| 1 | /* |
| 2 | * AC-3 encoder float/fixed template |
| 3 | * Copyright (c) 2000 Fabrice Bellard |
| 4 | * Copyright (c) 2006-2011 Justin Ruggles <justin.ruggles@gmail.com> |
| 5 | * Copyright (c) 2006-2010 Prakash Punnoor <prakash@punnoor.de> |
| 6 | * |
| 7 | * This file is part of FFmpeg. |
| 8 | * |
| 9 | * FFmpeg is free software; you can redistribute it and/or |
| 10 | * modify it under the terms of the GNU Lesser General Public |
| 11 | * License as published by the Free Software Foundation; either |
| 12 | * version 2.1 of the License, or (at your option) any later version. |
| 13 | * |
| 14 | * FFmpeg is distributed in the hope that it will be useful, |
| 15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 17 | * Lesser General Public License for more details. |
| 18 | * |
| 19 | * You should have received a copy of the GNU Lesser General Public |
| 20 | * License along with FFmpeg; if not, write to the Free Software |
| 21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 22 | */ |
| 23 | |
| 24 | /** |
| 25 | * @file |
| 26 | * AC-3 encoder float/fixed template |
| 27 | */ |
| 28 | |
| 29 | #include <stdint.h> |
| 30 | |
| 31 | #include "libavutil/attributes.h" |
| 32 | #include "libavutil/internal.h" |
| 33 | |
| 34 | #include "audiodsp.h" |
| 35 | #include "internal.h" |
| 36 | #include "ac3enc.h" |
| 37 | #include "eac3enc.h" |
| 38 | |
| 39 | /* prototypes for static functions in ac3enc_fixed.c and ac3enc_float.c */ |
| 40 | |
| 41 | static void scale_coefficients(AC3EncodeContext *s); |
| 42 | |
| 43 | static int normalize_samples(AC3EncodeContext *s); |
| 44 | |
| 45 | static void clip_coefficients(AudioDSPContext *adsp, CoefType *coef, |
| 46 | unsigned int len); |
| 47 | |
| 48 | static CoefType calc_cpl_coord(CoefSumType energy_ch, CoefSumType energy_cpl); |
| 49 | |
| 50 | static void sum_square_butterfly(AC3EncodeContext *s, CoefSumType sum[4], |
| 51 | const CoefType *coef0, const CoefType *coef1, |
| 52 | int len); |
| 53 | |
| 54 | int AC3_NAME(allocate_sample_buffers)(AC3EncodeContext *s) |
| 55 | { |
| 56 | int ch; |
| 57 | |
| 58 | FF_ALLOC_OR_GOTO(s->avctx, s->windowed_samples, AC3_WINDOW_SIZE * |
| 59 | sizeof(*s->windowed_samples), alloc_fail); |
| 60 | FF_ALLOC_ARRAY_OR_GOTO(s->avctx, s->planar_samples, s->channels, sizeof(*s->planar_samples), |
| 61 | alloc_fail); |
| 62 | for (ch = 0; ch < s->channels; ch++) { |
| 63 | FF_ALLOCZ_OR_GOTO(s->avctx, s->planar_samples[ch], |
| 64 | (AC3_FRAME_SIZE+AC3_BLOCK_SIZE) * sizeof(**s->planar_samples), |
| 65 | alloc_fail); |
| 66 | } |
| 67 | |
| 68 | return 0; |
| 69 | alloc_fail: |
| 70 | return AVERROR(ENOMEM); |
| 71 | } |
| 72 | |
| 73 | |
| 74 | /* |
| 75 | * Copy input samples. |
| 76 | * Channels are reordered from FFmpeg's default order to AC-3 order. |
