tor-browser

The Tor Browser
git clone https://git.dasho.dev/tor-browser.git
Log | Files | Refs | README | LICENSE

av1_inv_txfm2d.c (19490B)


      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                                 int stride, int bd) {
     22  /* 4-point reversible, orthonormal inverse Walsh-Hadamard in 3.5 adds,
     23     0.5 shifts per pixel. */
     24  int i;
     25  tran_low_t output[16];
     26  tran_low_t a1, b1, c1, d1, e1;
     27  const tran_low_t *ip = input;
     28  tran_low_t *op = output;
     29  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
     30 
     31  for (i = 0; i < 4; i++) {
     32    a1 = ip[4 * 0] >> UNIT_QUANT_SHIFT;
     33    c1 = ip[4 * 1] >> UNIT_QUANT_SHIFT;
     34    d1 = ip[4 * 2] >> UNIT_QUANT_SHIFT;
     35    b1 = ip[4 * 3] >> UNIT_QUANT_SHIFT;
     36    a1 += c1;
     37    d1 -= b1;
     38    e1 = (a1 - d1) >> 1;
     39    b1 = e1 - b1;
     40    c1 = e1 - c1;
     41    a1 -= b1;
     42    d1 += c1;
     43 
     44    op[4 * 0] = a1;
     45    op[4 * 1] = b1;
     46    op[4 * 2] = c1;
     47    op[4 * 3] = d1;
     48    ip++;
     49    op++;
     50  }
     51 
     52  ip = output;
     53  for (i = 0; i < 4; i++) {
     54    a1 = ip[0];
     55    c1 = ip[1];
     56    d1 = ip[2];
     57    b1 = ip[3];
     58    a1 += c1;
     59    d1 -= b1;
     60    e1 = (a1 - d1) >> 1;
     61    b1 = e1 - b1;
     62    c1 = e1 - c1;
     63    a1 -= b1;
     64    d1 += c1;
     65 
     66    range_check_value(a1, bd + 1);
     67    range_check_value(b1, bd + 1);
     68    range_check_value(c1, bd + 1);
     69    range_check_value(d1, bd + 1);
     70 
     71    dest[stride * 0] = highbd_clip_pixel_add(dest[stride * 0], a1, bd);
     72    dest[stride * 1] = highbd_clip_pixel_add(dest[stride * 1], b1, bd);
     73    dest[stride * 2] = highbd_clip_pixel_add(dest[stride * 2], c1, bd);
     74    dest[stride * 3] = highbd_clip_pixel_add(dest[stride * 3], d1, bd);
     75 
     76    ip += 4;
     77    dest++;
     78  }
     79 }
     80 
     81 void av1_highbd_iwht4x4_1_add_c(const tran_low_t *in, uint8_t *dest8,
     82                                int dest_stride, int bd) {
     83  int i;
     84  tran_low_t a1, e1;
     85  tran_low_t tmp[4];
     86  const tran_low_t *ip = in;
     87  tran_low_t *op = tmp;
     88  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
     89  (void)bd;
     90 
     91  a1 = ip[0 * 4] >> UNIT_QUANT_SHIFT;
     92  e1 = a1 >> 1;
     93  a1 -= e1;
     94  op[0] = a1;
     95  op[1] = op[2] = op[3] = e1;
     96 
     97  ip = tmp;
     98  for (i = 0; i < 4; i++) {
     99    e1 = ip[0] >> 1;
    100    a1 = ip[0] - e1;
    101    dest[dest_stride * 0] =
    102        highbd_clip_pixel_add(dest[dest_stride * 0], a1, bd);
    103    dest[dest_stride * 1] =
    104        highbd_clip_pixel_add(dest[dest_stride * 1], e1, bd);
    105    dest[dest_stride * 2] =
    106        highbd_clip_pixel_add(dest[dest_stride * 2], e1, bd);
    107    dest[dest_stride * 3] =
    108        highbd_clip_pixel_add(dest[dest_stride * 3], e1, bd);
    109    ip++;
    110    dest++;
    111  }
    112 }
    113 
    114 static inline TxfmFunc inv_txfm_type_to_func(TXFM_TYPE txfm_type) {
    115  switch (txfm_type) {
    116    case TXFM_TYPE_DCT4: return av1_idct4;
    117    case TXFM_TYPE_DCT8: return av1_idct8;
    118    case TXFM_TYPE_DCT16: return av1_idct16;
    119    case TXFM_TYPE_DCT32: return av1_idct32;
    120    case TXFM_TYPE_DCT64: return av1_idct64;
    121    case TXFM_TYPE_ADST4: return av1_iadst4;
