| 1 | /* |
| 2 | * Copyright (c) 2004 Michael Niedermayer <michaelni@gmx.at> |
| 3 | * Copyright (c) 2012 Justin Ruggles <justin.ruggles@gmail.com> |
| 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 | #include "libavutil/common.h" |
| 23 | #include "libavutil/libm.h" |
| 24 | #include "libavutil/log.h" |
| 25 | #include "internal.h" |
| 26 | #include "resample.h" |
| 27 | #include "audio_data.h" |
| 28 | |
| 29 | |
| 30 | /* double template */ |
| 31 | #define CONFIG_RESAMPLE_DBL |
| 32 | #include "resample_template.c" |
| 33 | #undef CONFIG_RESAMPLE_DBL |
| 34 | |
| 35 | /* float template */ |
| 36 | #define CONFIG_RESAMPLE_FLT |
| 37 | #include "resample_template.c" |
| 38 | #undef CONFIG_RESAMPLE_FLT |
| 39 | |
| 40 | /* s32 template */ |
| 41 | #define CONFIG_RESAMPLE_S32 |
| 42 | #include "resample_template.c" |
| 43 | #undef CONFIG_RESAMPLE_S32 |
| 44 | |
| 45 | /* s16 template */ |
| 46 | #include "resample_template.c" |
| 47 | |
| 48 | |
| 49 | /* 0th order modified bessel function of the first kind. */ |
| 50 | static double bessel(double x) |
| 51 | { |
| 52 | double v = 1; |
| 53 | double lastv = 0; |
| 54 | double t = 1; |
| 55 | int i; |
| 56 | |
| 57 | x = x * x / 4; |
| 58 | for (i = 1; v != lastv; i++) { |
| 59 | lastv = v; |
| 60 | t *= x / (i * i); |
| 61 | v += t; |
| 62 | } |
| 63 | return v; |
| 64 | } |
| 65 | |
| 66 | /* Build a polyphase filterbank. */ |
| 67 | static int build_filter(ResampleContext *c, double factor) |
| 68 | { |
| 69 | int ph, i; |
| 70 | double x, y, w; |
| 71 | double *tab; |
| 72 | int tap_count = c->filter_length; |
| 73 | int phase_count = 1 << c->phase_shift; |
| 74 | const int center = (tap_count - 1) / 2; |
| 75 | |
| 76 | tab = av_malloc(tap_count * sizeof(*tab)); |
| 77 | if (!tab) |
| 78 | return AVERROR(ENOMEM); |
| 79 | |
| 80 | for (ph = 0; ph < phase_count; ph++) { |
| 81 | double norm = 0; |
| 82 | for (i = 0; i < tap_count; i++) { |
| 83 | x = M_PI * ((double)(i - center) - (double)ph / phase_count) * factor; |
| 84 | if (x == 0) y = 1.0; |
| 85 | else y = sin(x) / x; |
| 86 | switch (c->filter_type) { |
| 87 | case AV_RESAMPLE_FILTER_TYPE_CUBIC: { |
| 88 | const float d = -0.5; //first order derivative = -0.5 |
| 89 | x = fabs(((double)(i - center) - (double)ph / phase_count) * factor); |
| 90 | if (x < 1.0) y = 1 - 3 * x*x + 2 * x*x*x + d * ( -x*x + x*x*x); |
| 91 | else y = d * (-4 + 8 * x - 5 * x*x + x*x*x); |
| 92 | break; |
| 93 | } |
| 94 | case AV_RESAMPLE_FILTER_TYPE_BLACKMAN_NUTTALL: |
| 95 | w = 2.0 * x / (factor * tap_count) + M_PI; |
| 96 | y *= 0.3635819 - 0.4891775 * cos( w) + |
| 97 | 0.1365995 * cos(2 * w) - |
| 98 | 0.0106411 * cos(3 * w); |
| 99 | break; |
| 100 | case AV_RESAMPLE_FILTER_TYPE_KAISER: |
| 101 | w = 2.0 * x / (factor * tap_count * M_PI); |
| 102 | y *= bessel(c->kaiser_beta * sqrt(FFMAX(1 - w * w, 0))); |
| 103 | break; |
| 104 | } |
| 105 | |