| 77 | */ |
| 78 | static void copy_input_samples(AC3EncodeContext *s, SampleType **samples) |
| 79 | { |
| 80 | int ch; |
| 81 | |
| 82 | /* copy and remap input samples */ |
| 83 | for (ch = 0; ch < s->channels; ch++) { |
| 84 | /* copy last 256 samples of previous frame to the start of the current frame */ |
| 85 | memcpy(&s->planar_samples[ch][0], &s->planar_samples[ch][AC3_BLOCK_SIZE * s->num_blocks], |
| 86 | AC3_BLOCK_SIZE * sizeof(s->planar_samples[0][0])); |
| 87 | |
| 88 | /* copy new samples for current frame */ |
| 89 | memcpy(&s->planar_samples[ch][AC3_BLOCK_SIZE], |
| 90 | samples[s->channel_map[ch]], |
| 91 | AC3_BLOCK_SIZE * s->num_blocks * sizeof(s->planar_samples[0][0])); |
| 92 | } |
| 93 | } |
| 94 | |
| 95 | |
| 96 | /* |
| 97 | * Apply the MDCT to input samples to generate frequency coefficients. |
| 98 | * This applies the KBD window and normalizes the input to reduce precision |
| 99 | * loss due to fixed-point calculations. |
| 100 | */ |
| 101 | static void apply_mdct(AC3EncodeContext *s) |
| 102 | { |
| 103 | int blk, ch; |
| 104 | |
| 105 | for (ch = 0; ch < s->channels; ch++) { |
| 106 | for (blk = 0; blk < s->num_blocks; blk++) { |
| 107 | AC3Block *block = &s->blocks[blk]; |
| 108 | const SampleType *input_samples = &s->planar_samples[ch][blk * AC3_BLOCK_SIZE]; |
| 109 | |
| 110 | #if CONFIG_AC3ENC_FLOAT |
| 111 | s->fdsp.vector_fmul(s->windowed_samples, input_samples, |
| 112 | s->mdct_window, AC3_WINDOW_SIZE); |
| 113 | #else |
| 114 | s->ac3dsp.apply_window_int16(s->windowed_samples, input_samples, |
| 115 | s->mdct_window, AC3_WINDOW_SIZE); |
| 116 | #endif |
| 117 | |
| 118 | if (s->fixed_point) |
| 119 | block->coeff_shift[ch+1] = normalize_samples(s); |
| 120 | |
| 121 | s->mdct.mdct_calcw(&s->mdct, block->mdct_coef[ch+1], |
| 122 | s->windowed_samples); |
| 123 | } |
| 124 | } |
| 125 | } |
| 126 | |
| 127 | |
| 128 | /* |
| 129 | * Calculate coupling channel and coupling coordinates. |
| 130 | */ |
| 131 | static void apply_channel_coupling(AC3EncodeContext *s) |
| 132 | { |
| 133 | LOCAL_ALIGNED_16(CoefType, cpl_coords, [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]); |
| 134 | #if CONFIG_AC3ENC_FLOAT |
| 135 | LOCAL_ALIGNED_16(int32_t, fixed_cpl_coords, [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]); |
| 136 | #else |
| 137 | int32_t (*fixed_cpl_coords)[AC3_MAX_CHANNELS][16] = cpl_coords; |
| 138 | #endif |
| 139 | int av_uninit(blk), ch, bnd, i, j; |
| 140 | CoefSumType energy[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][16] = {{{0}}}; |
| 141 | int cpl_start, num_cpl_coefs; |
| 142 | |
| 143 | memset(cpl_coords, 0, AC3_MAX_BLOCKS * sizeof(*cpl_coords)); |
| 144 | #if CONFIG_AC3ENC_FLOAT |
| 145 | memset(fixed_cpl_coords, 0, AC3_MAX_BLOCKS * sizeof(*cpl_coords)); |
