| 1 | /* |
| 2 | * AC-3 Audio Decoder |
| 3 | * This code was developed as part of Google Summer of Code 2006. |
| 4 | * E-AC-3 support was added as part of Google Summer of Code 2007. |
| 5 | * |
| 6 | * Copyright (c) 2006 Kartikey Mahendra BHATT (bhattkm at gmail dot com) |
| 7 | * Copyright (c) 2007-2008 Bartlomiej Wolowiec <bartek.wolowiec@gmail.com> |
| 8 | * Copyright (c) 2007 Justin Ruggles <justin.ruggles@gmail.com> |
| 9 | * |
| 10 | * This file is part of FFmpeg. |
| 11 | * |
| 12 | * FFmpeg is free software; you can redistribute it and/or |
| 13 | * modify it under the terms of the GNU Lesser General Public |
| 14 | * License as published by the Free Software Foundation; either |
| 15 | * version 2.1 of the License, or (at your option) any later version. |
| 16 | * |
| 17 | * FFmpeg is distributed in the hope that it will be useful, |
| 18 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 19 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 20 | * Lesser General Public License for more details. |
| 21 | * |
| 22 | * You should have received a copy of the GNU Lesser General Public |
| 23 | * License along with FFmpeg; if not, write to the Free Software |
| 24 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 25 | */ |
| 26 | |
| 27 | #include <stdio.h> |
| 28 | #include <stddef.h> |
| 29 | #include <math.h> |
| 30 | #include <string.h> |
| 31 | |
| 32 | #include "libavutil/channel_layout.h" |
| 33 | #include "libavutil/crc.h" |
| 34 | #include "libavutil/downmix_info.h" |
| 35 | #include "libavutil/opt.h" |
| 36 | #include "bswapdsp.h" |
| 37 | #include "internal.h" |
| 38 | #include "aac_ac3_parser.h" |
| 39 | #include "ac3_parser.h" |
| 40 | #include "ac3dec.h" |
| 41 | #include "ac3dec_data.h" |
| 42 | #include "kbdwin.h" |
| 43 | |
| 44 | /** |
| 45 | * table for ungrouping 3 values in 7 bits. |
| 46 | * used for exponents and bap=2 mantissas |
| 47 | */ |
| 48 | static uint8_t ungroup_3_in_7_bits_tab[128][3]; |
| 49 | |
| 50 | /** tables for ungrouping mantissas */ |
| 51 | static int b1_mantissas[32][3]; |
| 52 | static int b2_mantissas[128][3]; |
| 53 | static int b3_mantissas[8]; |
| 54 | static int b4_mantissas[128][2]; |
| 55 | static int b5_mantissas[16]; |
| 56 | |
| 57 | /** |
| 58 | * Quantization table: levels for symmetric. bits for asymmetric. |
| 59 | * reference: Table 7.18 Mapping of bap to Quantizer |
| 60 | */ |
| 61 | static const uint8_t quantization_tab[16] = { |
| 62 | 0, 3, 5, 7, 11, 15, |
| 63 | 5, 6, 7, 8, 9, 10, 11, 12, 14, 16 |
| 64 | }; |
| 65 | |
| 66 | /** dynamic range table. converts codes to scale factors. */ |
| 67 | static float dynamic_range_tab[256]; |
| 68 | static float heavy_dynamic_range_tab[256]; |
| 69 | |
| 70 | /** Adjustments in dB gain */ |
| 71 | static const float gain_levels[9] = { |
| 72 | LEVEL_PLUS_3DB, |
| 73 | LEVEL_PLUS_1POINT5DB, |
| 74 | LEVEL_ONE, |
| 75 | LEVEL_MINUS_1POINT5DB, |
| 76 | LEVEL_MINUS_3DB, |
| 77 | LEVEL_MINUS_4POINT5DB, |
| 78 | LEVEL_MINUS_6DB, |
| 79 | LEVEL_ZERO, |
| 80 | LEVEL_MINUS_9DB |
| 81 | }; |
| 82 | |
| 83 | /** Adjustments in dB gain (LFE, +10 to -21 dB) */ |
| 84 | static const float gain_levels_lfe[32] = { |
| 85 | 3.162275, 2.818382, 2.511886, 2.238719, 1.995261, 1.778278, 1.584893, |
| 86 | 1.412536, 1.258924, 1.122018, 1.000000, 0.891251, 0.794328, 0.707946, |
| 87 | 0.630957, 0.562341, 0.501187, 0.446683, 0.398107, 0.354813, 0.316227, |
| 88 | 0.281838, 0.251188, 0.223872, 0.199526, 0.177828, 0.158489, 0.141253, |
| 89 | 0.125892, 0.112201, 0.100000, 0.089125 |
| 90 | }; |
| 91 | |
| 92 | /** |
| 93 | * Table for default stereo downmixing coefficients |
| 94 | * reference: Section 7.8.2 Downmixing Into Two Channels |
| 95 | */ |
| 96 | static const uint8_t ac3_default_coeffs[8][5][2] = { |
| 97 | { { 2, 7 }, { 7, 2 }, }, |
| 98 | { { 4, 4 }, }, |
| 99 | { { 2, 7 }, { 7, 2 }, }, |
| 100 | { { 2, 7 }, { 5, 5 }, { 7, 2 }, }, |
| 101 | { { 2, 7 }, { 7, 2 }, { 6, 6 }, }, |
| 102 | { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 8, 8 }, }, |
| 103 | { { 2, 7 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, }, |
| 104 | { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, }, |
| 105 | }; |
| 106 | |
| 107 | /** |
| 108 | * Symmetrical Dequantization |
| 109 | * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization |
| 110 | * Tables 7.19 to 7.23 |
| 111 | */ |
| 112 | static inline int |
| 113 | symmetric_dequant(int code, int levels) |
| 114 | { |
| 115 | return ((code - (levels >> 1)) << 24) / levels; |
| 116 | } |
| 117 | |
| 118 | /* |
| 119 | * Initialize tables at runtime. |
| 120 | */ |
| 121 | static av_cold void ac3_tables_init(void) |
| 122 | { |
| 123 | int i; |
| 124 | |
| 125 | /* generate table for ungrouping 3 values in 7 bits |
| 126 | reference: Section 7.1.3 Exponent Decoding */ |
| 127 | for (i = 0; i < 128; i++) { |
| 128 | ungroup_3_in_7_bits_tab[i][0] = i / 25; |
| 129 | ungroup_3_in_7_bits_tab[i][1] = (i % 25) / 5; |
| 130 | ungroup_3_in_7_bits_tab[i][2] = (i % 25) % 5; |
| 131 | } |
| 132 | |
| 133 | /* generate grouped mantissa tables |
| 134 | reference: Section 7.3.5 Ungrouping of Mantissas */ |
| 135 | for (i = 0; i < 32; i++) { |
| 136 | /* bap=1 mantissas */ |
| 137 | b1_mantissas[i][0] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][0], 3); |
| 138 | b1_mantissas[i][1] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][1], 3); |
| 139 | b1_mantissas[i][2] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][2], 3); |
| 140 | } |
| 141 | for (i = 0; i < 128; i++) { |
| 142 | /* bap=2 mantissas */ |
| 143 | b2_mantissas[i][0] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][0], 5); |
| 144 | b2_mantissas[i][1] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][1], 5); |
| 145 | b2_mantissas[i][2] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][2], 5); |
| 146 | |
| 147 | /* bap=4 mantissas */ |
| 148 | b4_mantissas[i][0] = symmetric_dequant(i / 11, 11); |
| 149 | b4_mantissas[i][1] = symmetric_dequant(i % 11, 11); |
| 150 | } |
| 151 | /* generate ungrouped mantissa tables |
| 152 | reference: Tables 7.21 and 7.23 */ |
| 153 | for (i = 0; i < 7; i++) { |
| 154 | /* bap=3 mantissas */ |
| 155 | b3_mantissas[i] = symmetric_dequant(i, 7); |
| 156 | } |
| 157 | for (i = 0; i < 15; i++) { |
| 158 | /* bap=5 mantissas */ |
| 159 | b5_mantissas[i] = symmetric_dequant(i, 15); |
| 160 | } |
| 161 | |
| 162 | /* generate dynamic range table |
| 163 | reference: Section 7.7.1 Dynamic Range Control */ |
| 164 | for (i = 0; i < 256; i++) { |
| 165 | int v = (i >> 5) - ((i >> 7) << 3) - 5; |
| 166 | dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20); |
| 167 | } |
| 168 | |
| 169 | /* generate compr dynamic range table |
| 170 | reference: Section 7.7.2 Heavy Compression */ |
| 171 | for (i = 0; i < 256; i++) { |
| 172 | int v = (i >> 4) - ((i >> 7) << 4) - 4; |
| 173 | heavy_dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0xF) | 0x10); |
| 174 | } |
| 175 | |
| 176 | } |
| 177 | |
| 178 | /** |
| 179 | * AVCodec initialization |
| 180 | */ |
| 181 | static av_cold int ac3_decode_init(AVCodecContext *avctx) |
| 182 | { |
| 183 | AC3DecodeContext *s = avctx->priv_data; |
| 184 | int i; |
| 185 | |
| 186 | s->avctx = avctx; |
| 187 | |
| 188 | ff_ac3_common_init(); |
| 189 | ac3_tables_init(); |
| 190 | ff_mdct_init(&s->imdct_256, 8, 1, 1.0); |
| 191 | ff_mdct_init(&s->imdct_512, 9, 1, 1.0); |
| 192 | AC3_RENAME(ff_kbd_window_init)(s->window, 5.0, 256); |
| 193 | ff_bswapdsp_init(&s->bdsp); |
| 194 | |
| 195 | #if (USE_FIXED) |
| 196 | s->fdsp = avpriv_alloc_fixed_dsp(avctx->flags & CODEC_FLAG_BITEXACT); |
| 197 | #else |
| 198 | avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT); |
| 199 | #endif |
| 200 | |
| 201 | ff_ac3dsp_init(&s->ac3dsp, avctx->flags & CODEC_FLAG_BITEXACT); |
| 202 | ff_fmt_convert_init(&s->fmt_conv, avctx); |
| 203 | av_lfg_init(&s->dith_state, 0); |
| 204 | |
| 205 | if (USE_FIXED) |
| 206 | avctx->sample_fmt = AV_SAMPLE_FMT_S16P; |
| 207 | else |
| 208 | avctx->sample_fmt = AV_SAMPLE_FMT_FLTP; |
