av1_quantize.c 69 KB
Newer Older
Jingning Han's avatar
Jingning Han committed
1
/*
Yaowu Xu's avatar
Yaowu Xu committed
2
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
Jingning Han's avatar
Jingning Han committed
3
 *
Yaowu Xu's avatar
Yaowu Xu committed
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.
Jingning Han's avatar
Jingning Han committed
10 11 12
 */

#include <math.h>
Yaowu Xu's avatar
Yaowu Xu committed
13
#include "./aom_dsp_rtcd.h"
14
#include "aom_dsp/quantize.h"
Yaowu Xu's avatar
Yaowu Xu committed
15
#include "aom_mem/aom_mem.h"
16
#include "aom_ports/mem.h"
Jingning Han's avatar
Jingning Han committed
17

18
#include "av1/common/idct.h"
19 20 21
#include "av1/common/quant_common.h"
#include "av1/common/scan.h"
#include "av1/common/seg_common.h"
Jingning Han's avatar
Jingning Han committed
22

23
#include "av1/encoder/av1_quantize.h"
24 25
#include "av1/encoder/encoder.h"
#include "av1/encoder/rd.h"
Jingning Han's avatar
Jingning Han committed
26

27
#if CONFIG_NEW_QUANT
28 29 30 31 32
static INLINE int quantize_coeff_nuq(
    const tran_low_t coeffv, const int16_t quant, const int16_t quant_shift,
    const int16_t dequant, const tran_low_t *cuml_bins_ptr,
    const tran_low_t *dequant_val, tran_low_t *qcoeff_ptr,
    tran_low_t *dqcoeff_ptr) {
33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48
  const int coeff = coeffv;
  const int coeff_sign = (coeff >> 31);
  const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
  int i, q;
  int tmp = clamp(abs_coeff, INT16_MIN, INT16_MAX);
  for (i = 0; i < NUQ_KNOTS; i++) {
    if (tmp < cuml_bins_ptr[i]) {
      q = i;
      break;
    }
  }
  if (i == NUQ_KNOTS) {
    tmp -= cuml_bins_ptr[NUQ_KNOTS - 1];
    q = NUQ_KNOTS + (((((tmp * quant) >> 16) + tmp) * quant_shift) >> 16);
  }
  if (q) {
Yaowu Xu's avatar
Yaowu Xu committed
49
    *dqcoeff_ptr = av1_dequant_abscoeff_nuq(q, dequant, dequant_val);
50
    *qcoeff_ptr = (q ^ coeff_sign) - coeff_sign;
51 52 53 54 55 56 57 58
    *dqcoeff_ptr = *qcoeff_ptr < 0 ? -*dqcoeff_ptr : *dqcoeff_ptr;
  } else {
    *qcoeff_ptr = 0;
    *dqcoeff_ptr = 0;
  }
  return (q != 0);
}

59 60 61 62
static INLINE int quantize_coeff_bigtx_nuq(
    const tran_low_t coeffv, const int16_t quant, const int16_t quant_shift,
    const int16_t dequant, const tran_low_t *cuml_bins_ptr,
    const tran_low_t *dequant_val, tran_low_t *qcoeff_ptr,
63
    tran_low_t *dqcoeff_ptr, int logsizeby16) {
64 65 66 67 68 69
  const int coeff = coeffv;
  const int coeff_sign = (coeff >> 31);
  const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
  int i, q;
  int tmp = clamp(abs_coeff, INT16_MIN, INT16_MAX);
  for (i = 0; i < NUQ_KNOTS; i++) {
70
    if (tmp < ROUND_POWER_OF_TWO(cuml_bins_ptr[i], logsizeby16)) {
71 72 73 74 75
      q = i;
      break;
    }
  }
  if (i == NUQ_KNOTS) {
76
    tmp -= ROUND_POWER_OF_TWO(cuml_bins_ptr[NUQ_KNOTS - 1], logsizeby16);
77
    q = NUQ_KNOTS +
78
        (((((tmp * quant) >> 16) + tmp) * quant_shift) >> (16 - logsizeby16));
79 80
  }
  if (q) {
81
    *dqcoeff_ptr = ROUND_POWER_OF_TWO(
82
        av1_dequant_abscoeff_nuq(q, dequant, dequant_val), logsizeby16);
Yaowu Xu's avatar
Yaowu Xu committed
83
    // *dqcoeff_ptr = av1_dequant_abscoeff_nuq(q, dequant, dequant_val) >>
84
    // (logsizeby16);
85
    *qcoeff_ptr = (q ^ coeff_sign) - coeff_sign;
86 87 88 89 90 91 92 93
    *dqcoeff_ptr = *qcoeff_ptr < 0 ? -*dqcoeff_ptr : *dqcoeff_ptr;
  } else {
    *qcoeff_ptr = 0;
    *dqcoeff_ptr = 0;
  }
  return (q != 0);
}

94 95 96 97
static INLINE int quantize_coeff_fp_nuq(
    const tran_low_t coeffv, const int16_t quant, const int16_t dequant,
    const tran_low_t *cuml_bins_ptr, const tran_low_t *dequant_val,
    tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr) {
98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113
  const int coeff = coeffv;
  const int coeff_sign = (coeff >> 31);
  const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
  int i, q;
  int tmp = clamp(abs_coeff, INT16_MIN, INT16_MAX);
  for (i = 0; i < NUQ_KNOTS; i++) {
    if (tmp < cuml_bins_ptr[i]) {
      q = i;
      break;
    }
  }
  if (i == NUQ_KNOTS) {
    q = NUQ_KNOTS +
        ((((int64_t)tmp - cuml_bins_ptr[NUQ_KNOTS - 1]) * quant) >> 16);
  }
  if (q) {
Yaowu Xu's avatar
Yaowu Xu committed
114
    *dqcoeff_ptr = av1_dequant_abscoeff_nuq(q, dequant, dequant_val);
115
    *qcoeff_ptr = (q ^ coeff_sign) - coeff_sign;
116 117 118 119 120 121 122 123
    *dqcoeff_ptr = *qcoeff_ptr < 0 ? -*dqcoeff_ptr : *dqcoeff_ptr;
  } else {
    *qcoeff_ptr = 0;
    *dqcoeff_ptr = 0;
  }
  return (q != 0);
}

124 125 126
static INLINE int quantize_coeff_bigtx_fp_nuq(
    const tran_low_t coeffv, const int16_t quant, const int16_t dequant,
    const tran_low_t *cuml_bins_ptr, const tran_low_t *dequant_val,
127
    tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, int logsizeby16) {
128 129 130 131 132 133
  const int coeff = coeffv;
  const int coeff_sign = (coeff >> 31);
  const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
  int i, q;
  int tmp = clamp(abs_coeff, INT16_MIN, INT16_MAX);
  for (i = 0; i < NUQ_KNOTS; i++) {
134
    if (tmp < ROUND_POWER_OF_TWO(cuml_bins_ptr[i], logsizeby16)) {
135 136 137 138 139 140
      q = i;
      break;
    }
  }
  if (i == NUQ_KNOTS) {
    q = NUQ_KNOTS +
141
        ((((int64_t)tmp -
142
           ROUND_POWER_OF_TWO(cuml_bins_ptr[NUQ_KNOTS - 1], logsizeby16)) *
143
          quant) >>
144
         (16 - logsizeby16));
145 146
  }
  if (q) {
147
    *dqcoeff_ptr = ROUND_POWER_OF_TWO(
148
        av1_dequant_abscoeff_nuq(q, dequant, dequant_val), logsizeby16);
Yaowu Xu's avatar
Yaowu Xu committed
149
    // *dqcoeff_ptr = av1_dequant_abscoeff_nuq(q, dequant, dequant_val) >>
150
    // (logsizeby16);
151
    *qcoeff_ptr = (q ^ coeff_sign) - coeff_sign;
152 153 154 155 156 157 158 159
    *dqcoeff_ptr = *qcoeff_ptr < 0 ? -*dqcoeff_ptr : *dqcoeff_ptr;
  } else {
    *qcoeff_ptr = 0;
    *dqcoeff_ptr = 0;
  }
  return (q != 0);
}

