select more sensible default windows (= attenuation beyond the dynamic range of your...
[libav.git] / libavcodec / resample2.c
CommitLineData
aaaf1635
MN
1/*
2 * audio resampling
3 * Copyright (c) 2004 Michael Niedermayer <michaelni@gmx.at>
4 *
b78e7197
DB
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
aaaf1635
MN
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
b78e7197 10 * version 2.1 of the License, or (at your option) any later version.
aaaf1635 11 *
b78e7197 12 * FFmpeg is distributed in the hope that it will be useful,
aaaf1635
MN
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
b78e7197 18 * License along with FFmpeg; if not, write to the Free Software
5509bffa 19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
aaaf1635
MN
20 *
21 */
115329f1 22
aaaf1635
MN
23/**
24 * @file resample2.c
25 * audio resampling
26 * @author Michael Niedermayer <michaelni@gmx.at>
27 */
28
29#include "avcodec.h"
30#include "common.h"
1ac31760 31#include "dsputil.h"
aaaf1635 32
c252b26d 33#ifndef CONFIG_RESAMPLE_HP
aaaf1635
MN
34#define FILTER_SHIFT 15
35
51a88020
MN
36#define FELEM int16_t
37#define FELEM2 int32_t
38#define FELEM_MAX INT16_MAX
39#define FELEM_MIN INT16_MIN
14f887ef 40#define WINDOW_TYPE 9
51a88020 41#else
ca6940f8 42#define FILTER_SHIFT 30
51a88020
MN
43
44#define FELEM int32_t
45#define FELEM2 int64_t
46#define FELEM_MAX INT32_MAX
47#define FELEM_MIN INT32_MIN
14f887ef 48#define WINDOW_TYPE 12
51a88020
MN
49#endif
50
51
aaaf1635 52typedef struct AVResampleContext{
51a88020 53 FELEM *filter_bank;
aaaf1635
MN
54 int filter_length;
55 int ideal_dst_incr;
56 int dst_incr;
57 int index;
58 int frac;
59 int src_incr;
60 int compensation_distance;
ed861c6b
MN
61 int phase_shift;
62 int phase_mask;
63 int linear;
aaaf1635
MN
64}AVResampleContext;
65
66/**
67 * 0th order modified bessel function of the first kind.
68 */
7b49ce2e 69static double bessel(double x){
aaaf1635
MN
70 double v=1;
71 double t=1;
72 int i;
115329f1 73
aaaf1635
MN
74 for(i=1; i<50; i++){
75 t *= i;
76 v += pow(x*x/4, i)/(t*t);
77 }
78 return v;
79}
80
81/**
82 * builds a polyphase filterbank.
83 * @param factor resampling factor
84 * @param scale wanted sum of coefficients for each filter
20cf58c3 85 * @param type 0->cubic, 1->blackman nuttall windowed sinc, 2..16->kaiser windowed sinc beta=2..16
aaaf1635 86 */
51a88020 87void av_build_filter(FELEM *filter, double factor, int tap_count, int phase_count, int scale, int type){
aaaf1635
MN
88 int ph, i, v;
89 double x, y, w, tab[tap_count];
90 const int center= (tap_count-1)/2;
91
92 /* if upsampling, only need to interpolate, no filter */
93 if (factor > 1.0)
94 factor = 1.0;
95
96 for(ph=0;ph<phase_count;ph++) {
97 double norm = 0;
aaaf1635
MN
98 for(i=0;i<tap_count;i++) {
99 x = M_PI * ((double)(i - center) - (double)ph / phase_count) * factor;
100 if (x == 0) y = 1.0;
101 else y = sin(x) / x;
102 switch(type){
103 case 0:{
104 const float d= -0.5; //first order derivative = -0.5
105 x = fabs(((double)(i - center) - (double)ph / phase_count) * factor);
106 if(x<1.0) y= 1 - 3*x*x + 2*x*x*x + d*( -x*x + x*x*x);
107 else y= d*(-4 + 8*x - 5*x*x + x*x*x);
108 break;}
109 case 1:
110 w = 2.0*x / (factor*tap_count) + M_PI;
111 y *= 0.3635819 - 0.4891775 * cos(w) + 0.1365995 * cos(2*w) - 0.0106411 * cos(3*w);
112 break;
20cf58c3 113 default:
aaaf1635 114 w = 2.0*x / (factor*tap_count*M_PI);
20cf58c3 115 y *= bessel(type*sqrt(FFMAX(1-w*w, 0)));
aaaf1635
MN
116 break;
117 }
118
119 tab[i] = y;
120 norm += y;
121 }
122
123 /* normalize so that an uniform color remains the same */
124 for(i=0;i<tap_count;i++) {
66a148a1 125 v = av_clip(lrintf(tab[i] * scale / norm), FELEM_MIN, FELEM_MAX);
aaaf1635 126 filter[ph * tap_count + i] = v;
aaaf1635
MN
127 }
128 }
ae232dd7
MN
129#if 0
130 {
131#define LEN 1024
132 int j,k;
133 double sine[LEN + tap_count];
134 double filtered[LEN];
135 double maxff=-2, minff=2, maxsf=-2, minsf=2;
136 for(i=0; i<LEN; i++){
137 double ss=0, sf=0, ff=0;
138 for(j=0; j<LEN+tap_count; j++)
139 sine[j]= cos(i*j*M_PI/LEN);
140 for(j=0; j<LEN; j++){
141 double sum=0;
142 ph=0;
143 for(k=0; k<tap_count; k++)
144 sum += filter[ph * tap_count + k] * sine[k+j];
145 filtered[j]= sum / (1<<FILTER_SHIFT);
146 ss+= sine[j + center] * sine[j + center];
147 ff+= filtered[j] * filtered[j];
148 sf+= sine[j + center] * filtered[j];
149 }
150 ss= sqrt(2*ss/LEN);
151 ff= sqrt(2*ff/LEN);
152 sf= 2*sf/LEN;
153 maxff= FFMAX(maxff, ff);
154 minff= FFMIN(minff, ff);
155 maxsf= FFMAX(maxsf, sf);
156 minsf= FFMIN(minsf, sf);
157 if(i%11==0){
158 av_log(NULL, AV_LOG_ERROR, "i:%4d ss:%f ff:%f-%f sf:%f-%f\n", i, ss, maxff, minff, maxsf, minsf);
159 minff=minsf= 2;
160 maxff=maxsf= -2;
161 }
162 }
163 }
164#endif
aaaf1635
MN
165}
166
167/**
168 * initalizes a audio resampler.