| 209 | |
| 210 | /* allow downmixing to stereo or mono */ |
| 211 | #if FF_API_REQUEST_CHANNELS |
| 212 | FF_DISABLE_DEPRECATION_WARNINGS |
| 213 | if (avctx->request_channels == 1) |
| 214 | avctx->request_channel_layout = AV_CH_LAYOUT_MONO; |
| 215 | else if (avctx->request_channels == 2) |
| 216 | avctx->request_channel_layout = AV_CH_LAYOUT_STEREO; |
| 217 | FF_ENABLE_DEPRECATION_WARNINGS |
| 218 | #endif |
| 219 | if (avctx->channels > 1 && |
| 220 | avctx->request_channel_layout == AV_CH_LAYOUT_MONO) |
| 221 | avctx->channels = 1; |
| 222 | else if (avctx->channels > 2 && |
| 223 | avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) |
| 224 | avctx->channels = 2; |
| 225 | s->downmixed = 1; |
| 226 | |
| 227 | for (i = 0; i < AC3_MAX_CHANNELS; i++) { |
| 228 | s->xcfptr[i] = s->transform_coeffs[i]; |
| 229 | s->dlyptr[i] = s->delay[i]; |
| 230 | } |
| 231 | |
| 232 | return 0; |
| 233 | } |
| 234 | |
| 235 | /** |
| 236 | * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream. |
| 237 | * GetBitContext within AC3DecodeContext must point to |
| 238 | * the start of the synchronized AC-3 bitstream. |
| 239 | */ |
| 240 | static int ac3_parse_header(AC3DecodeContext *s) |
| 241 | { |
| 242 | GetBitContext *gbc = &s->gbc; |
| 243 | int i; |
| 244 | |
| 245 | /* read the rest of the bsi. read twice for dual mono mode. */ |
| 246 | i = !s->channel_mode; |
| 247 | do { |
| 248 | s->dialog_normalization[(!s->channel_mode)-i] = -get_bits(gbc, 5); |
| 249 | if (s->dialog_normalization[(!s->channel_mode)-i] == 0) { |
| 250 | s->dialog_normalization[(!s->channel_mode)-i] = -31; |
| 251 | } |
| 252 | if (s->target_level != 0) { |
| 253 | s->level_gain[(!s->channel_mode)-i] = powf(2.0f, |
| 254 | (float)(s->target_level - |
| 255 | s->dialog_normalization[(!s->channel_mode)-i])/6.0f); |
| 256 | } |
| 257 | if (s->compression_exists[(!s->channel_mode)-i] = get_bits1(gbc)) { |
| 258 | s->heavy_dynamic_range[(!s->channel_mode)-i] = |
| 259 | AC3_HEAVY_RANGE(get_bits(gbc, 8)); |
| 260 | } |
| 261 | if (get_bits1(gbc)) |
| 262 | skip_bits(gbc, 8); //skip language code |
| 263 | if (get_bits1(gbc)) |
| 264 | skip_bits(gbc, 7); //skip audio production information |
| 265 | } while (i--); |
| 266 | |
| 267 | skip_bits(gbc, 2); //skip copyright bit and original bitstream bit |
| 268 | |
| 269 | /* skip the timecodes or parse the Alternate Bit Stream Syntax */ |
| 270 | if (s->bitstream_id != 6) { |
| 271 | if (get_bits1(gbc)) |
| 272 | skip_bits(gbc, 14); //skip timecode1 |
| 273 | if (get_bits1(gbc)) |
| 274 | skip_bits(gbc, 14); //skip timecode2 |
| 275 | } else { |
| 276 | if (get_bits1(gbc)) { |
| 277 | s->preferred_downmix = get_bits(gbc, 2); |
| 278 | s->center_mix_level_ltrt = get_bits(gbc, 3); |
| 279 | s->surround_mix_level_ltrt = av_clip(get_bits(gbc, 3), 3, 7); |
| 280 | s->center_mix_level = get_bits(gbc, 3); |
| 281 | s->surround_mix_level = av_clip(get_bits(gbc, 3), 3, 7); |
| 282 | } |
| 283 | if (get_bits1(gbc)) { |
| 284 | s->dolby_surround_ex_mode = get_bits(gbc, 2); |
| 285 | s->dolby_headphone_mode = get_bits(gbc, 2); |
| 286 | skip_bits(gbc, 10); // skip adconvtyp (1), xbsi2 (8), encinfo (1) |
| 287 | } |
| 288 | } |
| 289 | |
| 290 | /* skip additional bitstream info */ |
| 291 | if (get_bits1(gbc)) { |
| 292 | i = get_bits(gbc, 6); |
| 293 | do { |
| 294 | skip_bits(gbc, 8); |
| 295 | } while (i--); |
| 296 | } |
| 297 | |
| 298 | return 0; |
| 299 | } |
| 300 | |
| 301 | /** |
| 302 | * Common function to parse AC-3 or E-AC-3 frame header |
| 303 | */ |
| 304 | static int parse_frame_header(AC3DecodeContext *s) |
| 305 | { |
| 306 | AC3HeaderInfo hdr, *phdr=&hdr; |
| 307 | int err; |
| 308 | |
| 309 | err = avpriv_ac3_parse_header2(&s->gbc, &phdr); |
| 310 | if (err) |
| 311 | return err; |
| 312 | |
| 313 | /* get decoding parameters from header info */ |
| 314 | s->bit_alloc_params.sr_code = hdr.sr_code; |
| 315 | s->bitstream_id = hdr.bitstream_id; |
| 316 | s->bitstream_mode = hdr.bitstream_mode; |
| 317 | s->channel_mode = hdr.channel_mode; |
| 318 | s->lfe_on = hdr.lfe_on; |
| 319 | s->bit_alloc_params.sr_shift = hdr.sr_shift; |
| 320 | s->sample_rate = hdr.sample_rate; |
| 321 | s->bit_rate = hdr.bit_rate; |
| 322 | s->channels = hdr.channels; |
| 323 | s->fbw_channels = s->channels - s->lfe_on; |
| 324 | s->lfe_ch = s->fbw_channels + 1; |
| 325 | s->frame_size = hdr.frame_size; |
| 326 | s->preferred_downmix = AC3_DMIXMOD_NOTINDICATED; |
| 327 | s->center_mix_level = hdr.center_mix_level; |
| 328 | s->center_mix_level_ltrt = 4; // -3.0dB |
| 329 | s->surround_mix_level = hdr.surround_mix_level; |
| 330 | s->surround_mix_level_ltrt = 4; // -3.0dB |
| 331 | s->lfe_mix_level_exists = 0; |
| 332 | s->num_blocks = hdr.num_blocks; |
| 333 | s->frame_type = hdr.frame_type; |
| 334 | s->substreamid = hdr.substreamid; |
| 335 | s->dolby_surround_mode = hdr.dolby_surround_mode; |
| 336 | s->dolby_surround_ex_mode = AC3_DSUREXMOD_NOTINDICATED; |
| 337 | s->dolby_headphone_mode = AC3_DHEADPHONMOD_NOTINDICATED; |
| 338 | |
| 339 | if (s->lfe_on) { |
| 340 | s->start_freq[s->lfe_ch] = 0; |
| 341 | s->end_freq[s->lfe_ch] = 7; |
| 342 | s->num_exp_groups[s->lfe_ch] = 2; |
| 343 | s->channel_in_cpl[s->lfe_ch] = 0; |
| 344 | } |
| 345 | |
| 346 | if (s->bitstream_id <= 10) { |
| 347 | s->eac3 = 0; |
| 348 | s->snr_offset_strategy = 2; |
| 349 | s->block_switch_syntax = 1; |
| 350 | s->dither_flag_syntax = 1; |
| 351 | s->bit_allocation_syntax = 1; |
| 352 | s->fast_gain_syntax = 0; |
| 353 | s->first_cpl_leak = 0; |
| 354 | s->dba_syntax = 1; |
| 355 | s->skip_syntax = 1; |
| 356 | memset(s->channel_uses_aht, 0, sizeof(s->channel_uses_aht)); |
| 357 | return ac3_parse_header(s); |
| 358 | } else if (CONFIG_EAC3_DECODER) { |
| 359 | s->eac3 = 1; |
| 360 | return ff_eac3_parse_header(s); |
| 361 | } else { |
| 362 | av_log(s->avctx, AV_LOG_ERROR, "E-AC-3 support not compiled in\n"); |
| 363 | return AVERROR(ENOSYS); |
| 364 | } |
| 365 | } |
| 366 | |
| 367 | /** |
| 368 | * Set stereo downmixing coefficients based on frame header info. |
| 369 | * reference: Section 7.8.2 Downmixing Into Two Channels |
| 370 | */ |
| 371 | static void set_downmix_coeffs(AC3DecodeContext *s) |
| 372 | { |
| 373 | int i; |
| 374 | float cmix = gain_levels[s-> center_mix_level]; |
| 375 | float smix = gain_levels[s->surround_mix_level]; |
| 376 | float norm0, norm1; |
| 377 | float downmix_coeffs[AC3_MAX_CHANNELS][2]; |
| 378 | |
| 379 | for (i = 0; i < s->fbw_channels; i++) { |
| 380 | downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]]; |
| 381 | downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]]; |
| 382 | } |
| 383 | if (s->channel_mode > 1 && s->channel_mode & 1) { |
| 384 | downmix_coeffs[1][0] = downmix_coeffs[1][1] = cmix; |
| 385 | } |
| 386 | if (s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) { |
| 387 | int nf = s->channel_mode - 2; |
| 388 | downmix_coeffs[nf][0] = downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB; |
| 389 | } |
| 390 | if (s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) { |
| 391 | int nf = s->channel_mode - 4; |
| 392 | downmix_coeffs[nf][0] = downmix_coeffs[nf+1][1] = smix; |
| 393 | } |
| 394 | |
| 395 | /* renormalize */ |
| 396 | norm0 = norm1 = 0.0; |
| 397 | for (i = 0; i < s->fbw_channels; i++) { |
| 398 | norm0 += downmix_coeffs[i][0]; |
| 399 | norm1 += downmix_coeffs[i][1]; |
| 400 | } |
| 401 | norm0 = 1.0f / norm0; |
| 402 | norm1 = 1.0f / norm1; |
| 403 | for (i = 0; i < s->fbw_channels; i++) { |
| 404 | downmix_coeffs[i][0] *= norm0; |
| 405 | downmix_coeffs[i][1] *= norm1; |
| 406 | } |
| 407 | |
| 408 | if (s->output_mode == AC3_CHMODE_MONO) { |
| 409 | for (i = 0; i < s->fbw_channels; i++) |
| 410 | downmix_coeffs[i][0] = (downmix_coeffs[i][0] + |
| 411 | downmix_coeffs[i][1]) * LEVEL_MINUS_3DB; |
| 412 | } |
| 413 | for (i = 0; i < s->fbw_channels; i++) { |
| 414 | s->downmix_coeffs[i][0] = FIXR12(downmix_coeffs[i][0]); |
| 415 | s->downmix_coeffs[i][1] = FIXR12(downmix_coeffs[i][1]); |
| 416 | } |
| 417 | } |
