av1_txfm.h 11.1 KB
Newer Older
Angie Chiang's avatar
Angie Chiang committed
1
/*
2
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
Angie Chiang's avatar
Angie Chiang committed
3
 *
4
5
6
7
8
9
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
Angie Chiang's avatar
Angie Chiang committed
10
 */
11

Yaowu Xu's avatar
Yaowu Xu committed
12
13
#ifndef AV1_TXFM_H_
#define AV1_TXFM_H_
Angie Chiang's avatar
Angie Chiang committed
14
15

#include <assert.h>
Angie Chiang's avatar
Angie Chiang committed
16
17
#include <math.h>
#include <stdio.h>
Angie Chiang's avatar
Angie Chiang committed
18

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

24
25
#define MAX_TXFM_STAGE_NUM 12

Angie Chiang's avatar
Angie Chiang committed
26
27
28
29
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)));
30
static const int32_t cospi_arr_data[7][64] = {
clang-format's avatar
clang-format committed
31
  { 1024, 1024, 1023, 1021, 1019, 1016, 1013, 1009, 1004, 999, 993, 987, 980,
clang-format's avatar
clang-format committed
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
    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
54
55
56
  { 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
57
58
59
    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
60
61
62
63
  { 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
64
65
    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
66
67
68
69
70
  { 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
71
    14359, 12785, 11204, 9616,  8022,  6424,  4821,  3216,  1608 }
clang-format's avatar
clang-format committed
72
};
Angie Chiang's avatar
Angie Chiang committed
73

74
75
76
77
static INLINE const int32_t *cospi_arr(int n) {
  return cospi_arr_data[n - cos_bit_min];
}

Angie Chiang's avatar
Angie Chiang committed
78
static INLINE int32_t round_shift(int32_t value, int bit) {
79
  assert(bit >= 1);
80
  return (value + (1 << (bit - 1))) >> bit;
Angie Chiang's avatar
Angie Chiang committed
81
82
83
84
85
86
87
}

static INLINE void round_shift_array(int32_t *arr, int size, int bit) {
  int i;
  if (bit == 0) {
    return;
  } else {
88
89
90
91
92
93
    if (bit > 0) {
      for (i = 0; i < size; i++) {
        arr[i] = round_shift(arr[i], bit);
      }
    } else {
      for (i = 0; i < size; i++) {
Yaowu Xu's avatar
Yaowu Xu committed
94
        arr[i] = arr[i] * (1 << (-bit));
95
      }
Angie Chiang's avatar
Angie Chiang committed
96
97
98
99
100
101
102
103
104
    }
  }
}

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;
105
  if (result_64 < INT32_MIN || result_64 > INT32_MAX) {
Angie Chiang's avatar
Angie Chiang committed
106
107
108
109
    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
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
    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
126
127
128
static INLINE void clamp_block(int16_t *block, int block_size_row,
                               int block_size_col, int stride, int low,
                               int high) {
Angie Chiang's avatar
Angie Chiang committed
129
  int i, j;
130
131
  for (i = 0; i < block_size_row; ++i) {
    for (j = 0; j < block_size_col; ++j) {
Angie Chiang's avatar
Angie Chiang committed
132
133
134
135
136
137
138
139
      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);

140
141
142
143
144
145
146
147
148
149
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,
150
151
152
153
  TXFM_TYPE_IDENTITY4,
  TXFM_TYPE_IDENTITY8,
  TXFM_TYPE_IDENTITY16,
  TXFM_TYPE_IDENTITY32,
154
155
} TXFM_TYPE;

156
typedef struct TXFM_1D_CFG {
Angie Chiang's avatar
Angie Chiang committed
157
  const int txfm_size;
158
  const int stage_num;
Angie Chiang's avatar
Angie Chiang committed
159

Angie Chiang's avatar
Angie Chiang committed
160
  const int8_t *shift;
161
162
163
164
  const int8_t *stage_range;
  const int8_t *cos_bit;
  const TXFM_TYPE txfm_type;
} TXFM_1D_CFG;
Angie Chiang's avatar
Angie Chiang committed
165

166
167
168
typedef struct TXFM_2D_FLIP_CFG {
  int ud_flip;  // flip upside down
  int lr_flip;  // flip left to right
169
170
  const TXFM_1D_CFG *col_cfg;
  const TXFM_1D_CFG *row_cfg;
171
172
} TXFM_2D_FLIP_CFG;

clang-format's avatar
clang-format committed
173
static INLINE void set_flip_cfg(int tx_type, TXFM_2D_FLIP_CFG *cfg) {
174
175
176
177
178
179
180
181
182
  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
183
184
185
186
187
188
189
190
    case IDTX:
    case V_DCT:
    case H_DCT:
    case V_ADST:
    case H_ADST:
      cfg->ud_flip = 0;
      cfg->lr_flip = 0;
      break;
191
    case FLIPADST_DCT:
192
    case FLIPADST_ADST:
193
    case V_FLIPADST:
194
195
196
197
      cfg->ud_flip = 1;
      cfg->lr_flip = 0;
      break;
    case DCT_FLIPADST:
198
    case ADST_FLIPADST:
199
    case H_FLIPADST:
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
      cfg->ud_flip = 0;
      cfg->lr_flip = 1;
      break;
    case FLIPADST_FLIPADST:
      cfg->ud_flip = 1;
      cfg->lr_flip = 1;
      break;
#endif  // CONFIG_EXT_TX
    default:
      cfg->ud_flip = 0;
      cfg->lr_flip = 0;
      assert(0);
  }
}

