3 * Copyright (c) 2004-2012 Michael Niedermayer <michaelni@gmx.at>
4 * bessel function: Copyright (c) 2006 Xiaogang Zhang
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 * @author Michael Niedermayer <michaelni@gmx.at>
29 #include "libavutil/avassert.h"
32 static inline double eval_poly(const double *coeff, int size, double x) {
33 double sum = coeff[size-1];
35 for (i = size-2; i >= 0; --i) {
43 * 0th order modified bessel function of the first kind.
44 * Algorithm taken from the Boost project, source:
45 * https://searchcode.com/codesearch/view/14918379/
46 * Use, modification and distribution are subject to the
47 * Boost Software License, Version 1.0 (see notice below).
48 * Boost Software License - Version 1.0 - August 17th, 2003
49 Permission is hereby granted, free of charge, to any person or organization
50 obtaining a copy of the software and accompanying documentation covered by
51 this license (the "Software") to use, reproduce, display, distribute,
52 execute, and transmit the Software, and to prepare derivative works of the
53 Software, and to permit third-parties to whom the Software is furnished to
54 do so, all subject to the following:
56 The copyright notices in the Software and this entire statement, including
57 the above license grant, this restriction and the following disclaimer,
58 must be included in all copies of the Software, in whole or in part, and
59 all derivative works of the Software, unless such copies or derivative
60 works are solely in the form of machine-executable object code generated by
61 a source language processor.
63 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
64 IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
65 FITNESS FOR A PARTICULAR PURPOSE, TITLE AND NON-INFRINGEMENT. IN NO EVENT
66 SHALL THE COPYRIGHT HOLDERS OR ANYONE DISTRIBUTING THE SOFTWARE BE LIABLE
67 FOR ANY DAMAGES OR OTHER LIABILITY, WHETHER IN CONTRACT, TORT OR OTHERWISE,
68 ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
69 DEALINGS IN THE SOFTWARE.
72 static double bessel(double x) {
73 // Modified Bessel function of the first kind of order zero
74 // minimax rational approximations on intervals, see
75 // Blair and Edwards, Chalk River Report AECL-4928, 1974
76 static const double p1[] = {
77 -2.2335582639474375249e+15,
78 -5.5050369673018427753e+14,
79 -3.2940087627407749166e+13,
80 -8.4925101247114157499e+11,
81 -1.1912746104985237192e+10,
82 -1.0313066708737980747e+08,
83 -5.9545626019847898221e+05,
84 -2.4125195876041896775e+03,
85 -7.0935347449210549190e+00,
86 -1.5453977791786851041e-02,
87 -2.5172644670688975051e-05,
88 -3.0517226450451067446e-08,
89 -2.6843448573468483278e-11,
90 -1.5982226675653184646e-14,
91 -5.2487866627945699800e-18,
93 static const double q1[] = {
94 -2.2335582639474375245e+15,
95 7.8858692566751002988e+12,
96 -1.2207067397808979846e+10,
97 1.0377081058062166144e+07,
98 -4.8527560179962773045e+03,
101 static const double p2[] = {
102 -2.2210262233306573296e-04,
103 1.3067392038106924055e-02,
104 -4.4700805721174453923e-01,
105 5.5674518371240761397e+00,
106 -2.3517945679239481621e+01,
107 3.1611322818701131207e+01,
108 -9.6090021968656180000e+00,
110 static const double q2[] = {
111 -5.5194330231005480228e-04,
112 3.2547697594819615062e-02,
113 -1.1151759188741312645e+00,
114 1.3982595353892851542e+01,
115 -6.0228002066743340583e+01,
116 8.5539563258012929600e+01,
117 -3.1446690275135491500e+01,
126 return eval_poly(p1, FF_ARRAY_ELEMS(p1), y) / eval_poly(q1, FF_ARRAY_ELEMS(q1), y);
129 y = 1 / x - 1.0 / 15;
130 r = eval_poly(p2, FF_ARRAY_ELEMS(p2), y) / eval_poly(q2, FF_ARRAY_ELEMS(q2), y);
131 factor = exp(x) / sqrt(x);
137 * builds a polyphase filterbank.
