av1_txfm.h 7.01 KB
Newer Older
Angie Chiang's avatar
Angie Chiang committed
1
2
3
4
5
6
7
8
9
/*
 *  Copyright (c) 2015 The WebM project authors. All Rights Reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS.  All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */
Yaowu Xu's avatar
Yaowu Xu committed
10
11
#ifndef AV1_TXFM_H_
#define AV1_TXFM_H_
Angie Chiang's avatar
Angie Chiang committed
12
13

#include <assert.h>
Angie Chiang's avatar
Angie Chiang committed
14
15
#include <math.h>
#include <stdio.h>
Angie Chiang's avatar
Angie Chiang committed
16

17
#include "av1/common/enums.h"
Yaowu Xu's avatar
Yaowu Xu committed
18
19
#include "aom/aom_integer.h"
#include "aom_dsp/aom_dsp_common.h"
Angie Chiang's avatar
Angie Chiang committed
20
21
22
23
24

static const int cos_bit_min = 10;
static const int cos_bit_max = 16;

// cospi_arr[i][j] = (int)round(cos(M_PI*j/128) * (1<<(cos_bit_min+i)));
clang-format's avatar
clang-format committed
25
26
static const int32_t cospi_arr[7][64] = {
  { 1024, 1024, 1023, 1021, 1019, 1016, 1013, 1009, 1004, 999, 993, 987, 980,
clang-format's avatar
clang-format committed
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
    972,  964,  955,  946,  936,  926,  915,  903,  891,  878, 865, 851, 837,
    822,  807,  792,  775,  759,  742,  724,  706,  688,  669, 650, 630, 610,
    590,  569,  548,  526,  505,  483,  460,  438,  415,  392, 369, 345, 321,
    297,  273,  249,  224,  200,  175,  150,  125,  100,  75,  50,  25 },
  { 2048, 2047, 2046, 2042, 2038, 2033, 2026, 2018, 2009, 1998, 1987,
    1974, 1960, 1945, 1928, 1911, 1892, 1872, 1851, 1829, 1806, 1782,
    1757, 1730, 1703, 1674, 1645, 1615, 1583, 1551, 1517, 1483, 1448,
    1412, 1375, 1338, 1299, 1260, 1220, 1179, 1138, 1096, 1053, 1009,
    965,  921,  876,  830,  784,  737,  690,  642,  595,  546,  498,
    449,  400,  350,  301,  251,  201,  151,  100,  50 },
  { 4096, 4095, 4091, 4085, 4076, 4065, 4052, 4036, 4017, 3996, 3973,
    3948, 3920, 3889, 3857, 3822, 3784, 3745, 3703, 3659, 3612, 3564,
    3513, 3461, 3406, 3349, 3290, 3229, 3166, 3102, 3035, 2967, 2896,
    2824, 2751, 2675, 2598, 2520, 2440, 2359, 2276, 2191, 2106, 2019,
    1931, 1842, 1751, 1660, 1567, 1474, 1380, 1285, 1189, 1092, 995,
    897,  799,  700,  601,  501,  401,  301,  201,  101 },
  { 8192, 8190, 8182, 8170, 8153, 8130, 8103, 8071, 8035, 7993, 7946,
    7895, 7839, 7779, 7713, 7643, 7568, 7489, 7405, 7317, 7225, 7128,
    7027, 6921, 6811, 6698, 6580, 6458, 6333, 6203, 6070, 5933, 5793,
    5649, 5501, 5351, 5197, 5040, 4880, 4717, 4551, 4383, 4212, 4038,
    3862, 3683, 3503, 3320, 3135, 2948, 2760, 2570, 2378, 2185, 1990,
    1795, 1598, 1401, 1202, 1003, 803,  603,  402,  201 },
clang-format's avatar
clang-format committed
49
50
51
  { 16384, 16379, 16364, 16340, 16305, 16261, 16207, 16143, 16069, 15986, 15893,
    15791, 15679, 15557, 15426, 15286, 15137, 14978, 14811, 14635, 14449, 14256,
    14053, 13842, 13623, 13395, 13160, 12916, 12665, 12406, 12140, 11866, 11585,
clang-format's avatar
clang-format committed
52
53
54
    11297, 11003, 10702, 10394, 10080, 9760,  9434,  9102,  8765,  8423,  8076,
    7723,  7366,  7005,  6639,  6270,  5897,  5520,  5139,  4756,  4370,  3981,
    3590,  3196,  2801,  2404,  2006,  1606,  1205,  804,   402 },
clang-format's avatar
clang-format committed
55
56
57
58
  { 32768, 32758, 32729, 32679, 32610, 32522, 32413, 32286, 32138, 31972, 31786,
    31581, 31357, 31114, 30853, 30572, 30274, 29957, 29622, 29269, 28899, 28511,
    28106, 27684, 27246, 26791, 26320, 25833, 25330, 24812, 24279, 23732, 23170,
    22595, 22006, 21403, 20788, 20160, 19520, 18868, 18205, 17531, 16846, 16151,
clang-format's avatar
clang-format committed
59
60
    15447, 14733, 14010, 13279, 12540, 11793, 11039, 10279, 9512,  8740,  7962,
    7180,  6393,  5602,  4808,  4011,  3212,  2411,  1608,  804 },
clang-format's avatar
clang-format committed
61
62
63
64
65
  { 65536, 65516, 65457, 65358, 65220, 65043, 64827, 64571, 64277, 63944, 63572,
    63162, 62714, 62228, 61705, 61145, 60547, 59914, 59244, 58538, 57798, 57022,
    56212, 55368, 54491, 53581, 52639, 51665, 50660, 49624, 48559, 47464, 46341,
    45190, 44011, 42806, 41576, 40320, 39040, 37736, 36410, 35062, 33692, 32303,
    30893, 29466, 28020, 26558, 25080, 23586, 22078, 20557, 19024, 17479, 15924,
clang-format's avatar
clang-format committed
66
    14359, 12785, 11204, 9616,  8022,  6424,  4821,  3216,  1608 }
clang-format's avatar
clang-format committed
67
};
Angie Chiang's avatar
Angie Chiang committed
68
69

