/*
 * Copyright (c) 2021, Alliance for Open Media. All rights reserved
 *
 * This source code is subject to the terms of the BSD 3-Clause Clear License
 * and the Alliance for Open Media Patent License 1.0. If the BSD 3-Clause Clear
 * License was not distributed with this source code in the LICENSE file, you
 * can obtain it at aomedia.org/license/software-license/bsd-3-c-c/.  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
 * aomedia.org/license/patent-license/.
 */

#include "config/aom_dsp_rtcd.h"
#include "config/av1_rtcd.h"

#include "av1/common/av1_txfm.h"
#if CONFIG_ADST8_TUNED || CONFIG_ADST16_TUNED
#include "av1/common/av1_inv_txfm1d.h"
#endif

// av1_cospi_arr[i][j] = (int)round(cos(PI*j/128) * (1<<(cos_bit_min+i)));
const int32_t av1_cospi_arr_data[7][64] = {
  { 1024, 1024, 1023, 1021, 1019, 1016, 1013, 1009, 1004, 999, 993, 987, 980,
    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 },
  { 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,
    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 },
  { 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,
    15447, 14733, 14010, 13279, 12540, 11793, 11039, 10279, 9512,  8740,  7962,
    7180,  6393,  5602,  4808,  4011,  3212,  2411,  1608,  804 },
  { 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,
    14359, 12785, 11204, 9616,  8022,  6424,  4821,  3216,  1608 }
};

#if CONFIG_CROSS_CHROMA_TX
// Given a rotation angle t, the CCTX transform matrix is defined as
// [cos(t), sin(t); -sin(t), cos(t)] * 1<<CCTX_PREC_BITS). The array below only
// stores two values: cos(t) and sin(t) for each rotation angle.
const int32_t cctx_mtx[CCTX_TYPES - 1][2] = {
  { 181, 181 },   // t = 45 degrees
  { 222, 128 },   // t = 30 degrees
  { 128, 222 },   // t = 60 degrees
  { 181, -181 },  // t = -45 degrees
  { 222, -128 },  // t = -30 degrees
  { 128, -222 },  // t = -60 degrees
};
#endif  // CONFIG_CROSS_CHROMA_TX

// av1_sinpi_arr_data[i][j] = (int)round((sqrt(2) * sin(j*Pi/9) * 2 / 3) * (1
// << (cos_bit_min + i))) modified so that elements j=1,2 sum to element j=4.
const int32_t av1_sinpi_arr_data[7][5] = {
  { 0, 330, 621, 836, 951 },        { 0, 660, 1241, 1672, 1901 },
  { 0, 1321, 2482, 3344, 3803 },    { 0, 2642, 4964, 6689, 7606 },
  { 0, 5283, 9929, 13377, 15212 },  { 0, 10566, 19858, 26755, 30424 },
  { 0, 21133, 39716, 53510, 60849 }
};

