
544 | | V := VERSHEAR_REDUCE_PREC_BITS |
545 | | (and note that we must have H + V = 2*F for the output to have the same |
546 | | scale as the input) |
547 | | |
548 | | then we end up with the following offsets and ranges: |
549 | | Horizontal filter: Apply an offset of 1 << (bd + F - 1), sum fits into a |
550 | | uint{bd + F + 1} |
551 | | After rounding: The values stored in 'tmp' fit into a uint{bd + F + 1 - H}. |
552 | | Vertical filter: Apply an offset of 1 << (bd + 2*F - H), sum fits into a |
553 | | uint{bd + 2*F + 2 - H} |
554 | | After rounding: The final value, before undoing the offset, fits into a |
555 | | uint{bd + 2}. |
556 | | |
557 | | Then we need to undo the offsets before clamping to a pixel. Note that, |
558 | | if we do this at the end, the amount to subtract is actually independent |
559 | | of H and V: |
560 | | |
561 | | offset to subtract = (1 << ((bd + F - 1) - H + F - V)) + |
562 | | (1 << ((bd + 2*F - H) - V)) |
563 | | == (1 << (bd - 1)) + (1 << bd) |
564 | | |
565 | | This allows us to entirely avoid clamping in both the warp filter and |
566 | | the convolve-round experiment. As of the time of writing, the Wiener filter |
567 | | from loop-restoration can encode a central coefficient up to 216, which |
568 | | leads to a maximum value of about 282 * 2^k after applying the offset. |
569 | | So in that case we still need to clamp. |
570 | | */ |
571 | | void svt_av1_warp_affine_c(const int32_t* mat, const uint8_t* ref, int width, int height, int stride, uint8_t* pred, |
572 | | int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, |
573 | | int subsampling_y, ConvolveParams* conv_params, int16_t alpha, int16_t beta, int16_t gamma, |
574 | 0 | int16_t delta) { |
575 | 0 | int32_t tmp[15 * 8]; |
576 | 0 | const int bd = 8; |
577 | 0 | const int reduce_bits_horiz = conv_params->round_0; |
578 | 0 | const int reduce_bits_vert = conv_params->is_compound ? conv_params->round_1 : 2 * FILTER_BITS - reduce_bits_horiz; |
579 | 0 | const int max_bits_horiz = bd + FILTER_BITS + 1 - reduce_bits_horiz; |
580 | 0 | const int offset_bits_horiz = bd + FILTER_BITS - 1; |
581 | 0 | const int offset_bits_vert = bd + 2 * FILTER_BITS - reduce_bits_horiz; |
582 | 0 | const int round_bits = 2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1; |
583 | 0 | const int offset_bits = bd + 2 * FILTER_BITS - conv_params->round_0; |
584 | 0 | (void)max_bits_horiz; |
585 | 0 | assert(IMPLIES(conv_params->is_compound, conv_params->dst != NULL)); |
586 | 0 | assert(IMPLIES(conv_params->do_average, conv_params->is_compound)); |
587 | |
|
588 | 0 | for (int i = p_row; i < p_row + p_height; i += 8) { |
589 | 0 | for (int j = p_col; j < p_col + p_width; j += 8) { |
590 | | // Calculate the center of this 8x8 block, |
591 | | // project to luma coordinates (if in a subsampled chroma plane), |
592 | | // apply the affine transformation, |
593 | | // then convert back to the original coordinates (if necessary) |
594 | 0 | const int32_t src_x = (j + 4) << subsampling_x; |
595 | 0 | const int32_t src_y = (i + 4) << subsampling_y; |
