| 1 | /* |
| 2 | * AAC encoder |
| 3 | * Copyright (C) 2008 Konstantin Shishkov |
| 4 | * |
| 5 | * This file is part of FFmpeg. |
| 6 | * |
| 7 | * FFmpeg is free software; you can redistribute it and/or |
| 8 | * modify it under the terms of the GNU Lesser General Public |
| 9 | * License as published by the Free Software Foundation; either |
| 10 | * version 2.1 of the License, or (at your option) any later version. |
| 11 | * |
| 12 | * FFmpeg is distributed in the hope that it will be useful, |
| 13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 15 | * Lesser General Public License for more details. |
| 16 | * |
| 17 | * You should have received a copy of the GNU Lesser General Public |
| 18 | * License along with FFmpeg; if not, write to the Free Software |
| 19 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 20 | */ |
| 21 | |
| 22 | /** |
| 23 | * @file |
| 24 | * AAC encoder |
| 25 | */ |
| 26 | |
| 27 | /*********************************** |
| 28 | * TODOs: |
| 29 | * add sane pulse detection |
| 30 | * add temporal noise shaping |
| 31 | ***********************************/ |
| 32 | |
| 33 | #include "libavutil/float_dsp.h" |
| 34 | #include "libavutil/opt.h" |
| 35 | #include "avcodec.h" |
| 36 | #include "put_bits.h" |
| 37 | #include "internal.h" |
| 38 | #include "mpeg4audio.h" |
| 39 | #include "kbdwin.h" |
| 40 | #include "sinewin.h" |
| 41 | |
| 42 | #include "aac.h" |
| 43 | #include "aactab.h" |
| 44 | #include "aacenc.h" |
| 45 | |
| 46 | #include "psymodel.h" |
| 47 | |
| 48 | #define AAC_MAX_CHANNELS 6 |
| 49 | |
| 50 | #define ERROR_IF(cond, ...) \ |
| 51 | if (cond) { \ |
| 52 | av_log(avctx, AV_LOG_ERROR, __VA_ARGS__); \ |
| 53 | return AVERROR(EINVAL); \ |
| 54 | } |
| 55 | |
| 56 | float ff_aac_pow34sf_tab[428]; |
| 57 | |
| 58 | static const uint8_t swb_size_1024_96[] = { |
| 59 | 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, |
| 60 | 12, 12, 12, 12, 12, 16, 16, 24, 28, 36, 44, |
| 61 | 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64 |
| 62 | }; |
| 63 | |
| 64 | static const uint8_t swb_size_1024_64[] = { |
| 65 | 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, |
| 66 | 12, 12, 12, 16, 16, 16, 20, 24, 24, 28, 36, |
| 67 | 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40 |
| 68 | }; |
| 69 | |
| 70 | static const uint8_t swb_size_1024_48[] = { |
| 71 | 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8, |
| 72 | 12, 12, 12, 12, 16, 16, 20, 20, 24, 24, 28, 28, |
| 73 | 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, |
| 74 | 96 |
| 75 | }; |
| 76 | |
| 77 | static const uint8_t swb_size_1024_32[] = { |
| 78 | 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8, |
| 79 | 12, 12, 12, 12, 16, 16, 20, 20, 24, 24, 28, 28, |
| 80 | 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32 |
| 81 | }; |
| 82 | |
| 83 | static const uint8_t swb_size_1024_24[] = { |
| 84 | 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, |
| 85 | 12, 12, 12, 12, 16, 16, 16, 20, 20, 24, 24, 28, 28, |
| 86 | 32, 36, 36, 40, 44, 48, 52, 52, 64, 64, 64, 64, 64 |
| 87 | }; |
| 88 | |
| 89 | static const uint8_t swb_size_1024_16[] = { |
| 90 | 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, |
| 91 | 12, 12, 12, 12, 12, 12, 12, 12, 12, 16, 16, 16, 16, 20, 20, 20, 24, 24, 28, 28, |
| 92 | 32, 36, 40, 40, 44, 48, 52, 56, 60, 64, 64, 64 |
| 93 | }; |
| 94 | |
| 95 | static const uint8_t swb_size_1024_8[] = { |
| 96 | 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, |
| 97 | 16, 16, 16, 16, 16, 16, 16, 20, 20, 20, 20, 24, 24, 24, 28, 28, |
| 98 | 32, 36, 36, 40, 44, 48, 52, 56, 60, 64, 80 |
| 99 | }; |
| 100 | |
| 101 | static const uint8_t *swb_size_1024[] = { |
| 102 | swb_size_1024_96, swb_size_1024_96, swb_size_1024_64, |
| 103 | swb_size_1024_48, swb_size_1024_48, swb_size_1024_32, |
| 104 | swb_size_1024_24, swb_size_1024_24, swb_size_1024_16, |
| 105 | swb_size_1024_16, swb_size_1024_16, swb_size_1024_8 |
| 106 | }; |
| 107 | |