160 161 162
void quantize_dc_nuq(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                     int skip_block, const int16_t quant,
                     const int16_t quant_shift, const int16_t dequant,
163
                     const tran_low_t *cuml_bins_ptr,
164 165
                     const tran_low_t *dequant_val, tran_low_t *qcoeff_ptr,
                     tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr) {
166 167 168 169 170
  int eob = -1;
  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
  if (!skip_block) {
    const int rc = 0;
171 172
    if (quantize_coeff_nuq(coeff_ptr[rc], quant, quant_shift, dequant,
                           cuml_bins_ptr, dequant_val, qcoeff_ptr, dqcoeff_ptr))
173 174 175 176 177
      eob = 0;
  }
  *eob_ptr = eob + 1;
}

178 179 180 181 182
void quantize_dc_fp_nuq(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                        int skip_block, const int16_t quant,
                        const int16_t dequant, const tran_low_t *cuml_bins_ptr,
                        const tran_low_t *dequant_val, tran_low_t *qcoeff_ptr,
                        tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr) {
183 184 185 186 187
  int eob = -1;
  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
  if (!skip_block) {
    const int rc = 0;
188 189
    if (quantize_coeff_fp_nuq(coeff_ptr[rc], quant, dequant, cuml_bins_ptr,
                              dequant_val, qcoeff_ptr, dqcoeff_ptr))
190 191 192 193 194
      eob = 0;
  }
  *eob_ptr = eob + 1;
}

195 196 197
void quantize_dc_32x32_nuq(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                           int skip_block, const int16_t quant,
                           const int16_t quant_shift, const int16_t dequant,
198 199
                           const tran_low_t *cuml_bins_ptr,
                           const tran_low_t *dequant_val,
200
                           tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
201 202 203 204 205 206
                           uint16_t *eob_ptr) {
  int eob = -1;
  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
  if (!skip_block) {
    const int rc = 0;
207 208
    if (quantize_coeff_bigtx_nuq(coeff_ptr[rc], quant, quant_shift, dequant,
                                 cuml_bins_ptr, dequant_val, qcoeff_ptr,
209
                                 dqcoeff_ptr, av1_get_tx_scale(TX_32X32)))
210 211 212 213 214
      eob = 0;
  }
  *eob_ptr = eob + 1;
}

215 216
void quantize_dc_32x32_fp_nuq(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                              int skip_block, const int16_t quant,
217 218 219
                              const int16_t dequant,
                              const tran_low_t *cuml_bins_ptr,
                              const tran_low_t *dequant_val,
220
                              tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
221 222 223 224 225 226
                              uint16_t *eob_ptr) {
  int eob = -1;
  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
  if (!skip_block) {
    const int rc = 0;
227 228
    if (quantize_coeff_bigtx_fp_nuq(coeff_ptr[rc], quant, dequant,
                                    cuml_bins_ptr, dequant_val, qcoeff_ptr,
229
                                    dqcoeff_ptr, av1_get_tx_scale(TX_32X32)))
230 231 232 233 234
      eob = 0;
  }
  *eob_ptr = eob + 1;
}

235 236 237 238 239 240 241 242 243 244 245 246 247 248 249
#if CONFIG_TX64X64
void quantize_dc_64x64_nuq(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                           int skip_block, const int16_t quant,
                           const int16_t quant_shift, const int16_t dequant,
                           const tran_low_t *cuml_bins_ptr,
                           const tran_low_t *dequant_val,
                           tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                           uint16_t *eob_ptr) {
  int eob = -1;
  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
  if (!skip_block) {
    const int rc = 0;
    if (quantize_coeff_bigtx_nuq(coeff_ptr[rc], quant, quant_shift, dequant,
                                 cuml_bins_ptr, dequant_val, qcoeff_ptr,
250
                                 dqcoeff_ptr, av1_get_tx_scale(TX_64X64)))
251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269
      eob = 0;
  }
  *eob_ptr = eob + 1;
}

void quantize_dc_64x64_fp_nuq(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                              int skip_block, const int16_t quant,
                              const int16_t dequant,
                              const tran_low_t *cuml_bins_ptr,
                              const tran_low_t *dequant_val,
                              tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                              uint16_t *eob_ptr) {
  int eob = -1;
  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
  if (!skip_block) {
    const int rc = 0;
    if (quantize_coeff_bigtx_fp_nuq(coeff_ptr[rc], quant, dequant,
                                    cuml_bins_ptr, dequant_val, qcoeff_ptr,
270
                                    dqcoeff_ptr, av1_get_tx_scale(TX_64X64)))
271 272 273 274 275 276
      eob = 0;
  }
  *eob_ptr = eob + 1;
}
#endif  // CONFIG_TX64X64

277 278 279
void quantize_nuq_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                    int skip_block, const int16_t *quant_ptr,
                    const int16_t *quant_shift_ptr, const int16_t *dequant_ptr,
280 281
                    const cuml_bins_type_nuq *cuml_bins_ptr,
                    const dequant_val_type_nuq *dequant_val,
282 283
                    tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                    uint16_t *eob_ptr, const int16_t *scan,
284 285 286 287 288 289 290 291
                    const uint8_t *band) {
  int eob = -1;
  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
  if (!skip_block) {
    int i;
    for (i = 0; i < n_coeffs; i++) {
      const int rc = scan[i];
292 293 294 295
      if (quantize_coeff_nuq(coeff_ptr[rc], quant_ptr[rc != 0],
                             quant_shift_ptr[rc != 0], dequant_ptr[rc != 0],
                             cuml_bins_ptr[band[i]], dequant_val[band[i]],
                             &qcoeff_ptr[rc], &dqcoeff_ptr[rc]))
296 297 298 299 300 301
        eob = i;
    }
  }
  *eob_ptr = eob + 1;
}

302 303
void quantize_fp_nuq_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                       int skip_block, const int16_t *quant_ptr,
304 305 306
                       const int16_t *dequant_ptr,
                       const cuml_bins_type_nuq *cuml_bins_ptr,
                       const dequant_val_type_nuq *dequant_val,
307 308
                       tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                       uint16_t *eob_ptr, const int16_t *scan,
309 310 311 312 313 314 315 316
                       const uint8_t *band) {
  int eob = -1;
  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
  if (!skip_block) {
    int i;
    for (i = 0; i < n_coeffs; i++) {
      const int rc = scan[i];
317 318 319
      if (quantize_coeff_fp_nuq(coeff_ptr[rc], quant_ptr[rc != 0],
                                dequant_ptr[rc != 0], cuml_bins_ptr[band[i]],
                                dequant_val[band[i]], &qcoeff_ptr[rc],
320 321 322 323 324 325 326
                                &dqcoeff_ptr[rc]))
        eob = i;
    }
  }
  *eob_ptr = eob + 1;
}

