Coverage Report

Created: 2025-06-13 07:07

/src/aom/av1/common/av1_inv_txfm2d.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 "config/aom_dsp_rtcd.h"
13
#include "config/av1_rtcd.h"
14
15
#include "av1/common/enums.h"
16
#include "av1/common/av1_txfm.h"
17
#include "av1/common/av1_inv_txfm1d.h"
18
#include "av1/common/av1_inv_txfm1d_cfg.h"
19
20
void av1_highbd_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest8,
21
0
                                 int stride, int bd) {
22
  /* 4-point reversible, orthonormal inverse Walsh-Hadamard in 3.5 adds,
23
     0.5 shifts per pixel. */
24
0
  int i;
25
0
  tran_low_t output[16];
26
0
  tran_low_t a1, b1, c1, d1, e1;
27
0
  const tran_low_t *ip = input;
28
0
  tran_low_t *op = output;
29
0
  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
30
31
0
  for (i = 0; i < 4; i++) {
32
0
    a1 = ip[4 * 0] >> UNIT_QUANT_SHIFT;
33
0
    c1 = ip[4 * 1] >> UNIT_QUANT_SHIFT;
34
0
    d1 = ip[4 * 2] >> UNIT_QUANT_SHIFT;
35
0
    b1 = ip[4 * 3] >> UNIT_QUANT_SHIFT;
36
0
    a1 += c1;
37
0
    d1 -= b1;
38
0
    e1 = (a1 - d1) >> 1;
39
0
    b1 = e1 - b1;
40
0
    c1 = e1 - c1;
41
0
    a1 -= b1;
42
0
    d1 += c1;
43
44
0
    op[4 * 0] = a1;
45
0
    op[4 * 1] = b1;
46
0
    op[4 * 2] = c1;
47
0
    op[4 * 3] = d1;
48
0
    ip++;
49
0
    op++;
50
0
  }
51
52
0
  ip = output;
53
0
  for (i = 0; i < 4; i++) {
54
0
    a1 = ip[0];
55
0
    c1 = ip[1];
56
0
    d1 = ip[2];
57
0
    b1 = ip[3];
58
0
    a1 += c1;
59
0
    d1 -= b1;
60
0
    e1 = (a1 - d1) >> 1;
61
0
    b1 = e1 - b1;
62
0
    c1 = e1 - c1;
63
0
    a1 -= b1;
64
0
    d1 += c1;
65
66
0
    range_check_value(a1, bd + 1);
67
0
    range_check_value(b1, bd + 1);
68
0
    range_check_value(c1, bd + 1);
69
0
    range_check_value(d1, bd + 1);
70
71
0
    dest[stride * 0] = highbd_clip_pixel_add(dest[stride * 0], a1, bd);
72
0
    dest[stride * 1] = highbd_clip_pixel_add(dest[stride * 1], b1, bd);
73
0
    dest[stride * 2] = highbd_clip_pixel_add(dest[stride * 2], c1, bd);
74
0
    dest[stride * 3] = highbd_clip_pixel_add(dest[stride * 3], d1, bd);
75
76
0
    ip += 4;
77
0
    dest++;
78
0
  }
79
0
}
80
81
void av1_highbd_iwht4x4_1_add_c(const tran_low_t *in, uint8_t *dest8,
82
630k
                                int dest_stride, int bd) {
83
630k
  int i;
84
630k
  tran_low_t a1, e1;
85
630k
  tran_low_t tmp[4];
86
630k
  const tran_low_t *ip = in;
87
630k
  tran_low_t *op = tmp;
88
630k
  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
89
630k
  (void)bd;
90
91
630k
  a1 = ip[0 * 4] >> UNIT_QUANT_SHIFT;
92
630k
  e1 = a1 >> 1;
93
630k
  a1 -= e1;
94
630k
  op[0] = a1;
95
630k
  op[1] = op[2] = op[3] = e1;
96
97
630k
  ip = tmp;
98
3.15M
  for (i = 0; i < 4; i++) {
99
2.52M
    e1 = ip[0] >> 1;
100
2.52M
    a1 = ip[0] - e1;
101
2.52M
    dest[dest_stride * 0] =
102
2.52M
        highbd_clip_pixel_add(dest[dest_stride * 0], a1, bd);
103
2.52M
    dest[dest_stride * 1] =
104
2.52M
        highbd_clip_pixel_add(dest[dest_stride * 1], e1, bd);
105
2.52M
    dest[dest_stride * 2] =
