3 * Copyright (c) 2004-2012 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
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.
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.
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
25 * @author Michael Niedermayer <michaelni@gmx.at>
28 #include "libavutil/log.h"
29 #include "libavutil/avassert.h"
30 #include "swresample_internal.h"
36 typedef struct ResampleContext {
37 const AVClass *av_class;
46 int compensation_distance;
51 enum AVSampleFormat format;
57 * 0th order modified bessel function of the first kind.
59 static double bessel(double x){
64 static const double inv[100]={
65 1.0/( 1* 1), 1.0/( 2* 2), 1.0/( 3* 3), 1.0/( 4* 4), 1.0/( 5* 5), 1.0/( 6* 6), 1.0/( 7* 7), 1.0/( 8* 8), 1.0/( 9* 9), 1.0/(10*10),
66 1.0/(11*11), 1.0/(12*12), 1.0/(13*13), 1.0/(14*14), 1.0/(15*15), 1.0/(16*16), 1.0/(17*17), 1.0/(18*18), 1.0/(19*19), 1.0/(20*20),
67 1.0/(21*21), 1.0/(22*22), 1.0/(23*23), 1.0/(24*24), 1.0/(25*25), 1.0/(26*26), 1.0/(27*27), 1.0/(28*28), 1.0/(29*29), 1.0/(30*30),
68 1.0/(31*31), 1.0/(32*32), 1.0/(33*33), 1.0/(34*34), 1.0/(35*35), 1.0/(36*36), 1.0/(37*37), 1.0/(38*38), 1.0/(39*39), 1.0/(40*40),
69 1.0/(41*41), 1.0/(42*42), 1.0/(43*43), 1.0/(44*44), 1.0/(45*45), 1.0/(46*46), 1.0/(47*47), 1.0/(48*48), 1.0/(49*49), 1.0/(50*50),
70 1.0/(51*51), 1.0/(52*52), 1.0/(53*53), 1.0/(54*54), 1.0/(55*55), 1.0/(56*56), 1.0/(57*57), 1.0/(58*58), 1.0/(59*59), 1.0/(60*60),
71 1.0/(61*61), 1.0/(62*62), 1.0/(63*63), 1.0/(64*64), 1.0/(65*65), 1.0/(66*66), 1.0/(67*67), 1.0/(68*68), 1.0/(69*69), 1.0/(70*70),
72 1.0/(71*71), 1.0/(72*72), 1.0/(73*73), 1.0/(74*74), 1.0/(75*75), 1.0/(76*76), 1.0/(77*77), 1.0/(78*78), 1.0/(79*79), 1.0/(80*80),
73 1.0/(81*81), 1.0/(82*82), 1.0/(83*83), 1.0/(84*84), 1.0/(85*85), 1.0/(86*86), 1.0/(87*87), 1.0/(88*88), 1.0/(89*89), 1.0/(90*90),
74 1.0/(91*91), 1.0/(92*92), 1.0/(93*93), 1.0/(94*94), 1.0/(95*95), 1.0/(96*96), 1.0/(97*97), 1.0/(98*98), 1.0/(99*99), 1.0/(10000)
78 for(i=0; v != lastv; i++){
87 * builds a polyphase filterbank.
88 * @param factor resampling factor
89 * @param scale wanted sum of coefficients for each filter
90 * @param type 0->cubic, 1->blackman nuttall windowed sinc, 2..16->kaiser windowed sinc beta=2..16
91 * @return 0 on success, negative on error
93 static int build_filter(ResampleContext *c, void *filter, double factor, int tap_count, int alloc, int phase_count, int scale, int type){
96 double *tab = av_malloc(tap_count * sizeof(*tab));
97 const int center= (tap_count-1)/2;
100 return AVERROR(ENOMEM);
102 /* if upsampling, only need to interpolate, no filter */
106 for(ph=0;ph<phase_count;ph++) {
108 for(i=0;i<tap_count;i++) {
109 x = M_PI * ((double)(i - center) - (double)ph / phase_count) * factor;
114 const float d= -0.5; //first order derivative = -0.5
115 x = fabs(((double)(i - center) - (double)ph / phase_count) * factor);
116 if(x<1.0) y= 1 - 3*x*x + 2*x*x*x + d*( -x*x + x*x*x);
117 else y= d*(-4 + 8*x - 5*x*x + x*x*x);
120 w = 2.0*x / (factor*tap_count) + M_PI;
121 y *= 0.3635819 - 0.4891775 * cos(w) + 0.1365995 * cos(2*w) - 0.0106411 * cos(3*w);
124 w = 2.0*x / (factor*tap_count*M_PI);
125 y *= bessel(type*sqrt(FFMAX(1-w*w, 0)));
133 /* normalize so that an uniform color remains the same */
135 case AV_SAMPLE_FMT_S16P:
136 for(i=0;i<tap_count;i++)
137 ((int16_t*)filter)[ph * alloc + i] = av_clip(lrintf(tab[i] * scale / norm), INT16_MIN, INT16_MAX);
139 case AV_SAMPLE_FMT_S32P:
140 for(i=0;i<tap_count;i++)