327 328
void quantize_32x32_nuq_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                          int skip_block, const int16_t *quant_ptr,
329 330 331 332
                          const int16_t *quant_shift_ptr,
                          const int16_t *dequant_ptr,
                          const cuml_bins_type_nuq *cuml_bins_ptr,
                          const dequant_val_type_nuq *dequant_val,
333 334
                          tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                          uint16_t *eob_ptr, const int16_t *scan,
335 336 337 338 339 340 341 342
                          const uint8_t *band) {
  int eob = -1;
  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
  if (!skip_block) {
    int i;
    for (i = 0; i < n_coeffs; i++) {
      const int rc = scan[i];
343 344 345 346 347
      if (quantize_coeff_bigtx_nuq(
              coeff_ptr[rc], quant_ptr[rc != 0], quant_shift_ptr[rc != 0],
              dequant_ptr[rc != 0], cuml_bins_ptr[band[i]],
              dequant_val[band[i]], &qcoeff_ptr[rc], &dqcoeff_ptr[rc],
              av1_get_tx_scale(TX_32X32)))
348 349 350 351 352 353
        eob = i;
    }
  }
  *eob_ptr = eob + 1;
}

354 355
void quantize_32x32_fp_nuq_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                             int skip_block, const int16_t *quant_ptr,
356 357 358
                             const int16_t *dequant_ptr,
                             const cuml_bins_type_nuq *cuml_bins_ptr,
                             const dequant_val_type_nuq *dequant_val,
359 360
                             tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                             uint16_t *eob_ptr, const int16_t *scan,
361 362 363 364 365 366 367 368
                             const uint8_t *band) {
  int eob = -1;
  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
  if (!skip_block) {
    int i;
    for (i = 0; i < n_coeffs; i++) {
      const int rc = scan[i];
369 370 371
      if (quantize_coeff_bigtx_fp_nuq(
              coeff_ptr[rc], quant_ptr[rc != 0], dequant_ptr[rc != 0],
              cuml_bins_ptr[band[i]], dequant_val[band[i]], &qcoeff_ptr[rc],
372
              &dqcoeff_ptr[rc], av1_get_tx_scale(TX_32X32)))
373 374 375 376 377
        eob = i;
    }
  }
  *eob_ptr = eob + 1;
}
378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395

#if CONFIG_TX64X64
void quantize_64x64_nuq_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                          int skip_block, const int16_t *quant_ptr,
                          const int16_t *quant_shift_ptr,
                          const int16_t *dequant_ptr,
                          const cuml_bins_type_nuq *cuml_bins_ptr,
                          const dequant_val_type_nuq *dequant_val,
                          tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                          uint16_t *eob_ptr, const int16_t *scan,
                          const uint8_t *band) {
  int eob = -1;
  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
  if (!skip_block) {
    int i;
    for (i = 0; i < n_coeffs; i++) {
      const int rc = scan[i];
396 397 398 399 400
      if (quantize_coeff_bigtx_nuq(
              coeff_ptr[rc], quant_ptr[rc != 0], quant_shift_ptr[rc != 0],
              dequant_ptr[rc != 0], cuml_bins_ptr[band[i]],
              dequant_val[band[i]], &qcoeff_ptr[rc], &dqcoeff_ptr[rc],
              av1_get_tx_scale(TX_64X64)))
401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424
        eob = i;
    }
  }
  *eob_ptr = eob + 1;
}

void quantize_64x64_fp_nuq_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                             int skip_block, const int16_t *quant_ptr,
                             const int16_t *dequant_ptr,
                             const cuml_bins_type_nuq *cuml_bins_ptr,
                             const dequant_val_type_nuq *dequant_val,
                             tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                             uint16_t *eob_ptr, const int16_t *scan,
                             const uint8_t *band) {
  int eob = -1;
  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
  if (!skip_block) {
    int i;
    for (i = 0; i < n_coeffs; i++) {
      const int rc = scan[i];
      if (quantize_coeff_bigtx_fp_nuq(
              coeff_ptr[rc], quant_ptr[rc != 0], dequant_ptr[rc != 0],
              cuml_bins_ptr[band[i]], dequant_val[band[i]], &qcoeff_ptr[rc],
425
              &dqcoeff_ptr[rc], av1_get_tx_scale(TX_64X64)))
426 427 428 429 430 431
        eob = i;
    }
  }
  *eob_ptr = eob + 1;
}
#endif  // CONFIG_TX64X64
432 433
#endif  // CONFIG_NEW_QUANT

Yaowu Xu's avatar
Yaowu Xu committed
434 435
void av1_quantize_skip(intptr_t n_coeffs, tran_low_t *qcoeff_ptr,
                       tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr) {
Angie Chiang's avatar
Angie Chiang committed
436 437 438 439 440
  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
  *eob_ptr = 0;
}

441 442 443 444 445
static void quantize_fp_helper_c(
    const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block,
    const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr,
    const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
    tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
Thomas Davies's avatar
Thomas Davies committed
446 447
    const int16_t *scan, const int16_t *iscan, const qm_val_t *qm_ptr,
    const qm_val_t *iqm_ptr, int log_scale) {
448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463
  int i, eob = -1;
  // TODO(jingning) Decide the need of these arguments after the
  // quantization process is completed.
  (void)zbin_ptr;
  (void)quant_shift_ptr;
  (void)iscan;

  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));

  if (!skip_block) {
    // Quantization pass: All coefficients with index >= zero_flag are
    // skippable. Note: zero_flag can be zero.
    for (i = 0; i < n_coeffs; i++) {
      const int rc = scan[i];
      const int coeff = coeff_ptr[rc];
Thomas Davies's avatar
Thomas Davies committed
464 465
      const qm_val_t wt = qm_ptr ? qm_ptr[rc] : (1 << AOM_QM_BITS);
      const qm_val_t iwt = iqm_ptr ? iqm_ptr[rc] : (1 << AOM_QM_BITS);
466 467 468 469
      const int dequant =
          (dequant_ptr[rc != 0] * iwt + (1 << (AOM_QM_BITS - 1))) >>
          AOM_QM_BITS;
      const int coeff_sign = (coeff >> 31);
Thomas Davies's avatar
Thomas Davies committed
470
      int64_t abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
471 472 473 474
      int tmp32 = 0;
      if (abs_coeff * wt >=
          (dequant_ptr[rc != 0] << (AOM_QM_BITS - (1 + log_scale)))) {
        abs_coeff += ROUND_POWER_OF_TWO(round_ptr[rc != 0], log_scale);
Thomas Davies's avatar
Thomas Davies committed
475
        abs_coeff = clamp64(abs_coeff, INT16_MIN, INT16_MAX);
476
        tmp32 = (int)((abs_coeff * wt * quant_ptr[rc != 0]) >>
Thomas Davies's avatar
Thomas Davies committed
477
                      (16 - log_scale + AOM_QM_BITS));
478 479 480 481 482 483 484 485 486 487
        qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
        dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant / (1 << log_scale);
      }

      if (tmp32) eob = i;
    }
  }
  *eob_ptr = eob + 1;
}

Thomas Davies's avatar
Thomas Davies committed
488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531
static void highbd_quantize_fp_helper_c(
    const tran_low_t *coeff_ptr, intptr_t count, int skip_block,
    const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr,
    const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
    tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
    const int16_t *scan, const int16_t *iscan, const qm_val_t *qm_ptr,
    const qm_val_t *iqm_ptr, int log_scale) {
  int i;
  int eob = -1;
  const int scale = 1 << log_scale;
  const int shift = 16 - log_scale;
  // TODO(jingning) Decide the need of these arguments after the
  // quantization process is completed.
  (void)zbin_ptr;
  (void)quant_shift_ptr;
  (void)iscan;