| 106 | tab[i] = y; |
| 107 | norm += y; |
| 108 | } |
| 109 | /* normalize so that an uniform color remains the same */ |
| 110 | for (i = 0; i < tap_count; i++) |
| 111 | tab[i] = tab[i] / norm; |
| 112 | |
| 113 | c->set_filter(c->filter_bank, tab, ph, tap_count); |
| 114 | } |
| 115 | |
| 116 | av_free(tab); |
| 117 | return 0; |
| 118 | } |
| 119 | |
| 120 | ResampleContext *ff_audio_resample_init(AVAudioResampleContext *avr) |
| 121 | { |
| 122 | ResampleContext *c; |
| 123 | int out_rate = avr->out_sample_rate; |
| 124 | int in_rate = avr->in_sample_rate; |
| 125 | double factor = FFMIN(out_rate * avr->cutoff / in_rate, 1.0); |
| 126 | int phase_count = 1 << avr->phase_shift; |
| 127 | int felem_size; |
| 128 | |
| 129 | if (avr->internal_sample_fmt != AV_SAMPLE_FMT_S16P && |
| 130 | avr->internal_sample_fmt != AV_SAMPLE_FMT_S32P && |
| 131 | avr->internal_sample_fmt != AV_SAMPLE_FMT_FLTP && |
| 132 | avr->internal_sample_fmt != AV_SAMPLE_FMT_DBLP) { |
| 133 | av_log(avr, AV_LOG_ERROR, "Unsupported internal format for " |
| 134 | "resampling: %s\n", |
| 135 | av_get_sample_fmt_name(avr->internal_sample_fmt)); |
| 136 | return NULL; |
| 137 | } |
| 138 | c = av_mallocz(sizeof(*c)); |
| 139 | if (!c) |
| 140 | return NULL; |
| 141 | |
| 142 | c->avr = avr; |
| 143 | c->phase_shift = avr->phase_shift; |
| 144 | c->phase_mask = phase_count - 1; |
| 145 | c->linear = avr->linear_interp; |
| 146 | c->filter_length = FFMAX((int)ceil(avr->filter_size / factor), 1); |
| 147 | c->filter_type = avr->filter_type; |
| 148 | c->kaiser_beta = avr->kaiser_beta; |
| 149 | |
| 150 | switch (avr->internal_sample_fmt) { |
| 151 | case AV_SAMPLE_FMT_DBLP: |
| 152 | c->resample_one = c->linear ? resample_linear_dbl : resample_one_dbl; |
| 153 | c->resample_nearest = resample_nearest_dbl; |
| 154 | c->set_filter = set_filter_dbl; |
| 155 | break; |
| 156 | case AV_SAMPLE_FMT_FLTP: |
| 157 | c->resample_one = c->linear ? resample_linear_flt : resample_one_flt; |
| 158 | c->resample_nearest = resample_nearest_flt; |
| 159 | c->set_filter = set_filter_flt; |
| 160 | break; |
| 161 | case AV_SAMPLE_FMT_S32P: |
| 162 | c->resample_one = c->linear ? resample_linear_s32 : resample_one_s32; |
| 163 | c->resample_nearest = resample_nearest_s32; |
| 164 | c->set_filter = set_filter_s32; |
| 165 | break; |
| 166 | case AV_SAMPLE_FMT_S16P: |
| 167 | c->resample_one = c->linear ? resample_linear_s16 : resample_one_s16; |
| 168 | c->resample_nearest = resample_nearest_s16; |
| 169 | c->set_filter = set_filter_s16; |
| 170 | break; |
| 171 | } |
| 172 | |
| 173 | if (ARCH_AARCH64) |
| 174 | ff_audio_resample_init_aarch64(c, avr->internal_sample_fmt); |
| 175 | |
| 176 | felem_size = av_get_bytes_per_sample(avr->internal_sample_fmt); |
| 177 | c->filter_bank = av_mallocz(c->filter_length * (phase_count + 1) * felem_size); |
| 178 | if (!c->filter_bank) |
| 179 | goto error; |