| 108 | static const uint8_t swb_size_128_96[] = { |
| 109 | 4, 4, 4, 4, 4, 4, 8, 8, 8, 16, 28, 36 |
| 110 | }; |
| 111 | |
| 112 | static const uint8_t swb_size_128_48[] = { |
| 113 | 4, 4, 4, 4, 4, 8, 8, 8, 12, 12, 12, 16, 16, 16 |
| 114 | }; |
| 115 | |
| 116 | static const uint8_t swb_size_128_24[] = { |
| 117 | 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 12, 12, 16, 16, 20 |
| 118 | }; |
| 119 | |
| 120 | static const uint8_t swb_size_128_16[] = { |
| 121 | 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 12, 12, 16, 20, 20 |
| 122 | }; |
| 123 | |
| 124 | static const uint8_t swb_size_128_8[] = { |
| 125 | 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 12, 16, 20, 20 |
| 126 | }; |
| 127 | |
| 128 | static const uint8_t *swb_size_128[] = { |
| 129 | /* the last entry on the following row is swb_size_128_64 but is a |
| 130 | duplicate of swb_size_128_96 */ |
| 131 | swb_size_128_96, swb_size_128_96, swb_size_128_96, |
| 132 | swb_size_128_48, swb_size_128_48, swb_size_128_48, |
| 133 | swb_size_128_24, swb_size_128_24, swb_size_128_16, |
| 134 | swb_size_128_16, swb_size_128_16, swb_size_128_8 |
| 135 | }; |
| 136 | |
| 137 | /** default channel configurations */ |
| 138 | static const uint8_t aac_chan_configs[6][5] = { |
| 139 | {1, TYPE_SCE}, // 1 channel - single channel element |
| 140 | {1, TYPE_CPE}, // 2 channels - channel pair |
| 141 | {2, TYPE_SCE, TYPE_CPE}, // 3 channels - center + stereo |
| 142 | {3, TYPE_SCE, TYPE_CPE, TYPE_SCE}, // 4 channels - front center + stereo + back center |
| 143 | {3, TYPE_SCE, TYPE_CPE, TYPE_CPE}, // 5 channels - front center + stereo + back stereo |
| 144 | {4, TYPE_SCE, TYPE_CPE, TYPE_CPE, TYPE_LFE}, // 6 channels - front center + stereo + back stereo + LFE |
| 145 | }; |
| 146 | |
| 147 | /** |
| 148 | * Table to remap channels from libavcodec's default order to AAC order. |
| 149 | */ |
| 150 | static const uint8_t aac_chan_maps[AAC_MAX_CHANNELS][AAC_MAX_CHANNELS] = { |
| 151 | { 0 }, |
| 152 | { 0, 1 }, |
| 153 | { 2, 0, 1 }, |
| 154 | { 2, 0, 1, 3 }, |
| 155 | { 2, 0, 1, 3, 4 }, |
| 156 | { 2, 0, 1, 4, 5, 3 }, |
| 157 | }; |
| 158 | |
| 159 | /** |
| 160 | * Make AAC audio config object. |
| 161 | * @see 1.6.2.1 "Syntax - AudioSpecificConfig" |
| 162 | */ |
| 163 | static void put_audio_specific_config(AVCodecContext *avctx) |
| 164 | { |
| 165 | PutBitContext pb; |
| 166 | AACEncContext *s = avctx->priv_data; |
| 167 | |
| 168 | init_put_bits(&pb, avctx->extradata, avctx->extradata_size*8); |
| 169 | put_bits(&pb, 5, 2); //object type - AAC-LC |
| 170 | put_bits(&pb, 4, s->samplerate_index); //sample rate index |
| 171 | put_bits(&pb, 4, s->channels); |
| 172 | //GASpecificConfig |
| 173 | put_bits(&pb, 1, 0); //frame length - 1024 samples |
| 174 | put_bits(&pb, 1, 0); //does not depend on core coder |
| 175 | put_bits(&pb, 1, 0); //is not extension |
| 176 | |
| 177 | //Explicitly Mark SBR absent |
| 178 | put_bits(&pb, 11, 0x2b7); //sync extension |
| 179 | put_bits(&pb, 5, AOT_SBR); |
| 180 | put_bits(&pb, 1, 0); |
| 181 | flush_put_bits(&pb); |
| 182 | } |
| 183 | |
| 184 | #define WINDOW_FUNC(type) \ |
| 185 | static void apply_ ##type ##_window(AVFloatDSPContext *fdsp, \ |
| 186 | SingleChannelElement *sce, \ |
| 187 | const float *audio) |
| 188 | |
| 189 | WINDOW_FUNC(only_long) |
| 190 | { |
| 191 | const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024; |
| 192 | const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024; |
| 193 | float *out = sce->ret_buf; |
| 194 | |
| 195 | fdsp->vector_fmul (out, audio, lwindow, 1024); |
| 196 | fdsp->vector_fmul_reverse(out + 1024, audio + 1024, pwindow, 1024); |
| 197 | } |
| 198 | |
| 199 | WINDOW_FUNC(long_start) |
| 200 | { |
| 201 | const float *lwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024; |
| 202 | const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128; |
| 203 | float *out = sce->ret_buf; |
| 204 | |
| 205 | fdsp->vector_fmul(out, audio, lwindow, 1024); |