  memset(qcoeff_ptr, 0, count * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, count * sizeof(*dqcoeff_ptr));

  if (!skip_block) {
    // Quantization pass: All coefficients with index >= zero_flag are
    // skippable. Note: zero_flag can be zero.
    for (i = 0; i < count; i++) {
      const int rc = scan[i];
      const int coeff = coeff_ptr[rc];
      const qm_val_t wt = qm_ptr != NULL ? qm_ptr[rc] : (1 << AOM_QM_BITS);
      const qm_val_t iwt = iqm_ptr != NULL ? iqm_ptr[rc] : (1 << AOM_QM_BITS);
      const int dequant =
          (dequant_ptr[rc != 0] * iwt + (1 << (AOM_QM_BITS - 1))) >>
          AOM_QM_BITS;
      const int coeff_sign = (coeff >> 31);
      const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
      const int64_t tmp = abs_coeff + (round_ptr[rc != 0] >> log_scale);
      const int abs_qcoeff =
          (int)((tmp * quant_ptr[rc != 0] * wt) >> (shift + AOM_QM_BITS));
      qcoeff_ptr[rc] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
      dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant / scale;
      if (abs_qcoeff) eob = i;
    }
  }
  *eob_ptr = eob + 1;
}

532 533 534 535 536 537
void av1_quantize_fp_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                       int skip_block, const int16_t *zbin_ptr,
                       const int16_t *round_ptr, const int16_t *quant_ptr,
                       const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
                       tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr,
                       uint16_t *eob_ptr, const int16_t *scan,
Thomas Davies's avatar
Thomas Davies committed
538
                       const int16_t *iscan) {
539 540
  quantize_fp_helper_c(coeff_ptr, n_coeffs, skip_block, zbin_ptr, round_ptr,
                       quant_ptr, quant_shift_ptr, qcoeff_ptr, dqcoeff_ptr,
Thomas Davies's avatar
Thomas Davies committed
541
                       dequant_ptr, eob_ptr, scan, iscan, NULL, NULL, 0);
542 543 544 545 546 547 548 549
}

void av1_quantize_fp_32x32_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                             int skip_block, const int16_t *zbin_ptr,
                             const int16_t *round_ptr, const int16_t *quant_ptr,
                             const int16_t *quant_shift_ptr,
                             tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                             const int16_t *dequant_ptr, uint16_t *eob_ptr,
Thomas Davies's avatar
Thomas Davies committed
550
                             const int16_t *scan, const int16_t *iscan) {
551 552
  quantize_fp_helper_c(coeff_ptr, n_coeffs, skip_block, zbin_ptr, round_ptr,
                       quant_ptr, quant_shift_ptr, qcoeff_ptr, dqcoeff_ptr,
Thomas Davies's avatar
Thomas Davies committed
553
                       dequant_ptr, eob_ptr, scan, iscan, NULL, NULL, 1);
554 555 556 557 558 559 560 561 562
}

#if CONFIG_TX64X64
void av1_quantize_fp_64x64_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                             int skip_block, const int16_t *zbin_ptr,
                             const int16_t *round_ptr, const int16_t *quant_ptr,
                             const int16_t *quant_shift_ptr,
                             tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                             const int16_t *dequant_ptr, uint16_t *eob_ptr,
Thomas Davies's avatar
Thomas Davies committed
563
                             const int16_t *scan, const int16_t *iscan) {
564 565
  quantize_fp_helper_c(coeff_ptr, n_coeffs, skip_block, zbin_ptr, round_ptr,
                       quant_ptr, quant_shift_ptr, qcoeff_ptr, dqcoeff_ptr,
Thomas Davies's avatar
Thomas Davies committed
566
                       dequant_ptr, eob_ptr, scan, iscan, NULL, NULL, 2);
567 568 569
}
#endif  // CONFIG_TX64X64

Yaowu Xu's avatar
Yaowu Xu committed
570 571 572 573
void av1_quantize_fp_facade(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                            const MACROBLOCK_PLANE *p, tran_low_t *qcoeff_ptr,
                            const MACROBLOCKD_PLANE *pd,
                            tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr,
574
                            const SCAN_ORDER *sc, const QUANT_PARAM *qparam) {
Angie Chiang's avatar
Angie Chiang committed
575 576
  // obsolete skip_block
  const int skip_block = 0;
577 578 579
#if CONFIG_AOM_QM
  const qm_val_t *qm_ptr = qparam->qmatrix;
  const qm_val_t *iqm_ptr = qparam->iqmatrix;
580
  if (qm_ptr != NULL && iqm_ptr != NULL) {
Thomas Davies's avatar
Thomas Davies committed
581 582 583 584 585
    quantize_fp_helper_c(coeff_ptr, n_coeffs, skip_block, p->zbin, p->round_fp,
                         p->quant_fp, p->quant_shift, qcoeff_ptr, dqcoeff_ptr,
                         pd->dequant, eob_ptr, sc->scan, sc->iscan, qm_ptr,
                         iqm_ptr, qparam->log_scale);
  } else {
586
#endif
Thomas Davies's avatar
Thomas Davies committed
587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607
    switch (qparam->log_scale) {
      case 0:
        if (n_coeffs < 16) {
          // TODO(jingning): Need SIMD implementation for smaller block size
          // quantization.
          quantize_fp_helper_c(
              coeff_ptr, n_coeffs, skip_block, p->zbin, p->round_fp,
              p->quant_fp, p->quant_shift, qcoeff_ptr, dqcoeff_ptr, pd->dequant,
              eob_ptr, sc->scan, sc->iscan, NULL, NULL, qparam->log_scale);
        } else {
          av1_quantize_fp(coeff_ptr, n_coeffs, skip_block, p->zbin, p->round_fp,
                          p->quant_fp, p->quant_shift, qcoeff_ptr, dqcoeff_ptr,
                          pd->dequant, eob_ptr, sc->scan, sc->iscan);
        }
        break;
      case 1:
        av1_quantize_fp_32x32(coeff_ptr, n_coeffs, skip_block, p->zbin,
                              p->round_fp, p->quant_fp, p->quant_shift,
                              qcoeff_ptr, dqcoeff_ptr, pd->dequant, eob_ptr,
                              sc->scan, sc->iscan);
        break;
608
#if CONFIG_TX64X64
Thomas Davies's avatar
Thomas Davies committed
609 610 611 612 613 614
      case 2:
        av1_quantize_fp_64x64(coeff_ptr, n_coeffs, skip_block, p->zbin,
                              p->round_fp, p->quant_fp, p->quant_shift,
                              qcoeff_ptr, dqcoeff_ptr, pd->dequant, eob_ptr,
                              sc->scan, sc->iscan);
        break;
615
#endif  // CONFIG_TX64X64
Thomas Davies's avatar
Thomas Davies committed
616 617 618
      default: assert(0);
    }
#if CONFIG_AOM_QM
619
  }
Thomas Davies's avatar
Thomas Davies committed
620
#endif
Angie Chiang's avatar
Angie Chiang committed
621 622
}