    122    case TXFM_TYPE_ADST8: return av1_iadst8;
    123    case TXFM_TYPE_ADST16: return av1_iadst16;
    124    case TXFM_TYPE_IDENTITY4: return av1_iidentity4_c;
    125    case TXFM_TYPE_IDENTITY8: return av1_iidentity8_c;
    126    case TXFM_TYPE_IDENTITY16: return av1_iidentity16_c;
    127    case TXFM_TYPE_IDENTITY32: return av1_iidentity32_c;
    128    default: assert(0); return NULL;
    129  }
    130 }
    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                          TXFM_2D_FLIP_CFG *cfg) {
    164  assert(cfg != NULL);
    165  cfg->tx_size = tx_size;
    166  av1_zero(cfg->stage_range_col);
    167  av1_zero(cfg->stage_range_row);
    168  set_flip_cfg(tx_type, cfg);
    169  const TX_TYPE_1D tx_type_1d_col = vtx_tab[tx_type];
    170  const TX_TYPE_1D tx_type_1d_row = htx_tab[tx_type];
    171  cfg->shift = av1_inv_txfm_shift_ls[tx_size];
    172  const int txw_idx = get_txw_idx(tx_size);
    173  const int txh_idx = get_txh_idx(tx_size);
    174  cfg->cos_bit_col = INV_COS_BIT;
    175  cfg->cos_bit_row = INV_COS_BIT;
    176  cfg->txfm_type_col = av1_txfm_type_ls[txh_idx][tx_type_1d_col];
    177  if (cfg->txfm_type_col == TXFM_TYPE_ADST4) {
    178    memcpy(cfg->stage_range_col, iadst4_range, sizeof(iadst4_range));
    179  }
    180  cfg->txfm_type_row = av1_txfm_type_ls[txw_idx][tx_type_1d_row];
    181  if (cfg->txfm_type_row == TXFM_TYPE_ADST4) {
    182    memcpy(cfg->stage_range_row, iadst4_range, sizeof(iadst4_range));
    183  }
    184  cfg->stage_num_col = av1_txfm_stage_num_list[cfg->txfm_type_col];
    185  cfg->stage_num_row = av1_txfm_stage_num_list[cfg->txfm_type_row];
    186 }
    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                             int bd) {
    191  const int fwd_shift = inv_start_range[tx_size];
    192  const int8_t *shift = cfg->shift;
    193  int8_t opt_range_row, opt_range_col;
    194  if (bd == 8) {
    195    opt_range_row = 16;
    196    opt_range_col = 16;
    197  } else if (bd == 10) {
    198    opt_range_row = 18;
    199    opt_range_col = 16;
    200  } else {
    201    assert(bd == 12);
    202    opt_range_row = 20;
    203    opt_range_col = 18;
    204  }
    205  // i < MAX_TXFM_STAGE_NUM will mute above array bounds warning
    206  for (int i = 0; i < cfg->stage_num_row && i < MAX_TXFM_STAGE_NUM; ++i) {
    207    int real_range_row = cfg->stage_range_row[i] + fwd_shift + bd + 1;
    208    (void)real_range_row;
    209    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      stage_range_row[i] = opt_range_row;
    213    } else {
    214      assert(opt_range_row >= real_range_row);
    215      stage_range_row[i] = opt_range_row;
    216    }
    217  }
    218  // i < MAX_TXFM_STAGE_NUM will mute above array bounds warning
    219  for (int i = 0; i < cfg->stage_num_col && i < MAX_TXFM_STAGE_NUM; ++i) {
    220    int real_range_col =
    221        cfg->stage_range_col[i] + fwd_shift + shift[0] + bd + 1;
    222    (void)real_range_col;
    223    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      stage_range_col[i] = opt_range_col;
    227    } else {
    228      assert(opt_range_col >= real_range_col);
    229      stage_range_col[i] = opt_range_col;
    230    }
    231  }
    232 }
    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                                    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  const int txfm_size_col = tx_size_wide[cfg->tx_size];
    245  const int txfm_size_row = tx_size_high[cfg->tx_size];
    246  // Take the shift from the larger dimension in the rectangular case.
