Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 1 | /* |
Yaowu Xu | bde4ac8 | 2016-11-28 15:26:06 -0800 | [diff] [blame] | 2 | * Copyright (c) 2016, Alliance for Open Media. All rights reserved |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 3 | * |
Yaowu Xu | bde4ac8 | 2016-11-28 15:26:06 -0800 | [diff] [blame] | 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. |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 10 | */ |
Yaowu Xu | bde4ac8 | 2016-11-28 15:26:06 -0800 | [diff] [blame] | 11 | |
Yaowu Xu | f883b42 | 2016-08-30 14:01:10 -0700 | [diff] [blame] | 12 | #ifndef AV1_TXFM_H_ |
| 13 | #define AV1_TXFM_H_ |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 14 | |
| 15 | #include <assert.h> |
| 16 | #include <math.h> |
| 17 | #include <stdio.h> |
| 18 | |
| 19 | #include "av1/common/enums.h" |
Angie Chiang | 155bf9a | 2017-08-06 19:52:57 -0700 | [diff] [blame] | 20 | #include "av1/common/blockd.h" |
Yaowu Xu | f883b42 | 2016-08-30 14:01:10 -0700 | [diff] [blame] | 21 | #include "aom/aom_integer.h" |
| 22 | #include "aom_dsp/aom_dsp_common.h" |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 23 | |
Angie Chiang | 9c7089a | 2017-08-08 16:21:11 -0700 | [diff] [blame] | 24 | #ifdef __cplusplus |
| 25 | extern "C" { |
| 26 | #endif |
| 27 | |
Angie Chiang | ce3ad28 | 2017-08-08 09:51:54 -0700 | [diff] [blame] | 28 | #define MAX_TXFM_STAGE_NUM 12 |
Sarah Parker | a097204 | 2017-11-10 16:35:46 -0800 | [diff] [blame] | 29 | #define DIVIDE_AND_ROUND(x, y) (((x) + ((y) >> 1)) / (y)) |
Angie Chiang | ce3ad28 | 2017-08-08 09:51:54 -0700 | [diff] [blame] | 30 | |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 31 | static const int cos_bit_min = 10; |
| 32 | static const int cos_bit_max = 16; |
| 33 | |
| 34 | // cospi_arr[i][j] = (int)round(cos(M_PI*j/128) * (1<<(cos_bit_min+i))); |
Frederic Barbier | bbf7bb0 | 2017-05-05 09:37:40 +0200 | [diff] [blame] | 35 | static const int32_t cospi_arr_data[7][64] = { |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 36 | { 1024, 1024, 1023, 1021, 1019, 1016, 1013, 1009, 1004, 999, 993, 987, 980, |
clang-format | 67948d3 | 2016-09-07 22:40:40 -0700 | [diff] [blame] | 37 | 972, 964, 955, 946, 936, 926, 915, 903, 891, 878, 865, 851, 837, |
| 38 | 822, 807, 792, 775, 759, 742, 724, 706, 688, 669, 650, 630, 610, |
| 39 | 590, 569, 548, 526, 505, 483, 460, 438, 415, 392, 369, 345, 321, |
| 40 | 297, 273, 249, 224, 200, 175, 150, 125, 100, 75, 50, 25 }, |
| 41 | { 2048, 2047, 2046, 2042, 2038, 2033, 2026, 2018, 2009, 1998, 1987, |
| 42 | 1974, 1960, 1945, 1928, 1911, 1892, 1872, 1851, 1829, 1806, 1782, |
| 43 | 1757, 1730, 1703, 1674, 1645, 1615, 1583, 1551, 1517, 1483, 1448, |
| 44 | 1412, 1375, 1338, 1299, 1260, 1220, 1179, 1138, 1096, 1053, 1009, |
| 45 | 965, 921, 876, 830, 784, 737, 690, 642, 595, 546, 498, |
| 46 | 449, 400, 350, 301, 251, 201, 151, 100, 50 }, |
| 47 | { 4096, 4095, 4091, 4085, 4076, 4065, 4052, 4036, 4017, 3996, 3973, |