| 146 | #endif |
| 147 | |
| 148 | /* align start to 16-byte boundary. align length to multiple of 32. |
| 149 | note: coupling start bin % 4 will always be 1 */ |
| 150 | cpl_start = s->start_freq[CPL_CH] - 1; |
| 151 | num_cpl_coefs = FFALIGN(s->num_cpl_subbands * 12 + 1, 32); |
| 152 | cpl_start = FFMIN(256, cpl_start + num_cpl_coefs) - num_cpl_coefs; |
| 153 | |
| 154 | /* calculate coupling channel from fbw channels */ |
| 155 | for (blk = 0; blk < s->num_blocks; blk++) { |
| 156 | AC3Block *block = &s->blocks[blk]; |
| 157 | CoefType *cpl_coef = &block->mdct_coef[CPL_CH][cpl_start]; |
| 158 | if (!block->cpl_in_use) |
| 159 | continue; |
| 160 | memset(cpl_coef, 0, num_cpl_coefs * sizeof(*cpl_coef)); |
| 161 | for (ch = 1; ch <= s->fbw_channels; ch++) { |
| 162 | CoefType *ch_coef = &block->mdct_coef[ch][cpl_start]; |
| 163 | if (!block->channel_in_cpl[ch]) |
| 164 | continue; |
| 165 | for (i = 0; i < num_cpl_coefs; i++) |
| 166 | cpl_coef[i] += ch_coef[i]; |
| 167 | } |
| 168 | |
| 169 | /* coefficients must be clipped in order to be encoded */ |
| 170 | clip_coefficients(&s->adsp, cpl_coef, num_cpl_coefs); |
| 171 | } |
| 172 | |
| 173 | /* calculate energy in each band in coupling channel and each fbw channel */ |
| 174 | /* TODO: possibly use SIMD to speed up energy calculation */ |
| 175 | bnd = 0; |
| 176 | i = s->start_freq[CPL_CH]; |
| 177 | while (i < s->cpl_end_freq) { |
| 178 | int band_size = s->cpl_band_sizes[bnd]; |
| 179 | for (ch = CPL_CH; ch <= s->fbw_channels; ch++) { |
| 180 | for (blk = 0; blk < s->num_blocks; blk++) { |
| 181 | AC3Block *block = &s->blocks[blk]; |
| 182 | if (!block->cpl_in_use || (ch > CPL_CH && !block->channel_in_cpl[ch])) |
| 183 | continue; |
| 184 | for (j = 0; j < band_size; j++) { |
| 185 | CoefType v = block->mdct_coef[ch][i+j]; |
| 186 | MAC_COEF(energy[blk][ch][bnd], v, v); |
| 187 | } |
| 188 | } |
| 189 | } |
| 190 | i += band_size; |
| 191 | bnd++; |
| 192 | } |
| 193 | |
| 194 | /* calculate coupling coordinates for all blocks for all channels */ |
| 195 | for (blk = 0; blk < s->num_blocks; blk++) { |
| 196 | AC3Block *block = &s->blocks[blk]; |
| 197 | if (!block->cpl_in_use) |
| 198 | continue; |
| 199 | for (ch = 1; ch <= s->fbw_channels; ch++) { |
| 200 | if (!block->channel_in_cpl[ch]) |
| 201 | continue; |
| 202 | for (bnd = 0; bnd < s->num_cpl_bands; bnd++) { |
| 203 | cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy[blk][ch][bnd], |
| 204 | energy[blk][CPL_CH][bnd]); |
| 205 | } |
| 206 | } |
| 207 | } |
| 208 | |
| 209 | /* determine which blocks to send new coupling coordinates for */ |
| 210 | for (blk = 0; blk < s->num_blocks; blk++) { |
| 211 | AC3Block *block = &s->blocks[blk]; |
| 212 | AC3Block *block0 = blk ? &s->blocks[blk-1] : NULL; |