| 180 | |
| 181 | if (build_filter(c, factor) < 0) |
| 182 | goto error; |
| 183 | |
| 184 | memcpy(&c->filter_bank[(c->filter_length * phase_count + 1) * felem_size], |
| 185 | c->filter_bank, (c->filter_length - 1) * felem_size); |
| 186 | memcpy(&c->filter_bank[c->filter_length * phase_count * felem_size], |
| 187 | &c->filter_bank[(c->filter_length - 1) * felem_size], felem_size); |
| 188 | |
| 189 | c->compensation_distance = 0; |
| 190 | if (!av_reduce(&c->src_incr, &c->dst_incr, out_rate, |
| 191 | in_rate * (int64_t)phase_count, INT32_MAX / 2)) |
| 192 | goto error; |
| 193 | c->ideal_dst_incr = c->dst_incr; |
| 194 | |
| 195 | c->padding_size = (c->filter_length - 1) / 2; |
| 196 | c->initial_padding_filled = 0; |
| 197 | c->index = 0; |
| 198 | c->frac = 0; |
| 199 | |
| 200 | /* allocate internal buffer */ |
| 201 | c->buffer = ff_audio_data_alloc(avr->resample_channels, c->padding_size, |
| 202 | avr->internal_sample_fmt, |
| 203 | "resample buffer"); |
| 204 | if (!c->buffer) |
| 205 | goto error; |
| 206 | c->buffer->nb_samples = c->padding_size; |
| 207 | c->initial_padding_samples = c->padding_size; |
| 208 | |
| 209 | av_log(avr, AV_LOG_DEBUG, "resample: %s from %d Hz to %d Hz\n", |
| 210 | av_get_sample_fmt_name(avr->internal_sample_fmt), |
| 211 | avr->in_sample_rate, avr->out_sample_rate); |
| 212 | |
| 213 | return c; |
| 214 | |
| 215 | error: |
| 216 | ff_audio_data_free(&c->buffer); |
| 217 | av_free(c->filter_bank); |
| 218 | av_free(c); |
| 219 | return NULL; |
| 220 | } |
| 221 | |
| 222 | void ff_audio_resample_free(ResampleContext **c) |
| 223 | { |
| 224 | if (!*c) |
| 225 | return; |
| 226 | ff_audio_data_free(&(*c)->buffer); |
| 227 | av_free((*c)->filter_bank); |
| 228 | av_freep(c); |
| 229 | } |
| 230 | |
| 231 | int avresample_set_compensation(AVAudioResampleContext *avr, int sample_delta, |
| 232 | int compensation_distance) |
| 233 | { |
| 234 | ResampleContext *c; |
| 235 | AudioData *fifo_buf = NULL; |
| 236 | int ret = 0; |
| 237 | |
| 238 | if (compensation_distance < 0) |
| 239 | return AVERROR(EINVAL); |
| 240 | if (!compensation_distance && sample_delta) |
| 241 | return AVERROR(EINVAL); |
| 242 | |
| 243 | if (!avr->resample_needed) { |
| 244 | #if FF_API_RESAMPLE_CLOSE_OPEN |
| 245 | /* if resampling was not enabled previously, re-initialize the |
| 246 | AVAudioResampleContext and force resampling */ |
| 247 | int fifo_samples; |
| 248 | int restore_matrix = 0; |
| 249 | double matrix[AVRESAMPLE_MAX_CHANNELS * AVRESAMPLE_MAX_CHANNELS] = { 0 }; |
| 250 | |
| 251 | /* buffer any remaining samples in the output FIFO before closing */ |
| 252 | fifo_samples = av_audio_fifo_size(avr->out_fifo); |
| 253 | if (fifo_samples > 0) { |
| 254 | fifo_buf = ff_audio_data_alloc(avr->out_channels, fifo_samples, |
| 255 | avr->out_sample_fmt, NULL); |
| 256 | if (!fifo_buf) |
| 257 | return AVERROR(EINVAL); |