138 * @param factor resampling factor
139 * @param scale wanted sum of coefficients for each filter
140 * @param filter_type filter type
141 * @param kaiser_beta kaiser window beta
142 * @return 0 on success, negative on error
144 static int build_filter(ResampleContext *c, void *filter, double factor, int tap_count, int alloc, int phase_count, int scale,
145 int filter_type, double kaiser_beta){
147 int ph_nb = phase_count % 2 ? phase_count : phase_count / 2 + 1;
148 double x, y, w, t, s;
149 double *tab = av_malloc_array(tap_count+1, sizeof(*tab));
150 double *sin_lut = av_malloc_array(ph_nb, sizeof(*sin_lut));
151 const int center= (tap_count-1)/2;
153 int ret = AVERROR(ENOMEM);
155 if (!tab || !sin_lut)
158 av_assert0(tap_count == 1 || tap_count % 2 == 0);
160 /* if upsampling, only need to interpolate, no filter */
165 for (ph = 0; ph < ph_nb; ph++)
166 sin_lut[ph] = sin(M_PI * ph / phase_count) * (center & 1 ? 1 : -1);
168 for(ph = 0; ph < ph_nb; ph++) {
170 for(i=0;i<tap_count;i++) {
171 x = M_PI * ((double)(i - center) - (double)ph / phase_count) * factor;
173 else if (factor == 1.0)
178 case SWR_FILTER_TYPE_CUBIC:{
179 const float d= -0.5; //first order derivative = -0.5
180 x = fabs(((double)(i - center) - (double)ph / phase_count) * factor);
181 if(x<1.0) y= 1 - 3*x*x + 2*x*x*x + d*( -x*x + x*x*x);
182 else y= d*(-4 + 8*x - 5*x*x + x*x*x);
184 case SWR_FILTER_TYPE_BLACKMAN_NUTTALL:
185 w = 2.0*x / (factor*tap_count);
187 y *= 0.3635819 - 0.4891775 * t + 0.1365995 * (2*t*t-1) - 0.0106411 * (4*t*t*t - 3*t);
189 case SWR_FILTER_TYPE_KAISER:
190 w = 2.0*x / (factor*tap_count*M_PI);
191 y *= bessel(kaiser_beta*sqrt(FFMAX(1-w*w, 0)));
203 /* normalize so that an uniform color remains the same */
205 case AV_SAMPLE_FMT_S16P:
206 for(i=0;i<tap_count;i++)
207 ((int16_t*)filter)[ph * alloc + i] = av_clip_int16(lrintf(tab[i] * scale / norm));
208 if (phase_count % 2) break;
209 for (i = 0; i < tap_count; i++)
210 ((int16_t*)filter)[(phase_count-ph) * alloc + tap_count-1-i] = ((int16_t*)filter)[ph * alloc + i];
212 case AV_SAMPLE_FMT_S32P:
213 for(i=0;i<tap_count;i++)
214 ((int32_t*)filter)[ph * alloc + i] = av_clipl_int32(llrint(tab[i] * scale / norm));
215 if (phase_count % 2) break;
216 for (i = 0; i < tap_count; i++)
217 ((int32_t*)filter)[(phase_count-ph) * alloc + tap_count-1-i] = ((int32_t*)filter)[ph * alloc + i];
219 case AV_SAMPLE_FMT_FLTP:
220 for(i=0;i<tap_count;i++)
221 ((float*)filter)[ph * alloc + i] = tab[i] * scale / norm;
222 if (phase_count % 2) break;
223 for (i = 0; i < tap_count; i++)
224 ((float*)filter)[(phase_count-ph) * alloc + tap_count-1-i] = ((float*)filter)[ph * alloc + i];
226 case AV_SAMPLE_FMT_DBLP:
227 for(i=0;i<tap_count;i++)
228 ((double*)filter)[ph * alloc + i] = tab[i] * scale / norm;
229 if (phase_count % 2) break;
230 for (i = 0; i < tap_count; i++)
231 ((double*)filter)[(phase_count-ph) * alloc + tap_count-1-i] = ((double*)filter)[ph * alloc + i];
239 double sine[LEN + tap_count];
240 double filtered[LEN];
241 double maxff=-2, minff=2, maxsf=-2, minsf=2;
242 for(i=0; i<LEN; i++){
243 double ss=0, sf=0, ff=0;
244 for(j=0; j<LEN+tap_count; j++)
245 sine[j]= cos(i*j*M_PI/LEN);
246 for(j=0; j<LEN; j++){
249 for(k=0; k<tap_count; k++)
250 sum += filter[ph * tap_count + k] * sine[k+j];
251 filtered[j]= sum / (1<<FILTER_SHIFT);
252 ss+= sine[j + center] * sine[j + center];
253 ff+= filtered[j] * filtered[j];
254 sf+= sine[j + center] * filtered[j];
259 maxff= FFMAX(maxff, ff);
