2 * Copyright (c) 2004 Michael Niedermayer <michaelni@gmx.at>
3 * Copyright (c) 2012 Justin Ruggles <justin.ruggles@gmail.com>
5 * This file is part of Libav.
7 * Libav 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.
12 * Libav 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.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 #include "libavutil/libm.h"
23 #include "libavutil/log.h"
25 #include "audio_data.h"
27 struct ResampleContext {
28 AVAudioResampleContext *avr;
37 int compensation_distance;
41 enum AVResampleFilterType filter_type;
44 void (*set_filter)(void *filter, double *tab, int phase, int tap_count);
45 void (*resample_one)(struct ResampleContext *c, int no_filter, void *dst0,
46 int dst_index, const void *src0, int src_size,
52 #define CONFIG_RESAMPLE_DBL
53 #include "resample_template.c"
54 #undef CONFIG_RESAMPLE_DBL
57 #define CONFIG_RESAMPLE_FLT
58 #include "resample_template.c"
59 #undef CONFIG_RESAMPLE_FLT
62 #define CONFIG_RESAMPLE_S32
63 #include "resample_template.c"
64 #undef CONFIG_RESAMPLE_S32
67 #include "resample_template.c"
71 * 0th order modified bessel function of the first kind.
73 static double bessel(double x)
81 for (i = 1; v != lastv; i++) {
90 * Build a polyphase filterbank.
92 * @param[out] filter filter coefficients
93 * @param factor resampling factor
94 * @param tap_count tap count
95 * @param phase_count phase count
96 * @param scale wanted sum of coefficients for each filter
97 * @param filter_type filter type
98 * @param kaiser_beta kaiser window beta
99 * @return 0 on success, negative AVERROR code on failure
101 static int build_filter(ResampleContext *c)
104 double x, y, w, factor;
106 int tap_count = c->filter_length;
107 int phase_count = 1 << c->phase_shift;
108 const int center = (tap_count - 1) / 2;
110 tab = av_malloc(tap_count * sizeof(*tab));
112 return AVERROR(ENOMEM);
114 /* if upsampling, only need to interpolate, no filter */
115 factor = FFMIN(c->factor, 1.0);
117 for (ph = 0; ph < phase_count; ph++) {
119 for (i = 0; i < tap_count; i++) {
120 x = M_PI * ((double)(i - center) - (double)ph / phase_count) * factor;
123 switch (c->filter_type) {
124 case AV_RESAMPLE_FILTER_TYPE_CUBIC: {
125 const float d = -0.5; //first order derivative = -0.5
126 x = fabs(((double)(i - center) - (double)ph / phase_count) * factor);
127 if (x < 1.0) y = 1 - 3 * x*x + 2 * x*x*x + d * ( -x*x + x*x*x);
128 else y = d * (-4 + 8 * x - 5 * x*x + x*x*x);
131 case AV_RESAMPLE_FILTER_TYPE_BLACKMAN_NUTTALL:
132 w = 2.0 * x / (factor * tap_count) + M_PI;
133 y *= 0.3635819 - 0.4891775 * cos( w) +
134 0.1365995 * cos(2 * w) -
135 0.0106411 * cos(3 * w);
137 case AV_RESAMPLE_FILTER_TYPE_KAISER:
138 w = 2.0 * x / (factor * tap_count * M_PI);
139 y *= bessel(c->kaiser_beta * sqrt(FFMAX(1 - w * w, 0)));
146 /* normalize so that an uniform color remains the same */
147 for (i = 0; i < tap_count; i++)
148 tab[i] = tab[i] / norm;
150 c->set_filter(c->filter_bank, tab, ph, tap_count);
157 ResampleContext *ff_audio_resample_init(AVAudioResampleContext *avr)
160 int out_rate = avr->out_sample_rate;
161 int in_rate = avr->in_sample_rate;
162 double factor = FFMIN(out_rate * avr->cutoff / in_rate, 1.0);
163 int phase_count = 1 << avr->phase_shift;
166 if (avr->internal_sample_fmt != AV_SAMPLE_FMT_S16P &&
167 avr->internal_sample_fmt != AV_SAMPLE_FMT_S32P &&
168 avr->internal_sample_fmt != AV_SAMPLE_FMT_FLTP &&
169 avr->internal_sample_fmt != AV_SAMPLE_FMT_DBLP) {
170 av_log(avr, AV_LOG_ERROR, "Unsupported internal format for "
172 av_get_sample_fmt_name(avr->internal_sample_fmt));