| 258 | ret = ff_audio_data_read_from_fifo(avr->out_fifo, fifo_buf, |
| 259 | fifo_samples); |
| 260 | if (ret < 0) |
| 261 | goto reinit_fail; |
| 262 | } |
| 263 | /* save the channel mixing matrix */ |
| 264 | if (avr->am) { |
| 265 | ret = avresample_get_matrix(avr, matrix, AVRESAMPLE_MAX_CHANNELS); |
| 266 | if (ret < 0) |
| 267 | goto reinit_fail; |
| 268 | restore_matrix = 1; |
| 269 | } |
| 270 | |
| 271 | /* close the AVAudioResampleContext */ |
| 272 | avresample_close(avr); |
| 273 | |
| 274 | avr->force_resampling = 1; |
| 275 | |
| 276 | /* restore the channel mixing matrix */ |
| 277 | if (restore_matrix) { |
| 278 | ret = avresample_set_matrix(avr, matrix, AVRESAMPLE_MAX_CHANNELS); |
| 279 | if (ret < 0) |
| 280 | goto reinit_fail; |
| 281 | } |
| 282 | |
| 283 | /* re-open the AVAudioResampleContext */ |
| 284 | ret = avresample_open(avr); |
| 285 | if (ret < 0) |
| 286 | goto reinit_fail; |
| 287 | |
| 288 | /* restore buffered samples to the output FIFO */ |
| 289 | if (fifo_samples > 0) { |
| 290 | ret = ff_audio_data_add_to_fifo(avr->out_fifo, fifo_buf, 0, |
| 291 | fifo_samples); |
| 292 | if (ret < 0) |
| 293 | goto reinit_fail; |
| 294 | ff_audio_data_free(&fifo_buf); |
| 295 | } |
| 296 | #else |
| 297 | av_log(avr, AV_LOG_ERROR, "Unable to set resampling compensation\n"); |
| 298 | return AVERROR(EINVAL); |
| 299 | #endif |
| 300 | } |
| 301 | c = avr->resample; |
| 302 | c->compensation_distance = compensation_distance; |
| 303 | if (compensation_distance) { |
| 304 | c->dst_incr = c->ideal_dst_incr - c->ideal_dst_incr * |
| 305 | (int64_t)sample_delta / compensation_distance; |
| 306 | } else { |
| 307 | c->dst_incr = c->ideal_dst_incr; |
| 308 | } |
| 309 | return 0; |
| 310 | |
| 311 | reinit_fail: |
| 312 | ff_audio_data_free(&fifo_buf); |
| 313 | return ret; |
| 314 | } |
| 315 | |
| 316 | static int resample(ResampleContext *c, void *dst, const void *src, |
| 317 | int *consumed, int src_size, int dst_size, int update_ctx, |
| 318 | int nearest_neighbour) |
| 319 | { |
| 320 | int dst_index; |
| 321 | unsigned int index = c->index; |
| 322 | int frac = c->frac; |
| 323 | int dst_incr_frac = c->dst_incr % c->src_incr; |
| 324 | int dst_incr = c->dst_incr / c->src_incr; |
| 325 | int compensation_distance = c->compensation_distance; |
| 326 | |
| 327 | if (!dst != !src) |
| 328 | return AVERROR(EINVAL); |
| 329 | |
| 330 | if (nearest_neighbour) { |
| 331 | uint64_t index2 = ((uint64_t)index) << 32; |
| 332 | int64_t incr = (1LL << 32) * c->dst_incr / c->src_incr; |
| 333 | dst_size = FFMIN(dst_size, |
| 334 | (src_size-1-index) * (int64_t)c->src_incr / |
| 335 | c->dst_incr); |
| 336 | |
| 337 | if (dst) { |
| 338 | for(dst_index = 0; dst_index < dst_size; dst_index++) { |
| 339 | c->resample_nearest(dst, dst_index, src, index2 >> 32); |
| 340 | index2 += incr; |
| 341 | } |
| 342 | } else { |
| 343 | dst_index = dst_size; |