215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
#if CONFIG_TXMG
static INLINE int av1_rotate_tx_size(int tx_size) {
  switch (tx_size) {
#if CONFIG_CHROMA_2X2
    case TX_2X2: return TX_2X2;
#endif
    case TX_4X4: return TX_4X4;
    case TX_8X8: return TX_8X8;
    case TX_16X16: return TX_16X16;
    case TX_32X32: return TX_32X32;
#if CONFIG_TX64X64
    case TX_64X64: return TX_64X64;
#endif
    case TX_4X8: return TX_8X4;
    case TX_8X4: return TX_4X8;
    case TX_8X16: return TX_16X8;
    case TX_16X8: return TX_8X16;
    case TX_16X32: return TX_32X16;
    case TX_32X16: return TX_16X32;
    case TX_4X16: return TX_16X4;
    case TX_16X4: return TX_4X16;
    case TX_8X32: return TX_32X8;
    case TX_32X8: return TX_8X32;
    default: assert(0); return TX_INVALID;
  }
}

static INLINE int av1_rotate_tx_type(int tx_type) {
  switch (tx_type) {
    case DCT_DCT: return DCT_DCT;
    case ADST_DCT: return DCT_ADST;
    case DCT_ADST: return ADST_DCT;
    case ADST_ADST: return ADST_ADST;
#if CONFIG_EXT_TX
    case FLIPADST_DCT: return DCT_FLIPADST;
    case DCT_FLIPADST: return FLIPADST_DCT;
    case FLIPADST_FLIPADST: return FLIPADST_FLIPADST;
    case ADST_FLIPADST: return FLIPADST_ADST;
    case FLIPADST_ADST: return ADST_FLIPADST;
    case IDTX: return IDTX;
    case V_DCT: return H_DCT;
    case H_DCT: return V_DCT;
    case V_ADST: return H_ADST;
    case H_ADST: return V_ADST;
    case V_FLIPADST: return H_FLIPADST;
    case H_FLIPADST: return V_FLIPADST;
#endif  // CONFIG_EXT_TX
#if CONFIG_MRC_TX
    case MRC_DCT: return MRC_DCT;
#endif  // CONFIG_MRC_TX
    default: assert(0); return TX_TYPES;
  }
}
#endif  // CONFIG_TXMG

270
#if CONFIG_MRC_TX
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
static INLINE int get_mrc_mask_inter(const uint8_t *pred, int pred_stride,
                                     int *mask, int mask_stride, int width,
                                     int height) {
  // placeholder mask generation function
  int n_masked_vals = 0;
  for (int i = 0; i < height; ++i) {
    for (int j = 0; j < width; ++j) {
      mask[i * mask_stride + j] = pred[i * pred_stride + j] > 100 ? 1 : 0;
      n_masked_vals += mask[i * mask_stride + j];
    }
  }
  return n_masked_vals;
}

static INLINE int get_mrc_mask_intra(const uint8_t *pred, int pred_stride,
                                     int *mask, int mask_stride, int width,
                                     int height) {
  // placeholder mask generation function
289
  int n_masked_vals = 0;
290
  for (int i = 0; i < height; ++i) {
291
    for (int j = 0; j < width; ++j) {
292
      mask[i * mask_stride + j] = pred[i * pred_stride + j] > 100 ? 1 : 0;
293
294
      n_masked_vals += mask[i * mask_stride + j];
    }
295
  }
296
297
298
  return n_masked_vals;
}

299
300
301
302
303
304
305
306
307
308
309
310
311
312
static INLINE int get_mrc_mask(const uint8_t *pred, int pred_stride, int *mask,
                               int mask_stride, int width, int height,
                               int is_inter) {
  if (is_inter) {
    assert(USE_MRC_INTER && "MRC invalid for inter blocks");
    return get_mrc_mask_inter(pred, pred_stride, mask, mask_stride, width,
                              height);
  } else {
    assert(USE_MRC_INTRA && "MRC invalid for intra blocks");
    return get_mrc_mask_intra(pred, pred_stride, mask, mask_stride, width,
                              height);
  }
}

313
314
static INLINE int is_valid_mrc_mask(int n_masked_vals, int width, int height) {
  return !(n_masked_vals == 0 || n_masked_vals == (width * height));
315
316
317
}
#endif  // CONFIG_MRC_TX

318
319
320
321
322
323
324
void av1_gen_fwd_stage_range(int8_t *stage_range_col, int8_t *stage_range_row,
                             const TXFM_2D_FLIP_CFG *cfg, int bd);

void av1_gen_inv_stage_range(int8_t *stage_range_col, int8_t *stage_range_row,
                             const TXFM_2D_FLIP_CFG *cfg, int8_t fwd_shift,
                             int bd);

325
326
327
#ifdef __cplusplus
extern "C" {
#endif
Yaowu Xu's avatar
Yaowu Xu committed
328
329
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);
330
331
332
333
#ifdef __cplusplus
}
#endif  // __cplusplus

Yaowu Xu's avatar
Yaowu Xu committed
334
#endif  // AV1_TXFM_H_