| 418 | |
| 419 | /** |
| 420 | * Decode the grouped exponents according to exponent strategy. |
| 421 | * reference: Section 7.1.3 Exponent Decoding |
| 422 | */ |
| 423 | static int decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps, |
| 424 | uint8_t absexp, int8_t *dexps) |
| 425 | { |
| 426 | int i, j, grp, group_size; |
| 427 | int dexp[256]; |
| 428 | int expacc, prevexp; |
| 429 | |
| 430 | /* unpack groups */ |
| 431 | group_size = exp_strategy + (exp_strategy == EXP_D45); |
| 432 | for (grp = 0, i = 0; grp < ngrps; grp++) { |
| 433 | expacc = get_bits(gbc, 7); |
| 434 | dexp[i++] = ungroup_3_in_7_bits_tab[expacc][0]; |
| 435 | dexp[i++] = ungroup_3_in_7_bits_tab[expacc][1]; |
| 436 | dexp[i++] = ungroup_3_in_7_bits_tab[expacc][2]; |
| 437 | } |
| 438 | |
| 439 | /* convert to absolute exps and expand groups */ |
| 440 | prevexp = absexp; |
| 441 | for (i = 0, j = 0; i < ngrps * 3; i++) { |
| 442 | prevexp += dexp[i] - 2; |
| 443 | if (prevexp > 24U) |
| 444 | return -1; |
| 445 | switch (group_size) { |
| 446 | case 4: dexps[j++] = prevexp; |
| 447 | dexps[j++] = prevexp; |
| 448 | case 2: dexps[j++] = prevexp; |
| 449 | case 1: dexps[j++] = prevexp; |
| 450 | } |
| 451 | } |
| 452 | return 0; |
| 453 | } |
| 454 | |
| 455 | /** |
| 456 | * Generate transform coefficients for each coupled channel in the coupling |
| 457 | * range using the coupling coefficients and coupling coordinates. |
| 458 | * reference: Section 7.4.3 Coupling Coordinate Format |
| 459 | */ |
| 460 | static void calc_transform_coeffs_cpl(AC3DecodeContext *s) |
| 461 | { |
| 462 | int bin, band, ch; |
| 463 | |
| 464 | bin = s->start_freq[CPL_CH]; |
| 465 | for (band = 0; band < s->num_cpl_bands; band++) { |
| 466 | int band_start = bin; |
| 467 | int band_end = bin + s->cpl_band_sizes[band]; |
| 468 | for (ch = 1; ch <= s->fbw_channels; ch++) { |
| 469 | if (s->channel_in_cpl[ch]) { |
| 470 | int cpl_coord = s->cpl_coords[ch][band] << 5; |
| 471 | for (bin = band_start; bin < band_end; bin++) { |
| 472 | s->fixed_coeffs[ch][bin] = |
| 473 | MULH(s->fixed_coeffs[CPL_CH][bin] << 4, cpl_coord); |
| 474 | } |
| 475 | if (ch == 2 && s->phase_flags[band]) { |
| 476 | for (bin = band_start; bin < band_end; bin++) |
| 477 | s->fixed_coeffs[2][bin] = -s->fixed_coeffs[2][bin]; |
| 478 | } |
| 479 | } |
| 480 | } |
| 481 | bin = band_end; |
| 482 | } |
| 483 | } |
| 484 | |
| 485 | /** |
| 486 | * Grouped mantissas for 3-level 5-level and 11-level quantization |
| 487 | */ |
| 488 | typedef struct { |
| 489 | int b1_mant[2]; |
| 490 | int b2_mant[2]; |
| 491 | int b4_mant; |
| 492 | int b1; |
| 493 | int b2; |
| 494 | int b4; |
| 495 | } mant_groups; |
| 496 | |
| 497 | /** |
| 498 | * Decode the transform coefficients for a particular channel |
| 499 | * reference: Section 7.3 Quantization and Decoding of Mantissas |
| 500 | */ |
| 501 | static void ac3_decode_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m) |
| 502 | { |
| 503 | int start_freq = s->start_freq[ch_index]; |
| 504 | int end_freq = s->end_freq[ch_index]; |
| 505 | uint8_t *baps = s->bap[ch_index]; |
| 506 | int8_t *exps = s->dexps[ch_index]; |
| 507 | int32_t *coeffs = s->fixed_coeffs[ch_index]; |
| 508 | int dither = (ch_index == CPL_CH) || s->dither_flag[ch_index]; |
| 509 | GetBitContext *gbc = &s->gbc; |
| 510 | int freq; |
| 511 | |
| 512 | for (freq = start_freq; freq < end_freq; freq++) { |
| 513 | int bap = baps[freq]; |
| 514 | int mantissa; |
| 515 | switch (bap) { |
| 516 | case 0: |
| 517 | /* random noise with approximate range of -0.707 to 0.707 */ |
| 518 | if (dither) |
| 519 | mantissa = (((av_lfg_get(&s->dith_state)>>8)*181)>>8) - 5931008; |
| 520 | else |
| 521 | mantissa = 0; |
| 522 | break; |
| 523 | case 1: |
| 524 | if (m->b1) { |
| 525 | m->b1--; |
| 526 | mantissa = m->b1_mant[m->b1]; |
| 527 | } else { |
| 528 | int bits = get_bits(gbc, 5); |
| 529 | mantissa = b1_mantissas[bits][0]; |
| 530 | m->b1_mant[1] = b1_mantissas[bits][1]; |
| 531 | m->b1_mant[0] = b1_mantissas[bits][2]; |
| 532 | m->b1 = 2; |
| 533 | } |
| 534 | break; |
| 535 | case 2: |
| 536 | if (m->b2) { |
| 537 | m->b2--; |
| 538 | mantissa = m->b2_mant[m->b2]; |
| 539 | } else { |
| 540 | int bits = get_bits(gbc, 7); |
| 541 | mantissa = b2_mantissas[bits][0]; |
| 542 | m->b2_mant[1] = b2_mantissas[bits][1]; |
| 543 | m->b2_mant[0] = b2_mantissas[bits][2]; |
| 544 | m->b2 = 2; |
| 545 | } |
| 546 | break; |
| 547 | case 3: |
| 548 | mantissa = b3_mantissas[get_bits(gbc, 3)]; |
| 549 | break; |
| 550 | case 4: |
| 551 | if (m->b4) { |
| 552 | m->b4 = 0; |
| 553 | mantissa = m->b4_mant; |
| 554 | } else { |
| 555 | int bits = get_bits(gbc, 7); |
| 556 | mantissa = b4_mantissas[bits][0]; |
| 557 | m->b4_mant = b4_mantissas[bits][1]; |
| 558 | m->b4 = 1; |
| 559 | } |
| 560 | break; |
| 561 | case 5: |
| 562 | mantissa = b5_mantissas[get_bits(gbc, 4)]; |
| 563 | break; |
| 564 | default: /* 6 to 15 */ |
| 565 | /* Shift mantissa and sign-extend it. */ |
| 566 | if (bap > 15) { |
| 567 | av_log(s->avctx, AV_LOG_ERROR, "bap %d is invalid in plain AC-3\n", bap); |
| 568 | bap = 15; |
| 569 | } |
| 570 | mantissa = get_sbits(gbc, quantization_tab[bap]); |
| 571 | mantissa <<= 24 - quantization_tab[bap]; |
| 572 | break; |
| 573 | } |
| 574 | coeffs[freq] = mantissa >> exps[freq]; |
| 575 | } |
| 576 | } |
| 577 | |
| 578 | /** |
| 579 | * Remove random dithering from coupling range coefficients with zero-bit |
| 580 | * mantissas for coupled channels which do not use dithering. |
| 581 | * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0) |
| 582 | */ |
| 583 | static void remove_dithering(AC3DecodeContext *s) { |
| 584 | int ch, i; |
| 585 | |
| 586 | for (ch = 1; ch <= s->fbw_channels; ch++) { |
| 587 | if (!s->dither_flag[ch] && s->channel_in_cpl[ch]) { |
| 588 | for (i = s->start_freq[CPL_CH]; i < s->end_freq[CPL_CH]; i++) { |
| 589 | if (!s->bap[CPL_CH][i]) |
| 590 | s->fixed_coeffs[ch][i] = 0; |
| 591 | } |
| 592 | } |
| 593 | } |
| 594 | } |
| 595 | |
| 596 | static void decode_transform_coeffs_ch(AC3DecodeContext *s, int blk, int ch, |
| 597 | mant_groups *m) |
| 598 | { |
| 599 | if (!s->channel_uses_aht[ch]) { |
| 600 | ac3_decode_transform_coeffs_ch(s, ch, m); |
| 601 | } else { |
| 602 | /* if AHT is used, mantissas for all blocks are encoded in the first |
| 603 | block of the frame. */ |
| 604 | int bin; |
| 605 | if (!blk && CONFIG_EAC3_DECODER) |
| 606 | ff_eac3_decode_transform_coeffs_aht_ch(s, ch); |
| 607 | for (bin = s->start_freq[ch]; bin < s->end_freq[ch]; bin++) { |
| 608 | s->fixed_coeffs[ch][bin] = s->pre_mantissa[ch][bin][blk] >> s->dexps[ch][bin]; |
| 609 | } |
| 610 | } |
| 611 | } |
| 612 | |
| 613 | /** |
| 614 | * Decode the transform coefficients. |
| 615 | */ |
| 616 | static void decode_transform_coeffs(AC3DecodeContext *s, int blk) |
| 617 | { |
| 618 | int ch, end; |
| 619 | int got_cplchan = 0; |
| 620 | mant_groups m; |
| 621 | |
| 622 | m.b1 = m.b2 = m.b4 = 0; |
| 623 | |
| 624 | for (ch = 1; ch <= s->channels; ch++) { |
| 625 | /* transform coefficients for full-bandwidth channel */ |
| 626 | decode_transform_coeffs_ch(s, blk, ch, &m); |
| 627 | /* transform coefficients for coupling channel come right after the |
| 628 | coefficients for the first coupled channel*/ |
| 629 | if (s->channel_in_cpl[ch]) { |
| 630 | if (!got_cplchan) { |
| 631 | decode_transform_coeffs_ch(s, blk, CPL_CH, &m); |
| 632 | calc_transform_coeffs_cpl(s); |
| 633 | got_cplchan = 1; |
| 634 | } |
| 635 | end = s->end_freq[CPL_CH]; |
| 636 | } else { |
| 637 | end = s->end_freq[ch]; |
| 638 | } |
| 639 | do |
| 640 | s->fixed_coeffs[ch][end] = 0; |
| 641 | while (++end < 256); |
| 642 | } |
| 643 | |
| 644 | /* zero the dithered coefficients for appropriate channels */ |
| 645 | remove_dithering(s); |
| 646 | } |
| 647 | |
| 648 | /** |
| 649 | * Stereo rematrixing. |
| 650 | * reference: Section 7.5.4 Rematrixing : Decoding Technique |
| 651 | */ |
| 652 | static void do_rematrixing(AC3DecodeContext *s) |
| 653 | { |
| 654 | int bnd, i; |
| 655 | int end, bndend; |