596 | 0 | const int32_t dst_x = mat[2] * src_x + mat[3] * src_y + mat[0]; |
597 | 0 | const int32_t dst_y = mat[4] * src_x + mat[5] * src_y + mat[1]; |
598 | 0 | const int32_t x4 = dst_x >> subsampling_x; |
599 | 0 | const int32_t y4 = dst_y >> subsampling_y; |
600 | |
|
601 | 0 | int32_t ix4 = x4 >> WARPEDMODEL_PREC_BITS; |
602 | 0 | int32_t sx4 = x4 & ((1 << WARPEDMODEL_PREC_BITS) - 1); |
603 | 0 | int32_t iy4 = y4 >> WARPEDMODEL_PREC_BITS; |
604 | 0 | int32_t sy4 = y4 & ((1 << WARPEDMODEL_PREC_BITS) - 1); |
605 | |
|
606 | 0 | sx4 += alpha * (-4) + beta * (-4); |
607 | 0 | sy4 += gamma * (-4) + delta * (-4); |
608 | |
|
609 | 0 | sx4 &= ~((1 << WARP_PARAM_REDUCE_BITS) - 1); |
610 | 0 | sy4 &= ~((1 << WARP_PARAM_REDUCE_BITS) - 1); |
611 | | |
612 | | // Horizontal filter |
613 | 0 | for (int k = -7; k < 8; ++k) { |
614 | | // Clamp to top/bottom edge of the frame |
615 | 0 | const int iy = clamp(iy4 + k, 0, height - 1); |
616 | |
|
617 | 0 | int sx = sx4 + beta * (k + 4); |
618 | |
|
619 | 0 | for (int l = -4; l < 4; ++l) { |
620 | 0 | int ix = ix4 + l - 3; |
621 | | // At this point, sx = sx4 + alpha * l + beta * k |
622 | 0 | const int offs = ROUND_POWER_OF_TWO(sx, WARPEDDIFF_PREC_BITS) + WARPEDPIXEL_PREC_SHIFTS; |
623 | 0 | assert(offs >= 0 && offs <= WARPEDPIXEL_PREC_SHIFTS * 3); |
624 | 0 | const int16_t* coeffs = svt_aom_warped_filter[offs]; |
625 | |
|
626 | 0 | int32_t sum = 1 << offset_bits_horiz; |
627 | 0 | for (int m = 0; m < 8; ++m) { |
628 | | // Clamp to left/right edge of the frame |
629 | 0 | const int sample_x = clamp(ix + m, 0, width - 1); |
630 | |
|
631 | 0 | sum += ref[iy * stride + sample_x] * coeffs[m]; |
632 | 0 | } |
633 | 0 | sum = ROUND_POWER_OF_TWO(sum, reduce_bits_horiz); |
634 | 0 | assert(0 <= sum && sum < (1 << max_bits_horiz)); |
635 | 0 | tmp[(k + 7) * 8 + (l + 4)] = sum; |
636 | 0 | sx += alpha; |
637 | 0 | } |
638 | 0 | } |
639 | | |
640 | | // Vertical filter |
641 | 0 | for (int k = -4; k < AOMMIN(4, p_row + p_height - i - 4); ++k) { |
642 | 0 | int sy = sy4 + delta * (k + 4); |
643 | 0 | for (int l = -4; l < AOMMIN(4, p_col + p_width - j - 4); ++l) { |
644 | | // At this point, sy = sy4 + gamma * l + delta * k |
645 | 0 | const int offs = ROUND_POWER_OF_TWO(sy, WARPEDDIFF_PREC_BITS) + WARPEDPIXEL_PREC_SHIFTS; |
646 | 0 | assert(offs >= 0 && offs <= WARPEDPIXEL_PREC_SHIFTS * 3); |
647 | 0 | const int16_t* coeffs = svt_aom_warped_filter[offs]; |
648 | |
|
649 | 0 | int32_t sum = 1 << offset_bits_vert; |
650 | 0 | for (int m = 0; m < 8; ++m) { |
651 | 0 | sum += tmp[(k + m + 4) * 8 + (l + 4)] * coeffs[m]; |
652 | 0 | } |
653 | 0 | if (conv_params->is_compound) { |
654 | 0 | ConvBufType* p = |
655 | 0 | &conv_params->dst[(i - p_row + k + 4) * conv_params->dst_stride + (j - p_col + l + 4)]; |
656 | 0 | sum = ROUND_POWER_OF_TWO(sum, reduce_bits_vert); |
657 | 0 | if (conv_params->do_average) { |
658 | 0 | uint8_t* dst8 = &pred[(i - p_row + k + 4) * p_stride + (j - p_col + l + 4)]; |
659 | 0 | int32_t tmp32 = *p; |
660 | 0 | if (conv_params->use_jnt_comp_avg) { |
661 | 0 | tmp32 = tmp32 * conv_params->fwd_offset + sum * conv_params->bck_offset; |