| 206 | memcpy(out + 1024, audio + 1024, sizeof(out[0]) * 448); |
| 207 | fdsp->vector_fmul_reverse(out + 1024 + 448, audio + 1024 + 448, swindow, 128); |
| 208 | memset(out + 1024 + 576, 0, sizeof(out[0]) * 448); |
| 209 | } |
| 210 | |
| 211 | WINDOW_FUNC(long_stop) |
| 212 | { |
| 213 | const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024; |
| 214 | const float *swindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128; |
| 215 | float *out = sce->ret_buf; |
| 216 | |
| 217 | memset(out, 0, sizeof(out[0]) * 448); |
| 218 | fdsp->vector_fmul(out + 448, audio + 448, swindow, 128); |
| 219 | memcpy(out + 576, audio + 576, sizeof(out[0]) * 448); |
| 220 | fdsp->vector_fmul_reverse(out + 1024, audio + 1024, lwindow, 1024); |
| 221 | } |
| 222 | |
| 223 | WINDOW_FUNC(eight_short) |
| 224 | { |
| 225 | const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128; |
| 226 | const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128; |
| 227 | const float *in = audio + 448; |
| 228 | float *out = sce->ret_buf; |
| 229 | int w; |
| 230 | |
| 231 | for (w = 0; w < 8; w++) { |
| 232 | fdsp->vector_fmul (out, in, w ? pwindow : swindow, 128); |
| 233 | out += 128; |
| 234 | in += 128; |
| 235 | fdsp->vector_fmul_reverse(out, in, swindow, 128); |
| 236 | out += 128; |
| 237 | } |
| 238 | } |
| 239 | |
| 240 | static void (*const apply_window[4])(AVFloatDSPContext *fdsp, |
| 241 | SingleChannelElement *sce, |
| 242 | const float *audio) = { |
| 243 | [ONLY_LONG_SEQUENCE] = apply_only_long_window, |
| 244 | [LONG_START_SEQUENCE] = apply_long_start_window, |
| 245 | [EIGHT_SHORT_SEQUENCE] = apply_eight_short_window, |
| 246 | [LONG_STOP_SEQUENCE] = apply_long_stop_window |
| 247 | }; |
| 248 | |
| 249 | static void apply_window_and_mdct(AACEncContext *s, SingleChannelElement *sce, |
| 250 | float *audio) |
| 251 | { |
| 252 | int i; |
| 253 | float *output = sce->ret_buf; |
| 254 | |
| 255 | apply_window[sce->ics.window_sequence[0]](&s->fdsp, sce, audio); |
| 256 | |
| 257 | if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) |
| 258 | s->mdct1024.mdct_calc(&s->mdct1024, sce->coeffs, output); |
| 259 | else |
| 260 | for (i = 0; i < 1024; i += 128) |
| 261 | s->mdct128.mdct_calc(&s->mdct128, sce->coeffs + i, output + i*2); |
| 262 | memcpy(audio, audio + 1024, sizeof(audio[0]) * 1024); |
| 263 | } |
| 264 | |
| 265 | /** |
| 266 | * Encode ics_info element. |
| 267 | * @see Table 4.6 (syntax of ics_info) |
| 268 | */ |
| 269 | static void put_ics_info(AACEncContext *s, IndividualChannelStream *info) |
| 270 | { |
| 271 | int w; |
| 272 | |
| 273 | put_bits(&s->pb, 1, 0); // ics_reserved bit |
| 274 | put_bits(&s->pb, 2, info->window_sequence[0]); |
| 275 | put_bits(&s->pb, 1, info->use_kb_window[0]); |
| 276 | if (info->window_sequence[0] != EIGHT_SHORT_SEQUENCE) { |
| 277 | put_bits(&s->pb, 6, info->max_sfb); |
| 278 | put_bits(&s->pb, 1, 0); // no prediction |
| 279 | } else { |
| 280 | put_bits(&s->pb, 4, info->max_sfb); |
| 281 | for (w = 1; w < 8; w++) |
| 282 | put_bits(&s->pb, 1, !info->group_len[w]); |
| 283 | } |
| 284 | } |
| 285 | |
| 286 | /** |
| 287 | * Encode MS data. |
| 288 | * @see 4.6.8.1 "Joint Coding - M/S Stereo" |
| 289 | */ |
| 290 | static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe) |
| 291 | { |
| 292 | int i, w; |
| 293 | |
| 294 | put_bits(pb, 2, cpe->ms_mode); |
| 295 | if (cpe->ms_mode == 1) |
| 296 | for (w = 0; w < cpe->ch[0].ics.num_windows; w += cpe->ch[0].ics.group_len[w]) |
| 297 | for (i = 0; i < cpe->ch[0].ics.max_sfb; i++) |
| 298 | put_bits(pb, 1, cpe->ms_mask[w*16 + i]); |
| 299 | } |
| 300 | |
| 301 | /** |
| 302 | * Produce integer coefficients from scalefactors provided by the model. |
| 303 | */ |
| 304 | static void adjust_frame_information(ChannelElement *cpe, int chans) |
| 305 | { |
| 306 | int i, w, w2, g, ch; |
| 307 | int start, maxsfb, cmaxsfb; |
| 308 | |
| 309 | for (ch = 0; ch < chans; ch++) { |
| 310 | IndividualChannelStream *ics = &cpe->ch[ch].ics; |
| 311 | start = 0; |