| 48 | 3948, 3920, 3889, 3857, 3822, 3784, 3745, 3703, 3659, 3612, 3564, |
| 49 | 3513, 3461, 3406, 3349, 3290, 3229, 3166, 3102, 3035, 2967, 2896, |
| 50 | 2824, 2751, 2675, 2598, 2520, 2440, 2359, 2276, 2191, 2106, 2019, |
| 51 | 1931, 1842, 1751, 1660, 1567, 1474, 1380, 1285, 1189, 1092, 995, |
| 52 | 897, 799, 700, 601, 501, 401, 301, 201, 101 }, |
| 53 | { 8192, 8190, 8182, 8170, 8153, 8130, 8103, 8071, 8035, 7993, 7946, |
| 54 | 7895, 7839, 7779, 7713, 7643, 7568, 7489, 7405, 7317, 7225, 7128, |
| 55 | 7027, 6921, 6811, 6698, 6580, 6458, 6333, 6203, 6070, 5933, 5793, |
| 56 | 5649, 5501, 5351, 5197, 5040, 4880, 4717, 4551, 4383, 4212, 4038, |
| 57 | 3862, 3683, 3503, 3320, 3135, 2948, 2760, 2570, 2378, 2185, 1990, |
| 58 | 1795, 1598, 1401, 1202, 1003, 803, 603, 402, 201 }, |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 59 | { 16384, 16379, 16364, 16340, 16305, 16261, 16207, 16143, 16069, 15986, 15893, |
| 60 | 15791, 15679, 15557, 15426, 15286, 15137, 14978, 14811, 14635, 14449, 14256, |
| 61 | 14053, 13842, 13623, 13395, 13160, 12916, 12665, 12406, 12140, 11866, 11585, |
clang-format | 67948d3 | 2016-09-07 22:40:40 -0700 | [diff] [blame] | 62 | 11297, 11003, 10702, 10394, 10080, 9760, 9434, 9102, 8765, 8423, 8076, |
| 63 | 7723, 7366, 7005, 6639, 6270, 5897, 5520, 5139, 4756, 4370, 3981, |
| 64 | 3590, 3196, 2801, 2404, 2006, 1606, 1205, 804, 402 }, |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 65 | { 32768, 32758, 32729, 32679, 32610, 32522, 32413, 32286, 32138, 31972, 31786, |
| 66 | 31581, 31357, 31114, 30853, 30572, 30274, 29957, 29622, 29269, 28899, 28511, |
| 67 | 28106, 27684, 27246, 26791, 26320, 25833, 25330, 24812, 24279, 23732, 23170, |
| 68 | 22595, 22006, 21403, 20788, 20160, 19520, 18868, 18205, 17531, 16846, 16151, |
clang-format | 67948d3 | 2016-09-07 22:40:40 -0700 | [diff] [blame] | 69 | 15447, 14733, 14010, 13279, 12540, 11793, 11039, 10279, 9512, 8740, 7962, |
| 70 | 7180, 6393, 5602, 4808, 4011, 3212, 2411, 1608, 804 }, |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 71 | { 65536, 65516, 65457, 65358, 65220, 65043, 64827, 64571, 64277, 63944, 63572, |
| 72 | 63162, 62714, 62228, 61705, 61145, 60547, 59914, 59244, 58538, 57798, 57022, |
| 73 | 56212, 55368, 54491, 53581, 52639, 51665, 50660, 49624, 48559, 47464, 46341, |
| 74 | 45190, 44011, 42806, 41576, 40320, 39040, 37736, 36410, 35062, 33692, 32303, |
| 75 | 30893, 29466, 28020, 26558, 25080, 23586, 22078, 20557, 19024, 17479, 15924, |
clang-format | 67948d3 | 2016-09-07 22:40:40 -0700 | [diff] [blame] | 76 | 14359, 12785, 11204, 9616, 8022, 6424, 4821, 3216, 1608 } |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 77 | }; |
| 78 | |
Frederic Barbier | bbf7bb0 | 2017-05-05 09:37:40 +0200 | [diff] [blame] | 79 | static INLINE const int32_t *cospi_arr(int n) { |
| 80 | return cospi_arr_data[n - cos_bit_min]; |
| 81 | } |
| 82 | |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 83 | static INLINE int32_t round_shift(int32_t value, int bit) { |