106
2.52M
        highbd_clip_pixel_add(dest[dest_stride * 2], e1, bd);
107
2.52M
    dest[dest_stride * 3] =
108
2.52M
        highbd_clip_pixel_add(dest[dest_stride * 3], e1, bd);
109
2.52M
    ip++;
110
2.52M
    dest++;
111
2.52M
  }
112
630k
}
113
114
0
static inline TxfmFunc inv_txfm_type_to_func(TXFM_TYPE txfm_type) {
115
0
  switch (txfm_type) {
116
0
    case TXFM_TYPE_DCT4: return av1_idct4;
117
0
    case TXFM_TYPE_DCT8: return av1_idct8;
118
0
    case TXFM_TYPE_DCT16: return av1_idct16;
119
0
    case TXFM_TYPE_DCT32: return av1_idct32;
120
0
    case TXFM_TYPE_DCT64: return av1_idct64;
121
0
    case TXFM_TYPE_ADST4: return av1_iadst4;
122
0
    case TXFM_TYPE_ADST8: return av1_iadst8;
123
0
    case TXFM_TYPE_ADST16: return av1_iadst16;
124
0
    case TXFM_TYPE_IDENTITY4: return av1_iidentity4_c;
125
0
    case TXFM_TYPE_IDENTITY8: return av1_iidentity8_c;
126
0
    case TXFM_TYPE_IDENTITY16: return av1_iidentity16_c;
127
0
    case TXFM_TYPE_IDENTITY32: return av1_iidentity32_c;
128
0
    default: assert(0); return NULL;
129
0
  }
130
0
}
131
132
static const int8_t inv_shift_4x4[2] = { 0, -4 };
133
static const int8_t inv_shift_8x8[2] = { -1, -4 };
134
static const int8_t inv_shift_16x16[2] = { -2, -4 };
135
static const int8_t inv_shift_32x32[2] = { -2, -4 };
136
static const int8_t inv_shift_64x64[2] = { -2, -4 };
137
static const int8_t inv_shift_4x8[2] = { 0, -4 };
138
static const int8_t inv_shift_8x4[2] = { 0, -4 };
139
static const int8_t inv_shift_8x16[2] = { -1, -4 };
140
static const int8_t inv_shift_16x8[2] = { -1, -4 };
141
static const int8_t inv_shift_16x32[2] = { -1, -4 };
142
static const int8_t inv_shift_32x16[2] = { -1, -4 };
143
static const int8_t inv_shift_32x64[2] = { -1, -4 };
144
static const int8_t inv_shift_64x32[2] = { -1, -4 };
145
static const int8_t inv_shift_4x16[2] = { -1, -4 };
146
static const int8_t inv_shift_16x4[2] = { -1, -4 };
147
static const int8_t inv_shift_8x32[2] = { -2, -4 };
148
static const int8_t inv_shift_32x8[2] = { -2, -4 };
149
static const int8_t inv_shift_16x64[2] = { -2, -4 };
150
static const int8_t inv_shift_64x16[2] = { -2, -4 };
151
152
const int8_t *av1_inv_txfm_shift_ls[TX_SIZES_ALL] = {
153
  inv_shift_4x4,   inv_shift_8x8,   inv_shift_16x16, inv_shift_32x32,
154
  inv_shift_64x64, inv_shift_4x8,   inv_shift_8x4,   inv_shift_8x16,
155
  inv_shift_16x8,  inv_shift_16x32, inv_shift_32x16, inv_shift_32x64,
156
  inv_shift_64x32, inv_shift_4x16,  inv_shift_16x4,  inv_shift_8x32,
157
  inv_shift_32x8,  inv_shift_16x64, inv_shift_64x16,
158
};
159
160
static const int8_t iadst4_range[7] = { 0, 1, 0, 0, 0, 0, 0 };
161
162
void av1_get_inv_txfm_cfg(TX_TYPE tx_type, TX_SIZE tx_size,
163
0
                          TXFM_2D_FLIP_CFG *cfg) {
164
0
  assert(cfg != NULL);
165
0
  cfg->tx_size = tx_size;
166
0
  av1_zero(cfg->stage_range_col);
167
0
  av1_zero(cfg->stage_range_row);
168
0
  set_flip_cfg(tx_type, cfg);
169
0
  const TX_TYPE_1D tx_type_1d_col = vtx_tab[tx_type];
170
0
  const TX_TYPE_1D tx_type_1d_row = htx_tab[tx_type];