| 312 | maxsfb = 0; |
| 313 | cpe->ch[ch].pulse.num_pulse = 0; |
| 314 | for (w = 0; w < ics->num_windows*16; w += 16) { |
| 315 | for (g = 0; g < ics->num_swb; g++) { |
| 316 | //apply M/S |
| 317 | if (cpe->common_window && !ch && cpe->ms_mask[w + g]) { |
| 318 | for (i = 0; i < ics->swb_sizes[g]; i++) { |
| 319 | cpe->ch[0].coeffs[start+i] = (cpe->ch[0].coeffs[start+i] + cpe->ch[1].coeffs[start+i]) / 2.0; |
| 320 | cpe->ch[1].coeffs[start+i] = cpe->ch[0].coeffs[start+i] - cpe->ch[1].coeffs[start+i]; |
| 321 | } |
| 322 | } |
| 323 | start += ics->swb_sizes[g]; |
| 324 | } |
| 325 | for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w+cmaxsfb-1]; cmaxsfb--) |
| 326 | ; |
| 327 | maxsfb = FFMAX(maxsfb, cmaxsfb); |
| 328 | } |
| 329 | ics->max_sfb = maxsfb; |
| 330 | |
| 331 | //adjust zero bands for window groups |
| 332 | for (w = 0; w < ics->num_windows; w += ics->group_len[w]) { |
| 333 | for (g = 0; g < ics->max_sfb; g++) { |
| 334 | i = 1; |
| 335 | for (w2 = w; w2 < w + ics->group_len[w]; w2++) { |
| 336 | if (!cpe->ch[ch].zeroes[w2*16 + g]) { |
| 337 | i = 0; |
| 338 | break; |
| 339 | } |
| 340 | } |
| 341 | cpe->ch[ch].zeroes[w*16 + g] = i; |
| 342 | } |
| 343 | } |
| 344 | } |
| 345 | |
| 346 | if (chans > 1 && cpe->common_window) { |
| 347 | IndividualChannelStream *ics0 = &cpe->ch[0].ics; |
| 348 | IndividualChannelStream *ics1 = &cpe->ch[1].ics; |
| 349 | int msc = 0; |
| 350 | ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb); |
| 351 | ics1->max_sfb = ics0->max_sfb; |
| 352 | for (w = 0; w < ics0->num_windows*16; w += 16) |
| 353 | for (i = 0; i < ics0->max_sfb; i++) |
| 354 | if (cpe->ms_mask[w+i]) |
| 355 | msc++; |
| 356 | if (msc == 0 || ics0->max_sfb == 0) |
| 357 | cpe->ms_mode = 0; |
| 358 | else |
| 359 | cpe->ms_mode = msc < ics0->max_sfb * ics0->num_windows ? 1 : 2; |
| 360 | } |
| 361 | } |
| 362 | |
| 363 | /** |
| 364 | * Encode scalefactor band coding type. |
| 365 | */ |
| 366 | static void encode_band_info(AACEncContext *s, SingleChannelElement *sce) |
| 367 | { |
| 368 | int w; |
| 369 | |
| 370 | for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) |
| 371 | s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda); |
| 372 | } |
| 373 | |
| 374 | /** |
| 375 | * Encode scalefactors. |
| 376 | */ |
| 377 | static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s, |
| 378 | SingleChannelElement *sce) |
| 379 | { |
| 380 | int off = sce->sf_idx[0], diff; |
| 381 | int i, w; |
| 382 | |
| 383 | for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) { |
| 384 | for (i = 0; i < sce->ics.max_sfb; i++) { |
| 385 | if (!sce->zeroes[w*16 + i]) { |
| 386 | diff = sce->sf_idx[w*16 + i] - off + SCALE_DIFF_ZERO; |
| 387 | av_assert0(diff >= 0 && diff <= 120); |
| 388 | off = sce->sf_idx[w*16 + i]; |
| 389 | put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]); |
| 390 | } |
| 391 | } |
| 392 | } |
| 393 | } |
| 394 | |
| 395 | /** |
| 396 | * Encode pulse data. |
| 397 | */ |
| 398 | static void encode_pulses(AACEncContext *s, Pulse *pulse) |
| 399 | { |
| 400 | int i; |
| 401 | |
| 402 | put_bits(&s->pb, 1, !!pulse->num_pulse); |
| 403 | if (!pulse->num_pulse) |
| 404 | return; |
| 405 | |
| 406 | put_bits(&s->pb, 2, pulse->num_pulse - 1); |
| 407 | put_bits(&s->pb, 6, pulse->start); |
| 408 | for (i = 0; i < pulse->num_pulse; i++) { |
| 409 | put_bits(&s->pb, 5, pulse->pos[i]); |
| 410 | put_bits(&s->pb, 4, pulse->amp[i]); |
| 411 | } |
| 412 | } |
| 413 | |
| 414 | /** |
| 415 | * Encode spectral coefficients processed by psychoacoustic model. |
| 416 | */ |
| 417 | static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce) |
| 418 | { |
| 419 | int start, i, w, w2; |
| 420 | |
| 421 | for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) { |
| 422 | start = 0; |
| 423 | for (i = 0; i < sce->ics.max_sfb; i++) { |
| 424 | if (sce->zeroes[w*16 + i]) { |
| 425 | start += sce->ics.swb_sizes[i]; |
| 426 | continue; |
| 427 | } |
| 428 | for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++) |