260 minff= FFMIN(minff, ff);
261 maxsf= FFMAX(maxsf, sf);
262 minsf= FFMIN(minsf, sf);
264 av_log(NULL, AV_LOG_ERROR, "i:%4d ss:%f ff:%13.6e-%13.6e sf:%13.6e-%13.6e\n", i, ss, maxff, minff, maxsf, minsf);
279 static void resample_free(ResampleContext **cc){
280 ResampleContext *c = *cc;
283 av_freep(&c->filter_bank);
287 static ResampleContext *resample_init(ResampleContext *c, int out_rate, int in_rate, int filter_size, int phase_shift, int linear,
288 double cutoff0, enum AVSampleFormat format, enum SwrFilterType filter_type, double kaiser_beta,
289 double precision, int cheby, int exact_rational)
291 double cutoff = cutoff0? cutoff0 : 0.97;
292 double factor= FFMIN(out_rate * cutoff / in_rate, 1.0);
293 int phase_count= 1<<phase_shift;
294 int phase_count_compensation = phase_count;
295 int filter_length = FFMAX((int)ceil(filter_size/factor), 1);
297 if (filter_length > 1)
298 filter_length = FFALIGN(filter_length, 2);
300 if (exact_rational) {
301 int phase_count_exact, phase_count_exact_den;
303 av_reduce(&phase_count_exact, &phase_count_exact_den, out_rate, in_rate, INT_MAX);
304 if (phase_count_exact <= phase_count) {
305 phase_count_compensation = phase_count_exact * (phase_count / phase_count_exact);
306 phase_count = phase_count_exact;
310 if (!c || c->phase_count != phase_count || c->linear!=linear || c->factor != factor
311 || c->filter_length != filter_length || c->format != format
312 || c->filter_type != filter_type || c->kaiser_beta != kaiser_beta) {
314 c = av_mallocz(sizeof(*c));
320 c->felem_size= av_get_bytes_per_sample(c->format);
323 case AV_SAMPLE_FMT_S16P:
324 c->filter_shift = 15;
326 case AV_SAMPLE_FMT_S32P:
327 c->filter_shift = 30;
329 case AV_SAMPLE_FMT_FLTP:
330 case AV_SAMPLE_FMT_DBLP:
334 av_log(NULL, AV_LOG_ERROR, "Unsupported sample format\n");
338 if (filter_size/factor > INT32_MAX/256) {
339 av_log(NULL, AV_LOG_ERROR, "Filter length too large\n");
343 c->phase_count = phase_count;
346 c->filter_length = filter_length;
347 c->filter_alloc = FFALIGN(c->filter_length, 8);
348 c->filter_bank = av_calloc(c->filter_alloc, (phase_count+1)*c->felem_size);
349 c->filter_type = filter_type;
350 c->kaiser_beta = kaiser_beta;
351 c->phase_count_compensation = phase_count_compensation;
354 if (build_filter(c, (void*)c->filter_bank, factor, c->filter_length, c->filter_alloc, phase_count, 1<<c->filter_shift, filter_type, kaiser_beta))
356 memcpy(c->filter_bank + (c->filter_alloc*phase_count+1)*c->felem_size, c->filter_bank, (c->filter_alloc-1)*c->felem_size);
357 memcpy(c->filter_bank + (c->filter_alloc*phase_count )*c->felem_size, c->filter_bank + (c->filter_alloc - 1)*c->felem_size, c->felem_size);
360 c->compensation_distance= 0;
361 if(!av_reduce(&c->src_incr, &c->dst_incr, out_rate, in_rate * (int64_t)phase_count, INT32_MAX/2))
363 while (c->dst_incr < (1<<20) && c->src_incr < (1<<20)) {
367 c->ideal_dst_incr = c->dst_incr;
368 c->dst_incr_div = c->dst_incr / c->src_incr;
369 c->dst_incr_mod = c->dst_incr % c->src_incr;
371 c->index= -phase_count*((c->filter_length-1)/2);
374 swri_resample_dsp_init(c);
378 av_freep(&c->filter_bank);
383 static int rebuild_filter_bank_with_compensation(ResampleContext *c)
385 uint8_t *new_filter_bank;
386 int new_src_incr, new_dst_incr;
387 int phase_count = c->phase_count_compensation;
390 if (phase_count == c->phase_count)
393 av_assert0(!c->frac && !c->dst_incr_mod);
395 new_filter_bank = av_calloc(c->filter_alloc, (phase_count + 1) * c->felem_size);