Yaowu Xu | 637590c | 2016-11-16 15:15:46 -0800 | [diff] [blame] | 84 | assert(bit >= 1); |
James Zern | a60e26d | 2017-11-15 12:11:17 -0800 | [diff] [blame] | 85 | return (int32_t)(((int64_t)value + (1ll << (bit - 1))) >> bit); |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 86 | } |
| 87 | |
| 88 | static INLINE void round_shift_array(int32_t *arr, int size, int bit) { |
| 89 | int i; |
| 90 | if (bit == 0) { |
| 91 | return; |
| 92 | } else { |
| 93 | if (bit > 0) { |
| 94 | for (i = 0; i < size; i++) { |
| 95 | arr[i] = round_shift(arr[i], bit); |
| 96 | } |
| 97 | } else { |
| 98 | for (i = 0; i < size; i++) { |
Yaowu Xu | 4d34154 | 2016-11-15 20:47:52 -0800 | [diff] [blame] | 99 | arr[i] = arr[i] * (1 << (-bit)); |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 100 | } |
| 101 | } |
| 102 | } |
| 103 | } |
| 104 | |
| 105 | static INLINE int32_t half_btf(int32_t w0, int32_t in0, int32_t w1, int32_t in1, |
| 106 | int bit) { |
Hui Su | 8e739bc | 2017-11-09 14:37:50 -0800 | [diff] [blame] | 107 | int32_t result_32 = (int32_t)clamp64((int64_t)w0 * in0 + (int64_t)w1 * in1, |
Yaowu Xu | ea69105 | 2017-11-26 16:46:50 -0800 | [diff] [blame] | 108 | (int64_t)INT32_MIN, (int64_t)INT32_MAX); |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 109 | return round_shift(result_32, bit); |
| 110 | } |
| 111 | |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 112 | typedef void (*TxfmFunc)(const int32_t *input, int32_t *output, |
| 113 | const int8_t *cos_bit, const int8_t *stage_range); |
| 114 | |
| 115 | typedef enum TXFM_TYPE { |
| 116 | TXFM_TYPE_DCT4, |
| 117 | TXFM_TYPE_DCT8, |
| 118 | TXFM_TYPE_DCT16, |
| 119 | TXFM_TYPE_DCT32, |
| 120 | TXFM_TYPE_DCT64, |
| 121 | TXFM_TYPE_ADST4, |
| 122 | TXFM_TYPE_ADST8, |
| 123 | TXFM_TYPE_ADST16, |
| 124 | TXFM_TYPE_ADST32, |
Sarah Parker | 3eed417 | 2017-05-15 20:49:22 -0700 | [diff] [blame] | 125 | TXFM_TYPE_IDENTITY4, |
| 126 | TXFM_TYPE_IDENTITY8, |
| 127 | TXFM_TYPE_IDENTITY16, |
| 128 | TXFM_TYPE_IDENTITY32, |
Debargha Mukherjee | 570423c | 2017-10-01 00:35:20 -0700 | [diff] [blame] | 129 | TXFM_TYPE_IDENTITY64, |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 130 | } TXFM_TYPE; |
| 131 | |
Sarah Parker | eec47e6 | 2017-05-15 20:49:22 -0700 | [diff] [blame] | 132 | typedef struct TXFM_1D_CFG { |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 133 | const int txfm_size; |
Sarah Parker | eec47e6 | 2017-05-15 20:49:22 -0700 | [diff] [blame] | 134 | const int stage_num; |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 135 | |
| 136 | const int8_t *shift; |
Sarah Parker | eec47e6 | 2017-05-15 20:49:22 -0700 | [diff] [blame] | 137 | const int8_t *stage_range; |
| 138 | const int8_t *cos_bit; |
| 139 | const TXFM_TYPE txfm_type; |
| 140 | } TXFM_1D_CFG; |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 141 | |
| 142 | typedef struct TXFM_2D_FLIP_CFG { |
| 143 | int ud_flip; // flip upside down |
| 144 | int lr_flip; // flip left to right |