| 656 | |
| 657 | end = FFMIN(s->end_freq[1], s->end_freq[2]); |
| 658 | |
| 659 | for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++) { |
| 660 | if (s->rematrixing_flags[bnd]) { |
| 661 | bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd + 1]); |
| 662 | for (i = ff_ac3_rematrix_band_tab[bnd]; i < bndend; i++) { |
| 663 | int tmp0 = s->fixed_coeffs[1][i]; |
| 664 | s->fixed_coeffs[1][i] += s->fixed_coeffs[2][i]; |
| 665 | s->fixed_coeffs[2][i] = tmp0 - s->fixed_coeffs[2][i]; |
| 666 | } |
| 667 | } |
| 668 | } |
| 669 | } |
| 670 | |
| 671 | /** |
| 672 | * Inverse MDCT Transform. |
| 673 | * Convert frequency domain coefficients to time-domain audio samples. |
| 674 | * reference: Section 7.9.4 Transformation Equations |
| 675 | */ |
| 676 | static inline void do_imdct(AC3DecodeContext *s, int channels) |
| 677 | { |
| 678 | int ch; |
| 679 | |
| 680 | for (ch = 1; ch <= channels; ch++) { |
| 681 | if (s->block_switch[ch]) { |
| 682 | int i; |
| 683 | FFTSample *x = s->tmp_output + 128; |
| 684 | for (i = 0; i < 128; i++) |
| 685 | x[i] = s->transform_coeffs[ch][2 * i]; |
| 686 | s->imdct_256.imdct_half(&s->imdct_256, s->tmp_output, x); |
| 687 | #if USE_FIXED |
| 688 | s->fdsp->vector_fmul_window_scaled(s->outptr[ch - 1], s->delay[ch - 1], |
| 689 | s->tmp_output, s->window, 128, 8); |
| 690 | #else |
| 691 | s->fdsp.vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1], |
| 692 | s->tmp_output, s->window, 128); |
| 693 | #endif |
| 694 | for (i = 0; i < 128; i++) |
| 695 | x[i] = s->transform_coeffs[ch][2 * i + 1]; |
| 696 | s->imdct_256.imdct_half(&s->imdct_256, s->delay[ch - 1], x); |
| 697 | } else { |
| 698 | s->imdct_512.imdct_half(&s->imdct_512, s->tmp_output, s->transform_coeffs[ch]); |
| 699 | #if USE_FIXED |
| 700 | s->fdsp->vector_fmul_window_scaled(s->outptr[ch - 1], s->delay[ch - 1], |
| 701 | s->tmp_output, s->window, 128, 8); |
| 702 | #else |
| 703 | s->fdsp.vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1], |
| 704 | s->tmp_output, s->window, 128); |
| 705 | #endif |
| 706 | memcpy(s->delay[ch - 1], s->tmp_output + 128, 128 * sizeof(FFTSample)); |
| 707 | } |
| 708 | } |
| 709 | } |
| 710 | |
| 711 | /** |
| 712 | * Upmix delay samples from stereo to original channel layout. |
| 713 | */ |
| 714 | static void ac3_upmix_delay(AC3DecodeContext *s) |
| 715 | { |
| 716 | int channel_data_size = sizeof(s->delay[0]); |
| 717 | switch (s->channel_mode) { |
| 718 | case AC3_CHMODE_DUALMONO: |
| 719 | case AC3_CHMODE_STEREO: |
| 720 | /* upmix mono to stereo */ |
| 721 | memcpy(s->delay[1], s->delay[0], channel_data_size); |
| 722 | break; |
| 723 | case AC3_CHMODE_2F2R: |
| 724 | memset(s->delay[3], 0, channel_data_size); |
| 725 | case AC3_CHMODE_2F1R: |
| 726 | memset(s->delay[2], 0, channel_data_size); |
| 727 | break; |
| 728 | case AC3_CHMODE_3F2R: |
| 729 | memset(s->delay[4], 0, channel_data_size); |
| 730 | case AC3_CHMODE_3F1R: |
| 731 | memset(s->delay[3], 0, channel_data_size); |
| 732 | case AC3_CHMODE_3F: |
| 733 | memcpy(s->delay[2], s->delay[1], channel_data_size); |
| 734 | memset(s->delay[1], 0, channel_data_size); |
| 735 | break; |
| 736 | } |
| 737 | } |
| 738 | |
| 739 | /** |
| 740 | * Decode band structure for coupling, spectral extension, or enhanced coupling. |
| 741 | * The band structure defines how many subbands are in each band. For each |
| 742 | * subband in the range, 1 means it is combined with the previous band, and 0 |
| 743 | * means that it starts a new band. |
| 744 | * |
| 745 | * @param[in] gbc bit reader context |
| 746 | * @param[in] blk block number |
| 747 | * @param[in] eac3 flag to indicate E-AC-3 |
| 748 | * @param[in] ecpl flag to indicate enhanced coupling |
| 749 | * @param[in] start_subband subband number for start of range |
| 750 | * @param[in] end_subband subband number for end of range |
| 751 | * @param[in] default_band_struct default band structure table |
| 752 | * @param[out] num_bands number of bands (optionally NULL) |
| 753 | * @param[out] band_sizes array containing the number of bins in each band (optionally NULL) |
| 754 | */ |
| 755 | static void decode_band_structure(GetBitContext *gbc, int blk, int eac3, |
| 756 | int ecpl, int start_subband, int end_subband, |
| 757 | const uint8_t *default_band_struct, |
| 758 | int *num_bands, uint8_t *band_sizes) |
| 759 | { |
| 760 | int subbnd, bnd, n_subbands, n_bands=0; |
| 761 | uint8_t bnd_sz[22]; |
| 762 | uint8_t coded_band_struct[22]; |
| 763 | const uint8_t *band_struct; |
| 764 | |
| 765 | n_subbands = end_subband - start_subband; |
| 766 | |
| 767 | /* decode band structure from bitstream or use default */ |
| 768 | if (!eac3 || get_bits1(gbc)) { |
| 769 | for (subbnd = 0; subbnd < n_subbands - 1; subbnd++) { |
| 770 | coded_band_struct[subbnd] = get_bits1(gbc); |
| 771 | } |
| 772 | band_struct = coded_band_struct; |
| 773 | } else if (!blk) { |
| 774 | band_struct = &default_band_struct[start_subband+1]; |
| 775 | } else { |
| 776 | /* no change in band structure */ |
| 777 | return; |
| 778 | } |
| 779 | |
| 780 | /* calculate number of bands and band sizes based on band structure. |
| 781 | note that the first 4 subbands in enhanced coupling span only 6 bins |
| 782 | instead of 12. */ |
| 783 | if (num_bands || band_sizes ) { |
| 784 | n_bands = n_subbands; |
| 785 | bnd_sz[0] = ecpl ? 6 : 12; |
| 786 | for (bnd = 0, subbnd = 1; subbnd < n_subbands; subbnd++) { |
| 787 | int subbnd_size = (ecpl && subbnd < 4) ? 6 : 12; |
| 788 | if (band_struct[subbnd - 1]) { |
| 789 | n_bands--; |
| 790 | bnd_sz[bnd] += subbnd_size; |
| 791 | } else { |
| 792 | bnd_sz[++bnd] = subbnd_size; |
| 793 | } |
| 794 | } |
| 795 | } |
| 796 | |
| 797 | /* set optional output params */ |
| 798 | if (num_bands) |
| 799 | *num_bands = n_bands; |
| 800 | if (band_sizes) |
| 801 | memcpy(band_sizes, bnd_sz, n_bands); |
| 802 | } |
| 803 | |
| 804 | /** |
| 805 | * Decode a single audio block from the AC-3 bitstream. |
| 806 | */ |
| 807 | static int decode_audio_block(AC3DecodeContext *s, int blk) |
| 808 | { |
| 809 | int fbw_channels = s->fbw_channels; |
| 810 | int channel_mode = s->channel_mode; |
| 811 | int i, bnd, seg, ch; |
| 812 | int different_transforms; |
| 813 | int downmix_output; |
| 814 | int cpl_in_use; |
| 815 | GetBitContext *gbc = &s->gbc; |
| 816 | uint8_t bit_alloc_stages[AC3_MAX_CHANNELS] = { 0 }; |
| 817 | |
| 818 | /* block switch flags */ |
| 819 | different_transforms = 0; |
| 820 | if (s->block_switch_syntax) { |
| 821 | for (ch = 1; ch <= fbw_channels; ch++) { |
| 822 | s->block_switch[ch] = get_bits1(gbc); |
| 823 | if (ch > 1 && s->block_switch[ch] != s->block_switch[1]) |
| 824 | different_transforms = 1; |
| 825 | } |
| 826 | } |
| 827 | |
| 828 | /* dithering flags */ |
| 829 | if (s->dither_flag_syntax) { |
| 830 | for (ch = 1; ch <= fbw_channels; ch++) { |
| 831 | s->dither_flag[ch] = get_bits1(gbc); |
| 832 | } |
| 833 | } |
| 834 | |
| 835 | /* dynamic range */ |
| 836 | i = !s->channel_mode; |
| 837 | do { |
| 838 | if (get_bits1(gbc)) { |
| 839 | /* Allow asymmetric application of DRC when drc_scale > 1. |
| 840 | Amplification of quiet sounds is enhanced */ |
| 841 | int range_bits = get_bits(gbc, 8); |
| 842 | INTFLOAT range = AC3_RANGE(range_bits); |
| 843 | if (range_bits <= 127 || s->drc_scale <= 1.0) |
| 844 | s->dynamic_range[i] = AC3_DYNAMIC_RANGE(range); |
| 845 | else |
| 846 | s->dynamic_range[i] = range; |
| 847 | } else if (blk == 0) { |
| 848 | s->dynamic_range[i] = AC3_DYNAMIC_RANGE1; |
| 849 | } |
| 850 | } while (i--); |
| 851 | |
| 852 | /* spectral extension strategy */ |
| 853 | if (s->eac3 && (!blk || get_bits1(gbc))) { |
| 854 | s->spx_in_use = get_bits1(gbc); |
| 855 | if (s->spx_in_use) { |
| 856 | int dst_start_freq, dst_end_freq, src_start_freq, |
| 857 | start_subband, end_subband; |
| 858 | |
| 859 | /* determine which channels use spx */ |
| 860 | if (s->channel_mode == AC3_CHMODE_MONO) { |
| 861 | s->channel_uses_spx[1] = 1; |
| 862 | } else { |
| 863 | for (ch = 1; ch <= fbw_channels; ch++) |