    247  const int8_t *shift = cfg->shift;
    248  const int rect_type = get_rect_tx_log_ratio(txfm_size_col, txfm_size_row);
    249  int8_t stage_range_row[MAX_TXFM_STAGE_NUM];
    250  int8_t stage_range_col[MAX_TXFM_STAGE_NUM];
    251  assert(cfg->stage_num_row <= MAX_TXFM_STAGE_NUM);
    252  assert(cfg->stage_num_col <= MAX_TXFM_STAGE_NUM);
    253  av1_gen_inv_stage_range(stage_range_col, stage_range_row, cfg, tx_size, bd);
    254 
    255  const int8_t cos_bit_col = cfg->cos_bit_col;
    256  const int8_t cos_bit_row = cfg->cos_bit_row;
    257  const TxfmFunc txfm_func_col = inv_txfm_type_to_func(cfg->txfm_type_col);
    258  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  const int buf_offset = AOMMAX(txfm_size_row, txfm_size_col);
    264  int32_t *temp_in = txfm_buf;
    265  int32_t *temp_out = temp_in + buf_offset;
    266  int32_t *buf = temp_out + buf_offset;
    267  int32_t *buf_ptr = buf;
    268  int c, r;
    269 
    270  // Rows
    271  for (r = 0; r < txfm_size_row; ++r) {
    272    if (abs(rect_type) == 1) {
    273      for (c = 0; c < txfm_size_col; ++c) {
    274        temp_in[c] = round_shift(
    275            (int64_t)input[c * txfm_size_row + r] * NewInvSqrt2, NewSqrt2Bits);
    276      }
    277      clamp_buf(temp_in, txfm_size_col, bd + 8);
    278      txfm_func_row(temp_in, buf_ptr, cos_bit_row, stage_range_row);
    279    } else {
    280      for (c = 0; c < txfm_size_col; ++c) {
    281        temp_in[c] = input[c * txfm_size_row + r];
    282      }
    283      clamp_buf(temp_in, txfm_size_col, bd + 8);
    284      txfm_func_row(temp_in, buf_ptr, cos_bit_row, stage_range_row);
    285    }
    286    av1_round_shift_array(buf_ptr, txfm_size_col, -shift[0]);
    287    buf_ptr += txfm_size_col;
    288  }
    289 
    290  // Columns
    291  for (c = 0; c < txfm_size_col; ++c) {
    292    if (cfg->lr_flip == 0) {
    293      for (r = 0; r < txfm_size_row; ++r)
    294        temp_in[r] = buf[r * txfm_size_col + c];
    295    } else {
    296      // flip left right
    297      for (r = 0; r < txfm_size_row; ++r)
    298        temp_in[r] = buf[r * txfm_size_col + (txfm_size_col - c - 1)];
    299    }
    300    clamp_buf(temp_in, txfm_size_row, AOMMAX(bd + 6, 16));
    301    txfm_func_col(temp_in, temp_out, cos_bit_col, stage_range_col);
    302    av1_round_shift_array(temp_out, txfm_size_row, -shift[1]);
    303    if (cfg->ud_flip == 0) {
    304      for (r = 0; r < txfm_size_row; ++r) {
    305        output[r * stride + c] =
    306            highbd_clip_pixel_add(output[r * stride + c], temp_out[r], bd);
    307      }
    308    } else {
    309      // flip upside down
    310      for (r = 0; r < txfm_size_row; ++r) {
    311        output[r * stride + c] = highbd_clip_pixel_add(
    312            output[r * stride + c], temp_out[txfm_size_row - r - 1], bd);
    313      }
    314    }
    315  }
    316 }
    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                                         int bd) {
    322  TXFM_2D_FLIP_CFG cfg;
    323  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  inv_txfm2d_add_c(input, output, stride, &cfg, txfm_buf, tx_size, bd);
    327 }
    328 
    329 void av1_inv_txfm2d_add_4x8_c(const int32_t *input, uint16_t *output,
    330                              int stride, TX_TYPE tx_type, int bd) {
    331  DECLARE_ALIGNED(32, int, txfm_buf[4 * 8 + 8 + 8]);
    332  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_4X8, bd);