396 if (!new_filter_bank)
397 return AVERROR(ENOMEM);
399 ret = build_filter(c, new_filter_bank, c->factor, c->filter_length, c->filter_alloc,
400 phase_count, 1 << c->filter_shift, c->filter_type, c->kaiser_beta);
402 av_freep(&new_filter_bank);
405 memcpy(new_filter_bank + (c->filter_alloc*phase_count+1)*c->felem_size, new_filter_bank, (c->filter_alloc-1)*c->felem_size);
406 memcpy(new_filter_bank + (c->filter_alloc*phase_count )*c->felem_size, new_filter_bank + (c->filter_alloc - 1)*c->felem_size, c->felem_size);
408 if (!av_reduce(&new_src_incr, &new_dst_incr, c->src_incr,
409 c->dst_incr * (int64_t)(phase_count/c->phase_count), INT32_MAX/2))
411 av_freep(&new_filter_bank);
412 return AVERROR(EINVAL);
415 c->src_incr = new_src_incr;
416 c->dst_incr = new_dst_incr;
417 while (c->dst_incr < (1<<20) && c->src_incr < (1<<20)) {
421 c->ideal_dst_incr = c->dst_incr;
422 c->dst_incr_div = c->dst_incr / c->src_incr;
423 c->dst_incr_mod = c->dst_incr % c->src_incr;
424 c->index *= phase_count / c->phase_count;
425 c->phase_count = phase_count;
426 av_freep(&c->filter_bank);
427 c->filter_bank = new_filter_bank;
431 static int set_compensation(ResampleContext *c, int sample_delta, int compensation_distance){
434 if (compensation_distance && sample_delta) {
435 ret = rebuild_filter_bank_with_compensation(c);
440 c->compensation_distance= compensation_distance;
441 if (compensation_distance)
442 c->dst_incr = c->ideal_dst_incr - c->ideal_dst_incr * (int64_t)sample_delta / compensation_distance;
444 c->dst_incr = c->ideal_dst_incr;
446 c->dst_incr_div = c->dst_incr / c->src_incr;
447 c->dst_incr_mod = c->dst_incr % c->src_incr;
452 static int multiple_resample(ResampleContext *c, AudioData *dst, int dst_size, AudioData *src, int src_size, int *consumed){
454 int av_unused mm_flags = av_get_cpu_flags();
455 int need_emms = c->format == AV_SAMPLE_FMT_S16P && ARCH_X86_32 &&
456 (mm_flags & (AV_CPU_FLAG_MMX2 | AV_CPU_FLAG_SSE2)) == AV_CPU_FLAG_MMX2;
457 int64_t max_src_size = (INT64_MAX/2 / c->phase_count) / c->src_incr;
459 if (c->compensation_distance)
460 dst_size = FFMIN(dst_size, c->compensation_distance);
461 src_size = FFMIN(src_size, max_src_size);
465 if (c->filter_length == 1 && c->phase_count == 1) {
466 int64_t index2= (1LL<<32)*c->frac/c->src_incr + (1LL<<32)*c->index;
467 int64_t incr= (1LL<<32) * c->dst_incr / c->src_incr;
468 int new_size = (src_size * (int64_t)c->src_incr - c->frac + c->dst_incr - 1) / c->dst_incr;
470 dst_size = FFMAX(FFMIN(dst_size, new_size), 0);
472 for (i = 0; i < dst->ch_count; i++) {
473 c->dsp.resample_one(dst->ch[i], src->ch[i], dst_size, index2, incr);
474 if (i+1 == dst->ch_count) {
475 c->index += dst_size * c->dst_incr_div;
476 c->index += (c->frac + dst_size * (int64_t)c->dst_incr_mod) / c->src_incr;
477 av_assert2(c->index >= 0);
478 *consumed = c->index;
479 c->frac = (c->frac + dst_size * (int64_t)c->dst_incr_mod) % c->src_incr;
485 int64_t end_index = (1LL + src_size - c->filter_length) * c->phase_count;
486 int64_t delta_frac = (end_index - c->index) * c->src_incr - c->frac;
487 int delta_n = (delta_frac + c->dst_incr - 1) / c->dst_incr;
488 int (*resample_func)(struct ResampleContext *c, void *dst,
489 const void *src, int n, int update_ctx);
491 dst_size = FFMAX(FFMIN(dst_size, delta_n), 0);
493 /* resample_linear and resample_common should have same behavior
494 * when frac and dst_incr_mod are zero */
495 resample_func = (c->linear && (c->frac || c->dst_incr_mod)) ?