| 429 | s->coder->quantize_and_encode_band(s, &s->pb, sce->coeffs + start + w2*128, |
| 430 | sce->ics.swb_sizes[i], |
| 431 | sce->sf_idx[w*16 + i], |
| 432 | sce->band_type[w*16 + i], |
| 433 | s->lambda); |
| 434 | start += sce->ics.swb_sizes[i]; |
| 435 | } |
| 436 | } |
| 437 | } |
| 438 | |
| 439 | /** |
| 440 | * Encode one channel of audio data. |
| 441 | */ |
| 442 | static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s, |
| 443 | SingleChannelElement *sce, |
| 444 | int common_window) |
| 445 | { |
| 446 | put_bits(&s->pb, 8, sce->sf_idx[0]); |
| 447 | if (!common_window) |
| 448 | put_ics_info(s, &sce->ics); |
| 449 | encode_band_info(s, sce); |
| 450 | encode_scale_factors(avctx, s, sce); |
| 451 | encode_pulses(s, &sce->pulse); |
| 452 | put_bits(&s->pb, 1, 0); //tns |
| 453 | put_bits(&s->pb, 1, 0); //ssr |
| 454 | encode_spectral_coeffs(s, sce); |
| 455 | return 0; |
| 456 | } |
| 457 | |
| 458 | /** |
| 459 | * Write some auxiliary information about the created AAC file. |
| 460 | */ |
| 461 | static void put_bitstream_info(AACEncContext *s, const char *name) |
| 462 | { |
| 463 | int i, namelen, padbits; |
| 464 | |
| 465 | namelen = strlen(name) + 2; |
| 466 | put_bits(&s->pb, 3, TYPE_FIL); |
| 467 | put_bits(&s->pb, 4, FFMIN(namelen, 15)); |
| 468 | if (namelen >= 15) |
| 469 | put_bits(&s->pb, 8, namelen - 14); |
| 470 | put_bits(&s->pb, 4, 0); //extension type - filler |
| 471 | padbits = -put_bits_count(&s->pb) & 7; |
| 472 | avpriv_align_put_bits(&s->pb); |
| 473 | for (i = 0; i < namelen - 2; i++) |
| 474 | put_bits(&s->pb, 8, name[i]); |
| 475 | put_bits(&s->pb, 12 - padbits, 0); |
| 476 | } |
| 477 | |
| 478 | /* |
| 479 | * Copy input samples. |
| 480 | * Channels are reordered from libavcodec's default order to AAC order. |
| 481 | */ |
| 482 | static void copy_input_samples(AACEncContext *s, const AVFrame *frame) |
| 483 | { |
| 484 | int ch; |
| 485 | int end = 2048 + (frame ? frame->nb_samples : 0); |
| 486 | const uint8_t *channel_map = aac_chan_maps[s->channels - 1]; |
| 487 | |
| 488 | /* copy and remap input samples */ |
| 489 | for (ch = 0; ch < s->channels; ch++) { |
| 490 | /* copy last 1024 samples of previous frame to the start of the current frame */ |
| 491 | memcpy(&s->planar_samples[ch][1024], &s->planar_samples[ch][2048], 1024 * sizeof(s->planar_samples[0][0])); |
| 492 | |
| 493 | /* copy new samples and zero any remaining samples */ |
| 494 | if (frame) { |
| 495 | memcpy(&s->planar_samples[ch][2048], |
| 496 | frame->extended_data[channel_map[ch]], |
| 497 | frame->nb_samples * sizeof(s->planar_samples[0][0])); |
| 498 | } |
| 499 | memset(&s->planar_samples[ch][end], 0, |
| 500 | (3072 - end) * sizeof(s->planar_samples[0][0])); |
| 501 | } |
| 502 | } |
| 503 | |
| 504 | static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt, |
| 505 | const AVFrame *frame, int *got_packet_ptr) |
| 506 | { |
| 507 | AACEncContext *s = avctx->priv_data; |
| 508 | float **samples = s->planar_samples, *samples2, *la, *overlap; |
| 509 | ChannelElement *cpe; |
| 510 | int i, ch, w, g, chans, tag, start_ch, ret; |
| 511 | int chan_el_counter[4]; |
| 512 | FFPsyWindowInfo windows[AAC_MAX_CHANNELS]; |
| 513 | |
| 514 | if (s->last_frame == 2) |
| 515 | return 0; |
| 516 | |
| 517 | /* add current frame to queue */ |
| 518 | if (frame) { |
| 519 | if ((ret = ff_af_queue_add(&s->afq, frame)) < 0) |
| 520 | return ret; |
| 521 | } |
| 522 | |
| 523 | copy_input_samples(s, frame); |
| 524 | if (s->psypp) |
| 525 | ff_psy_preprocess(s->psypp, s->planar_samples, s->channels); |
| 526 | |
| 527 | if (!avctx->frame_number) |
| 528 | return 0; |
| 529 | |
| 530 | start_ch = 0; |
| 531 | for (i = 0; i < s->chan_map[0]; i++) { |
| 532 | FFPsyWindowInfo* wi = windows + start_ch; |
| 533 | tag = s->chan_map[i+1]; |
| 534 | chans = tag == TYPE_CPE ? 2 : 1; |
| 535 | cpe = &s->cpe[i]; |
| 536 | for (ch = 0; ch < chans; ch++) { |
| 537 | IndividualChannelStream *ics = &cpe->ch[ch].ics; |