| 213 | |
| 214 | memset(block->new_cpl_coords, 0, sizeof(block->new_cpl_coords)); |
| 215 | |
| 216 | if (block->cpl_in_use) { |
| 217 | /* send new coordinates if this is the first block, if previous |
| 218 | * block did not use coupling but this block does, the channels |
| 219 | * using coupling has changed from the previous block, or the |
| 220 | * coordinate difference from the last block for any channel is |
| 221 | * greater than a threshold value. */ |
| 222 | if (blk == 0 || !block0->cpl_in_use) { |
| 223 | for (ch = 1; ch <= s->fbw_channels; ch++) |
| 224 | block->new_cpl_coords[ch] = 1; |
| 225 | } else { |
| 226 | for (ch = 1; ch <= s->fbw_channels; ch++) { |
| 227 | if (!block->channel_in_cpl[ch]) |
| 228 | continue; |
| 229 | if (!block0->channel_in_cpl[ch]) { |
| 230 | block->new_cpl_coords[ch] = 1; |
| 231 | } else { |
| 232 | CoefSumType coord_diff = 0; |
| 233 | for (bnd = 0; bnd < s->num_cpl_bands; bnd++) { |
| 234 | coord_diff += FFABS(cpl_coords[blk-1][ch][bnd] - |
| 235 | cpl_coords[blk ][ch][bnd]); |
| 236 | } |
| 237 | coord_diff /= s->num_cpl_bands; |
| 238 | if (coord_diff > NEW_CPL_COORD_THRESHOLD) |
| 239 | block->new_cpl_coords[ch] = 1; |
| 240 | } |
| 241 | } |
| 242 | } |
| 243 | } |
| 244 | } |
| 245 | |
| 246 | /* calculate final coupling coordinates, taking into account reusing of |
| 247 | coordinates in successive blocks */ |
| 248 | for (bnd = 0; bnd < s->num_cpl_bands; bnd++) { |
| 249 | blk = 0; |
| 250 | while (blk < s->num_blocks) { |
| 251 | int av_uninit(blk1); |
| 252 | AC3Block *block = &s->blocks[blk]; |
| 253 | |
| 254 | if (!block->cpl_in_use) { |
| 255 | blk++; |
| 256 | continue; |
| 257 | } |
| 258 | |
| 259 | for (ch = 1; ch <= s->fbw_channels; ch++) { |
| 260 | CoefSumType energy_ch, energy_cpl; |
| 261 | if (!block->channel_in_cpl[ch]) |
| 262 | continue; |
| 263 | energy_cpl = energy[blk][CPL_CH][bnd]; |
| 264 | energy_ch = energy[blk][ch][bnd]; |
| 265 | blk1 = blk+1; |
| 266 | while (blk1 < s->num_blocks && !s->blocks[blk1].new_cpl_coords[ch]) { |
| 267 | if (s->blocks[blk1].cpl_in_use) { |
| 268 | energy_cpl += energy[blk1][CPL_CH][bnd]; |
| 269 | energy_ch += energy[blk1][ch][bnd]; |
| 270 | } |
| 271 | blk1++; |
| 272 | } |
| 273 | cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy_ch, energy_cpl); |
| 274 | } |
| 275 | blk = blk1; |
| 276 | } |
| 277 | } |
| 278 | |
| 279 | /* calculate exponents/mantissas for coupling coordinates */ |
| 280 | for (blk = 0; blk < s->num_blocks; blk++) { |
| 281 | AC3Block *block = &s->blocks[blk]; |
| 282 | if (!block->cpl_in_use) |
| 283 | continue; |
| 284 | |
| 285 | #if CONFIG_AC3ENC_FLOAT |
| 286 | s->ac3dsp.float_to_fixed24(fixed_cpl_coords[blk][1], |
| 287 | cpl_coords[blk][1], |