662 | 0 | tmp32 = tmp32 >> DIST_PRECISION_BITS; |
663 | 0 | } else { |
664 | 0 | tmp32 += sum; |
665 | 0 | tmp32 = tmp32 >> 1; |
666 | 0 | } |
667 | 0 | tmp32 = tmp32 - (1 << (offset_bits - conv_params->round_1)) - |
668 | 0 | (1 << (offset_bits - conv_params->round_1 - 1)); |
669 | 0 | *dst8 = clip_pixel(ROUND_POWER_OF_TWO(tmp32, round_bits)); |
670 | 0 | } else { |
671 | 0 | *p = sum; |
672 | 0 | } |
673 | 0 | } else { |
674 | 0 | uint8_t* p = &pred[(i - p_row + k + 4) * p_stride + (j - p_col + l + 4)]; |
675 | 0 | sum = ROUND_POWER_OF_TWO(sum, reduce_bits_vert); |
676 | 0 | assert(0 <= sum && sum < (1 << (bd + 2))); |
677 | 0 | *p = clip_pixel(sum - (1 << (bd - 1)) - (1 << bd)); |
678 | 0 | } |
679 | 0 | sy += gamma; |
680 | 0 | } |
681 | 0 | } |
682 | 0 | } |
683 | 0 | } |
684 | 0 | } |
685 | | |
686 | | void svt_warp_plane(WarpedMotionParams* wm, const uint8_t* const ref, int width, int height, int stride, uint8_t* pred, |
687 | | int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, |
688 | 0 | ConvolveParams* conv_params) { |
689 | 0 | assert(wm->wmtype <= AFFINE); |
690 | 0 | if (wm->wmtype == ROTZOOM) { |
691 | 0 | wm->wmmat[5] = wm->wmmat[2]; |
692 | 0 | wm->wmmat[4] = -wm->wmmat[3]; |
693 | 0 | } |
694 | 0 | const int32_t* const mat = wm->wmmat; |
695 | 0 | const int16_t alpha = wm->alpha; |
696 | 0 | const int16_t beta = wm->beta; |
697 | 0 | const int16_t gamma = wm->gamma; |
698 | 0 | const int16_t delta = wm->delta; |
699 | 0 | svt_av1_warp_affine(mat, |
700 | 0 | ref, |
701 | 0 | width, |
702 | 0 | height, |
703 | 0 | stride, |
704 | 0 | pred, |
705 | 0 | p_col, |
706 | 0 | p_row, |
707 | 0 | p_width, |
708 | 0 | p_height, |
709 | 0 | p_stride, |
710 | 0 | subsampling_x, |
711 | 0 | subsampling_y, |
712 | 0 | conv_params, |
713 | 0 | alpha, |
714 | 0 | beta, |
715 | 0 | gamma, |
716 | 0 | delta); |
717 | 0 | } |
718 | | |
719 | | void svt_av1_highbd_warp_affine_c(const int32_t* mat, const uint8_t* ref8b, const uint8_t* ref2b, int width, int height, |
720 | | int stride8b, int stride2b, uint16_t* pred, int p_col, int p_row, int p_width, |
721 | | int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, |
722 | | ConvolveParams* conv_params, int16_t alpha, int16_t beta, int16_t gamma, |
723 | 0 | int16_t delta) { |
724 | 0 | int32_t tmp[15 * 8]; |
725 | 0 | const int reduce_bits_horiz = conv_params->round_0 + AOMMAX(bd + FILTER_BITS - conv_params->round_0 - 14, 0); |
726 | 0 | const int reduce_bits_vert = conv_params->is_compound ? conv_params->round_1 : 2 * FILTER_BITS - reduce_bits_horiz; |
727 | 0 | const int max_bits_horiz = bd + FILTER_BITS + 1 - reduce_bits_horiz; |
728 | 0 | const int offset_bits_horiz = bd + FILTER_BITS - 1; |
729 | 0 | const int offset_bits_vert = bd + 2 * FILTER_BITS - reduce_bits_horiz; |
730 | 0 | const int round_bits = 2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1; |
731 | 0 | const int offset_bits = bd + 2 * FILTER_BITS - conv_params->round_0; |
732 | 0 | (void)max_bits_horiz; |
733 | 0 | assert(IMPLIES(conv_params->is_compound, conv_params->dst != NULL)); |