171
0
  cfg->shift = av1_inv_txfm_shift_ls[tx_size];
172
0
  const int txw_idx = get_txw_idx(tx_size);
173
0
  const int txh_idx = get_txh_idx(tx_size);
174
0
  cfg->cos_bit_col = INV_COS_BIT;
175
0
  cfg->cos_bit_row = INV_COS_BIT;
176
0
  cfg->txfm_type_col = av1_txfm_type_ls[txh_idx][tx_type_1d_col];
177
0
  if (cfg->txfm_type_col == TXFM_TYPE_ADST4) {
178
0
    memcpy(cfg->stage_range_col, iadst4_range, sizeof(iadst4_range));
179
0
  }
180
0
  cfg->txfm_type_row = av1_txfm_type_ls[txw_idx][tx_type_1d_row];
181
0
  if (cfg->txfm_type_row == TXFM_TYPE_ADST4) {
182
0
    memcpy(cfg->stage_range_row, iadst4_range, sizeof(iadst4_range));
183
0
  }
184
0
  cfg->stage_num_col = av1_txfm_stage_num_list[cfg->txfm_type_col];
185
0
  cfg->stage_num_row = av1_txfm_stage_num_list[cfg->txfm_type_row];
186
0
}
187
188
void av1_gen_inv_stage_range(int8_t *stage_range_col, int8_t *stage_range_row,
189
                             const TXFM_2D_FLIP_CFG *cfg, TX_SIZE tx_size,
190
0
                             int bd) {
191
0
  const int fwd_shift = inv_start_range[tx_size];
192
0
  const int8_t *shift = cfg->shift;
193
0
  int8_t opt_range_row, opt_range_col;
194
0
  if (bd == 8) {
195
0
    opt_range_row = 16;
196
0
    opt_range_col = 16;
197
0
  } else if (bd == 10) {
198
0
    opt_range_row = 18;
199
0
    opt_range_col = 16;
200
0
  } else {
201
0
    assert(bd == 12);
202
0
    opt_range_row = 20;
203
0
    opt_range_col = 18;
204
0
  }
205
  // i < MAX_TXFM_STAGE_NUM will mute above array bounds warning
206
0
  for (int i = 0; i < cfg->stage_num_row && i < MAX_TXFM_STAGE_NUM; ++i) {
207
0
    int real_range_row = cfg->stage_range_row[i] + fwd_shift + bd + 1;
208
0
    (void)real_range_row;
209
0
    if (cfg->txfm_type_row == TXFM_TYPE_ADST4 && i == 1) {
210
      // the adst4 may use 1 extra bit on top of opt_range_row at stage 1
211
      // so opt_range_row >= real_range_row will not hold
212
0
      stage_range_row[i] = opt_range_row;
213
0
    } else {
214
0
      assert(opt_range_row >= real_range_row);
215
0
      stage_range_row[i] = opt_range_row;
216
0
    }
217
0
  }
218
  // i < MAX_TXFM_STAGE_NUM will mute above array bounds warning
219
0
  for (int i = 0; i < cfg->stage_num_col && i < MAX_TXFM_STAGE_NUM; ++i) {
220
0
    int real_range_col =
221
0
        cfg->stage_range_col[i] + fwd_shift + shift[0] + bd + 1;
222
0
    (void)real_range_col;
223
0
    if (cfg->txfm_type_col == TXFM_TYPE_ADST4 && i == 1) {
224
      // the adst4 may use 1 extra bit on top of opt_range_col at stage 1
225
      // so opt_range_col >= real_range_col will not hold
226
0
      stage_range_col[i] = opt_range_col;
227
0
    } else {
228
0
      assert(opt_range_col >= real_range_col);
229
0
      stage_range_col[i] = opt_range_col;
230
0
    }
231
0
  }
232
0
}
233
234
static inline void inv_txfm2d_add_c(const int32_t *input, uint16_t *output,
235
                                    int stride, TXFM_2D_FLIP_CFG *cfg,
236
                                    int32_t *txfm_buf, TX_SIZE tx_size,
237