| 538 | int cur_channel = start_ch + ch; |
| 539 | overlap = &samples[cur_channel][0]; |
| 540 | samples2 = overlap + 1024; |
| 541 | la = samples2 + (448+64); |
| 542 | if (!frame) |
| 543 | la = NULL; |
| 544 | if (tag == TYPE_LFE) { |
| 545 | wi[ch].window_type[0] = ONLY_LONG_SEQUENCE; |
| 546 | wi[ch].window_shape = 0; |
| 547 | wi[ch].num_windows = 1; |
| 548 | wi[ch].grouping[0] = 1; |
| 549 | |
| 550 | /* Only the lowest 12 coefficients are used in a LFE channel. |
| 551 | * The expression below results in only the bottom 8 coefficients |
| 552 | * being used for 11.025kHz to 16kHz sample rates. |
| 553 | */ |
| 554 | ics->num_swb = s->samplerate_index >= 8 ? 1 : 3; |
| 555 | } else { |
| 556 | wi[ch] = s->psy.model->window(&s->psy, samples2, la, cur_channel, |
| 557 | ics->window_sequence[0]); |
| 558 | } |
| 559 | ics->window_sequence[1] = ics->window_sequence[0]; |
| 560 | ics->window_sequence[0] = wi[ch].window_type[0]; |
| 561 | ics->use_kb_window[1] = ics->use_kb_window[0]; |
| 562 | ics->use_kb_window[0] = wi[ch].window_shape; |
| 563 | ics->num_windows = wi[ch].num_windows; |
| 564 | ics->swb_sizes = s->psy.bands [ics->num_windows == 8]; |
| 565 | ics->num_swb = tag == TYPE_LFE ? ics->num_swb : s->psy.num_bands[ics->num_windows == 8]; |
| 566 | for (w = 0; w < ics->num_windows; w++) |
| 567 | ics->group_len[w] = wi[ch].grouping[w]; |
| 568 | |
| 569 | apply_window_and_mdct(s, &cpe->ch[ch], overlap); |
| 570 | } |
| 571 | start_ch += chans; |
| 572 | } |
| 573 | if ((ret = ff_alloc_packet2(avctx, avpkt, 8192 * s->channels)) < 0) |
| 574 | return ret; |
| 575 | do { |
| 576 | int frame_bits; |
| 577 | |
| 578 | init_put_bits(&s->pb, avpkt->data, avpkt->size); |
| 579 | |
| 580 | if ((avctx->frame_number & 0xFF)==1 && !(avctx->flags & CODEC_FLAG_BITEXACT)) |
| 581 | put_bitstream_info(s, LIBAVCODEC_IDENT); |
| 582 | start_ch = 0; |
| 583 | memset(chan_el_counter, 0, sizeof(chan_el_counter)); |
| 584 | for (i = 0; i < s->chan_map[0]; i++) { |
| 585 | FFPsyWindowInfo* wi = windows + start_ch; |
| 586 | const float *coeffs[2]; |
| 587 | tag = s->chan_map[i+1]; |
| 588 | chans = tag == TYPE_CPE ? 2 : 1; |
| 589 | cpe = &s->cpe[i]; |
| 590 | put_bits(&s->pb, 3, tag); |
| 591 | put_bits(&s->pb, 4, chan_el_counter[tag]++); |
| 592 | for (ch = 0; ch < chans; ch++) |
| 593 | coeffs[ch] = cpe->ch[ch].coeffs; |
| 594 | s->psy.model->analyze(&s->psy, start_ch, coeffs, wi); |
| 595 | for (ch = 0; ch < chans; ch++) { |
| 596 | s->cur_channel = start_ch + ch; |
| 597 | s->coder->search_for_quantizers(avctx, s, &cpe->ch[ch], s->lambda); |
| 598 | } |
| 599 | cpe->common_window = 0; |
| 600 | if (chans > 1 |
| 601 | && wi[0].window_type[0] == wi[1].window_type[0] |
| 602 | && wi[0].window_shape == wi[1].window_shape) { |
| 603 | |
| 604 | cpe->common_window = 1; |
| 605 | for (w = 0; w < wi[0].num_windows; w++) { |
| 606 | if (wi[0].grouping[w] != wi[1].grouping[w]) { |
| 607 | cpe->common_window = 0; |
| 608 | break; |
| 609 | } |
| 610 | } |
| 611 | } |
| 612 | s->cur_channel = start_ch; |
| 613 | if (s->options.stereo_mode && cpe->common_window) { |
| 614 | if (s->options.stereo_mode > 0) { |
| 615 | IndividualChannelStream *ics = &cpe->ch[0].ics; |
| 616 | for (w = 0; w < ics->num_windows; w += ics->group_len[w]) |
| 617 | for (g = 0; g < ics->num_swb; g++) |
| 618 | cpe->ms_mask[w*16+g] = 1; |
| 619 | } else if (s->coder->search_for_ms) { |
| 620 | s->coder->search_for_ms(s, cpe, s->lambda); |
| 621 | } |
| 622 | } |
| 623 | adjust_frame_information(cpe, chans); |
| 624 | if (chans == 2) { |
| 625 | put_bits(&s->pb, 1, cpe->common_window); |
| 626 | if (cpe->common_window) { |
| 627 | put_ics_info(s, &cpe->ch[0].ics); |
| 628 | encode_ms_info(&s->pb, cpe); |
| 629 | } |
| 630 | } |
| 631 | for (ch = 0; ch < chans; ch++) { |
| 632 | s->cur_channel = start_ch + ch; |
| 633 | encode_individual_channel(avctx, s, &cpe->ch[ch], cpe->common_window); |
| 634 | } |
| 635 | start_ch += chans; |
| 636 | } |
| 637 | |