Yaowu Xu's avatar
Yaowu Xu committed
623 624 625
void av1_quantize_b_facade(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                           const MACROBLOCK_PLANE *p, tran_low_t *qcoeff_ptr,
                           const MACROBLOCKD_PLANE *pd, tran_low_t *dqcoeff_ptr,
626
                           uint16_t *eob_ptr, const SCAN_ORDER *sc,
627
                           const QUANT_PARAM *qparam) {
Angie Chiang's avatar
Angie Chiang committed
628 629
  // obsolete skip_block
  const int skip_block = 0;
630 631 632
#if CONFIG_AOM_QM
  const qm_val_t *qm_ptr = qparam->qmatrix;
  const qm_val_t *iqm_ptr = qparam->iqmatrix;
633
  if (qm_ptr != NULL && iqm_ptr != NULL) {
Thomas Davies's avatar
Thomas Davies committed
634 635 636 637 638
    quantize_b_helper_c(coeff_ptr, n_coeffs, skip_block, p->zbin, p->round,
                        p->quant, p->quant_shift, qcoeff_ptr, dqcoeff_ptr,
                        pd->dequant, eob_ptr, sc->scan, sc->iscan, qm_ptr,
                        iqm_ptr, qparam->log_scale);
  } else {
639
#endif  // CONFIG_AOM_QM
Angie Chiang's avatar
Angie Chiang committed
640

Thomas Davies's avatar
Thomas Davies committed
641 642 643 644 645 646 647 648 649 650 651
    switch (qparam->log_scale) {
      case 0:
        aom_quantize_b(coeff_ptr, n_coeffs, skip_block, p->zbin, p->round,
                       p->quant, p->quant_shift, qcoeff_ptr, dqcoeff_ptr,
                       pd->dequant, eob_ptr, sc->scan, sc->iscan);
        break;
      case 1:
        aom_quantize_b_32x32(coeff_ptr, n_coeffs, skip_block, p->zbin, p->round,
                             p->quant, p->quant_shift, qcoeff_ptr, dqcoeff_ptr,
                             pd->dequant, eob_ptr, sc->scan, sc->iscan);
        break;
652
#if CONFIG_TX64X64
Thomas Davies's avatar
Thomas Davies committed
653 654 655 656 657 658 659 660
      case 2:
        aom_quantize_b_64x64(coeff_ptr, n_coeffs, skip_block, p->zbin, p->round,
                             p->quant, p->quant_shift, qcoeff_ptr, dqcoeff_ptr,
                             pd->dequant, eob_ptr, sc->scan, sc->iscan);
        break;
#endif  // CONFIG_TX64X64
      default: assert(0);
    }
661
#if CONFIG_AOM_QM
Thomas Davies's avatar
Thomas Davies committed
662
  }
663
#endif
Thomas Davies's avatar
Thomas Davies committed
664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692
}

static void quantize_dc(const tran_low_t *coeff_ptr, int n_coeffs,
                        int skip_block, const int16_t *round_ptr,
                        const int16_t quant, tran_low_t *qcoeff_ptr,
                        tran_low_t *dqcoeff_ptr, const int16_t dequant_ptr,
                        uint16_t *eob_ptr, const qm_val_t *qm_ptr,
                        const qm_val_t *iqm_ptr, const int log_scale) {
  const int rc = 0;
  const int coeff = coeff_ptr[rc];
  const int coeff_sign = (coeff >> 31);
  const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
  int64_t tmp, eob = -1;
  int32_t tmp32;
  int dequant;

  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));

  if (!skip_block) {
    const int wt = qm_ptr != NULL ? qm_ptr[rc] : (1 << AOM_QM_BITS);
    const int iwt = iqm_ptr != NULL ? iqm_ptr[rc] : (1 << AOM_QM_BITS);
    tmp = clamp(abs_coeff + ROUND_POWER_OF_TWO(round_ptr[rc != 0], log_scale),
                INT16_MIN, INT16_MAX);
    tmp32 = (int32_t)((tmp * wt * quant) >> (16 - log_scale + AOM_QM_BITS));
    qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
    dequant = (dequant_ptr * iwt + (1 << (AOM_QM_BITS - 1))) >> AOM_QM_BITS;
    dqcoeff_ptr[rc] = (qcoeff_ptr[rc] * dequant) / (1 << log_scale);
    if (tmp32) eob = 0;
693
  }
Thomas Davies's avatar
Thomas Davies committed
694
  *eob_ptr = eob + 1;
Angie Chiang's avatar
Angie Chiang committed
695 696
}

Yaowu Xu's avatar
Yaowu Xu committed
697 698 699 700
void av1_quantize_dc_facade(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                            const MACROBLOCK_PLANE *p, tran_low_t *qcoeff_ptr,
                            const MACROBLOCKD_PLANE *pd,
                            tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr,
701
                            const SCAN_ORDER *sc, const QUANT_PARAM *qparam) {
Angie Chiang's avatar
Angie Chiang committed
702 703
  // obsolete skip_block
  const int skip_block = 0;
Thomas Davies's avatar
Thomas Davies committed
704 705
  (void)sc;
  assert(qparam->log_scale >= 0 && qparam->log_scale < (2 + CONFIG_TX64X64));
706 707 708
#if CONFIG_AOM_QM
  const qm_val_t *qm_ptr = qparam->qmatrix;
  const qm_val_t *iqm_ptr = qparam->iqmatrix;
Thomas Davies's avatar
Thomas Davies committed
709 710 711
#else
  const qm_val_t *qm_ptr = NULL;
  const qm_val_t *iqm_ptr = NULL;
712
#endif
Thomas Davies's avatar
Thomas Davies committed
713 714 715
  quantize_dc(coeff_ptr, (int)n_coeffs, skip_block, p->round, p->quant_fp[0],
              qcoeff_ptr, dqcoeff_ptr, pd->dequant[0], eob_ptr, qm_ptr, iqm_ptr,
              qparam->log_scale);
Angie Chiang's avatar
Angie Chiang committed
716 717
}

718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842
#if CONFIG_NEW_QUANT
void av1_quantize_b_nuq_facade(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                               const MACROBLOCK_PLANE *p,
                               tran_low_t *qcoeff_ptr,
                               const MACROBLOCKD_PLANE *pd,
                               tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr,
                               const SCAN_ORDER *sc,
                               const QUANT_PARAM *qparam) {
  // obsolete skip_block
  const int skip_block = 0;
  const uint8_t *band = get_band_translate(qparam->tx_size);
  int dq = qparam->dq;

  switch (qparam->log_scale) {
    case 0:
      quantize_nuq(coeff_ptr, n_coeffs, skip_block, p->quant, p->quant_shift,
                   pd->dequant,
                   (const cuml_bins_type_nuq *)p->cuml_bins_nuq[dq],
                   (const dequant_val_type_nuq *)pd->dequant_val_nuq[dq],
                   qcoeff_ptr, dqcoeff_ptr, eob_ptr, sc->scan, band);
      break;
    case 1:
      quantize_32x32_nuq(coeff_ptr, n_coeffs, skip_block, p->quant,
                         p->quant_shift, pd->dequant,
                         (const cuml_bins_type_nuq *)p->cuml_bins_nuq[dq],
                         (const dequant_val_type_nuq *)pd->dequant_val_nuq[dq],
                         qcoeff_ptr, dqcoeff_ptr, eob_ptr, sc->scan, band);
      break;
#if CONFIG_TX64X64
    case 2:
      quantize_64x64_nuq(coeff_ptr, n_coeffs, skip_block, p->quant,
                         p->quant_shift, pd->dequant,
                         (const cuml_bins_type_nuq *)p->cuml_bins_nuq[dq],
                         (const dequant_val_type_nuq *)pd->dequant_val_nuq[dq],
                         qcoeff_ptr, dqcoeff_ptr, eob_ptr, sc->scan, band);
      break;
#endif  // CONFIG_TX64X64
    default: assert(0);
  }
}