0
                                    int bd) {
238
  // Note when assigning txfm_size_col, we use the txfm_size from the
239
  // row configuration and vice versa. This is intentionally done to
240
  // accurately perform rectangular transforms. When the transform is
241
  // rectangular, the number of columns will be the same as the
242
  // txfm_size stored in the row cfg struct. It will make no difference
243
  // for square transforms.
244
0
  const int txfm_size_col = tx_size_wide[cfg->tx_size];
245
0
  const int txfm_size_row = tx_size_high[cfg->tx_size];
246
  // Take the shift from the larger dimension in the rectangular case.
247
0
  const int8_t *shift = cfg->shift;
248
0
  const int rect_type = get_rect_tx_log_ratio(txfm_size_col, txfm_size_row);
249
0
  int8_t stage_range_row[MAX_TXFM_STAGE_NUM];
250
0
  int8_t stage_range_col[MAX_TXFM_STAGE_NUM];
251
0
  assert(cfg->stage_num_row <= MAX_TXFM_STAGE_NUM);
252
0
  assert(cfg->stage_num_col <= MAX_TXFM_STAGE_NUM);
253
0
  av1_gen_inv_stage_range(stage_range_col, stage_range_row, cfg, tx_size, bd);
254
255
0
  const int8_t cos_bit_col = cfg->cos_bit_col;
256
0
  const int8_t cos_bit_row = cfg->cos_bit_row;
257
0
  const TxfmFunc txfm_func_col = inv_txfm_type_to_func(cfg->txfm_type_col);
258
0
  const TxfmFunc txfm_func_row = inv_txfm_type_to_func(cfg->txfm_type_row);
259
260
  // txfm_buf's length is  txfm_size_row * txfm_size_col + 2 *
261
  // AOMMAX(txfm_size_row, txfm_size_col)
262
  // it is used for intermediate data buffering
263
0
  const int buf_offset = AOMMAX(txfm_size_row, txfm_size_col);
264
0
  int32_t *temp_in = txfm_buf;
265
0
  int32_t *temp_out = temp_in + buf_offset;
266
0
  int32_t *buf = temp_out + buf_offset;
267
0
  int32_t *buf_ptr = buf;
268
0
  int c, r;
269
270
  // Rows
271
0
  for (r = 0; r < txfm_size_row; ++r) {
272
0
    if (abs(rect_type) == 1) {
273
0
      for (c = 0; c < txfm_size_col; ++c) {
274
0
        temp_in[c] = round_shift(
275
0
            (int64_t)input[c * txfm_size_row + r] * NewInvSqrt2, NewSqrt2Bits);
276
0
      }
277
0
      clamp_buf(temp_in, txfm_size_col, bd + 8);
278
0
      txfm_func_row(temp_in, buf_ptr, cos_bit_row, stage_range_row);
279
0
    } else {
280
0
      for (c = 0; c < txfm_size_col; ++c) {
281
0
        temp_in[c] = input[c * txfm_size_row + r];
282
0
      }
283
0
      clamp_buf(temp_in, txfm_size_col, bd + 8);
284
0
      txfm_func_row(temp_in, buf_ptr, cos_bit_row, stage_range_row);
285
0
    }
286
0
    av1_round_shift_array(buf_ptr, txfm_size_col, -shift[0]);
287
0
    buf_ptr += txfm_size_col;
288
0
  }
289
290
  // Columns
291
0
  for (c = 0; c < txfm_size_col; ++c) {
292
0
    if (cfg->lr_flip == 0) {
293
0
      for (r = 0; r < txfm_size_row; ++r)
294
0
        temp_in[r] = buf[r * txfm_size_col + c];
295
0
    } else {
296
      // flip left right
297
0
      for (r = 0; r < txfm_size_row; ++r)
298
0
        temp_in[r] = buf[r * txfm_size_col + (txfm_size_col - c - 1)];
299
0
    }
300
0
    clamp_buf(temp_in, txfm_size_row, AOMMAX(bd + 6, 16));
301