175 c = av_mallocz(sizeof(*c));
180 c->phase_shift = avr->phase_shift;
181 c->phase_mask = phase_count - 1;
182 c->linear = avr->linear_interp;
184 c->filter_length = FFMAX((int)ceil(avr->filter_size / factor), 1);
185 c->filter_type = avr->filter_type;
186 c->kaiser_beta = avr->kaiser_beta;
188 switch (avr->internal_sample_fmt) {
189 case AV_SAMPLE_FMT_DBLP:
190 c->resample_one = resample_one_dbl;
191 c->set_filter = set_filter_dbl;
193 case AV_SAMPLE_FMT_FLTP:
194 c->resample_one = resample_one_flt;
195 c->set_filter = set_filter_flt;
197 case AV_SAMPLE_FMT_S32P:
198 c->resample_one = resample_one_s32;
199 c->set_filter = set_filter_s32;
201 case AV_SAMPLE_FMT_S16P:
202 c->resample_one = resample_one_s16;
203 c->set_filter = set_filter_s16;
207 felem_size = av_get_bytes_per_sample(avr->internal_sample_fmt);
208 c->filter_bank = av_mallocz(c->filter_length * (phase_count + 1) * felem_size);
212 if (build_filter(c) < 0)
215 memcpy(&c->filter_bank[(c->filter_length * phase_count + 1) * felem_size],
216 c->filter_bank, (c->filter_length - 1) * felem_size);
217 memcpy(&c->filter_bank[c->filter_length * phase_count * felem_size],
218 &c->filter_bank[(c->filter_length - 1) * felem_size], felem_size);
220 c->compensation_distance = 0;
221 if (!av_reduce(&c->src_incr, &c->dst_incr, out_rate,
222 in_rate * (int64_t)phase_count, INT32_MAX / 2))
224 c->ideal_dst_incr = c->dst_incr;
226 c->index = -phase_count * ((c->filter_length - 1) / 2);
229 /* allocate internal buffer */
230 c->buffer = ff_audio_data_alloc(avr->resample_channels, 0,
231 avr->internal_sample_fmt,
236 av_log(avr, AV_LOG_DEBUG, "resample: %s from %d Hz to %d Hz\n",
237 av_get_sample_fmt_name(avr->internal_sample_fmt),
238 avr->in_sample_rate, avr->out_sample_rate);
243 ff_audio_data_free(&c->buffer);
244 av_free(c->filter_bank);
249 void ff_audio_resample_free(ResampleContext **c)
253 ff_audio_data_free(&(*c)->buffer);
254 av_free((*c)->filter_bank);
258 int avresample_set_compensation(AVAudioResampleContext *avr, int sample_delta,
259 int compensation_distance)
262 AudioData *fifo_buf = NULL;
265 if (compensation_distance < 0)
266 return AVERROR(EINVAL);
267 if (!compensation_distance && sample_delta)
268 return AVERROR(EINVAL);
270 /* if resampling was not enabled previously, re-initialize the
271 AVAudioResampleContext and force resampling */
272 if (!avr->resample_needed) {
274 double matrix[AVRESAMPLE_MAX_CHANNELS * AVRESAMPLE_MAX_CHANNELS] = { 0 };
276 /* buffer any remaining samples in the output FIFO before closing */
277 fifo_samples = av_audio_fifo_size(avr->out_fifo);
278 if (fifo_samples > 0) {
279 fifo_buf = ff_audio_data_alloc(avr->out_channels, fifo_samples,
280 avr->out_sample_fmt, NULL);
282 return AVERROR(EINVAL);
283 ret = ff_audio_data_read_from_fifo(avr->out_fifo, fifo_buf,
288 /* save the channel mixing matrix */
289 ret = avresample_get_matrix(avr, matrix, AVRESAMPLE_MAX_CHANNELS);
293 /* close the AVAudioResampleContext */
294 avresample_close(avr);
296 avr->force_resampling = 1;
298 /* restore the channel mixing matrix */
299 ret = avresample_set_matrix(avr, matrix, AVRESAMPLE_MAX_CHANNELS);
303 /* re-open the AVAudioResampleContext */
304 ret = avresample_open(avr);
308 /* restore buffered samples to the output FIFO */
309 if (fifo_samples > 0) {
310 ret = ff_audio_data_add_to_fifo(avr->out_fifo, fifo_buf, 0,
314 ff_audio_data_free(&fifo_buf);
318 c->compensation_distance = compensation_distance;
319 if (compensation_distance) {
320 c->dst_incr = c->ideal_dst_incr - c->ideal_dst_incr *