    333 }
    334 
    335 void av1_inv_txfm2d_add_8x4_c(const int32_t *input, uint16_t *output,
    336                              int stride, TX_TYPE tx_type, int bd) {
    337  DECLARE_ALIGNED(32, int, txfm_buf[8 * 4 + 8 + 8]);
    338  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_8X4, bd);
    339 }
    340 
    341 void av1_inv_txfm2d_add_8x16_c(const int32_t *input, uint16_t *output,
    342                               int stride, TX_TYPE tx_type, int bd) {
    343  DECLARE_ALIGNED(32, int, txfm_buf[8 * 16 + 16 + 16]);
    344  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_8X16, bd);
    345 }
    346 
    347 void av1_inv_txfm2d_add_16x8_c(const int32_t *input, uint16_t *output,
    348                               int stride, TX_TYPE tx_type, int bd) {
    349  DECLARE_ALIGNED(32, int, txfm_buf[16 * 8 + 16 + 16]);
    350  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_16X8, bd);
    351 }
    352 
    353 void av1_inv_txfm2d_add_16x32_c(const int32_t *input, uint16_t *output,
    354                                int stride, TX_TYPE tx_type, int bd) {
    355  DECLARE_ALIGNED(32, int, txfm_buf[16 * 32 + 32 + 32]);
    356  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_16X32, bd);
    357 }
    358 
    359 void av1_inv_txfm2d_add_32x16_c(const int32_t *input, uint16_t *output,
    360                                int stride, TX_TYPE tx_type, int bd) {
    361  DECLARE_ALIGNED(32, int, txfm_buf[32 * 16 + 32 + 32]);
    362  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_32X16, bd);
    363 }
    364 
    365 void av1_inv_txfm2d_add_4x4_c(const int32_t *input, uint16_t *output,
    366                              int stride, TX_TYPE tx_type, int bd) {
    367  DECLARE_ALIGNED(32, int, txfm_buf[4 * 4 + 4 + 4]);
    368  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_4X4, bd);
    369 }
    370 
    371 void av1_inv_txfm2d_add_8x8_c(const int32_t *input, uint16_t *output,
    372                              int stride, TX_TYPE tx_type, int bd) {
    373  DECLARE_ALIGNED(32, int, txfm_buf[8 * 8 + 8 + 8]);
    374  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_8X8, bd);
    375 }
    376 
    377 void av1_inv_txfm2d_add_16x16_c(const int32_t *input, uint16_t *output,
    378                                int stride, TX_TYPE tx_type, int bd) {
    379  DECLARE_ALIGNED(32, int, txfm_buf[16 * 16 + 16 + 16]);
    380  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_16X16, bd);
    381 }
    382 
    383 void av1_inv_txfm2d_add_32x32_c(const int32_t *input, uint16_t *output,
    384                                int stride, TX_TYPE tx_type, int bd) {
    385  DECLARE_ALIGNED(32, int, txfm_buf[32 * 32 + 32 + 32]);
    386  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_32X32, bd);
    387 }
    388 
    389 void av1_inv_txfm2d_add_64x64_c(const int32_t *input, uint16_t *output,
    390                                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  int32_t mod_input[64 * 64];
    396  for (int col = 0; col < 32; ++col) {
    397    memcpy(mod_input + col * 64, input + col * 32, 32 * sizeof(*mod_input));
    398    memset(mod_input + col * 64 + 32, 0, 32 * sizeof(*mod_input));
    399  }
    400  memset(mod_input + 32 * 64, 0, 32 * 64 * sizeof(*mod_input));
    401  DECLARE_ALIGNED(32, int, txfm_buf[64 * 64 + 64 + 64]);
    402  inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_64X64,
    403                        bd);
    404 }
    405 