734 | |
|
735 | 0 | for (int i = p_row; i < p_row + p_height; i += 8) { |
736 | 0 | for (int j = p_col; j < p_col + p_width; j += 8) { |
737 | | // Calculate the center of this 8x8 block, |
738 | | // project to luma coordinates (if in a subsampled chroma plane), |
739 | | // apply the affine transformation, |
740 | | // then convert back to the original coordinates (if necessary) |
741 | 0 | const int32_t src_x = (j + 4) << subsampling_x; |
742 | 0 | const int32_t src_y = (i + 4) << subsampling_y; |
743 | 0 | const int32_t dst_x = mat[2] * src_x + mat[3] * src_y + mat[0]; |
744 | 0 | const int32_t dst_y = mat[4] * src_x + mat[5] * src_y + mat[1]; |
745 | 0 | const int32_t x4 = dst_x >> subsampling_x; |
746 | 0 | const int32_t y4 = dst_y >> subsampling_y; |
747 | |
|
748 | 0 | const int32_t ix4 = x4 >> WARPEDMODEL_PREC_BITS; |
749 | 0 | int32_t sx4 = x4 & ((1 << WARPEDMODEL_PREC_BITS) - 1); |
750 | 0 | const int32_t iy4 = y4 >> WARPEDMODEL_PREC_BITS; |
751 | 0 | int32_t sy4 = y4 & ((1 << WARPEDMODEL_PREC_BITS) - 1); |
752 | |
|
753 | 0 | sx4 += alpha * (-4) + beta * (-4); |
754 | 0 | sy4 += gamma * (-4) + delta * (-4); |
755 | |
|
756 | 0 | sx4 &= ~((1 << WARP_PARAM_REDUCE_BITS) - 1); |
757 | 0 | sy4 &= ~((1 << WARP_PARAM_REDUCE_BITS) - 1); |
758 | | |
759 | | // Horizontal filter |
760 | 0 | for (int k = -7; k < 8; ++k) { |
761 | 0 | const int iy = clamp(iy4 + k, 0, height - 1); |
762 | |
|
763 | 0 | int sx = sx4 + beta * (k + 4); |
764 | 0 | for (int l = -4; l < 4; ++l) { |
765 | 0 | int ix = ix4 + l - 3; |
766 | 0 | const int offs = ROUND_POWER_OF_TWO(sx, WARPEDDIFF_PREC_BITS) + WARPEDPIXEL_PREC_SHIFTS; |
767 | 0 | assert(offs >= 0 && offs <= WARPEDPIXEL_PREC_SHIFTS * 3); |
768 | 0 | const int16_t* coeffs = svt_aom_warped_filter[offs]; |
769 | |
|
770 | 0 | int32_t sum = 1 << offset_bits_horiz; |
771 | 0 | for (int m = 0; m < 8; ++m) { |
772 | 0 | const int sample_x = clamp(ix + m, 0, width - 1); |
773 | 0 | uint16_t ref = (ref8b[iy * stride8b + sample_x] << 2) | |
774 | 0 | ((ref2b[iy * stride2b + sample_x] >> 6) & 3); |
775 | 0 | sum += ref * coeffs[m]; |
776 | 0 | } |
777 | 0 | sum = ROUND_POWER_OF_TWO(sum, reduce_bits_horiz); |
778 | 0 | assert(0 <= sum && sum < (1 << max_bits_horiz)); |
779 | 0 | tmp[(k + 7) * 8 + (l + 4)] = sum; |
780 | 0 | sx += alpha; |
781 | 0 | } |
782 | 0 | } |
783 | | |
784 | | // Vertical filter |
785 | 0 | for (int k = -4; k < AOMMIN(4, p_row + p_height - i - 4); ++k) { |
786 | 0 | int sy = sy4 + delta * (k + 4); |
787 | 0 | for (int l = -4; l < AOMMIN(4, p_col + p_width - j - 4); ++l) { |
788 | 0 | const int offs = ROUND_POWER_OF_TWO(sy, WARPEDDIFF_PREC_BITS) + WARPEDPIXEL_PREC_SHIFTS; |
789 | 0 | assert(offs >= 0 && offs <= WARPEDPIXEL_PREC_SHIFTS * 3); |
790 | 0 | const int16_t* coeffs = svt_aom_warped_filter[offs]; |
791 | |
|
792 | 0 | int32_t sum = 1 << offset_bits_vert; |
793 | 0 | for (int m = 0; m < 8; ++m) { |
794 | 0 | sum += tmp[(k + m + 4) * 8 + (l + 4)] * coeffs[m]; |
795 | 0 | } |
796 | 0 | if (conv_params->is_compound) { |
797 | 0 | ConvBufType* p = |