#if CONFIG_ADST8_TUNED
const int32_t av2_adst_kernel8[TXFM_DIRECTIONS][TXFM_KERNEL_SIZE8] =
{
  {
    519, 1278, 1989, 2628, 3169, 3594, 3886, 4035,
    1529, 3327, 4049, 3461, 1754, -521, -2627, -3884,
    2454, 4041, 2179, -1542, -3947, -2984, 526, 3587,
    3232, 3081, -1835, -3913, 61, 3941, 1726, -3158,
    3781, 759, -4008, 440, 3877, -1599, -3398, 2616,
    3974, -1987, -1987, 3974, -1987, -1987, 3974, -1987,
    3581, -3764, 2258, 262, -2665, 3871, -3339, 1309,
    2264, -3145, 3679, -3805, 3511, -2828, 1832, -634,
  },
  {
    16, 48, 77, 101, 118, 124, 112, 71,
    40, 104, 126, 96, 24, -62, -118, -98,
    62, 127, 68, -57, -125, -62, 71, 115,
    82, 108, -48, -122, 14, 124, 8, -119,
    99, 55, -123, 2, 121, -62, -83, 110,
    112, -16, -93, 123, -50, -62, 121, -88,
    121, -82, 16, 54, -106, 124, -104, 57,
    126, -121, 112, -99, 82, -62, 41, -20,
  }
};
#endif
#if CONFIG_ADST16_TUNED
const int32_t av2_adst_kernel16[TXFM_DIRECTIONS][TXFM_KERNEL_SIZE16] =
{
  {
    383, 763, 1136, 1499, 1848, 2181, 2493, 2783, 3048, 3286, 3493, 3669, 3812, 3920, 3992, 4029,
    1136, 2181, 3048, 3669, 3992, 3992, 3669, 3048, 2181, 1136, 0, -1136, -2181, -3048, -3669, -3992,
    1848, 3286, 3992, 3812, 2783, 1136, -763, -2493, -3669, -4029, -3493, -2181, -383, 1499, 3048, 3920,
    2493, 3920, 3669, 1848, -763, -3048, -4029, -3286, -1136, 1499, 3493, 3992, 2783, 383, -2181, -3812,
    3048, 3992, 2181, -1136, -3669, -3669, -1136, 2181, 3992, 3048, 0, -3048, -3992, -2181, 1136, 3669,
    3493, 3493, 0, -3493, -3493, 0, 3493, 3493, 0, -3493, -3493, 0, 3493, 3493, 0, -3493,
    3812, 2493, -2181, -3920, -383, 3669, 2783, -1848, -3992, -763, 3493, 3048, -1499, -4029, -1136, 3286,
    3992, 1136, -3669, -2181, 3048, 3048, -2181, -3669, 1136, 3992, 0, -3992, -1136, 3669, 2181, -3048,
    4029, -383, -3992, 763, 3920, -1136, -3812, 1499, 3669, -1848, -3493, 2181, 3286, -2493, -3048, 2783,
    3920, -1848, -3048, 3286, 1499, -3992, 383, 3812, -2181, -2783, 3493, 1136, -4029, 763, 3669, -2493,
    3669, -3048, -1136, 3992, -2181, -2181, 3992, -1136, -3048, 3669, 0, -3669, 3048, 1136, -3992, 2181,
    3286, -3812, 1136, 2493, -4029, 2181, 1499, -3920, 3048, 383, -3493, 3669, -763, -2783, 3992, -1848,
    2783, -4029, 3048, -383, -2493, 3992, -3286, 763, 2181, -3920, 3493, -1136, -1848, 3812, -3669, 1499,
    2181, -3669, 3992, -3048, 1136, 1136, -3048, 3992, -3669, 2181, 0, -2181, 3669, -3992, 3048, -1136,
    1499, -2783, 3669, -4029, 3812, -3048, 1848, -383, -1136, 2493, -3493, 3992, -3920, 3286, -2181, 763,
    763, -1499, 2181, -2783, 3286, -3669, 3920, -4029, 3992, -3812, 3493, -3048, 2493, -1848, 1136, -383,
  },
  {
    12, 36, 58, 78, 95, 109, 119, 125, 126, 122, 115, 103, 87, 68, 47, 24,
    24, 68, 103, 122, 125, 109, 78, 36, -12, -58, -95, -119, -126, -115, -87, -47,
    36, 95, 125, 115, 68, 0, -68, -115, -125, -95, -36, 36, 95, 125, 115, 68,
    47, 115, 119, 58, -36, -109, -122, -68, 24, 103, 125, 78, -12, -95, -126, -87,
    58, 125, 87, -24, -115, -109, -12, 95, 122, 47, -68, -126, -78, 36, 119, 103,
    68, 125, 36, -95, -115, 0, 115, 95, -36, -125, -68, 68, 125, 36, -95, -115,
    78, 115, -24, -126, -36, 109, 87, -68, -119, 12, 125, 47, -103, -95, 58, 122,
    87, 95, -78, -103, 68, 109, -58, -115, 47, 119, -36, -122, 24, 125, -12, -126,
    95, 68, -115, -36, 125, 0, -125, 36, 115, -68, -95, 95, 68, -115, -36, 125,
    103, 36, -126, 47, 95, -109, -24, 125, -58, -87, 115, 12, -122, 68, 78, -119,
    109, 0, -109, 109, 0, -109, 109, 0, -109, 109, 0, -109, 109, 0, -109, 109,
    115, -36, -68, 125, -95, 0, 95, -125, 68, 36, -115, 115, -36, -68, 125, -95,
    119, -68, -12, 87, -125, 109, -47, -36, 103, -126, 95, -24, -58, 115, -122, 78,
    122, -95, 47, 12, -68, 109, -126, 115, -78, 24, 36, -87, 119, -125, 103, -58,
    125, -115, 95, -68, 36, 0, -36, 68, -95, 115, -125, 125, -115, 95, -68, 36,
    126, -125, 122, -119, 115, -109, 103, -95, 87, -78, 68, -58, 47, -36, 24, -12,
  }
};
#endif