0
    txfm_func_col(temp_in, temp_out, cos_bit_col, stage_range_col);
302
0
    av1_round_shift_array(temp_out, txfm_size_row, -shift[1]);
303
0
    if (cfg->ud_flip == 0) {
304
0
      for (r = 0; r < txfm_size_row; ++r) {
305
0
        output[r * stride + c] =
306
0
            highbd_clip_pixel_add(output[r * stride + c], temp_out[r], bd);
307
0
      }
308
0
    } else {
309
      // flip upside down
310
0
      for (r = 0; r < txfm_size_row; ++r) {
311
0
        output[r * stride + c] = highbd_clip_pixel_add(
312
0
            output[r * stride + c], temp_out[txfm_size_row - r - 1], bd);
313
0
      }
314
0
    }
315
0
  }
316
0
}
317
318
static inline void inv_txfm2d_add_facade(const int32_t *input, uint16_t *output,
319
                                         int stride, int32_t *txfm_buf,
320
                                         TX_TYPE tx_type, TX_SIZE tx_size,
321
0
                                         int bd) {
322
0
  TXFM_2D_FLIP_CFG cfg;
323
0
  av1_get_inv_txfm_cfg(tx_type, tx_size, &cfg);
324
  // Forward shift sum uses larger square size, to be consistent with what
325
  // av1_gen_inv_stage_range() does for inverse shifts.
326
0
  inv_txfm2d_add_c(input, output, stride, &cfg, txfm_buf, tx_size, bd);
327
0
}
328
329
void av1_inv_txfm2d_add_4x8_c(const int32_t *input, uint16_t *output,
330
0
                              int stride, TX_TYPE tx_type, int bd) {
331
0
  DECLARE_ALIGNED(32, int, txfm_buf[4 * 8 + 8 + 8]);
332
0
  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_4X8, bd);
333
0
}
334
335
void av1_inv_txfm2d_add_8x4_c(const int32_t *input, uint16_t *output,
336
0
                              int stride, TX_TYPE tx_type, int bd) {
337
0
  DECLARE_ALIGNED(32, int, txfm_buf[8 * 4 + 8 + 8]);
338
0
  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_8X4, bd);
339
0
}
340
341
void av1_inv_txfm2d_add_8x16_c(const int32_t *input, uint16_t *output,
342
0
                               int stride, TX_TYPE tx_type, int bd) {
343
0
  DECLARE_ALIGNED(32, int, txfm_buf[8 * 16 + 16 + 16]);
344
0
  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_8X16, bd);
345
0
}
346
347
void av1_inv_txfm2d_add_16x8_c(const int32_t *input, uint16_t *output,
348
0
                               int stride, TX_TYPE tx_type, int bd) {
349
0
  DECLARE_ALIGNED(32, int, txfm_buf[16 * 8 + 16 + 16]);
350
0
  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_16X8, bd);
351
0
}
352
353
void av1_inv_txfm2d_add_16x32_c(const int32_t *input, uint16_t *output,
354
0
                                int stride, TX_TYPE tx_type, int bd) {
355
0
  DECLARE_ALIGNED(32, int, txfm_buf[16 * 32 + 32 + 32]);
356
0
  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_16X32, bd);
357
0
}
358
359
void av1_inv_txfm2d_add_32x16_c(const int32_t *input, uint16_t *output,
360
0
                                int stride, TX_TYPE tx_type, int bd) {
361
0
  DECLARE_ALIGNED(32, int, txfm_buf[32 * 16 + 32 + 32]);
362
0
  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_32X16, bd);
363
0
}
364