798 | 0 | &conv_params->dst[(i - p_row + k + 4) * conv_params->dst_stride + (j - p_col + l + 4)]; |
799 | 0 | sum = ROUND_POWER_OF_TWO(sum, reduce_bits_vert); |
800 | 0 | if (conv_params->do_average) { |
801 | 0 | uint16_t* dst16 = &pred[(i - p_row + k + 4) * p_stride + (j - p_col + l + 4)]; |
802 | 0 | int32_t tmp32 = *p; |
803 | 0 | if (conv_params->use_jnt_comp_avg) { |
804 | 0 | tmp32 = tmp32 * conv_params->fwd_offset + sum * conv_params->bck_offset; |
805 | 0 | tmp32 = tmp32 >> DIST_PRECISION_BITS; |
806 | 0 | } else { |
807 | 0 | tmp32 += sum; |
808 | 0 | tmp32 = tmp32 >> 1; |
809 | 0 | } |
810 | 0 | tmp32 = tmp32 - (1 << (offset_bits - conv_params->round_1)) - |
811 | 0 | (1 << (offset_bits - conv_params->round_1 - 1)); |
812 | 0 | *dst16 = clip_pixel_highbd(ROUND_POWER_OF_TWO(tmp32, round_bits), bd); |
813 | 0 | } else { |
814 | 0 | *p = sum; |
815 | 0 | } |
816 | 0 | } else { |
817 | 0 | uint16_t* p = &pred[(i - p_row + k + 4) * p_stride + (j - p_col + l + 4)]; |
818 | 0 | sum = ROUND_POWER_OF_TWO(sum, reduce_bits_vert); |
819 | 0 | assert(0 <= sum && sum < (1 << (bd + 2))); |
820 | 0 | *p = clip_pixel_highbd(sum - (1 << (bd - 1)) - (1 << bd), bd); |
821 | 0 | } |
822 | 0 | sy += gamma; |
823 | 0 | } |
824 | 0 | } |
825 | 0 | } |
826 | 0 | } |
827 | 0 | } |
828 | | |
829 | | static void highbd_warp_plane(WarpedMotionParams* wm, const uint8_t* const ref8, const uint8_t* const ref_2b, int width, |
830 | | int height, int stride, const uint8_t* const pred8, int p_col, int p_row, int p_width, |
831 | | int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, |
832 | 0 | ConvolveParams* conv_params) { |
833 | 0 | assert(wm->wmtype <= AFFINE); |
834 | 0 | if (wm->wmtype == ROTZOOM) { |
835 | 0 | wm->wmmat[5] = wm->wmmat[2]; |
836 | 0 | wm->wmmat[4] = -wm->wmmat[3]; |
837 | 0 | } |
838 | 0 | const int32_t* const mat = wm->wmmat; |
839 | 0 | const int16_t alpha = wm->alpha; |
840 | 0 | const int16_t beta = wm->beta; |
841 | 0 | const int16_t gamma = wm->gamma; |
842 | 0 | const int16_t delta = wm->delta; |
843 | |
|
844 | 0 | uint16_t* pred = (uint16_t*)pred8; |
845 | 0 | svt_av1_highbd_warp_affine(mat, |
846 | 0 | ref8, |
847 | 0 | ref_2b, |
848 | 0 | width, |
849 | 0 | height, |
850 | 0 | stride, |
851 | 0 | stride, |
852 | 0 | pred, |
853 | 0 | p_col, |
854 | 0 | p_row, |
855 | 0 | p_width, |
856 | 0 | p_height, |
857 | 0 | p_stride, |
858 | 0 | subsampling_x, |
859 | 0 | subsampling_y, |
860 | 0 | bd, |
861 | 0 | conv_params, |
862 | 0 | alpha, |
863 | 0 | beta, |
864 | 0 | gamma, |
865 | 0 | delta); |
866 | 0 | } |
867 | | |
868 | | void svt_av1_warp_plane(WarpedMotionParams* wm, int use_hbd, int bd, const uint8_t* ref, const uint8_t* ref_2b, |
869 | | int width, int height, int stride, uint8_t* pred, int p_col, int p_row, int p_width, |
870 | 0 | int p_height, int p_stride, int subsampling_x, int subsampling_y, ConvolveParams* conv_params) { |
871 | 0 | if (use_hbd) { |
872 | 0 | highbd_warp_plane(wm, |
873 | 0 | ref, |
874 | 0 | ref_2b, |
875 | 0 | width, |
876 | 0 | height, |
877 | 0 | stride, |
878 | 0 | pred, |
879 | 0 | p_col, |
880 | 0 | p_row, |