| 638 | frame_bits = put_bits_count(&s->pb); |
| 639 | if (frame_bits <= 6144 * s->channels - 3) { |
| 640 | s->psy.bitres.bits = frame_bits / s->channels; |
| 641 | break; |
| 642 | } |
| 643 | |
| 644 | s->lambda *= avctx->bit_rate * 1024.0f / avctx->sample_rate / frame_bits; |
| 645 | |
| 646 | } while (1); |
| 647 | |
| 648 | put_bits(&s->pb, 3, TYPE_END); |
| 649 | flush_put_bits(&s->pb); |
| 650 | avctx->frame_bits = put_bits_count(&s->pb); |
| 651 | |
| 652 | // rate control stuff |
| 653 | if (!(avctx->flags & CODEC_FLAG_QSCALE)) { |
| 654 | float ratio = avctx->bit_rate * 1024.0f / avctx->sample_rate / avctx->frame_bits; |
| 655 | s->lambda *= ratio; |
| 656 | s->lambda = FFMIN(s->lambda, 65536.f); |
| 657 | } |
| 658 | |
| 659 | if (!frame) |
| 660 | s->last_frame++; |
| 661 | |
| 662 | ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts, |
| 663 | &avpkt->duration); |
| 664 | |
| 665 | avpkt->size = put_bits_count(&s->pb) >> 3; |
| 666 | *got_packet_ptr = 1; |
| 667 | return 0; |
| 668 | } |
| 669 | |
| 670 | static av_cold int aac_encode_end(AVCodecContext *avctx) |
| 671 | { |
| 672 | AACEncContext *s = avctx->priv_data; |
| 673 | |
| 674 | ff_mdct_end(&s->mdct1024); |
| 675 | ff_mdct_end(&s->mdct128); |
| 676 | ff_psy_end(&s->psy); |
| 677 | if (s->psypp) |
| 678 | ff_psy_preprocess_end(s->psypp); |
| 679 | av_freep(&s->buffer.samples); |
| 680 | av_freep(&s->cpe); |
| 681 | ff_af_queue_close(&s->afq); |
| 682 | return 0; |
| 683 | } |
| 684 | |
| 685 | static av_cold int dsp_init(AVCodecContext *avctx, AACEncContext *s) |
| 686 | { |
| 687 | int ret = 0; |
| 688 | |
| 689 | avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT); |
| 690 | |
| 691 | // window init |
| 692 | ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024); |
| 693 | ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128); |
| 694 | ff_init_ff_sine_windows(10); |
| 695 | ff_init_ff_sine_windows(7); |
| 696 | |
| 697 | if (ret = ff_mdct_init(&s->mdct1024, 11, 0, 32768.0)) |
| 698 | return ret; |
| 699 | if (ret = ff_mdct_init(&s->mdct128, 8, 0, 32768.0)) |
| 700 | return ret; |
| 701 | |
| 702 | return 0; |
| 703 | } |
| 704 | |
| 705 | static av_cold int alloc_buffers(AVCodecContext *avctx, AACEncContext *s) |
| 706 | { |
| 707 | int ch; |
| 708 | FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->buffer.samples, s->channels, 3 * 1024 * sizeof(s->buffer.samples[0]), alloc_fail); |
| 709 | FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->cpe, s->chan_map[0], sizeof(ChannelElement), alloc_fail); |
| 710 | FF_ALLOCZ_OR_GOTO(avctx, avctx->extradata, 5 + FF_INPUT_BUFFER_PADDING_SIZE, alloc_fail); |
| 711 | |
| 712 | for(ch = 0; ch < s->channels; ch++) |
| 713 | s->planar_samples[ch] = s->buffer.samples + 3 * 1024 * ch; |
| 714 | |
| 715 | return 0; |
| 716 | alloc_fail: |
| 717 | return AVERROR(ENOMEM); |
| 718 | } |
| 719 | |
| 720 | static av_cold int aac_encode_init(AVCodecContext *avctx) |
| 721 | { |
| 722 | AACEncContext *s = avctx->priv_data; |
| 723 | int i, ret = 0; |
| 724 | const uint8_t *sizes[2]; |
| 725 | uint8_t grouping[AAC_MAX_CHANNELS]; |
| 726 | int lengths[2]; |
| 727 | |
| 728 | avctx->frame_size = 1024; |
| 729 | |
| 730 | for (i = 0; i < 16; i++) |
| 731 | if (avctx->sample_rate == avpriv_mpeg4audio_sample_rates[i]) |
| 732 | break; |
| 733 | |
| 734 | s->channels = avctx->channels; |
| 735 | |
| 736 | ERROR_IF(i == 16, |
| 737 | "Unsupported sample rate %d\n", avctx->sample_rate); |
| 738 | ERROR_IF(s->channels > AAC_MAX_CHANNELS, |
| 739 | "Unsupported number of channels: %d\n", s->channels); |
| 740 | ERROR_IF(avctx->profile != FF_PROFILE_UNKNOWN && avctx->profile != FF_PROFILE_AAC_LOW, |
| 741 | "Unsupported profile %d\n", avctx->profile); |
| 742 | ERROR_IF(1024.0 * avctx->bit_rate / avctx->sample_rate > 6144 * s->channels, |
| 743 | "Too many bits per frame requested\n"); |
| 744 | |
| 745 | s->samplerate_index = i; |
| 746 | |
| 747 | s->chan_map = aac_chan_configs[s->channels-1]; |
| 748 | |