Sarah Parker | eec47e6 | 2017-05-15 20:49:22 -0700 | [diff] [blame] | 145 | const TXFM_1D_CFG *col_cfg; |
| 146 | const TXFM_1D_CFG *row_cfg; |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 147 | } TXFM_2D_FLIP_CFG; |
| 148 | |
Urvang Joshi | 2283d37 | 2017-10-02 17:16:45 -0700 | [diff] [blame] | 149 | static INLINE void set_flip_cfg(TX_TYPE tx_type, TXFM_2D_FLIP_CFG *cfg) { |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 150 | switch (tx_type) { |
| 151 | case DCT_DCT: |
| 152 | case ADST_DCT: |
| 153 | case DCT_ADST: |
| 154 | case ADST_ADST: |
| 155 | cfg->ud_flip = 0; |
| 156 | cfg->lr_flip = 0; |
| 157 | break; |
Sarah Parker | 3eed417 | 2017-05-15 20:49:22 -0700 | [diff] [blame] | 158 | case IDTX: |
| 159 | case V_DCT: |
| 160 | case H_DCT: |
| 161 | case V_ADST: |
| 162 | case H_ADST: |
| 163 | cfg->ud_flip = 0; |
| 164 | cfg->lr_flip = 0; |
| 165 | break; |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 166 | case FLIPADST_DCT: |
Sarah Parker | eec47e6 | 2017-05-15 20:49:22 -0700 | [diff] [blame] | 167 | case FLIPADST_ADST: |
Sarah Parker | 3eed417 | 2017-05-15 20:49:22 -0700 | [diff] [blame] | 168 | case V_FLIPADST: |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 169 | cfg->ud_flip = 1; |
| 170 | cfg->lr_flip = 0; |
| 171 | break; |
| 172 | case DCT_FLIPADST: |
Sarah Parker | eec47e6 | 2017-05-15 20:49:22 -0700 | [diff] [blame] | 173 | case ADST_FLIPADST: |
Sarah Parker | 3eed417 | 2017-05-15 20:49:22 -0700 | [diff] [blame] | 174 | case H_FLIPADST: |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 175 | cfg->ud_flip = 0; |
| 176 | cfg->lr_flip = 1; |
| 177 | break; |
| 178 | case FLIPADST_FLIPADST: |
| 179 | cfg->ud_flip = 1; |
| 180 | cfg->lr_flip = 1; |
| 181 | break; |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 182 | default: |
| 183 | cfg->ud_flip = 0; |
| 184 | cfg->lr_flip = 0; |
| 185 | assert(0); |
| 186 | } |
| 187 | } |
| 188 | |
Angie Chiang | 155bf9a | 2017-08-06 19:52:57 -0700 | [diff] [blame] | 189 | #if CONFIG_TXMG |
Urvang Joshi | 9752a2e | 2017-10-02 17:32:27 -0700 | [diff] [blame] | 190 | static INLINE TX_SIZE av1_rotate_tx_size(TX_SIZE tx_size) { |
Angie Chiang | 155bf9a | 2017-08-06 19:52:57 -0700 | [diff] [blame] | 191 | switch (tx_size) { |
Angie Chiang | 155bf9a | 2017-08-06 19:52:57 -0700 | [diff] [blame] | 192 | case TX_4X4: return TX_4X4; |
| 193 | case TX_8X8: return TX_8X8; |
| 194 | case TX_16X16: return TX_16X16; |
| 195 | case TX_32X32: return TX_32X32; |
| 196 | #if CONFIG_TX64X64 |
| 197 | case TX_64X64: return TX_64X64; |
Debargha Mukherjee | 2b43501 | 2017-09-28 08:30:35 -0700 | [diff] [blame] | 198 | case TX_32X64: return TX_64X32; |
| 199 | case TX_64X32: return TX_32X64; |
Angie Chiang | 155bf9a | 2017-08-06 19:52:57 -0700 | [diff] [blame] | 200 | #endif |
| 201 | case TX_4X8: return TX_8X4; |
| 202 | case TX_8X4: return TX_4X8; |
| 203 | case TX_8X16: return TX_16X8; |
| 204 | case TX_16X8: return TX_8X16; |
| 205 | case TX_16X32: return TX_32X16; |
| 206 | case TX_32X16: return TX_16X32; |