| 288 | s->fbw_channels * 16); |
| 289 | #endif |
| 290 | s->ac3dsp.extract_exponents(block->cpl_coord_exp[1], |
| 291 | fixed_cpl_coords[blk][1], |
| 292 | s->fbw_channels * 16); |
| 293 | |
| 294 | for (ch = 1; ch <= s->fbw_channels; ch++) { |
| 295 | int bnd, min_exp, max_exp, master_exp; |
| 296 | |
| 297 | if (!block->new_cpl_coords[ch]) |
| 298 | continue; |
| 299 | |
| 300 | /* determine master exponent */ |
| 301 | min_exp = max_exp = block->cpl_coord_exp[ch][0]; |
| 302 | for (bnd = 1; bnd < s->num_cpl_bands; bnd++) { |
| 303 | int exp = block->cpl_coord_exp[ch][bnd]; |
| 304 | min_exp = FFMIN(exp, min_exp); |
| 305 | max_exp = FFMAX(exp, max_exp); |
| 306 | } |
| 307 | master_exp = ((max_exp - 15) + 2) / 3; |
| 308 | master_exp = FFMAX(master_exp, 0); |
| 309 | while (min_exp < master_exp * 3) |
| 310 | master_exp--; |
| 311 | for (bnd = 0; bnd < s->num_cpl_bands; bnd++) { |
| 312 | block->cpl_coord_exp[ch][bnd] = av_clip(block->cpl_coord_exp[ch][bnd] - |
| 313 | master_exp * 3, 0, 15); |
| 314 | } |
| 315 | block->cpl_master_exp[ch] = master_exp; |
| 316 | |
| 317 | /* quantize mantissas */ |
| 318 | for (bnd = 0; bnd < s->num_cpl_bands; bnd++) { |
| 319 | int cpl_exp = block->cpl_coord_exp[ch][bnd]; |
| 320 | int cpl_mant = (fixed_cpl_coords[blk][ch][bnd] << (5 + cpl_exp + master_exp * 3)) >> 24; |
| 321 | if (cpl_exp == 15) |
| 322 | cpl_mant >>= 1; |
| 323 | else |
| 324 | cpl_mant -= 16; |
| 325 | |
| 326 | block->cpl_coord_mant[ch][bnd] = cpl_mant; |
| 327 | } |
| 328 | } |
| 329 | } |
| 330 | |
| 331 | if (CONFIG_EAC3_ENCODER && s->eac3) |
| 332 | ff_eac3_set_cpl_states(s); |
| 333 | } |
| 334 | |
| 335 | |
| 336 | /* |
| 337 | * Determine rematrixing flags for each block and band. |
| 338 | */ |
| 339 | static void compute_rematrixing_strategy(AC3EncodeContext *s) |
| 340 | { |
| 341 | int nb_coefs; |
| 342 | int blk, bnd; |
| 343 | AC3Block *block, *block0 = NULL; |
| 344 | |
| 345 | if (s->channel_mode != AC3_CHMODE_STEREO) |
| 346 | return; |
| 347 | |
| 348 | for (blk = 0; blk < s->num_blocks; blk++) { |
| 349 | block = &s->blocks[blk]; |
| 350 | block->new_rematrixing_strategy = !blk; |
| 351 | |
| 352 | block->num_rematrixing_bands = 4; |
| 353 | if (block->cpl_in_use) { |
| 354 | block->num_rematrixing_bands -= (s->start_freq[CPL_CH] <= 61); |
| 355 | block->num_rematrixing_bands -= (s->start_freq[CPL_CH] == 37); |
| 356 | if (blk && block->num_rematrixing_bands != block0->num_rematrixing_bands) |
| 357 | block->new_rematrixing_strategy = 1; |
| 358 | } |
| 359 | nb_coefs = FFMIN(block->end_freq[1], block->end_freq[2]); |
| 360 | |
| 361 | if (!s->rematrixing_enabled) { |
| 362 | block0 = block; |
| 363 | continue; |
| 364 | } |
| 365 | |
| 366 | for (bnd = 0; bnd < block->num_rematrixing_bands; bnd++) { |