| 749 | if (ret = dsp_init(avctx, s)) |
| 750 | goto fail; |
| 751 | |
| 752 | if (ret = alloc_buffers(avctx, s)) |
| 753 | goto fail; |
| 754 | |
| 755 | avctx->extradata_size = 5; |
| 756 | put_audio_specific_config(avctx); |
| 757 | |
| 758 | sizes[0] = swb_size_1024[i]; |
| 759 | sizes[1] = swb_size_128[i]; |
| 760 | lengths[0] = ff_aac_num_swb_1024[i]; |
| 761 | lengths[1] = ff_aac_num_swb_128[i]; |
| 762 | for (i = 0; i < s->chan_map[0]; i++) |
| 763 | grouping[i] = s->chan_map[i + 1] == TYPE_CPE; |
| 764 | if (ret = ff_psy_init(&s->psy, avctx, 2, sizes, lengths, s->chan_map[0], grouping)) |
| 765 | goto fail; |
| 766 | s->psypp = ff_psy_preprocess_init(avctx); |
| 767 | s->coder = &ff_aac_coders[s->options.aac_coder]; |
| 768 | |
| 769 | if (HAVE_MIPSDSPR1) |
| 770 | ff_aac_coder_init_mips(s); |
| 771 | |
| 772 | s->lambda = avctx->global_quality > 0 ? avctx->global_quality : 120; |
| 773 | |
| 774 | ff_aac_tableinit(); |
| 775 | |
| 776 | for (i = 0; i < 428; i++) |
| 777 | ff_aac_pow34sf_tab[i] = sqrt(ff_aac_pow2sf_tab[i] * sqrt(ff_aac_pow2sf_tab[i])); |
| 778 | |
| 779 | avctx->delay = 1024; |
| 780 | ff_af_queue_init(avctx, &s->afq); |
| 781 | |
| 782 | return 0; |
| 783 | fail: |
| 784 | aac_encode_end(avctx); |
| 785 | return ret; |
| 786 | } |
| 787 | |
| 788 | #define AACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM |
| 789 | static const AVOption aacenc_options[] = { |
| 790 | {"stereo_mode", "Stereo coding method", offsetof(AACEncContext, options.stereo_mode), AV_OPT_TYPE_INT, {.i64 = 0}, -1, 1, AACENC_FLAGS, "stereo_mode"}, |
| 791 | {"auto", "Selected by the Encoder", 0, AV_OPT_TYPE_CONST, {.i64 = -1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"}, |
| 792 | {"ms_off", "Disable Mid/Side coding", 0, AV_OPT_TYPE_CONST, {.i64 = 0 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"}, |
| 793 | {"ms_force", "Force Mid/Side for the whole frame if possible", 0, AV_OPT_TYPE_CONST, {.i64 = 1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"}, |
| 794 | {"aac_coder", "", offsetof(AACEncContext, options.aac_coder), AV_OPT_TYPE_INT, {.i64 = AAC_CODER_TWOLOOP}, 0, AAC_CODER_NB-1, AACENC_FLAGS, "aac_coder"}, |
| 795 | {"faac", "FAAC-inspired method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAAC}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"}, |
| 796 | {"anmr", "ANMR method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_ANMR}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"}, |
| 797 | {"twoloop", "Two loop searching method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_TWOLOOP}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"}, |
| 798 | {"fast", "Constant quantizer", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAST}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"}, |
| 799 | {NULL} |
| 800 | }; |
| 801 | |
| 802 | static const AVClass aacenc_class = { |
| 803 | "AAC encoder", |
| 804 | av_default_item_name, |
| 805 | aacenc_options, |
| 806 | LIBAVUTIL_VERSION_INT, |
| 807 | }; |
| 808 | |
| 809 | /* duplicated from avpriv_mpeg4audio_sample_rates to avoid shared build |
| 810 | * failures */ |
| 811 | static const int mpeg4audio_sample_rates[16] = { |
| 812 | 96000, 88200, 64000, 48000, 44100, 32000, |
| 813 | 24000, 22050, 16000, 12000, 11025, 8000, 7350 |
| 814 | }; |
| 815 | |
| 816 | AVCodec ff_aac_encoder = { |
| 817 | .name = "aac", |
| 818 | .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"), |
| 819 | .type = AVMEDIA_TYPE_AUDIO, |
| 820 | .id = AV_CODEC_ID_AAC, |
| 821 | .priv_data_size = sizeof(AACEncContext), |
| 822 | .init = aac_encode_init, |
| 823 | .encode2 = aac_encode_frame, |
| 824 | .close = aac_encode_end, |
| 825 | .supported_samplerates = mpeg4audio_sample_rates, |
| 826 | .capabilities = CODEC_CAP_SMALL_LAST_FRAME | CODEC_CAP_DELAY | |
| 827 | CODEC_CAP_EXPERIMENTAL, |
| 828 | .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP, |
| 829 | AV_SAMPLE_FMT_NONE }, |
| 830 | .priv_class = &aacenc_class, |
| 831 | }; |