| 864 | s->channel_uses_spx[ch] = get_bits1(gbc); |
| 865 | } |
| 866 | |
| 867 | /* get the frequency bins of the spx copy region and the spx start |
| 868 | and end subbands */ |
| 869 | dst_start_freq = get_bits(gbc, 2); |
| 870 | start_subband = get_bits(gbc, 3) + 2; |
| 871 | if (start_subband > 7) |
| 872 | start_subband += start_subband - 7; |
| 873 | end_subband = get_bits(gbc, 3) + 5; |
| 874 | #if USE_FIXED |
| 875 | s->spx_dst_end_freq = end_freq_inv_tab[end_subband]; |
| 876 | #endif |
| 877 | if (end_subband > 7) |
| 878 | end_subband += end_subband - 7; |
| 879 | dst_start_freq = dst_start_freq * 12 + 25; |
| 880 | src_start_freq = start_subband * 12 + 25; |
| 881 | dst_end_freq = end_subband * 12 + 25; |
| 882 | |
| 883 | /* check validity of spx ranges */ |
| 884 | if (start_subband >= end_subband) { |
| 885 | av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension " |
| 886 | "range (%d >= %d)\n", start_subband, end_subband); |
| 887 | return AVERROR_INVALIDDATA; |
| 888 | } |
| 889 | if (dst_start_freq >= src_start_freq) { |
| 890 | av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension " |
| 891 | "copy start bin (%d >= %d)\n", dst_start_freq, src_start_freq); |
| 892 | return AVERROR_INVALIDDATA; |
| 893 | } |
| 894 | |
| 895 | s->spx_dst_start_freq = dst_start_freq; |
| 896 | s->spx_src_start_freq = src_start_freq; |
| 897 | if (!USE_FIXED) |
| 898 | s->spx_dst_end_freq = dst_end_freq; |
| 899 | |
| 900 | decode_band_structure(gbc, blk, s->eac3, 0, |
| 901 | start_subband, end_subband, |
| 902 | ff_eac3_default_spx_band_struct, |
| 903 | &s->num_spx_bands, |
| 904 | s->spx_band_sizes); |
| 905 | } else { |
| 906 | for (ch = 1; ch <= fbw_channels; ch++) { |
| 907 | s->channel_uses_spx[ch] = 0; |
| 908 | s->first_spx_coords[ch] = 1; |
| 909 | } |
| 910 | } |
| 911 | } |
| 912 | |
| 913 | /* spectral extension coordinates */ |
| 914 | if (s->spx_in_use) { |
| 915 | for (ch = 1; ch <= fbw_channels; ch++) { |
| 916 | if (s->channel_uses_spx[ch]) { |
| 917 | if (s->first_spx_coords[ch] || get_bits1(gbc)) { |
| 918 | INTFLOAT spx_blend; |
| 919 | int bin, master_spx_coord; |
| 920 | |
| 921 | s->first_spx_coords[ch] = 0; |
| 922 | spx_blend = AC3_SPX_BLEND(get_bits(gbc, 5)); |
| 923 | master_spx_coord = get_bits(gbc, 2) * 3; |
| 924 | |
| 925 | bin = s->spx_src_start_freq; |
| 926 | for (bnd = 0; bnd < s->num_spx_bands; bnd++) { |
| 927 | int bandsize; |
| 928 | int spx_coord_exp, spx_coord_mant; |
| 929 | INTFLOAT nratio, sblend, nblend; |
| 930 | #if USE_FIXED |
| 931 | int64_t accu; |
| 932 | /* calculate blending factors */ |
| 933 | bandsize = s->spx_band_sizes[bnd]; |
| 934 | accu = (int64_t)((bin << 23) + (bandsize << 22)) * s->spx_dst_end_freq; |
| 935 | nratio = (int)(accu >> 32); |
| 936 | nratio -= spx_blend << 18; |
| 937 | |
| 938 | if (nratio < 0) { |
| 939 | nblend = 0; |
| 940 | sblend = 0x800000; |
| 941 | } else if (nratio > 0x7fffff) { |
| 942 | nblend = 0x800000; |
| 943 | sblend = 0; |
| 944 | } else { |
| 945 | nblend = fixed_sqrt(nratio, 23); |
| 946 | accu = (int64_t)nblend * 1859775393; |
| 947 | nblend = (int)((accu + (1<<29)) >> 30); |
| 948 | sblend = fixed_sqrt(0x800000 - nratio, 23); |
| 949 | } |
| 950 | #else |
| 951 | float spx_coord; |
| 952 | |
| 953 | /* calculate blending factors */ |
| 954 | bandsize = s->spx_band_sizes[bnd]; |
| 955 | nratio = ((float)((bin + (bandsize >> 1))) / s->spx_dst_end_freq) - spx_blend; |
| 956 | nratio = av_clipf(nratio, 0.0f, 1.0f); |
| 957 | nblend = sqrtf(3.0f * nratio); // noise is scaled by sqrt(3) |
| 958 | // to give unity variance |
| 959 | sblend = sqrtf(1.0f - nratio); |
| 960 | #endif |
| 961 | bin += bandsize; |
| 962 | |
| 963 | /* decode spx coordinates */ |
| 964 | spx_coord_exp = get_bits(gbc, 4); |
| 965 | spx_coord_mant = get_bits(gbc, 2); |
| 966 | if (spx_coord_exp == 15) spx_coord_mant <<= 1; |
| 967 | else spx_coord_mant += 4; |
| 968 | spx_coord_mant <<= (25 - spx_coord_exp - master_spx_coord); |
| 969 | |
| 970 | /* multiply noise and signal blending factors by spx coordinate */ |
| 971 | #if USE_FIXED |
| 972 | accu = (int64_t)nblend * spx_coord_mant; |
| 973 | s->spx_noise_blend[ch][bnd] = (int)((accu + (1<<22)) >> 23); |
| 974 | accu = (int64_t)sblend * spx_coord_mant; |
| 975 | s->spx_signal_blend[ch][bnd] = (int)((accu + (1<<22)) >> 23); |
| 976 | #else |
| 977 | spx_coord = spx_coord_mant * (1.0f / (1 << 23)); |
| 978 | s->spx_noise_blend [ch][bnd] = nblend * spx_coord; |
| 979 | s->spx_signal_blend[ch][bnd] = sblend * spx_coord; |
| 980 | #endif |
| 981 | } |
| 982 | } |
| 983 | } else { |
| 984 | s->first_spx_coords[ch] = 1; |
| 985 | } |
| 986 | } |
| 987 | } |
| 988 | |
| 989 | /* coupling strategy */ |
| 990 | if (s->eac3 ? s->cpl_strategy_exists[blk] : get_bits1(gbc)) { |
| 991 | memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS); |
| 992 | if (!s->eac3) |
| 993 | s->cpl_in_use[blk] = get_bits1(gbc); |
| 994 | if (s->cpl_in_use[blk]) { |
| 995 | /* coupling in use */ |
| 996 | int cpl_start_subband, cpl_end_subband; |
| 997 | |
| 998 | if (channel_mode < AC3_CHMODE_STEREO) { |
| 999 | av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n"); |
| 1000 | return AVERROR_INVALIDDATA; |
| 1001 | } |
| 1002 | |
| 1003 | /* check for enhanced coupling */ |
| 1004 | if (s->eac3 && get_bits1(gbc)) { |
| 1005 | /* TODO: parse enhanced coupling strategy info */ |
| 1006 | avpriv_request_sample(s->avctx, "Enhanced coupling"); |
| 1007 | return AVERROR_PATCHWELCOME; |
| 1008 | } |
| 1009 | |
| 1010 | /* determine which channels are coupled */ |
| 1011 | if (s->eac3 && s->channel_mode == AC3_CHMODE_STEREO) { |
| 1012 | s->channel_in_cpl[1] = 1; |
| 1013 | s->channel_in_cpl[2] = 1; |
| 1014 | } else { |
| 1015 | for (ch = 1; ch <= fbw_channels; ch++) |
| 1016 | s->channel_in_cpl[ch] = get_bits1(gbc); |
| 1017 | } |
| 1018 | |
| 1019 | /* phase flags in use */ |
| 1020 | if (channel_mode == AC3_CHMODE_STEREO) |
| 1021 | s->phase_flags_in_use = get_bits1(gbc); |
| 1022 | |
| 1023 | /* coupling frequency range */ |
| 1024 | cpl_start_subband = get_bits(gbc, 4); |
| 1025 | cpl_end_subband = s->spx_in_use ? (s->spx_src_start_freq - 37) / 12 : |
| 1026 | get_bits(gbc, 4) + 3; |
| 1027 | if (cpl_start_subband >= cpl_end_subband) { |
| 1028 | av_log(s->avctx, AV_LOG_ERROR, "invalid coupling range (%d >= %d)\n", |
| 1029 | cpl_start_subband, cpl_end_subband); |
| 1030 | return AVERROR_INVALIDDATA; |
| 1031 | } |
| 1032 | s->start_freq[CPL_CH] = cpl_start_subband * 12 + 37; |
| 1033 | s->end_freq[CPL_CH] = cpl_end_subband * 12 + 37; |
| 1034 | |
| 1035 | decode_band_structure(gbc, blk, s->eac3, 0, cpl_start_subband, |
| 1036 | cpl_end_subband, |
| 1037 | ff_eac3_default_cpl_band_struct, |
| 1038 | &s->num_cpl_bands, s->cpl_band_sizes); |
| 1039 | } else { |
| 1040 | /* coupling not in use */ |
| 1041 | for (ch = 1; ch <= fbw_channels; ch++) { |
| 1042 | s->channel_in_cpl[ch] = 0; |
| 1043 | s->first_cpl_coords[ch] = 1; |
| 1044 | } |
| 1045 | s->first_cpl_leak = s->eac3; |
| 1046 | s->phase_flags_in_use = 0; |
| 1047 | } |
| 1048 | } else if (!s->eac3) { |
| 1049 | if (!blk) { |
| 1050 | av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must " |
| 1051 | "be present in block 0\n"); |
| 1052 | return AVERROR_INVALIDDATA; |
| 1053 | } else { |
| 1054 | s->cpl_in_use[blk] = s->cpl_in_use[blk-1]; |
| 1055 | } |
| 1056 | } |
| 1057 | cpl_in_use = s->cpl_in_use[blk]; |
| 1058 | |
| 1059 | /* coupling coordinates */ |
| 1060 | if (cpl_in_use) { |
| 1061 | int cpl_coords_exist = 0; |
| 1062 | |
| 1063 | for (ch = 1; ch <= fbw_channels; ch++) { |
| 1064 | if (s->channel_in_cpl[ch]) { |
| 1065 | if ((s->eac3 && s->first_cpl_coords[ch]) || get_bits1(gbc)) { |
| 1066 | int master_cpl_coord, cpl_coord_exp, cpl_coord_mant; |
| 1067 | s->first_cpl_coords[ch] = 0; |
| 1068 | cpl_coords_exist = 1; |
| 1069 | master_cpl_coord = 3 * get_bits(gbc, 2); |
| 1070 | for (bnd = 0; bnd < s->num_cpl_bands; bnd++) { |
| 1071 | cpl_coord_exp = get_bits(gbc, 4); |
| 1072 | cpl_coord_mant = get_bits(gbc, 4); |