static INLINE int32_t round_shift(int32_t value, int bit) {
70
  return (value + (1 << (bit - 1))) >> bit;
Angie Chiang's avatar
Angie Chiang committed
71
72
73
74
75
76
77
}

static INLINE void round_shift_array(int32_t *arr, int size, int bit) {
  int i;
  if (bit == 0) {
    return;
  } else {
78
79
80
81
82
83
84
85
    if (bit > 0) {
      for (i = 0; i < size; i++) {
        arr[i] = round_shift(arr[i], bit);
      }
    } else {
      for (i = 0; i < size; i++) {
        arr[i] = arr[i] << (-bit);
      }
Angie Chiang's avatar
Angie Chiang committed
86
87
88
89
90
91
92
93
94
95
    }
  }
}

static INLINE int32_t half_btf(int32_t w0, int32_t in0, int32_t w1, int32_t in1,
                               int bit) {
  int32_t result_32 = w0 * in0 + w1 * in1;
#if CONFIG_COEFFICIENT_RANGE_CHECKING
  int64_t result_64 = (int64_t)w0 * (int64_t)in0 + (int64_t)w1 * (int64_t)in1;
  if (result_32 != result_64) {
Angie Chiang's avatar
Angie Chiang committed
96
97
98
99
    printf("%s %d overflow result_32: %d result_64: %" PRId64
           " w0: %d in0: %d w1: %d in1: "
           "%d\n",
           __FILE__, __LINE__, result_32, result_64, w0, in0, w1, in1);
Angie Chiang's avatar
Angie Chiang committed
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
    assert(0 && "half_btf overflow");
  }
#endif
  return round_shift(result_32, bit);
}

static INLINE int get_max_bit(int x) {
  int max_bit = -1;
  while (x) {
    x = x >> 1;
    max_bit++;
  }
  return max_bit;
}