496 c->dsp.resample_linear : c->dsp.resample_common;
497 for (i = 0; i < dst->ch_count; i++)
498 *consumed = resample_func(c, dst->ch[i], src->ch[i], dst_size, i+1 == dst->ch_count);
505 if (c->compensation_distance) {
506 c->compensation_distance -= dst_size;
507 if (!c->compensation_distance) {
508 c->dst_incr = c->ideal_dst_incr;
509 c->dst_incr_div = c->dst_incr / c->src_incr;
510 c->dst_incr_mod = c->dst_incr % c->src_incr;
517 static int64_t get_delay(struct SwrContext *s, int64_t base){
518 ResampleContext *c = s->resample;
519 int64_t num = s->in_buffer_count - (c->filter_length-1)/2;
520 num *= c->phase_count;
524 return av_rescale(num, base, s->in_sample_rate*(int64_t)c->src_incr * c->phase_count);
527 static int64_t get_out_samples(struct SwrContext *s, int in_samples) {
528 ResampleContext *c = s->resample;
529 // The + 2 are added to allow implementations to be slightly inaccurate, they should not be needed currently.
530 // They also make it easier to proof that changes and optimizations do not
531 // break the upper bound.
532 int64_t num = s->in_buffer_count + 2LL + in_samples;
533 num *= c->phase_count;
535 num = av_rescale_rnd(num, s->out_sample_rate, ((int64_t)s->in_sample_rate) * c->phase_count, AV_ROUND_UP) + 2;
537 if (c->compensation_distance) {
539 return AVERROR(EINVAL);
541 num = FFMAX(num, (num * c->ideal_dst_incr - 1) / c->dst_incr + 1);
546 static int resample_flush(struct SwrContext *s) {
547 ResampleContext *c = s->resample;
548 AudioData *a= &s->in_buffer;
550 int reflection = (FFMIN(s->in_buffer_count, c->filter_length) + 1) / 2;
552 if((ret = swri_realloc_audio(a, s->in_buffer_index + s->in_buffer_count + reflection)) < 0)
554 av_assert0(a->planar);
555 for(i=0; i<a->ch_count; i++){
556 for(j=0; j<reflection; j++){
557 memcpy(a->ch[i] + (s->in_buffer_index+s->in_buffer_count+j )*a->bps,
558 a->ch[i] + (s->in_buffer_index+s->in_buffer_count-j-1)*a->bps, a->bps);
561 s->in_buffer_count += reflection;
565 // in fact the whole handle multiple ridiculously small buffers might need more thinking...
566 static int invert_initial_buffer(ResampleContext *c, AudioData *dst, const AudioData *src,
567 int in_count, int *out_idx, int *out_sz)
569 int n, ch, num = FFMIN(in_count + *out_sz, c->filter_length + 1), res;
574 if ((res = swri_realloc_audio(dst, c->filter_length * 2 + 1)) < 0)
578 for (n = *out_sz; n < num; n++) {
579 for (ch = 0; ch < src->ch_count; ch++) {
580 memcpy(dst->ch[ch] + ((c->filter_length + n) * c->felem_size),
581 src->ch[ch] + ((n - *out_sz) * c->felem_size), c->felem_size);
585 // if not enough data is in, return and wait for more
586 if (num < c->filter_length + 1) {
588 *out_idx = c->filter_length;
593 for (n = 1; n <= c->filter_length; n++) {
594 for (ch = 0; ch < src->ch_count; ch++) {
595 memcpy(dst->ch[ch] + ((c->filter_length - n) * c->felem_size),
596 dst->ch[ch] + ((c->filter_length + n) * c->felem_size),
602 *out_idx = c->filter_length;
603 while (c->index < 0) {
605 c->index += c->phase_count;
607 *out_sz = FFMAX(*out_sz + c->filter_length,
608 1 + c->filter_length * 2) - *out_idx;
610 return FFMAX(res, 0);
613 struct Resampler const swri_resampler={
620 invert_initial_buffer,