| 1073 | if (cpl_coord_exp == 15) |
| 1074 | s->cpl_coords[ch][bnd] = cpl_coord_mant << 22; |
| 1075 | else |
| 1076 | s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21; |
| 1077 | s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord); |
| 1078 | } |
| 1079 | } else if (!blk) { |
| 1080 | av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must " |
| 1081 | "be present in block 0\n"); |
| 1082 | return AVERROR_INVALIDDATA; |
| 1083 | } |
| 1084 | } else { |
| 1085 | /* channel not in coupling */ |
| 1086 | s->first_cpl_coords[ch] = 1; |
| 1087 | } |
| 1088 | } |
| 1089 | /* phase flags */ |
| 1090 | if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) { |
| 1091 | for (bnd = 0; bnd < s->num_cpl_bands; bnd++) { |
| 1092 | s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0; |
| 1093 | } |
| 1094 | } |
| 1095 | } |
| 1096 | |
| 1097 | /* stereo rematrixing strategy and band structure */ |
| 1098 | if (channel_mode == AC3_CHMODE_STEREO) { |
| 1099 | if ((s->eac3 && !blk) || get_bits1(gbc)) { |
| 1100 | s->num_rematrixing_bands = 4; |
| 1101 | if (cpl_in_use && s->start_freq[CPL_CH] <= 61) { |
| 1102 | s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37); |
| 1103 | } else if (s->spx_in_use && s->spx_src_start_freq <= 61) { |
| 1104 | s->num_rematrixing_bands--; |
| 1105 | } |
| 1106 | for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++) |
| 1107 | s->rematrixing_flags[bnd] = get_bits1(gbc); |
| 1108 | } else if (!blk) { |
| 1109 | av_log(s->avctx, AV_LOG_WARNING, "Warning: " |
| 1110 | "new rematrixing strategy not present in block 0\n"); |
| 1111 | s->num_rematrixing_bands = 0; |
| 1112 | } |
| 1113 | } |
| 1114 | |
| 1115 | /* exponent strategies for each channel */ |
| 1116 | for (ch = !cpl_in_use; ch <= s->channels; ch++) { |
| 1117 | if (!s->eac3) |
| 1118 | s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch)); |
| 1119 | if (s->exp_strategy[blk][ch] != EXP_REUSE) |
| 1120 | bit_alloc_stages[ch] = 3; |
| 1121 | } |
| 1122 | |
| 1123 | /* channel bandwidth */ |
| 1124 | for (ch = 1; ch <= fbw_channels; ch++) { |
| 1125 | s->start_freq[ch] = 0; |
| 1126 | if (s->exp_strategy[blk][ch] != EXP_REUSE) { |
| 1127 | int group_size; |
| 1128 | int prev = s->end_freq[ch]; |
| 1129 | if (s->channel_in_cpl[ch]) |
| 1130 | s->end_freq[ch] = s->start_freq[CPL_CH]; |
| 1131 | else if (s->channel_uses_spx[ch]) |
| 1132 | s->end_freq[ch] = s->spx_src_start_freq; |
| 1133 | else { |
| 1134 | int bandwidth_code = get_bits(gbc, 6); |
| 1135 | if (bandwidth_code > 60) { |
| 1136 | av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60\n", bandwidth_code); |
| 1137 | return AVERROR_INVALIDDATA; |
| 1138 | } |
| 1139 | s->end_freq[ch] = bandwidth_code * 3 + 73; |
| 1140 | } |
| 1141 | group_size = 3 << (s->exp_strategy[blk][ch] - 1); |
| 1142 | s->num_exp_groups[ch] = (s->end_freq[ch] + group_size-4) / group_size; |
| 1143 | if (blk > 0 && s->end_freq[ch] != prev) |
| 1144 | memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS); |
| 1145 | } |
| 1146 | } |
| 1147 | if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) { |
| 1148 | s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) / |
| 1149 | (3 << (s->exp_strategy[blk][CPL_CH] - 1)); |
| 1150 | } |
| 1151 | |
| 1152 | /* decode exponents for each channel */ |
| 1153 | for (ch = !cpl_in_use; ch <= s->channels; ch++) { |
| 1154 | if (s->exp_strategy[blk][ch] != EXP_REUSE) { |
| 1155 | s->dexps[ch][0] = get_bits(gbc, 4) << !ch; |
| 1156 | if (decode_exponents(gbc, s->exp_strategy[blk][ch], |
| 1157 | s->num_exp_groups[ch], s->dexps[ch][0], |
| 1158 | &s->dexps[ch][s->start_freq[ch]+!!ch])) { |
| 1159 | av_log(s->avctx, AV_LOG_ERROR, "exponent out-of-range\n"); |
| 1160 | return AVERROR_INVALIDDATA; |
| 1161 | } |
| 1162 | if (ch != CPL_CH && ch != s->lfe_ch) |
| 1163 | skip_bits(gbc, 2); /* skip gainrng */ |
| 1164 | } |
| 1165 | } |
| 1166 | |
| 1167 | /* bit allocation information */ |
| 1168 | if (s->bit_allocation_syntax) { |
| 1169 | if (get_bits1(gbc)) { |
| 1170 | s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift; |
| 1171 | s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift; |
| 1172 | s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)]; |
| 1173 | s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)]; |
| 1174 | s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)]; |
| 1175 | for (ch = !cpl_in_use; ch <= s->channels; ch++) |
| 1176 | bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2); |
| 1177 | } else if (!blk) { |
| 1178 | av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must " |
| 1179 | "be present in block 0\n"); |
| 1180 | return AVERROR_INVALIDDATA; |
| 1181 | } |
| 1182 | } |
| 1183 | |
| 1184 | /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */ |
| 1185 | if (!s->eac3 || !blk) { |
| 1186 | if (s->snr_offset_strategy && get_bits1(gbc)) { |
| 1187 | int snr = 0; |
| 1188 | int csnr; |
| 1189 | csnr = (get_bits(gbc, 6) - 15) << 4; |
| 1190 | for (i = ch = !cpl_in_use; ch <= s->channels; ch++) { |
| 1191 | /* snr offset */ |
| 1192 | if (ch == i || s->snr_offset_strategy == 2) |
| 1193 | snr = (csnr + get_bits(gbc, 4)) << 2; |
| 1194 | /* run at least last bit allocation stage if snr offset changes */ |
| 1195 | if (blk && s->snr_offset[ch] != snr) { |
| 1196 | bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 1); |
| 1197 | } |
| 1198 | s->snr_offset[ch] = snr; |
| 1199 | |
| 1200 | /* fast gain (normal AC-3 only) */ |
| 1201 | if (!s->eac3) { |
| 1202 | int prev = s->fast_gain[ch]; |
| 1203 | s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)]; |
| 1204 | /* run last 2 bit allocation stages if fast gain changes */ |
| 1205 | if (blk && prev != s->fast_gain[ch]) |
| 1206 | bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2); |
| 1207 | } |
| 1208 | } |
| 1209 | } else if (!s->eac3 && !blk) { |
| 1210 | av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n"); |
| 1211 | return AVERROR_INVALIDDATA; |
| 1212 | } |
| 1213 | } |
| 1214 | |
| 1215 | /* fast gain (E-AC-3 only) */ |
| 1216 | if (s->fast_gain_syntax && get_bits1(gbc)) { |
| 1217 | for (ch = !cpl_in_use; ch <= s->channels; ch++) { |
| 1218 | int prev = s->fast_gain[ch]; |
| 1219 | s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)]; |
| 1220 | /* run last 2 bit allocation stages if fast gain changes */ |
| 1221 | if (blk && prev != s->fast_gain[ch]) |
| 1222 | bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2); |
| 1223 | } |
| 1224 | } else if (s->eac3 && !blk) { |
| 1225 | for (ch = !cpl_in_use; ch <= s->channels; ch++) |
| 1226 | s->fast_gain[ch] = ff_ac3_fast_gain_tab[4]; |
| 1227 | } |
| 1228 | |
| 1229 | /* E-AC-3 to AC-3 converter SNR offset */ |
| 1230 | if (s->frame_type == EAC3_FRAME_TYPE_INDEPENDENT && get_bits1(gbc)) { |
| 1231 | skip_bits(gbc, 10); // skip converter snr offset |
| 1232 | } |
| 1233 | |
| 1234 | /* coupling leak information */ |
| 1235 | if (cpl_in_use) { |
| 1236 | if (s->first_cpl_leak || get_bits1(gbc)) { |
| 1237 | int fl = get_bits(gbc, 3); |
| 1238 | int sl = get_bits(gbc, 3); |
| 1239 | /* run last 2 bit allocation stages for coupling channel if |
| 1240 | coupling leak changes */ |
| 1241 | if (blk && (fl != s->bit_alloc_params.cpl_fast_leak || |
| 1242 | sl != s->bit_alloc_params.cpl_slow_leak)) { |
| 1243 | bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2); |
| 1244 | } |
| 1245 | s->bit_alloc_params.cpl_fast_leak = fl; |
| 1246 | s->bit_alloc_params.cpl_slow_leak = sl; |
| 1247 | } else if (!s->eac3 && !blk) { |
| 1248 | av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must " |
| 1249 | "be present in block 0\n"); |
| 1250 | return AVERROR_INVALIDDATA; |
| 1251 | } |
| 1252 | s->first_cpl_leak = 0; |
| 1253 | } |
| 1254 | |
| 1255 | /* delta bit allocation information */ |
| 1256 | if (s->dba_syntax && get_bits1(gbc)) { |
| 1257 | /* delta bit allocation exists (strategy) */ |
| 1258 | for (ch = !cpl_in_use; ch <= fbw_channels; ch++) { |
| 1259 | s->dba_mode[ch] = get_bits(gbc, 2); |
| 1260 | if (s->dba_mode[ch] == DBA_RESERVED) { |