141 ((int32_t*)filter)[ph * alloc + i] = av_clip(lrintf(tab[i] * scale / norm), INT32_MIN, INT32_MAX);
143 case AV_SAMPLE_FMT_FLTP:
144 for(i=0;i<tap_count;i++)
145 ((float*)filter)[ph * alloc + i] = tab[i] * scale / norm;
147 case AV_SAMPLE_FMT_DBLP:
148 for(i=0;i<tap_count;i++)
149 ((double*)filter)[ph * alloc + i] = tab[i] * scale / norm;
157 double sine[LEN + tap_count];
158 double filtered[LEN];
159 double maxff=-2, minff=2, maxsf=-2, minsf=2;
160 for(i=0; i<LEN; i++){
161 double ss=0, sf=0, ff=0;
162 for(j=0; j<LEN+tap_count; j++)
163 sine[j]= cos(i*j*M_PI/LEN);
164 for(j=0; j<LEN; j++){
167 for(k=0; k<tap_count; k++)
168 sum += filter[ph * tap_count + k] * sine[k+j];
169 filtered[j]= sum / (1<<FILTER_SHIFT);
170 ss+= sine[j + center] * sine[j + center];
171 ff+= filtered[j] * filtered[j];
172 sf+= sine[j + center] * filtered[j];
177 maxff= FFMAX(maxff, ff);
178 minff= FFMIN(minff, ff);
179 maxsf= FFMAX(maxsf, sf);
180 minsf= FFMIN(minsf, sf);
182 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);
194 ResampleContext *swri_resample_init(ResampleContext *c, int out_rate, int in_rate, int filter_size, int phase_shift, int linear, double cutoff, enum AVSampleFormat format){
195 double factor= FFMIN(out_rate * cutoff / in_rate, 1.0);
196 int phase_count= 1<<phase_shift;
198 if (!c || c->phase_shift != phase_shift || c->linear!=linear || c->factor != factor
199 || c->filter_length != FFMAX((int)ceil(filter_size/factor), 1) || c->format != format) {
200 c = av_mallocz(sizeof(*c));
206 c->felem_size= av_get_bytes_per_sample(c->format);
209 case AV_SAMPLE_FMT_S16P:
210 c->filter_shift = 15;
212 case AV_SAMPLE_FMT_S32P:
213 c->filter_shift = 30;
215 case AV_SAMPLE_FMT_FLTP:
216 case AV_SAMPLE_FMT_DBLP:
220 av_log(NULL, AV_LOG_ERROR, "Unsupported sample format\n");
224 c->phase_shift = phase_shift;
225 c->phase_mask = phase_count - 1;
228 c->filter_length = FFMAX((int)ceil(filter_size/factor), 1);
229 c->filter_alloc = FFALIGN(c->filter_length, 8);
230 c->filter_bank = av_mallocz(c->filter_alloc*(phase_count+1)*c->felem_size);
233 if (build_filter(c, (void*)c->filter_bank, factor, c->filter_length, c->filter_alloc, phase_count, 1<<c->filter_shift, WINDOW_TYPE))
235 memcpy(c->filter_bank + (c->filter_alloc*phase_count+1)*c->felem_size, c->filter_bank, (c->filter_alloc-1)*c->felem_size);
236 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);
239 c->compensation_distance= 0;
240 if(!av_reduce(&c->src_incr, &c->dst_incr, out_rate, in_rate * (int64_t)phase_count, INT32_MAX/2))
242 c->ideal_dst_incr= c->dst_incr;
244 c->index= -phase_count*((c->filter_length-1)/2);
249 av_free(c->filter_bank);
254 void swri_resample_free(ResampleContext **c){
257 av_freep(&(*c)->filter_bank);
261 int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
265 if (!s || compensation_distance < 0)
266 return AVERROR(EINVAL);
267 if (!compensation_distance && sample_delta)
268 return AVERROR(EINVAL);
270 s->flags |= SWR_FLAG_RESAMPLE;
276 c->compensation_distance= compensation_distance;
277 if (compensation_distance)
278 c->dst_incr = c->ideal_dst_incr - c->ideal_dst_incr * (int64_t)sample_delta / compensation_distance;
280 c->dst_incr = c->ideal_dst_incr;
284 #define RENAME(N) N ## _int16
285 #define FILTER_SHIFT 15
286 #define DELEM int16_t
287 #define FELEM int16_t
288 #define FELEM2 int32_t
289 #define FELEML int64_t
290 #define FELEM_MAX INT16_MAX
291 #define FELEM_MIN INT16_MIN
292 #define OUT(d, v) v = (v + (1<<(FILTER_SHIFT-1)))>>FILTER_SHIFT;\