    406 void av1_inv_txfm2d_add_64x32_c(const int32_t *input, uint16_t *output,
    407                                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  int32_t mod_input[32 * 64];
    412  memcpy(mod_input, input, 32 * 32 * sizeof(*mod_input));
    413  memset(mod_input + 32 * 32, 0, 32 * 32 * sizeof(*mod_input));
    414  DECLARE_ALIGNED(32, int, txfm_buf[64 * 32 + 64 + 64]);
    415  inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_64X32,
    416                        bd);
    417 }
    418 
    419 void av1_inv_txfm2d_add_32x64_c(const int32_t *input, uint16_t *output,
    420                                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  int32_t mod_input[64 * 32];
    425  for (int col = 0; col < 32; ++col) {
    426    memcpy(mod_input + col * 64, input + col * 32, 32 * sizeof(*mod_input));
    427    memset(mod_input + col * 64 + 32, 0, 32 * sizeof(*mod_input));
    428  }
    429  DECLARE_ALIGNED(32, int, txfm_buf[64 * 32 + 64 + 64]);
    430  inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_32X64,
    431                        bd);
    432 }
    433 
    434 void av1_inv_txfm2d_add_16x64_c(const int32_t *input, uint16_t *output,
    435                                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  int32_t mod_input[64 * 16];
    440  for (int col = 0; col < 16; ++col) {
    441    memcpy(mod_input + col * 64, input + col * 32, 32 * sizeof(*mod_input));
    442    memset(mod_input + col * 64 + 32, 0, 32 * sizeof(*mod_input));
    443  }
    444  DECLARE_ALIGNED(32, int, txfm_buf[16 * 64 + 64 + 64]);
    445  inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_16X64,
    446                        bd);
    447 }
    448 
    449 void av1_inv_txfm2d_add_64x16_c(const int32_t *input, uint16_t *output,
    450                                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  int32_t mod_input[16 * 64];
    455  memcpy(mod_input, input, 16 * 32 * sizeof(*mod_input));
    456  memset(mod_input + 16 * 32, 0, 16 * 32 * sizeof(*mod_input));
    457  DECLARE_ALIGNED(32, int, txfm_buf[16 * 64 + 64 + 64]);
    458  inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_64X16,
    459                        bd);
    460 }
    461 
    462 void av1_inv_txfm2d_add_4x16_c(const int32_t *input, uint16_t *output,
    463                               int stride, TX_TYPE tx_type, int bd) {
    464  DECLARE_ALIGNED(32, int, txfm_buf[4 * 16 + 16 + 16]);
    465  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_4X16, bd);
    466 }
    467 
    468 void av1_inv_txfm2d_add_16x4_c(const int32_t *input, uint16_t *output,
    469                               int stride, TX_TYPE tx_type, int bd) {
    470  DECLARE_ALIGNED(32, int, txfm_buf[4 * 16 + 16 + 16]);
    471  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_16X4, bd);
    472 }
    473 
    474 void av1_inv_txfm2d_add_8x32_c(const int32_t *input, uint16_t *output,
    475                               int stride, TX_TYPE tx_type, int bd) {
    476  DECLARE_ALIGNED(32, int, txfm_buf[8 * 32 + 32 + 32]);
    477  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_8X32, bd);
    478 }
    479 
    480 void av1_inv_txfm2d_add_32x8_c(const int32_t *input, uint16_t *output,
    481                               int stride, TX_TYPE tx_type, int bd) {
    482  DECLARE_ALIGNED(32, int, txfm_buf[8 * 32 + 32 + 32]);
    483  inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_32X8, bd);
    484 }