void av1_quantize_fp_nuq_facade(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                                const MACROBLOCK_PLANE *p,
                                tran_low_t *qcoeff_ptr,
                                const MACROBLOCKD_PLANE *pd,
                                tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr,
                                const SCAN_ORDER *sc,
                                const QUANT_PARAM *qparam) {
  // obsolete skip_block
  const int skip_block = 0;
  const uint8_t *band = get_band_translate(qparam->tx_size);
  int dq = qparam->dq;

  switch (qparam->log_scale) {
    case 0:
      quantize_fp_nuq(coeff_ptr, n_coeffs, skip_block, p->quant_fp, pd->dequant,
                      (const cuml_bins_type_nuq *)p->cuml_bins_nuq[dq],
                      (const dequant_val_type_nuq *)pd->dequant_val_nuq[dq],
                      qcoeff_ptr, dqcoeff_ptr, eob_ptr, sc->scan, band);
      break;
    case 1:
      quantize_32x32_fp_nuq(
          coeff_ptr, n_coeffs, skip_block, p->quant_fp, pd->dequant,
          (const cuml_bins_type_nuq *)p->cuml_bins_nuq[dq],
          (const dequant_val_type_nuq *)pd->dequant_val_nuq[dq], qcoeff_ptr,
          dqcoeff_ptr, eob_ptr, sc->scan, band);
      break;
#if CONFIG_TX64X64
    case 2:
      quantize_64x64_fp_nuq(
          coeff_ptr, n_coeffs, skip_block, p->quant_fp, pd->dequant,
          (const cuml_bins_type_nuq *)p->cuml_bins_nuq[dq],
          (const dequant_val_type_nuq *)pd->dequant_val_nuq[dq], qcoeff_ptr,
          dqcoeff_ptr, eob_ptr, sc->scan, band);
      break;
#endif  // CONFIG_TX64X64
    default: assert(0);
  }
}

void av1_quantize_dc_nuq_facade(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                                const MACROBLOCK_PLANE *p,
                                tran_low_t *qcoeff_ptr,
                                const MACROBLOCKD_PLANE *pd,
                                tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr,
                                const SCAN_ORDER *sc,
                                const QUANT_PARAM *qparam) {
  // obsolete skip_block
  const int skip_block = 0;
  int dq = qparam->dq;
  (void)sc;

  switch (qparam->log_scale) {
    case 0:
      quantize_dc_fp_nuq(coeff_ptr, n_coeffs, skip_block, p->quant_fp[0],
                         pd->dequant[0], p->cuml_bins_nuq[dq][0],
                         pd->dequant_val_nuq[dq][0], qcoeff_ptr, dqcoeff_ptr,
                         eob_ptr);
      break;
    case 1:
      quantize_dc_32x32_fp_nuq(coeff_ptr, n_coeffs, skip_block, p->quant_fp[0],
                               pd->dequant[0], p->cuml_bins_nuq[dq][0],
                               pd->dequant_val_nuq[dq][0], qcoeff_ptr,
                               dqcoeff_ptr, eob_ptr);
      break;
#if CONFIG_TX64X64
    case 2:
      quantize_dc_64x64_fp_nuq(coeff_ptr, n_coeffs, skip_block, p->quant_fp[0],
                               pd->dequant[0], p->cuml_bins_nuq[dq][0],
                               pd->dequant_val_nuq[dq][0], qcoeff_ptr,
                               dqcoeff_ptr, eob_ptr);
      break;
#endif  // CONFIG_TX64X64
    default: assert(0);
  }
}
#endif  // CONFIG_NEW_QUANT

void av1_highbd_quantize_fp_facade(const tran_low_t *coeff_ptr,
                                   intptr_t n_coeffs, const MACROBLOCK_PLANE *p,
                                   tran_low_t *qcoeff_ptr,
                                   const MACROBLOCKD_PLANE *pd,
                                   tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr,
                                   const SCAN_ORDER *sc,
                                   const QUANT_PARAM *qparam) {
Angie Chiang's avatar
Angie Chiang committed
843 844
  // obsolete skip_block
  const int skip_block = 0;
845 846 847
#if CONFIG_AOM_QM
  const qm_val_t *qm_ptr = qparam->qmatrix;
  const qm_val_t *iqm_ptr = qparam->iqmatrix;
848
  if (qm_ptr != NULL && iqm_ptr != NULL) {
Thomas Davies's avatar
Thomas Davies committed
849 850 851 852 853
    highbd_quantize_fp_helper_c(
        coeff_ptr, n_coeffs, skip_block, p->zbin, p->round_fp, p->quant_fp,
        p->quant_shift, qcoeff_ptr, dqcoeff_ptr, pd->dequant, eob_ptr, sc->scan,
        sc->iscan, qm_ptr, iqm_ptr, qparam->log_scale);
  } else {
854
#endif  // CONFIG_AOM_QM
Angie Chiang's avatar
Angie Chiang committed
855

Thomas Davies's avatar
Thomas Davies committed
856 857 858 859 860 861 862 863 864
    if (n_coeffs < 16) {
      // TODO(jingning): Need SIMD implementation for smaller block size
      // quantization.
      av1_highbd_quantize_fp_c(coeff_ptr, n_coeffs, skip_block, p->zbin,
                               p->round_fp, p->quant_fp, p->quant_shift,
                               qcoeff_ptr, dqcoeff_ptr, pd->dequant, eob_ptr,
                               sc->scan, sc->iscan, qparam->log_scale);
      return;
    }
865

Thomas Davies's avatar
Thomas Davies committed
866 867 868 869
    av1_highbd_quantize_fp(coeff_ptr, n_coeffs, skip_block, p->zbin,
                           p->round_fp, p->quant_fp, p->quant_shift, qcoeff_ptr,
                           dqcoeff_ptr, pd->dequant, eob_ptr, sc->scan,
                           sc->iscan, qparam->log_scale);
870
#if CONFIG_AOM_QM
Thomas Davies's avatar
Thomas Davies committed
871
  }
872
#endif
Angie Chiang's avatar
Angie Chiang committed
873 874
}

875 876 877 878 879 880 881
void av1_highbd_quantize_b_facade(const tran_low_t *coeff_ptr,
                                  intptr_t n_coeffs, const MACROBLOCK_PLANE *p,
                                  tran_low_t *qcoeff_ptr,
                                  const MACROBLOCKD_PLANE *pd,
                                  tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr,
                                  const SCAN_ORDER *sc,
                                  const QUANT_PARAM *qparam) {
Yaowu Xu's avatar
Yaowu Xu committed
882 883
  // obsolete skip_block
  const int skip_block = 0;
884 885 886
#if CONFIG_AOM_QM
  const qm_val_t *qm_ptr = qparam->qmatrix;
  const qm_val_t *iqm_ptr = qparam->iqmatrix;
887
  if (qm_ptr != NULL && iqm_ptr != NULL) {
Thomas Davies's avatar
Thomas Davies committed
888 889 890 891 892
    highbd_quantize_b_helper_c(coeff_ptr, n_coeffs, skip_block, p->zbin,
                               p->round, p->quant, p->quant_shift, qcoeff_ptr,
                               dqcoeff_ptr, pd->dequant, eob_ptr, sc->scan,
                               sc->iscan, qm_ptr, iqm_ptr, qparam->log_scale);
  } else {
893
#endif  // CONFIG_AOM_QM
Yaowu Xu's avatar
Yaowu Xu committed
894