293 d = (unsigned)(v + 32768) > 65535 ? (v>>31) ^ 32767 : v
294 #include "resample_template.c"
307 #define RENAME(N) N ## _int32
308 #define FILTER_SHIFT 30
309 #define DELEM int32_t
310 #define FELEM int32_t
311 #define FELEM2 int64_t
312 #define FELEML int64_t
313 #define FELEM_MAX INT32_MAX
314 #define FELEM_MIN INT32_MIN
315 #define OUT(d, v) v = (v + (1<<(FILTER_SHIFT-1)))>>FILTER_SHIFT;\
316 d = (uint64_t)(v + 0x80000000) > 0xFFFFFFFF ? (v>>63) ^ 0x7FFFFFFF : v
317 #include "resample_template.c"
330 #define RENAME(N) N ## _float
331 #define FILTER_SHIFT 0
336 #define OUT(d, v) d = v
337 #include "resample_template.c"
350 #define RENAME(N) N ## _double
351 #define FILTER_SHIFT 0
354 #define FELEM2 double
355 #define FELEML double
356 #define OUT(d, v) d = v
357 #include "resample_template.c"
369 // XXX FIXME the whole C loop should be written in asm so this x86 specific code here isnt needed
371 #include "x86/resample_mmx.h"
372 #define COMMON_CORE COMMON_CORE_INT16_MMX2
373 #define RENAME(N) N ## _int16_mmx2
374 #define FILTER_SHIFT 15
375 #define DELEM int16_t
376 #define FELEM int16_t
377 #define FELEM2 int32_t
378 #define FELEML int64_t
379 #define FELEM_MAX INT16_MAX
380 #define FELEM_MIN INT16_MIN
381 #define OUT(d, v) v = (v + (1<<(FILTER_SHIFT-1)))>>FILTER_SHIFT;\
382 d = (unsigned)(v + 32768) > 65535 ? (v>>31) ^ 32767 : v
383 #include "resample_template.c"
397 #define COMMON_CORE COMMON_CORE_INT16_SSSE3
398 #define RENAME(N) N ## _int16_ssse3
399 #define FILTER_SHIFT 15
400 #define DELEM int16_t
401 #define FELEM int16_t
402 #define FELEM2 int32_t
403 #define FELEML int64_t
404 #define FELEM_MAX INT16_MAX
405 #define FELEM_MIN INT16_MIN
406 #define OUT(d, v) v = (v + (1<<(FILTER_SHIFT-1)))>>FILTER_SHIFT;\
407 d = (unsigned)(v + 32768) > 65535 ? (v>>31) ^ 32767 : v
408 #include "resample_template.c"
412 int swri_multiple_resample(ResampleContext *c, AudioData *dst, int dst_size, AudioData *src, int src_size, int *consumed){
414 int mm_flags = av_get_cpu_flags();
417 for(i=0; i<dst->ch_count; i++){
420 if(c->format == AV_SAMPLE_FMT_S16P && (mm_flags&AV_CPU_FLAG_SSSE3)) ret= swri_resample_int16_ssse3(c, (int16_t*)dst->ch[i], (const int16_t*)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
423 if(c->format == AV_SAMPLE_FMT_S16P && (mm_flags&AV_CPU_FLAG_MMX2 )){
424 ret= swri_resample_int16_mmx2 (c, (int16_t*)dst->ch[i], (const int16_t*)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
428 if(c->format == AV_SAMPLE_FMT_S16P) ret= swri_resample_int16(c, (int16_t*)dst->ch[i], (const int16_t*)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
429 else if(c->format == AV_SAMPLE_FMT_S32P) ret= swri_resample_int32(c, (int32_t*)dst->ch[i], (const int32_t*)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
430 else if(c->format == AV_SAMPLE_FMT_FLTP) ret= swri_resample_float(c, (float *)dst->ch[i], (const float *)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
431 else if(c->format == AV_SAMPLE_FMT_DBLP) ret= swri_resample_double(c,(double *)dst->ch[i], (const double *)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
438 int64_t swr_get_delay(struct SwrContext *s, int64_t base){
439 ResampleContext *c = s->resample;
441 int64_t num = s->in_buffer_count - (c->filter_length-1)/2;
442 num <<= c->phase_shift;
447 return av_rescale(num, base, s->in_sample_rate*(int64_t)c->src_incr << c->phase_shift);
449 return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;