| 1261 | av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n"); |
| 1262 | return AVERROR_INVALIDDATA; |
| 1263 | } |
| 1264 | bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2); |
| 1265 | } |
| 1266 | /* channel delta offset, len and bit allocation */ |
| 1267 | for (ch = !cpl_in_use; ch <= fbw_channels; ch++) { |
| 1268 | if (s->dba_mode[ch] == DBA_NEW) { |
| 1269 | s->dba_nsegs[ch] = get_bits(gbc, 3) + 1; |
| 1270 | for (seg = 0; seg < s->dba_nsegs[ch]; seg++) { |
| 1271 | s->dba_offsets[ch][seg] = get_bits(gbc, 5); |
| 1272 | s->dba_lengths[ch][seg] = get_bits(gbc, 4); |
| 1273 | s->dba_values[ch][seg] = get_bits(gbc, 3); |
| 1274 | } |
| 1275 | /* run last 2 bit allocation stages if new dba values */ |
| 1276 | bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2); |
| 1277 | } |
| 1278 | } |
| 1279 | } else if (blk == 0) { |
| 1280 | for (ch = 0; ch <= s->channels; ch++) { |
| 1281 | s->dba_mode[ch] = DBA_NONE; |
| 1282 | } |
| 1283 | } |
| 1284 | |
| 1285 | /* Bit allocation */ |
| 1286 | for (ch = !cpl_in_use; ch <= s->channels; ch++) { |
| 1287 | if (bit_alloc_stages[ch] > 2) { |
| 1288 | /* Exponent mapping into PSD and PSD integration */ |
| 1289 | ff_ac3_bit_alloc_calc_psd(s->dexps[ch], |
| 1290 | s->start_freq[ch], s->end_freq[ch], |
| 1291 | s->psd[ch], s->band_psd[ch]); |
| 1292 | } |
| 1293 | if (bit_alloc_stages[ch] > 1) { |
| 1294 | /* Compute excitation function, Compute masking curve, and |
| 1295 | Apply delta bit allocation */ |
| 1296 | if (ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch], |
| 1297 | s->start_freq[ch], s->end_freq[ch], |
| 1298 | s->fast_gain[ch], (ch == s->lfe_ch), |
| 1299 | s->dba_mode[ch], s->dba_nsegs[ch], |
| 1300 | s->dba_offsets[ch], s->dba_lengths[ch], |
| 1301 | s->dba_values[ch], s->mask[ch])) { |
| 1302 | av_log(s->avctx, AV_LOG_ERROR, "error in bit allocation\n"); |
| 1303 | return AVERROR_INVALIDDATA; |
| 1304 | } |
| 1305 | } |
| 1306 | if (bit_alloc_stages[ch] > 0) { |
| 1307 | /* Compute bit allocation */ |
| 1308 | const uint8_t *bap_tab = s->channel_uses_aht[ch] ? |
| 1309 | ff_eac3_hebap_tab : ff_ac3_bap_tab; |
| 1310 | s->ac3dsp.bit_alloc_calc_bap(s->mask[ch], s->psd[ch], |
| 1311 | s->start_freq[ch], s->end_freq[ch], |
| 1312 | s->snr_offset[ch], |
| 1313 | s->bit_alloc_params.floor, |
| 1314 | bap_tab, s->bap[ch]); |
| 1315 | } |
| 1316 | } |
| 1317 | |
| 1318 | /* unused dummy data */ |
| 1319 | if (s->skip_syntax && get_bits1(gbc)) { |
| 1320 | int skipl = get_bits(gbc, 9); |
| 1321 | while (skipl--) |
| 1322 | skip_bits(gbc, 8); |
| 1323 | } |
| 1324 | |
| 1325 | /* unpack the transform coefficients |
| 1326 | this also uncouples channels if coupling is in use. */ |
| 1327 | decode_transform_coeffs(s, blk); |
| 1328 | |
| 1329 | /* TODO: generate enhanced coupling coordinates and uncouple */ |
| 1330 | |
| 1331 | /* recover coefficients if rematrixing is in use */ |
| 1332 | if (s->channel_mode == AC3_CHMODE_STEREO) |
| 1333 | do_rematrixing(s); |
| 1334 | |
| 1335 | /* apply scaling to coefficients (headroom, dynrng) */ |
| 1336 | for (ch = 1; ch <= s->channels; ch++) { |
| 1337 | int audio_channel = 0; |
| 1338 | INTFLOAT gain; |
| 1339 | if (s->channel_mode == AC3_CHMODE_DUALMONO) |
| 1340 | audio_channel = 2-ch; |
| 1341 | if (s->heavy_compression && s->compression_exists[audio_channel]) |
| 1342 | gain = s->heavy_dynamic_range[audio_channel]; |
| 1343 | else |
| 1344 | gain = s->dynamic_range[audio_channel]; |
| 1345 | |
| 1346 | #if USE_FIXED |
| 1347 | scale_coefs(s->transform_coeffs[ch], s->fixed_coeffs[ch], gain, 256); |
| 1348 | #else |
| 1349 | if (s->target_level != 0) |
| 1350 | gain = gain * s->level_gain[audio_channel]; |
| 1351 | gain *= 1.0 / 4194304.0f; |
| 1352 | s->fmt_conv.int32_to_float_fmul_scalar(s->transform_coeffs[ch], |
| 1353 | s->fixed_coeffs[ch], gain, 256); |
| 1354 | #endif |
| 1355 | } |
| 1356 | |
| 1357 | /* apply spectral extension to high frequency bins */ |
| 1358 | if (s->spx_in_use && CONFIG_EAC3_DECODER) { |
| 1359 | ff_eac3_apply_spectral_extension(s); |
| 1360 | } |
| 1361 | |
| 1362 | /* downmix and MDCT. order depends on whether block switching is used for |
| 1363 | any channel in this block. this is because coefficients for the long |
| 1364 | and short transforms cannot be mixed. */ |
| 1365 | downmix_output = s->channels != s->out_channels && |
| 1366 | !((s->output_mode & AC3_OUTPUT_LFEON) && |
| 1367 | s->fbw_channels == s->out_channels); |
| 1368 | if (different_transforms) { |
| 1369 | /* the delay samples have already been downmixed, so we upmix the delay |
| 1370 | samples in order to reconstruct all channels before downmixing. */ |
| 1371 | if (s->downmixed) { |
| 1372 | s->downmixed = 0; |
| 1373 | ac3_upmix_delay(s); |
| 1374 | } |
| 1375 | |
| 1376 | do_imdct(s, s->channels); |
| 1377 | |
| 1378 | if (downmix_output) { |
| 1379 | #if USE_FIXED |
| 1380 | ac3_downmix_c_fixed16(s->outptr, s->downmix_coeffs, |
| 1381 | s->out_channels, s->fbw_channels, 256); |
| 1382 | #else |
| 1383 | s->ac3dsp.downmix(s->outptr, s->downmix_coeffs, |
| 1384 | s->out_channels, s->fbw_channels, 256); |
| 1385 | #endif |
| 1386 | } |
| 1387 | } else { |
| 1388 | if (downmix_output) { |
| 1389 | s->ac3dsp.AC3_RENAME(downmix)(s->xcfptr + 1, s->downmix_coeffs, |
| 1390 | s->out_channels, s->fbw_channels, 256); |
| 1391 | } |
| 1392 | |
| 1393 | if (downmix_output && !s->downmixed) { |
| 1394 | s->downmixed = 1; |
| 1395 | s->ac3dsp.AC3_RENAME(downmix)(s->dlyptr, s->downmix_coeffs, |
| 1396 | s->out_channels, s->fbw_channels, 128); |
| 1397 | } |
| 1398 | |
| 1399 | do_imdct(s, s->out_channels); |
| 1400 | } |
| 1401 | |
| 1402 | return 0; |
| 1403 | } |
| 1404 | |
| 1405 | /** |
| 1406 | * Decode a single AC-3 frame. |
| 1407 | */ |
| 1408 | static int ac3_decode_frame(AVCodecContext * avctx, void *data, |
| 1409 | int *got_frame_ptr, AVPacket *avpkt) |
| 1410 | { |
| 1411 | AVFrame *frame = data; |
| 1412 | const uint8_t *buf = avpkt->data; |
| 1413 | int buf_size = avpkt->size; |
| 1414 | AC3DecodeContext *s = avctx->priv_data; |
| 1415 | int blk, ch, err, ret; |
| 1416 | const uint8_t *channel_map; |
| 1417 | const SHORTFLOAT *output[AC3_MAX_CHANNELS]; |
| 1418 | enum AVMatrixEncoding matrix_encoding; |
| 1419 | AVDownmixInfo *downmix_info; |
| 1420 | |
| 1421 | /* copy input buffer to decoder context to avoid reading past the end |
| 1422 | of the buffer, which can be caused by a damaged input stream. */ |
| 1423 | if (buf_size >= 2 && AV_RB16(buf) == 0x770B) { |
| 1424 | // seems to be byte-swapped AC-3 |
| 1425 | int cnt = FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE) >> 1; |
| 1426 | s->bdsp.bswap16_buf((uint16_t *) s->input_buffer, |
| 1427 | (const uint16_t *) buf, cnt); |
| 1428 | } else |
| 1429 | memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE)); |
| 1430 | buf = s->input_buffer; |
| 1431 | /* initialize the GetBitContext with the start of valid AC-3 Frame */ |
| 1432 | init_get_bits(&s->gbc, buf, buf_size * 8); |
| 1433 | |
| 1434 | /* parse the syncinfo */ |
| 1435 | err = parse_frame_header(s); |
| 1436 | |
| 1437 | if (err) { |
| 1438 | switch (err) { |
| 1439 | case AAC_AC3_PARSE_ERROR_SYNC: |
| 1440 | av_log(avctx, AV_LOG_ERROR, "frame sync error\n"); |
| 1441 | return AVERROR_INVALIDDATA; |
| 1442 | case AAC_AC3_PARSE_ERROR_BSID: |
| 1443 | av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n"); |
| 1444 | break; |
| 1445 | case AAC_AC3_PARSE_ERROR_SAMPLE_RATE: |
| 1446 | av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n"); |
| 1447 | break; |
| 1448 | case AAC_AC3_PARSE_ERROR_FRAME_SIZE: |
| 1449 | av_log(avctx, AV_LOG_ERROR, "invalid frame size\n"); |
| 1450 | break; |
| 1451 | case AAC_AC3_PARSE_ERROR_FRAME_TYPE: |
| 1452 | /* skip frame if CRC is ok. otherwise use error concealment. */ |
| 1453 | /* TODO: add support for substreams and dependent frames */ |
| 1454 | if (s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) { |
| 1455 | av_log(avctx, AV_LOG_WARNING, "unsupported frame type : " |
| 1456 | "skipping frame\n"); |