Thomas Davies's avatar
Thomas Davies committed
895 896 897 898
    switch (qparam->log_scale) {
      case 0:
        if (LIKELY(n_coeffs >= 8)) {
          aom_highbd_quantize_b(coeff_ptr, n_coeffs, skip_block, p->zbin,
Yi Luo's avatar
Yi Luo committed
899 900
                                p->round, p->quant, p->quant_shift, qcoeff_ptr,
                                dqcoeff_ptr, pd->dequant, eob_ptr, sc->scan,
Thomas Davies's avatar
Thomas Davies committed
901 902 903 904 905
                                sc->iscan);
        } else {
          // TODO(luoyi): Need SIMD (e.g. sse2) for smaller block size
          // quantization
          aom_highbd_quantize_b_c(coeff_ptr, n_coeffs, skip_block, p->zbin,
Cheng Chen's avatar
Cheng Chen committed
906 907
                                  p->round, p->quant, p->quant_shift,
                                  qcoeff_ptr, dqcoeff_ptr, pd->dequant, eob_ptr,
Thomas Davies's avatar
Thomas Davies committed
908 909 910 911 912 913 914 915 916
                                  sc->scan, sc->iscan);
        }
        break;
      case 1:
        aom_highbd_quantize_b_32x32(coeff_ptr, n_coeffs, skip_block, p->zbin,
                                    p->round, p->quant, p->quant_shift,
                                    qcoeff_ptr, dqcoeff_ptr, pd->dequant,
                                    eob_ptr, sc->scan, sc->iscan);
        break;
Cheng Chen's avatar
Cheng Chen committed
917
#if CONFIG_TX64X64
Thomas Davies's avatar
Thomas Davies committed
918 919 920 921 922 923
      case 2:
        aom_highbd_quantize_b_64x64(coeff_ptr, n_coeffs, skip_block, p->zbin,
                                    p->round, p->quant, p->quant_shift,
                                    qcoeff_ptr, dqcoeff_ptr, pd->dequant,
                                    eob_ptr, sc->scan, sc->iscan);
        break;
Cheng Chen's avatar
Cheng Chen committed
924
#endif  // CONFIG_TX64X64
Thomas Davies's avatar
Thomas Davies committed
925 926 927
      default: assert(0);
    }
#if CONFIG_AOM_QM
Cheng Chen's avatar
Cheng Chen committed
928
  }
Thomas Davies's avatar
Thomas Davies committed
929
#endif
Yaowu Xu's avatar
Yaowu Xu committed
930 931
}

932 933 934 935
static INLINE void highbd_quantize_dc(
    const tran_low_t *coeff_ptr, int n_coeffs, int skip_block,
    const int16_t *round_ptr, const int16_t quant, tran_low_t *qcoeff_ptr,
    tran_low_t *dqcoeff_ptr, const int16_t dequant_ptr, uint16_t *eob_ptr,
Thomas Davies's avatar
Thomas Davies committed
936
    const qm_val_t *qm_ptr, const qm_val_t *iqm_ptr, const int log_scale) {
937 938 939 940
  int eob = -1;

  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
Thomas Davies's avatar
Thomas Davies committed
941

942
  if (!skip_block) {
Thomas Davies's avatar
Thomas Davies committed
943 944
    const qm_val_t wt = qm_ptr != NULL ? qm_ptr[0] : (1 << AOM_QM_BITS);
    const qm_val_t iwt = iqm_ptr != NULL ? iqm_ptr[0] : (1 << AOM_QM_BITS);
945 946 947
    const int coeff = coeff_ptr[0];
    const int coeff_sign = (coeff >> 31);
    const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
Thomas Davies's avatar
Thomas Davies committed
948 949 950 951
    const int64_t tmp = abs_coeff + ROUND_POWER_OF_TWO(round_ptr[0], log_scale);
    const int64_t tmpw = tmp * wt;
    const int abs_qcoeff =
        (int)((tmpw * quant) >> (16 - log_scale + AOM_QM_BITS));
952
    qcoeff_ptr[0] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
Thomas Davies's avatar
Thomas Davies committed
953 954 955 956
    const int dequant =
        (dequant_ptr * iwt + (1 << (AOM_QM_BITS - 1))) >> AOM_QM_BITS;

    dqcoeff_ptr[0] = (qcoeff_ptr[0] * dequant) / (1 << log_scale);
957 958 959 960 961 962 963 964 965 966 967 968
    if (abs_qcoeff) eob = 0;
  }
  *eob_ptr = eob + 1;
}

void av1_highbd_quantize_dc_facade(const tran_low_t *coeff_ptr,
                                   intptr_t n_coeffs, const MACROBLOCK_PLANE *p,
                                   tran_low_t *qcoeff_ptr,
                                   const MACROBLOCKD_PLANE *pd,
                                   tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr,
                                   const SCAN_ORDER *sc,
                                   const QUANT_PARAM *qparam) {
Angie Chiang's avatar
Angie Chiang committed
969 970
  // obsolete skip_block
  const int skip_block = 0;
971 972 973
#if CONFIG_AOM_QM
  const qm_val_t *qm_ptr = qparam->qmatrix;
  const qm_val_t *iqm_ptr = qparam->iqmatrix;
Thomas Davies's avatar
Thomas Davies committed
974 975 976
#else
  const qm_val_t *qm_ptr = NULL;
  const qm_val_t *iqm_ptr = NULL;
977
#endif  // CONFIG_AOM_QM
Angie Chiang's avatar
Angie Chiang committed
978 979 980

  (void)sc;

981 982
  highbd_quantize_dc(coeff_ptr, (int)n_coeffs, skip_block, p->round,
                     p->quant_fp[0], qcoeff_ptr, dqcoeff_ptr, pd->dequant[0],
Thomas Davies's avatar
Thomas Davies committed
983
                     eob_ptr, qm_ptr, iqm_ptr, qparam->log_scale);
Angie Chiang's avatar
Angie Chiang committed
984
}
985 986

#if CONFIG_NEW_QUANT
987 988 989 990 991
static INLINE int highbd_quantize_coeff_nuq(
    const tran_low_t coeffv, const int16_t quant, const int16_t quant_shift,
    const int16_t dequant, const tran_low_t *cuml_bins_ptr,
    const tran_low_t *dequant_val, tran_low_t *qcoeff_ptr,
    tran_low_t *dqcoeff_ptr) {
992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004
  const int coeff = coeffv;
  const int coeff_sign = (coeff >> 31);
  const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
  int i, q;
  int64_t tmp = clamp(abs_coeff, INT32_MIN, INT32_MAX);
  for (i = 0; i < NUQ_KNOTS; i++) {
    if (tmp < cuml_bins_ptr[i]) {
      q = i;
      break;
    }
  }
  if (i == NUQ_KNOTS) {
    tmp -= cuml_bins_ptr[NUQ_KNOTS - 1];
1005
    q = NUQ_KNOTS + (int)(((((tmp * quant) >> 16) + tmp) * quant_shift) >> 16);
1006 1007
  }
  if (q) {
Yaowu Xu's avatar
Yaowu Xu committed
1008
    *dqcoeff_ptr = av1_dequant_abscoeff_nuq(q, dequant, dequant_val);
1009
    *qcoeff_ptr = (q ^ coeff_sign) - coeff_sign;
1010 1011 1012 1013 1014 1015 1016 1017
    *dqcoeff_ptr = *qcoeff_ptr < 0 ? -*dqcoeff_ptr : *dqcoeff_ptr;
  } else {
    *qcoeff_ptr = 0;
    *dqcoeff_ptr = 0;
  }
  return (q != 0);
}