| 344 | } |
| 345 | index += dst_index * dst_incr; |
| 346 | index += (frac + dst_index * (int64_t)dst_incr_frac) / c->src_incr; |
| 347 | frac = (frac + dst_index * (int64_t)dst_incr_frac) % c->src_incr; |
| 348 | } else { |
| 349 | for (dst_index = 0; dst_index < dst_size; dst_index++) { |
| 350 | int sample_index = index >> c->phase_shift; |
| 351 | |
| 352 | if (sample_index + c->filter_length > src_size) |
| 353 | break; |
| 354 | |
| 355 | if (dst) |
| 356 | c->resample_one(c, dst, dst_index, src, index, frac); |
| 357 | |
| 358 | frac += dst_incr_frac; |
| 359 | index += dst_incr; |
| 360 | if (frac >= c->src_incr) { |
| 361 | frac -= c->src_incr; |
| 362 | index++; |
| 363 | } |
| 364 | if (dst_index + 1 == compensation_distance) { |
| 365 | compensation_distance = 0; |
| 366 | dst_incr_frac = c->ideal_dst_incr % c->src_incr; |
| 367 | dst_incr = c->ideal_dst_incr / c->src_incr; |
| 368 | } |
| 369 | } |
| 370 | } |
| 371 | if (consumed) |
| 372 | *consumed = index >> c->phase_shift; |
| 373 | |
| 374 | if (update_ctx) { |
| 375 | index &= c->phase_mask; |
| 376 | |
| 377 | if (compensation_distance) { |
| 378 | compensation_distance -= dst_index; |
| 379 | if (compensation_distance <= 0) |
| 380 | return AVERROR_BUG; |
| 381 | } |
| 382 | c->frac = frac; |
| 383 | c->index = index; |
| 384 | c->dst_incr = dst_incr_frac + c->src_incr*dst_incr; |
| 385 | c->compensation_distance = compensation_distance; |
| 386 | } |
| 387 | |
| 388 | return dst_index; |
| 389 | } |
| 390 | |
| 391 | int ff_audio_resample(ResampleContext *c, AudioData *dst, AudioData *src) |
| 392 | { |
| 393 | int ch, in_samples, in_leftover, consumed = 0, out_samples = 0; |
| 394 | int ret = AVERROR(EINVAL); |
| 395 | int nearest_neighbour = (c->compensation_distance == 0 && |
| 396 | c->filter_length == 1 && |
| 397 | c->phase_shift == 0); |
| 398 | |
| 399 | in_samples = src ? src->nb_samples : 0; |
| 400 | in_leftover = c->buffer->nb_samples; |
| 401 | |
| 402 | /* add input samples to the internal buffer */ |
| 403 | if (src) { |
| 404 | ret = ff_audio_data_combine(c->buffer, in_leftover, src, 0, in_samples); |
| 405 | if (ret < 0) |
| 406 | return ret; |
| 407 | } else if (in_leftover <= c->final_padding_samples) { |
| 408 | /* no remaining samples to flush */ |
| 409 | return 0; |
| 410 | } |
| 411 | |
| 412 | if (!c->initial_padding_filled) { |
| 413 | int bps = av_get_bytes_per_sample(c->avr->internal_sample_fmt); |
| 414 | int i; |
| 415 | |
| 416 | if (src && c->buffer->nb_samples < 2 * c->padding_size) |
| 417 | return 0; |
| 418 | |
| 419 | for (i = 0; i < c->padding_size; i++) |
| 420 | for (ch = 0; ch < c->buffer->channels; ch++) { |
| 421 | if (c->buffer->nb_samples > 2 * c->padding_size - i) { |
| 422 | memcpy(c->buffer->data[ch] + bps * i, |
| 423 | c->buffer->data[ch] + bps * (2 * c->padding_size - i), bps); |
| 424 | } else { |