| 1457 | *got_frame_ptr = 0; |
| 1458 | return buf_size; |
| 1459 | } else { |
| 1460 | av_log(avctx, AV_LOG_ERROR, "invalid frame type\n"); |
| 1461 | } |
| 1462 | break; |
| 1463 | case AAC_AC3_PARSE_ERROR_CRC: |
| 1464 | case AAC_AC3_PARSE_ERROR_CHANNEL_CFG: |
| 1465 | break; |
| 1466 | default: // Normal AVERROR do not try to recover. |
| 1467 | *got_frame_ptr = 0; |
| 1468 | return err; |
| 1469 | } |
| 1470 | } else { |
| 1471 | /* check that reported frame size fits in input buffer */ |
| 1472 | if (s->frame_size > buf_size) { |
| 1473 | av_log(avctx, AV_LOG_ERROR, "incomplete frame\n"); |
| 1474 | err = AAC_AC3_PARSE_ERROR_FRAME_SIZE; |
| 1475 | } else if (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL)) { |
| 1476 | /* check for crc mismatch */ |
| 1477 | if (av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], |
| 1478 | s->frame_size - 2)) { |
| 1479 | av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n"); |
| 1480 | if (avctx->err_recognition & AV_EF_EXPLODE) |
| 1481 | return AVERROR_INVALIDDATA; |
| 1482 | err = AAC_AC3_PARSE_ERROR_CRC; |
| 1483 | } |
| 1484 | } |
| 1485 | } |
| 1486 | |
| 1487 | /* if frame is ok, set audio parameters */ |
| 1488 | if (!err) { |
| 1489 | avctx->sample_rate = s->sample_rate; |
| 1490 | avctx->bit_rate = s->bit_rate; |
| 1491 | } |
| 1492 | |
| 1493 | /* channel config */ |
| 1494 | if (!err || (s->channels && s->out_channels != s->channels)) { |
| 1495 | s->out_channels = s->channels; |
| 1496 | s->output_mode = s->channel_mode; |
| 1497 | if (s->lfe_on) |
| 1498 | s->output_mode |= AC3_OUTPUT_LFEON; |
| 1499 | if (s->channels > 1 && |
| 1500 | avctx->request_channel_layout == AV_CH_LAYOUT_MONO) { |
| 1501 | s->out_channels = 1; |
| 1502 | s->output_mode = AC3_CHMODE_MONO; |
| 1503 | } else if (s->channels > 2 && |
| 1504 | avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) { |
| 1505 | s->out_channels = 2; |
| 1506 | s->output_mode = AC3_CHMODE_STEREO; |
| 1507 | } |
| 1508 | |
| 1509 | s->loro_center_mix_level = gain_levels[s-> center_mix_level]; |
| 1510 | s->loro_surround_mix_level = gain_levels[s->surround_mix_level]; |
| 1511 | s->ltrt_center_mix_level = LEVEL_MINUS_3DB; |
| 1512 | s->ltrt_surround_mix_level = LEVEL_MINUS_3DB; |
| 1513 | /* set downmixing coefficients if needed */ |
| 1514 | if (s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) && |
| 1515 | s->fbw_channels == s->out_channels)) { |
| 1516 | set_downmix_coeffs(s); |
| 1517 | } |
| 1518 | } else if (!s->channels) { |
| 1519 | av_log(avctx, AV_LOG_ERROR, "unable to determine channel mode\n"); |
| 1520 | return AVERROR_INVALIDDATA; |
| 1521 | } |
| 1522 | avctx->channels = s->out_channels; |
| 1523 | avctx->channel_layout = avpriv_ac3_channel_layout_tab[s->output_mode & ~AC3_OUTPUT_LFEON]; |
| 1524 | if (s->output_mode & AC3_OUTPUT_LFEON) |
| 1525 | avctx->channel_layout |= AV_CH_LOW_FREQUENCY; |
| 1526 | |
| 1527 | /* set audio service type based on bitstream mode for AC-3 */ |
| 1528 | avctx->audio_service_type = s->bitstream_mode; |
| 1529 | if (s->bitstream_mode == 0x7 && s->channels > 1) |
| 1530 | avctx->audio_service_type = AV_AUDIO_SERVICE_TYPE_KARAOKE; |
| 1531 | |
| 1532 | /* get output buffer */ |
| 1533 | frame->nb_samples = s->num_blocks * AC3_BLOCK_SIZE; |
| 1534 | if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) |
| 1535 | return ret; |
| 1536 | |
| 1537 | /* decode the audio blocks */ |
| 1538 | channel_map = ff_ac3_dec_channel_map[s->output_mode & ~AC3_OUTPUT_LFEON][s->lfe_on]; |
| 1539 | for (ch = 0; ch < AC3_MAX_CHANNELS; ch++) { |
| 1540 | output[ch] = s->output[ch]; |
| 1541 | s->outptr[ch] = s->output[ch]; |
| 1542 | } |
| 1543 | for (ch = 0; ch < s->channels; ch++) { |
| 1544 | if (ch < s->out_channels) |
| 1545 | s->outptr[channel_map[ch]] = (SHORTFLOAT *)frame->data[ch]; |
| 1546 | } |
| 1547 | for (blk = 0; blk < s->num_blocks; blk++) { |
| 1548 | if (!err && decode_audio_block(s, blk)) { |
| 1549 | av_log(avctx, AV_LOG_ERROR, "error decoding the audio block\n"); |
| 1550 | err = 1; |
| 1551 | } |
| 1552 | if (err) |
| 1553 | for (ch = 0; ch < s->out_channels; ch++) |
| 1554 | memcpy(((SHORTFLOAT*)frame->data[ch]) + AC3_BLOCK_SIZE*blk, output[ch], AC3_BLOCK_SIZE*sizeof(SHORTFLOAT)); |
| 1555 | for (ch = 0; ch < s->out_channels; ch++) |
| 1556 | output[ch] = s->outptr[channel_map[ch]]; |
| 1557 | for (ch = 0; ch < s->out_channels; ch++) { |
| 1558 | if (!ch || channel_map[ch]) |
| 1559 | s->outptr[channel_map[ch]] += AC3_BLOCK_SIZE; |
| 1560 | } |
| 1561 | } |
| 1562 | |
| 1563 | av_frame_set_decode_error_flags(frame, err ? FF_DECODE_ERROR_INVALID_BITSTREAM : 0); |
| 1564 | |
| 1565 | /* keep last block for error concealment in next frame */ |
| 1566 | for (ch = 0; ch < s->out_channels; ch++) |
| 1567 | memcpy(s->output[ch], output[ch], AC3_BLOCK_SIZE*sizeof(SHORTFLOAT)); |
| 1568 | |
| 1569 | /* |
| 1570 | * AVMatrixEncoding |
| 1571 | * |
| 1572 | * Check whether the input layout is compatible, and make sure we're not |
| 1573 | * downmixing (else the matrix encoding is no longer applicable). |
| 1574 | */ |
| 1575 | matrix_encoding = AV_MATRIX_ENCODING_NONE; |
| 1576 | if (s->channel_mode == AC3_CHMODE_STEREO && |
| 1577 | s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) { |
| 1578 | if (s->dolby_surround_mode == AC3_DSURMOD_ON) |
| 1579 | matrix_encoding = AV_MATRIX_ENCODING_DOLBY; |
| 1580 | else if (s->dolby_headphone_mode == AC3_DHEADPHONMOD_ON) |
| 1581 | matrix_encoding = AV_MATRIX_ENCODING_DOLBYHEADPHONE; |
| 1582 | } else if (s->channel_mode >= AC3_CHMODE_2F2R && |
| 1583 | s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) { |
| 1584 | switch (s->dolby_surround_ex_mode) { |
| 1585 | case AC3_DSUREXMOD_ON: // EX or PLIIx |
| 1586 | matrix_encoding = AV_MATRIX_ENCODING_DOLBYEX; |
| 1587 | break; |
| 1588 | case AC3_DSUREXMOD_PLIIZ: |
| 1589 | matrix_encoding = AV_MATRIX_ENCODING_DPLIIZ; |
| 1590 | break; |
| 1591 | default: // not indicated or off |
| 1592 | break; |
| 1593 | } |
| 1594 | } |
| 1595 | if ((ret = ff_side_data_update_matrix_encoding(frame, matrix_encoding)) < 0) |
| 1596 | return ret; |
| 1597 | |
| 1598 | /* AVDownmixInfo */ |
| 1599 | if ((downmix_info = av_downmix_info_update_side_data(frame))) { |
| 1600 | switch (s->preferred_downmix) { |
| 1601 | case AC3_DMIXMOD_LTRT: |
| 1602 | downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LTRT; |
| 1603 | break; |
| 1604 | case AC3_DMIXMOD_LORO: |
| 1605 | downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LORO; |
| 1606 | break; |
| 1607 | case AC3_DMIXMOD_DPLII: |
| 1608 | downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_DPLII; |
| 1609 | break; |
| 1610 | default: |
| 1611 | downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_UNKNOWN; |
| 1612 | break; |
| 1613 | } |
| 1614 | downmix_info->center_mix_level = gain_levels[s-> center_mix_level]; |
| 1615 | downmix_info->center_mix_level_ltrt = gain_levels[s-> center_mix_level_ltrt]; |
| 1616 | downmix_info->surround_mix_level = gain_levels[s-> surround_mix_level]; |
| 1617 | downmix_info->surround_mix_level_ltrt = gain_levels[s->surround_mix_level_ltrt]; |
| 1618 | if (s->lfe_mix_level_exists) |
| 1619 | downmix_info->lfe_mix_level = gain_levels_lfe[s->lfe_mix_level]; |
| 1620 | else |
| 1621 | downmix_info->lfe_mix_level = 0.0; // -inf dB |
| 1622 | } else |
| 1623 | return AVERROR(ENOMEM); |
| 1624 | |
| 1625 | *got_frame_ptr = 1; |
| 1626 | |
| 1627 | return FFMIN(buf_size, s->frame_size); |
| 1628 | } |
| 1629 | |
| 1630 | /** |
| 1631 | * Uninitialize the AC-3 decoder. |
| 1632 | */ |
| 1633 | static av_cold int ac3_decode_end(AVCodecContext *avctx) |
| 1634 | { |
| 1635 | AC3DecodeContext *s = avctx->priv_data; |
| 1636 | ff_mdct_end(&s->imdct_512); |
| 1637 | ff_mdct_end(&s->imdct_256); |
| 1638 | #if (USE_FIXED) |
| 1639 | av_freep(&s->fdsp); |
| 1640 | #endif |
| 1641 | |
| 1642 | return 0; |
| 1643 | } |
| 1644 | |
| 1645 | #define OFFSET(x) offsetof(AC3DecodeContext, x) |
| 1646 | #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM) |