// TODO(angiebird): implement SSE
static INLINE void clamp_block(int16_t *block, int block_size, int stride,
                               int low, int high) {
  int i, j;
  for (i = 0; i < block_size; ++i) {
    for (j = 0; j < block_size; ++j) {
      block[i * stride + j] = clamp(block[i * stride + j], low, high);
    }
  }
}

typedef void (*TxfmFunc)(const int32_t *input, int32_t *output,
                         const int8_t *cos_bit, const int8_t *stage_range);

129
130
131
132
133
134
135
136
137
138
139
140
typedef enum TXFM_TYPE {
  TXFM_TYPE_DCT4,
  TXFM_TYPE_DCT8,
  TXFM_TYPE_DCT16,
  TXFM_TYPE_DCT32,
  TXFM_TYPE_DCT64,
  TXFM_TYPE_ADST4,
  TXFM_TYPE_ADST8,
  TXFM_TYPE_ADST16,
  TXFM_TYPE_ADST32,
} TXFM_TYPE;

Angie Chiang's avatar
Angie Chiang committed
141
typedef struct TXFM_2D_CFG {
Angie Chiang's avatar
Angie Chiang committed
142
143
144
  const int txfm_size;
  const int stage_num_col;
  const int stage_num_row;
Angie Chiang's avatar
Angie Chiang committed
145

Angie Chiang's avatar
Angie Chiang committed
146
147
148
149
150
  const int8_t *shift;
  const int8_t *stage_range_col;
  const int8_t *stage_range_row;
  const int8_t *cos_bit_col;
  const int8_t *cos_bit_row;
151
152
  const TXFM_TYPE txfm_type_col;
  const TXFM_TYPE txfm_type_row;
Angie Chiang's avatar
Angie Chiang committed
153
154
} TXFM_2D_CFG;

155
156
157
typedef struct TXFM_2D_FLIP_CFG {
  int ud_flip;  // flip upside down
  int lr_flip;  // flip left to right
clang-format's avatar
clang-format committed
158
  const TXFM_2D_CFG *cfg;
159
160
} TXFM_2D_FLIP_CFG;

clang-format's avatar
clang-format committed
161
static INLINE void set_flip_cfg(int tx_type, TXFM_2D_FLIP_CFG *cfg) {
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
  switch (tx_type) {
    case DCT_DCT:
    case ADST_DCT:
    case DCT_ADST:
    case ADST_ADST:
      cfg->ud_flip = 0;
      cfg->lr_flip = 0;
      break;
#if CONFIG_EXT_TX
    case FLIPADST_DCT:
      cfg->ud_flip = 1;
      cfg->lr_flip = 0;
      break;
    case DCT_FLIPADST:
      cfg->ud_flip = 0;
      cfg->lr_flip = 1;
      break;
    case FLIPADST_FLIPADST:
      cfg->ud_flip = 1;
      cfg->lr_flip = 1;
      break;
    case ADST_FLIPADST:
      cfg->ud_flip = 0;
      cfg->lr_flip = 1;
      break;
    case FLIPADST_ADST:
      cfg->ud_flip = 1;
      cfg->lr_flip = 0;
      break;
#endif  // CONFIG_EXT_TX
    default:
      cfg->ud_flip = 0;
      cfg->lr_flip = 0;
      assert(0);
  }
}

199
200
201
#ifdef __cplusplus
extern "C" {
#endif
Yaowu Xu's avatar
Yaowu Xu committed
202
203
TXFM_2D_FLIP_CFG av1_get_fwd_txfm_cfg(int tx_type, int tx_size);
TXFM_2D_FLIP_CFG av1_get_fwd_txfm_64x64_cfg(int tx_type);
204
205
206
207
#ifdef __cplusplus
}
#endif  // __cplusplus

Yaowu Xu's avatar
Yaowu Xu committed
208
#endif  // AV1_TXFM_H_