| 425 | memset(c->buffer->data[ch] + bps * i, 0, bps); |
| 426 | } |
| 427 | } |
| 428 | c->initial_padding_filled = 1; |
| 429 | } |
| 430 | |
| 431 | if (!src && !c->final_padding_filled) { |
| 432 | int bps = av_get_bytes_per_sample(c->avr->internal_sample_fmt); |
| 433 | int i; |
| 434 | |
| 435 | ret = ff_audio_data_realloc(c->buffer, in_samples + c->padding_size); |
| 436 | if (ret < 0) { |
| 437 | av_log(c->avr, AV_LOG_ERROR, "Error reallocating resampling buffer\n"); |
| 438 | return AVERROR(ENOMEM); |
| 439 | } |
| 440 | |
| 441 | for (i = 0; i < c->padding_size; i++) |
| 442 | for (ch = 0; ch < c->buffer->channels; ch++) { |
| 443 | if (in_leftover > i) { |
| 444 | memcpy(c->buffer->data[ch] + bps * (in_leftover + i), |
| 445 | c->buffer->data[ch] + bps * (in_leftover - i - 1), |
| 446 | bps); |
| 447 | } else { |
| 448 | memset(c->buffer->data[ch] + bps * (in_leftover + i), |
| 449 | 0, bps); |
| 450 | } |
| 451 | } |
| 452 | c->buffer->nb_samples += c->padding_size; |
| 453 | c->final_padding_samples = c->padding_size; |
| 454 | c->final_padding_filled = 1; |
| 455 | } |
| 456 | |
| 457 | |
| 458 | /* calculate output size and reallocate output buffer if needed */ |
| 459 | /* TODO: try to calculate this without the dummy resample() run */ |
| 460 | if (!dst->read_only && dst->allow_realloc) { |
| 461 | out_samples = resample(c, NULL, NULL, NULL, c->buffer->nb_samples, |
| 462 | INT_MAX, 0, nearest_neighbour); |
| 463 | ret = ff_audio_data_realloc(dst, out_samples); |
| 464 | if (ret < 0) { |
| 465 | av_log(c->avr, AV_LOG_ERROR, "error reallocating output\n"); |
| 466 | return ret; |
| 467 | } |
| 468 | } |
| 469 | |
| 470 | /* resample each channel plane */ |
| 471 | for (ch = 0; ch < c->buffer->channels; ch++) { |
| 472 | out_samples = resample(c, (void *)dst->data[ch], |
| 473 | (const void *)c->buffer->data[ch], &consumed, |
| 474 | c->buffer->nb_samples, dst->allocated_samples, |
| 475 | ch + 1 == c->buffer->channels, nearest_neighbour); |
| 476 | } |
| 477 | if (out_samples < 0) { |
| 478 | av_log(c->avr, AV_LOG_ERROR, "error during resampling\n"); |
| 479 | return out_samples; |
| 480 | } |
| 481 | |
| 482 | /* drain consumed samples from the internal buffer */ |
| 483 | ff_audio_data_drain(c->buffer, consumed); |
| 484 | c->initial_padding_samples = FFMAX(c->initial_padding_samples - consumed, 0); |
| 485 | |
| 486 | av_dlog(c->avr, "resampled %d in + %d leftover to %d out + %d leftover\n", |
| 487 | in_samples, in_leftover, out_samples, c->buffer->nb_samples); |
| 488 | |
| 489 | dst->nb_samples = out_samples; |
| 490 | return 0; |
| 491 | } |
| 492 | |
| 493 | int avresample_get_delay(AVAudioResampleContext *avr) |
| 494 | { |
| 495 | ResampleContext *c = avr->resample; |
| 496 | |
| 497 | if (!avr->resample_needed || !avr->resample) |
| 498 | return 0; |
| 499 | |
| 500 | return FFMAX(c->buffer->nb_samples - c->padding_size, 0); |
| 501 | } |