#if CONFIG_ADST8_TUNED || CONFIG_ADST16_TUNED
void av2_txfm_matrix_mult(const int32_t *input, int32_t *output,
                          const int32_t *kernel, int kernel_size,
                          int8_t bit, int8_t clamp) {
    assert(kernel_size < TXFM_SIZE_MAX);

    int32_t result[TXFM_SIZE_MAX];
    memset(result, 0, sizeof(result[0]) * kernel_size);

    for(int i = 0; i < kernel_size; ++i) {
        for(int j = 0; j < kernel_size; ++j) {
            int32_t prod = kernel[kernel_size * i + j] * input[j];
            result[i] = range_check_value(prod + result[i], STAGE_RANGE_MAX);
        }
    }
    for(int i = 0; i < kernel_size; ++i)
        output[i] = round_shift(result[i], bit);

    if (clamp) {
      for(int i = 0; i < kernel_size; ++i)
        output[i] = clamp_value(output[i], clamp);
    }
}
#endif

void av1_round_shift_array_c(int32_t *arr, int size, int bit) {
  int i;
  if (bit == 0) {
    return;
  } else {
    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] = (int32_t)clamp64(((int64_t)1 << (-bit)) * arr[i], INT32_MIN,
                                  INT32_MAX);
      }
    }
  }
}

const TXFM_TYPE av1_txfm_type_ls[5][TX_TYPES_1D] = {
  { TXFM_TYPE_DCT4, TXFM_TYPE_ADST4, TXFM_TYPE_ADST4, TXFM_TYPE_IDENTITY4 },
  { TXFM_TYPE_DCT8, TXFM_TYPE_ADST8, TXFM_TYPE_ADST8, TXFM_TYPE_IDENTITY8 },
  { TXFM_TYPE_DCT16, TXFM_TYPE_ADST16, TXFM_TYPE_ADST16, TXFM_TYPE_IDENTITY16 },
  { TXFM_TYPE_DCT32, TXFM_TYPE_INVALID, TXFM_TYPE_INVALID,
    TXFM_TYPE_IDENTITY32 },
  { TXFM_TYPE_DCT64, TXFM_TYPE_INVALID, TXFM_TYPE_INVALID, TXFM_TYPE_INVALID }
};

const int8_t av1_txfm_stage_num_list[TXFM_TYPES] = {
  4,   // TXFM_TYPE_DCT4
  6,   // TXFM_TYPE_DCT8
  8,   // TXFM_TYPE_DCT16
  10,  // TXFM_TYPE_DCT32
  12,  // TXFM_TYPE_DCT64
  7,   // TXFM_TYPE_ADST4
  8,   // TXFM_TYPE_ADST8
  10,  // TXFM_TYPE_ADST16
  1,   // TXFM_TYPE_IDENTITY4
  1,   // TXFM_TYPE_IDENTITY8
  1,   // TXFM_TYPE_IDENTITY16
  1,   // TXFM_TYPE_IDENTITY32
};

void av1_range_check_buf(int32_t stage, const int32_t *input,
                         const int32_t *buf, int32_t size, int8_t bit) {
#if CONFIG_COEFFICIENT_RANGE_CHECKING
  const int64_t max_value = (1LL << (bit - 1)) - 1;
  const int64_t min_value = -(1LL << (bit - 1));

  int in_range = 1;

  for (int i = 0; i < size; ++i) {
    if (buf[i] < min_value || buf[i] > max_value) {
      in_range = 0;
    }
  }

  if (!in_range) {
    fprintf(stderr, "Error: coeffs contain out-of-range values\n");
    fprintf(stderr, "size: %d\n", size);
    fprintf(stderr, "stage: %d\n", stage);
    fprintf(stderr, "allowed range: [%" PRId64 ";%" PRId64 "]\n", min_value,
            max_value);

    fprintf(stderr, "coeffs: ");

    fprintf(stderr, "[");
    for (int j = 0; j < size; j++) {
      if (j > 0) fprintf(stderr, ", ");
      fprintf(stderr, "%d", input[j]);
    }
    fprintf(stderr, "]\n");

    fprintf(stderr, "   buf: ");

    fprintf(stderr, "[");
    for (int j = 0; j < size; j++) {
      if (j > 0) fprintf(stderr, ", ");
      fprintf(stderr, "%d", buf[j]);
    }
    fprintf(stderr, "]\n\n");
  }

  assert(in_range);
#else
  (void)stage;
  (void)input;
  (void)buf;
  (void)size;
  (void)bit;
#endif
}