| 207 | case TX_4X16: return TX_16X4; |
| 208 | case TX_16X4: return TX_4X16; |
| 209 | case TX_8X32: return TX_32X8; |
| 210 | case TX_32X8: return TX_8X32; |
| 211 | default: assert(0); return TX_INVALID; |
| 212 | } |
| 213 | } |
| 214 | |
Urvang Joshi | 2283d37 | 2017-10-02 17:16:45 -0700 | [diff] [blame] | 215 | static INLINE TX_TYPE av1_rotate_tx_type(TX_TYPE tx_type) { |
Angie Chiang | 155bf9a | 2017-08-06 19:52:57 -0700 | [diff] [blame] | 216 | switch (tx_type) { |
| 217 | case DCT_DCT: return DCT_DCT; |
| 218 | case ADST_DCT: return DCT_ADST; |
| 219 | case DCT_ADST: return ADST_DCT; |
| 220 | case ADST_ADST: return ADST_ADST; |
Angie Chiang | 155bf9a | 2017-08-06 19:52:57 -0700 | [diff] [blame] | 221 | case FLIPADST_DCT: return DCT_FLIPADST; |
| 222 | case DCT_FLIPADST: return FLIPADST_DCT; |
| 223 | case FLIPADST_FLIPADST: return FLIPADST_FLIPADST; |
| 224 | case ADST_FLIPADST: return FLIPADST_ADST; |
| 225 | case FLIPADST_ADST: return ADST_FLIPADST; |
| 226 | case IDTX: return IDTX; |
| 227 | case V_DCT: return H_DCT; |
| 228 | case H_DCT: return V_DCT; |
| 229 | case V_ADST: return H_ADST; |
| 230 | case H_ADST: return V_ADST; |
| 231 | case V_FLIPADST: return H_FLIPADST; |
| 232 | case H_FLIPADST: return V_FLIPADST; |
Angie Chiang | 155bf9a | 2017-08-06 19:52:57 -0700 | [diff] [blame] | 233 | default: assert(0); return TX_TYPES; |
| 234 | } |
| 235 | } |
| 236 | #endif // CONFIG_TXMG |
| 237 | |
Debargha Mukherjee | 9eabd69 | 2017-11-16 12:44:31 -0800 | [diff] [blame] | 238 | // Utility function that returns the log of the ratio of the col and row |
| 239 | // sizes. |
| 240 | static INLINE int get_rect_tx_log_ratio(int col, int row) { |
| 241 | if (col == row) return 0; |
| 242 | if (col > row) { |
| 243 | if (col == row * 2) return 1; |
| 244 | if (col == row * 4) return 2; |
| 245 | assert(0 && "Unsupported transform size"); |
| 246 | } else { |
| 247 | if (row == col * 2) return -1; |
| 248 | if (row == col * 4) return -2; |
| 249 | assert(0 && "Unsupported transform size"); |
| 250 | } |
| 251 | return 0; // Invalid |
| 252 | } |
| 253 | |
Angie Chiang | ce3ad28 | 2017-08-08 09:51:54 -0700 | [diff] [blame] | 254 | void av1_gen_fwd_stage_range(int8_t *stage_range_col, int8_t *stage_range_row, |
| 255 | const TXFM_2D_FLIP_CFG *cfg, int bd); |
| 256 | |
| 257 | void av1_gen_inv_stage_range(int8_t *stage_range_col, int8_t *stage_range_row, |
| 258 | const TXFM_2D_FLIP_CFG *cfg, int8_t fwd_shift, |
| 259 | int bd); |
| 260 | |
Urvang Joshi | c502216 | 2017-11-21 15:57:42 -0800 | [diff] [blame^] | 261 | void av1_get_fwd_txfm_cfg(TX_TYPE tx_type, TX_SIZE tx_size, |
| 262 | TXFM_2D_FLIP_CFG *cfg); |
| 263 | void av1_get_inv_txfm_cfg(TX_TYPE tx_type, TX_SIZE tx_size, |
| 264 | TXFM_2D_FLIP_CFG *cfg); |
Yaowu Xu | c27fc14 | 2016-08-22 16:08:15 -0700 | [diff] [blame] | 265 | #ifdef __cplusplus |
| 266 | } |
| 267 | #endif // __cplusplus |
| 268 | |
Yaowu Xu | f883b42 | 2016-08-30 14:01:10 -0700 | [diff] [blame] | 269 | #endif // AV1_TXFM_H_ |