881 | 0 | p_width, |
882 | 0 | p_height, |
883 | 0 | p_stride, |
884 | 0 | subsampling_x, |
885 | 0 | subsampling_y, |
886 | 0 | bd, |
887 | 0 | conv_params); |
888 | 0 | } else { |
889 | 0 | svt_warp_plane(wm, |
890 | 0 | ref, |
891 | 0 | width, |
892 | 0 | height, |
893 | 0 | stride, |
894 | 0 | pred, |
895 | 0 | p_col, |
896 | 0 | p_row, |
897 | 0 | p_width, |
898 | 0 | p_height, |
899 | 0 | p_stride, |
900 | 0 | subsampling_x, |
901 | 0 | subsampling_y, |
902 | 0 | conv_params); |
903 | 0 | } |
904 | 0 | } |
905 | | |
906 | | // Returns 1 on success or 0 on an invalid affine set |
907 | 0 | int svt_get_shear_params(WarpedMotionParams* wm) { |
908 | 0 | const int32_t* mat = wm->wmmat; |
909 | 0 | if (!is_affine_valid(wm)) { |
910 | 0 | return 0; |
911 | 0 | } |
912 | 0 | wm->alpha = clamp(mat[2] - (1 << WARPEDMODEL_PREC_BITS), INT16_MIN, INT16_MAX); |
913 | 0 | wm->beta = clamp(mat[3], INT16_MIN, INT16_MAX); |
914 | 0 | int16_t shift; |
915 | 0 | int16_t y = resolve_divisor_32(abs(mat[2]), &shift) * (mat[2] < 0 ? -1 : 1); |
916 | 0 | int64_t v = ((int64_t)mat[4] * (1 << WARPEDMODEL_PREC_BITS)) * y; |
917 | 0 | wm->gamma = clamp((int)ROUND_POWER_OF_TWO_SIGNED_64(v, shift), INT16_MIN, INT16_MAX); |
918 | 0 | v = ((int64_t)mat[3] * mat[4]) * y; |
919 | 0 | wm->delta = clamp( |
920 | 0 | mat[5] - (int)ROUND_POWER_OF_TWO_SIGNED_64(v, shift) - (1 << WARPEDMODEL_PREC_BITS), INT16_MIN, INT16_MAX); |
921 | |
|
922 | 0 | wm->alpha = ROUND_POWER_OF_TWO_SIGNED(wm->alpha, WARP_PARAM_REDUCE_BITS) * (1 << WARP_PARAM_REDUCE_BITS); |
923 | 0 | wm->beta = ROUND_POWER_OF_TWO_SIGNED(wm->beta, WARP_PARAM_REDUCE_BITS) * (1 << WARP_PARAM_REDUCE_BITS); |
924 | 0 | wm->gamma = ROUND_POWER_OF_TWO_SIGNED(wm->gamma, WARP_PARAM_REDUCE_BITS) * (1 << WARP_PARAM_REDUCE_BITS); |
925 | 0 | wm->delta = ROUND_POWER_OF_TWO_SIGNED(wm->delta, WARP_PARAM_REDUCE_BITS) * (1 << WARP_PARAM_REDUCE_BITS); |
926 | |
|
927 | 0 | if (!is_affine_shear_allowed(wm->alpha, wm->beta, wm->gamma, wm->delta)) { |
928 | 0 | return 0; |
929 | 0 | } |
930 | | |
931 | 0 | return 1; |
932 | 0 | } |
933 | | |
934 | | // Select samples according to the motion vector difference. |
935 | 0 | uint8_t svt_aom_select_samples(Mv mv, int* pts, int* pts_inref, int len, BlockSize bsize) { |
936 | 0 | const int bw = block_size_wide[bsize]; |
937 | 0 | const int bh = block_size_high[bsize]; |
938 | 0 | const int thresh = clamp(AOMMAX(bw, bh), 16, 112); |
939 | 0 | uint8_t ret = 0; |
940 | | |
941 | | // Only keep the samples with MV differences within threshold. |
942 | 0 | for (int i = 0; i < len; ++i) { |
943 | 0 | const int diff = abs(pts_inref[2 * i] - pts[2 * i] - mv.x) + abs(pts_inref[2 * i + 1] - pts[2 * i + 1] - mv.y); |
944 | 0 | if (diff > thresh) { |
945 | 0 | continue; |
946 | 0 | } |
947 | 0 | if (ret != i) { |
948 | 0 | memcpy(pts + 2 * ret, pts + 2 * i, 2 * sizeof(pts[0])); |
949 | 0 | memcpy(pts_inref + 2 * ret, pts_inref + 2 * i, 2 * sizeof(pts_inref[0])); |
950 | 0 | } |
951 | 0 | ++ret; |
952 | 0 | } |
953 | | // Keep at least 1 sample. |
954 | 0 | return AOMMAX(ret, 1); |
955 | 0 | } |