| 367 | /* calculate sum of squared coeffs for one band in one block */ |
| 368 | int start = ff_ac3_rematrix_band_tab[bnd]; |
| 369 | int end = FFMIN(nb_coefs, ff_ac3_rematrix_band_tab[bnd+1]); |
| 370 | CoefSumType sum[4]; |
| 371 | sum_square_butterfly(s, sum, block->mdct_coef[1] + start, |
| 372 | block->mdct_coef[2] + start, end - start); |
| 373 | |
| 374 | /* compare sums to determine if rematrixing will be used for this band */ |
| 375 | if (FFMIN(sum[2], sum[3]) < FFMIN(sum[0], sum[1])) |
| 376 | block->rematrixing_flags[bnd] = 1; |
| 377 | else |
| 378 | block->rematrixing_flags[bnd] = 0; |
| 379 | |
| 380 | /* determine if new rematrixing flags will be sent */ |
| 381 | if (blk && |
| 382 | block->rematrixing_flags[bnd] != block0->rematrixing_flags[bnd]) { |
| 383 | block->new_rematrixing_strategy = 1; |
| 384 | } |
| 385 | } |
| 386 | block0 = block; |
| 387 | } |
| 388 | } |
| 389 | |
| 390 | |
| 391 | int AC3_NAME(encode_frame)(AVCodecContext *avctx, AVPacket *avpkt, |
| 392 | const AVFrame *frame, int *got_packet_ptr) |
| 393 | { |
| 394 | AC3EncodeContext *s = avctx->priv_data; |
| 395 | int ret; |
| 396 | |
| 397 | if (s->options.allow_per_frame_metadata) { |
| 398 | ret = ff_ac3_validate_metadata(s); |
| 399 | if (ret) |
| 400 | return ret; |
| 401 | } |
| 402 | |
| 403 | if (s->bit_alloc.sr_code == 1 || s->eac3) |
| 404 | ff_ac3_adjust_frame_size(s); |
| 405 | |
| 406 | copy_input_samples(s, (SampleType **)frame->extended_data); |
| 407 | |
| 408 | apply_mdct(s); |
| 409 | |
| 410 | if (s->fixed_point) |
| 411 | scale_coefficients(s); |
| 412 | |
| 413 | clip_coefficients(&s->adsp, s->blocks[0].mdct_coef[1], |
| 414 | AC3_MAX_COEFS * s->num_blocks * s->channels); |
| 415 | |
| 416 | s->cpl_on = s->cpl_enabled; |
| 417 | ff_ac3_compute_coupling_strategy(s); |
| 418 | |
| 419 | if (s->cpl_on) |
| 420 | apply_channel_coupling(s); |
| 421 | |
| 422 | compute_rematrixing_strategy(s); |
| 423 | |
| 424 | if (!s->fixed_point) |
| 425 | scale_coefficients(s); |
| 426 | |
| 427 | ff_ac3_apply_rematrixing(s); |
| 428 | |
| 429 | ff_ac3_process_exponents(s); |
| 430 | |
| 431 | ret = ff_ac3_compute_bit_allocation(s); |
| 432 | if (ret) { |
| 433 | av_log(avctx, AV_LOG_ERROR, "Bit allocation failed. Try increasing the bitrate.\n"); |
| 434 | return ret; |
| 435 | } |
| 436 | |
| 437 | ff_ac3_group_exponents(s); |
| 438 | |
| 439 | ff_ac3_quantize_mantissas(s); |
| 440 | |
| 441 | if ((ret = ff_alloc_packet2(avctx, avpkt, s->frame_size)) < 0) |
| 442 | return ret; |
| 443 | ff_ac3_output_frame(s, avpkt->data); |
| 444 | |
| 445 | if (frame->pts != AV_NOPTS_VALUE) |
| 446 | avpkt->pts = frame->pts - ff_samples_to_time_base(avctx, avctx->delay); |
| 447 | |
| 448 | *got_packet_ptr = 1; |
| 449 | return 0; |
| 450 | } |