1018 1019 1020 1021
static INLINE int highbd_quantize_coeff_fp_nuq(
    const tran_low_t coeffv, const int16_t quant, const int16_t dequant,
    const tran_low_t *cuml_bins_ptr, const tran_low_t *dequant_val,
    tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr) {
1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033
  const int coeff = coeffv;
  const int coeff_sign = (coeff >> 31);
  const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
  int i, q;
  int64_t tmp = clamp(abs_coeff, INT32_MIN, INT32_MAX);
  for (i = 0; i < NUQ_KNOTS; i++) {
    if (tmp < cuml_bins_ptr[i]) {
      q = i;
      break;
    }
  }
  if (i == NUQ_KNOTS) {
1034
    q = NUQ_KNOTS + (int)(((tmp - cuml_bins_ptr[NUQ_KNOTS - 1]) * quant) >> 16);
1035 1036
  }
  if (q) {
Yaowu Xu's avatar
Yaowu Xu committed
1037
    *dqcoeff_ptr = av1_dequant_abscoeff_nuq(q, dequant, dequant_val);
1038
    *qcoeff_ptr = (q ^ coeff_sign) - coeff_sign;
1039 1040 1041 1042 1043 1044 1045 1046 1047
    *dqcoeff_ptr = *qcoeff_ptr < 0 ? -*dqcoeff_ptr : *dqcoeff_ptr;
  } else {
    *qcoeff_ptr = 0;
    *dqcoeff_ptr = 0;
  }
  return (q != 0);
}

static INLINE int highbd_quantize_coeff_bigtx_fp_nuq(
1048 1049
    const tran_low_t coeffv, const int16_t quant, const int16_t dequant,
    const tran_low_t *cuml_bins_ptr, const tran_low_t *dequant_val,
1050
    tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, int logsizeby16) {
1051 1052 1053 1054 1055 1056
  const int coeff = coeffv;
  const int coeff_sign = (coeff >> 31);
  const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
  int i, q;
  int64_t tmp = clamp(abs_coeff, INT32_MIN, INT32_MAX);
  for (i = 0; i < NUQ_KNOTS; i++) {
1057
    if (tmp < ROUND_POWER_OF_TWO(cuml_bins_ptr[i], logsizeby16)) {
1058 1059 1060 1061 1062
      q = i;
      break;
    }
  }
  if (i == NUQ_KNOTS) {
1063
    q = NUQ_KNOTS +
1064 1065
        (int)(((tmp -
                ROUND_POWER_OF_TWO(cuml_bins_ptr[NUQ_KNOTS - 1], logsizeby16)) *
1066
               quant) >>
1067
              (16 - logsizeby16));
1068 1069
  }
  if (q) {
1070
    *dqcoeff_ptr = ROUND_POWER_OF_TWO(
1071
        av1_dequant_abscoeff_nuq(q, dequant, dequant_val), logsizeby16);
1072
    *qcoeff_ptr = (q ^ coeff_sign) - coeff_sign;
1073 1074 1075 1076 1077 1078 1079 1080
    *dqcoeff_ptr = *qcoeff_ptr < 0 ? -*dqcoeff_ptr : *dqcoeff_ptr;
  } else {
    *qcoeff_ptr = 0;
    *dqcoeff_ptr = 0;
  }
  return (q != 0);
}

1081 1082 1083 1084
static INLINE int highbd_quantize_coeff_bigtx_nuq(
    const tran_low_t coeffv, const int16_t quant, const int16_t quant_shift,
    const int16_t dequant, const tran_low_t *cuml_bins_ptr,
    const tran_low_t *dequant_val, tran_low_t *qcoeff_ptr,
1085
    tran_low_t *dqcoeff_ptr, int logsizeby16) {
1086 1087 1088 1089 1090 1091
  const int coeff = coeffv;
  const int coeff_sign = (coeff >> 31);
  const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
  int i, q;
  int64_t tmp = clamp(abs_coeff, INT32_MIN, INT32_MAX);
  for (i = 0; i < NUQ_KNOTS; i++) {
1092
    if (tmp < ROUND_POWER_OF_TWO(cuml_bins_ptr[i], logsizeby16)) {
1093 1094 1095 1096 1097
      q = i;
      break;
    }
  }
  if (i == NUQ_KNOTS) {
1098
    tmp -= ROUND_POWER_OF_TWO(cuml_bins_ptr[NUQ_KNOTS - 1], logsizeby16);
1099
    q = NUQ_KNOTS + (int)(((((tmp * quant) >> 16) + tmp) * quant_shift) >>
1100
                          (16 - logsizeby16));
1101 1102
  }
  if (q) {
1103
    *dqcoeff_ptr = ROUND_POWER_OF_TWO(
1104
        av1_dequant_abscoeff_nuq(q, dequant, dequant_val), logsizeby16);
1105
    *qcoeff_ptr = (q ^ coeff_sign) - coeff_sign;
1106 1107 1108 1109 1110 1111 1112 1113
    *dqcoeff_ptr = *qcoeff_ptr < 0 ? -*dqcoeff_ptr : *dqcoeff_ptr;
  } else {
    *qcoeff_ptr = 0;
    *dqcoeff_ptr = 0;
  }
  return (q != 0);
}

1114 1115 1116
void highbd_quantize_dc_nuq(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                            int skip_block, const int16_t quant,
                            const int16_t quant_shift, const int16_t dequant,
1117 1118
                            const tran_low_t *cuml_bins_ptr,
                            const tran_low_t *dequant_val,
1119
                            tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
1120 1121 1122 1123 1124 1125
                            uint16_t *eob_ptr) {
  int eob = -1;
  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
  if (!skip_block) {
    const int rc = 0;
1126 1127
    if (highbd_quantize_coeff_nuq(coeff_ptr[rc], quant, quant_shift, dequant,
                                  cuml_bins_ptr, dequant_val, qcoeff_ptr,
1128 1129 1130 1131 1132 1133
                                  dqcoeff_ptr))
      eob = 0;
  }
  *eob_ptr = eob + 1;
}

1134 1135
void highbd_quantize_dc_fp_nuq(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                               int skip_block, const int16_t quant,
1136 1137 1138
                               const int16_t dequant,
                               const tran_low_t *cuml_bins_ptr,
                               const tran_low_t *dequant_val,
1139
                               tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
1140 1141 1142 1143 1144 1145
                               uint16_t *eob_ptr) {
  int eob = -1;
  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
  if (!skip_block) {
    const int rc = 0;
1146 1147
    if (highbd_quantize_coeff_fp_nuq(coeff_ptr[rc], quant, dequant,
                                     cuml_bins_ptr, dequant_val, qcoeff_ptr,
1148 1149 1150 1151 1152 1153
                                     dqcoeff_ptr))
      eob = 0;
  }
  *eob_ptr = eob + 1;
}

1154 1155
void highbd_quantize_nuq_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                           int skip_block, const int16_t *quant_ptr,
1156 1157 1158 1159
                           const int16_t *quant_shift_ptr,
                           const int16_t *dequant_ptr,
                           const cuml_bins_type_nuq *cuml_bins_ptr,
                           const dequant_val_type_nuq *dequant_val,
1160 1161
                           tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
                           uint16_t *eob_ptr, const int16_t *scan,
1162 1163 1164 1165 1166 1167 1168 1169
                           const uint8_t *band) {
  int eob = -1;
  memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
  memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
  if (!skip_block) {
    int i;
    for (i = 0; i < n_coeffs; i++) {
      const int rc = scan[i];
1170 1171 1172 1173
      if (highbd_quantize_coeff_nuq(
              coeff_ptr[rc], quant_ptr[rc != 0], quant_shift_ptr[rc != 0],
              dequant_ptr[rc != 0], cuml_bins_ptr[band[i]],
              dequant_val[band[i]], &qcoeff_ptr[rc], &dqcoeff_ptr[rc]))