321 (int64_t)sample_delta / compensation_distance;
323 c->dst_incr = c->ideal_dst_incr;
328 ff_audio_data_free(&fifo_buf);
332 static int resample(ResampleContext *c, void *dst, const void *src,
333 int *consumed, int src_size, int dst_size, int update_ctx)
336 int index = c->index;
338 int dst_incr_frac = c->dst_incr % c->src_incr;
339 int dst_incr = c->dst_incr / c->src_incr;
340 int compensation_distance = c->compensation_distance;
343 return AVERROR(EINVAL);
345 if (compensation_distance == 0 && c->filter_length == 1 &&
346 c->phase_shift == 0) {
347 int64_t index2 = ((int64_t)index) << 32;
348 int64_t incr = (1LL << 32) * c->dst_incr / c->src_incr;
349 dst_size = FFMIN(dst_size,
350 (src_size-1-index) * (int64_t)c->src_incr /
354 for(dst_index = 0; dst_index < dst_size; dst_index++) {
355 c->resample_one(c, 1, dst, dst_index, src, 0, index2 >> 32, 0);
359 dst_index = dst_size;
361 index += dst_index * dst_incr;
362 index += (frac + dst_index * (int64_t)dst_incr_frac) / c->src_incr;
363 frac = (frac + dst_index * (int64_t)dst_incr_frac) % c->src_incr;
365 for (dst_index = 0; dst_index < dst_size; dst_index++) {
366 int sample_index = index >> c->phase_shift;
368 if (sample_index + c->filter_length > src_size ||
369 -sample_index >= src_size)
373 c->resample_one(c, 0, dst, dst_index, src, src_size, index, frac);
375 frac += dst_incr_frac;
377 if (frac >= c->src_incr) {
381 if (dst_index + 1 == compensation_distance) {
382 compensation_distance = 0;
383 dst_incr_frac = c->ideal_dst_incr % c->src_incr;
384 dst_incr = c->ideal_dst_incr / c->src_incr;
389 *consumed = FFMAX(index, 0) >> c->phase_shift;
393 index &= c->phase_mask;
395 if (compensation_distance) {
396 compensation_distance -= dst_index;
397 if (compensation_distance <= 0)
402 c->dst_incr = dst_incr_frac + c->src_incr*dst_incr;
403 c->compensation_distance = compensation_distance;
409 int ff_audio_resample(ResampleContext *c, AudioData *dst, AudioData *src,
412 int ch, in_samples, in_leftover, out_samples = 0;
413 int ret = AVERROR(EINVAL);
415 in_samples = src ? src->nb_samples : 0;
416 in_leftover = c->buffer->nb_samples;
418 /* add input samples to the internal buffer */
420 ret = ff_audio_data_combine(c->buffer, in_leftover, src, 0, in_samples);
423 } else if (!in_leftover) {
424 /* no remaining samples to flush */
427 /* TODO: pad buffer to flush completely */
430 /* calculate output size and reallocate output buffer if needed */
431 /* TODO: try to calculate this without the dummy resample() run */
432 if (!dst->read_only && dst->allow_realloc) {
433 out_samples = resample(c, NULL, NULL, NULL, c->buffer->nb_samples,
435 ret = ff_audio_data_realloc(dst, out_samples);
437 av_log(c->avr, AV_LOG_ERROR, "error reallocating output\n");
442 /* resample each channel plane */
443 for (ch = 0; ch < c->buffer->channels; ch++) {
444 out_samples = resample(c, (void *)dst->data[ch],
445 (const void *)c->buffer->data[ch], consumed,
446 c->buffer->nb_samples, dst->allocated_samples,
447 ch + 1 == c->buffer->channels);
449 if (out_samples < 0) {
450 av_log(c->avr, AV_LOG_ERROR, "error during resampling\n");
454 /* drain consumed samples from the internal buffer */
455 ff_audio_data_drain(c->buffer, *consumed);
457 av_dlog(c->avr, "resampled %d in + %d leftover to %d out + %d leftover\n",
458 in_samples, in_leftover, out_samples, c->buffer->nb_samples);
460 dst->nb_samples = out_samples;
464 int avresample_get_delay(AVAudioResampleContext *avr)
466 if (!avr->resample_needed || !avr->resample)
469 return avr->resample->buffer->nb_samples;