365
void av1_inv_txfm2d_add_4x4_c(const int32_t *input, uint16_t *output,
366
0
                              int stride, TX_TYPE tx_type, int bd) {
367
0
  DECLARE_ALIGNED(32, int, txfm_buf[4 * 4 + 4 + 4]);
368
0
  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_4X4, bd);
369
0
}
370
371
void av1_inv_txfm2d_add_8x8_c(const int32_t *input, uint16_t *output,
372
0
                              int stride, TX_TYPE tx_type, int bd) {
373
0
  DECLARE_ALIGNED(32, int, txfm_buf[8 * 8 + 8 + 8]);
374
0
  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_8X8, bd);
375
0
}
376
377
void av1_inv_txfm2d_add_16x16_c(const int32_t *input, uint16_t *output,
378
0
                                int stride, TX_TYPE tx_type, int bd) {
379
0
  DECLARE_ALIGNED(32, int, txfm_buf[16 * 16 + 16 + 16]);
380
0
  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_16X16, bd);
381
0
}
382
383
void av1_inv_txfm2d_add_32x32_c(const int32_t *input, uint16_t *output,
384
0
                                int stride, TX_TYPE tx_type, int bd) {
385
0
  DECLARE_ALIGNED(32, int, txfm_buf[32 * 32 + 32 + 32]);
386
0
  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_32X32, bd);
387
0
}
388
389
void av1_inv_txfm2d_add_64x64_c(const int32_t *input, uint16_t *output,
390
0
                                int stride, TX_TYPE tx_type, int bd) {
391
  // TODO(urvang): Can the same array be reused, instead of using a new array?
392
  // Remap 32x32 input into a modified 64x64 by:
393
  // - Copying over these values in top-left 32x32 locations.
394
  // - Setting the rest of the locations to 0.
395
0
  int32_t mod_input[64 * 64];
396
0
  for (int col = 0; col < 32; ++col) {
397
0
    memcpy(mod_input + col * 64, input + col * 32, 32 * sizeof(*mod_input));
398
0
    memset(mod_input + col * 64 + 32, 0, 32 * sizeof(*mod_input));
399
0
  }
400
0
  memset(mod_input + 32 * 64, 0, 32 * 64 * sizeof(*mod_input));
401
0
  DECLARE_ALIGNED(32, int, txfm_buf[64 * 64 + 64 + 64]);
402
0
  inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_64X64,
403
0
                        bd);
404
0
}
405
406
void av1_inv_txfm2d_add_64x32_c(const int32_t *input, uint16_t *output,
407
0
                                int stride, TX_TYPE tx_type, int bd) {
408
  // Remap 32x32 input into a modified 64x32 by:
409
  // - Copying over these values in top-left 32x32 locations.
410
  // - Setting the rest of the locations to 0.
411
0
  int32_t mod_input[32 * 64];
412
0
  memcpy(mod_input, input, 32 * 32 * sizeof(*mod_input));
413
0
  memset(mod_input + 32 * 32, 0, 32 * 32 * sizeof(*mod_input));
414
0
  DECLARE_ALIGNED(32, int, txfm_buf[64 * 32 + 64 + 64]);
415
0
  inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_64X32,
416
0
                        bd);
417
0
}
418
419
void av1_inv_txfm2d_add_32x64_c(const int32_t *input, uint16_t *output,
420
0
                                int stride, TX_TYPE tx_type, int bd) {
421
  // Remap 32x32 input into a modified 32x64 input by:
422
  // - Copying over these values in top-left 32x32 locations.
423
  // - Setting the rest of the locations to 0.
424
0
  int32_t mod_input[64 * 32];
425
0
  for (int col = 0; col < 32; ++col) {
426
0
    memcpy(mod_input + col * 64, input + col * 32, 32 * sizeof(*mod_input));
427
0
    memset(mod_input + col * 64 + 32, 0, 32 * sizeof(*mod_input));
428
0
  }
429
0
  DECLARE_ALIGNED(32, int, txfm_buf[64 * 32 + 64 + 64]);
430
0
  inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_32X64,
431
0
                        bd);
432
0
}
433
434
void av1_inv_txfm2d_add_16x64_c(const int32_t *input, uint16_t *output,
435
0
                                int stride, TX_TYPE tx_type, int bd) {
436
  // Remap 16x32 input into a modified 16x64 input by:
437
  // - Copying over these values in top-left 16x32 locations.
438
  // - Setting the rest of the locations to 0.
439
0
  int32_t mod_input[64 * 16];
440
0
  for (int col = 0; col < 16; ++col) {
441
0
    memcpy(mod_input + col * 64, input + col * 32, 32 * sizeof(*mod_input));
442
0
    memset(mod_input + col * 64 + 32, 0, 32 * sizeof(*mod_input));
443
0
  }
444
0
  DECLARE_ALIGNED(32, int, txfm_buf[16 * 64 + 64 + 64]);
445
0
  inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_16X64,
446
0
                        bd);
447
0
}
448
449
void av1_inv_txfm2d_add_64x16_c(const int32_t *input, uint16_t *output,
450
0
                                int stride, TX_TYPE tx_type, int bd) {
451
  // Remap 32x16 input into a modified 64x16 by:
452
  // - Copying over these values in top-left 32x16 locations.
453
  // - Setting the rest of the locations to 0.
454
0
  int32_t mod_input[16 * 64];
455
0
  memcpy(mod_input, input, 16 * 32 * sizeof(*mod_input));
456
0
  memset(mod_input + 16 * 32, 0, 16 * 32 * sizeof(*mod_input));
457
0
  DECLARE_ALIGNED(32, int, txfm_buf[16 * 64 + 64 + 64]);
458
0
  inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_64X16,
459
0
                        bd);
460
0
}
461
462
void av1_inv_txfm2d_add_4x16_c(const int32_t *input, uint16_t *output,
463
0
                               int stride, TX_TYPE tx_type, int bd) {
464
0
  DECLARE_ALIGNED(32, int, txfm_buf[4 * 16 + 16 + 16]);
465
0
  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_4X16, bd);
466
0
}
467
468
void av1_inv_txfm2d_add_16x4_c(const int32_t *input, uint16_t *output,
469
0
                               int stride, TX_TYPE tx_type, int bd) {
470
0
  DECLARE_ALIGNED(32, int, txfm_buf[4 * 16 + 16 + 16]);
471
0
  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_16X4, bd);
472
0
}
473
474
void av1_inv_txfm2d_add_8x32_c(const int32_t *input, uint16_t *output,
475
0
                               int stride, TX_TYPE tx_type, int bd) {
476
0
  DECLARE_ALIGNED(32, int, txfm_buf[8 * 32 + 32 + 32]);
477
0
  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_8X32, bd);
478
0
}
479
480
void av1_inv_txfm2d_add_32x8_c(const int32_t *input, uint16_t *output,
481
0
                               int stride, TX_TYPE tx_type, int bd) {
482
0
  DECLARE_ALIGNED(32, int, txfm_buf[8 * 32 + 32 + 32]);
483
0
  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_32X8, bd);
484
0
}