169 * note, if either rate is not a integer then simply scale both rates up so they are
170 */
6e225de2 171AVResampleContext *av_resample_init(int out_rate, int in_rate, int filter_size, int phase_shift, int linear, double cutoff){
aaaf1635 172 AVResampleContext *c= av_mallocz(sizeof(AVResampleContext));
6e225de2 173 double factor= FFMIN(out_rate * cutoff / in_rate, 1.0);
ed861c6b 174 int phase_count= 1<<phase_shift;
115329f1 175
ed861c6b
MN
176 c->phase_shift= phase_shift;
177 c->phase_mask= phase_count-1;
178 c->linear= linear;
aaaf1635 179
f0ff20a1 180 c->filter_length= FFMAX((int)ceil(filter_size/factor), 1);
ed861c6b 181 c->filter_bank= av_mallocz(c->filter_length*(phase_count+1)*sizeof(FELEM));
14f887ef 182 av_build_filter(c->filter_bank, factor, c->filter_length, phase_count, 1<<FILTER_SHIFT, WINDOW_TYPE);
ed861c6b
MN
183 memcpy(&c->filter_bank[c->filter_length*phase_count+1], c->filter_bank, (c->filter_length-1)*sizeof(FELEM));
184 c->filter_bank[c->filter_length*phase_count]= c->filter_bank[c->filter_length - 1];
aaaf1635
MN
185
186 c->src_incr= out_rate;
ed861c6b
MN
187 c->ideal_dst_incr= c->dst_incr= in_rate * phase_count;
188 c->index= -phase_count*((c->filter_length-1)/2);
aaaf1635
MN
189
190 return c;
191}
192
193void av_resample_close(AVResampleContext *c){
194 av_freep(&c->filter_bank);
195 av_freep(&c);
196}
197
788d7a8c
MN
198/**
199 * Compensates samplerate/timestamp drift. The compensation is done by changing
200 * the resampler parameters, so no audible clicks or similar distortions ocur
201 * @param compensation_distance distance in output samples over which the compensation should be performed
202 * @param sample_delta number of output samples which should be output less
203 *
204 * example: av_resample_compensate(c, 10, 500)
205 * here instead of 510 samples only 500 samples would be output
206 *
115329f1 207 * note, due to rounding the actual compensation might be slightly different,
788d7a8c
MN
208 * especially if the compensation_distance is large and the in_rate used during init is small
209 */
aaaf1635 210void av_resample_compensate(AVResampleContext *c, int sample_delta, int compensation_distance){
08f7073a 211// sample_delta += (c->ideal_dst_incr - c->dst_incr)*(int64_t)c->compensation_distance / c->ideal_dst_incr;
aaaf1635 212 c->compensation_distance= compensation_distance;
08f7073a 213 c->dst_incr = c->ideal_dst_incr - c->ideal_dst_incr * (int64_t)sample_delta / compensation_distance;
aaaf1635
MN
214}
215
216/**
217 * resamples.
218 * @param src an array of unconsumed samples
219 * @param consumed the number of samples of src which have been consumed are returned here
220 * @param src_size the number of unconsumed samples available
221 * @param dst_size the amount of space in samples available in dst
222 * @param update_ctx if this is 0 then the context wont be modified, that way several channels can be resampled with the same context
223 * @return the number of samples written in dst or -1 if an error occured
224 */
225int av_resample(AVResampleContext *c, short *dst, short *src, int *consumed, int src_size, int dst_size, int update_ctx){
226 int dst_index, i;
227 int index= c->index;
228 int frac= c->frac;
229 int dst_incr_frac= c->dst_incr % c->src_incr;
230 int dst_incr= c->dst_incr / c->src_incr;
80e85288 231 int compensation_distance= c->compensation_distance;
53f0090d 232
6cb5dcb3 233 if(compensation_distance == 0 && c->filter_length == 1 && c->phase_shift==0){
53f0090d
MN
234 int64_t index2= ((int64_t)index)<<32;
235 int64_t incr= (1LL<<32) * c->dst_incr / c->src_incr;
236 dst_size= FFMIN(dst_size, (src_size-1-index) * (int64_t)c->src_incr / c->dst_incr);
115329f1 237
6cb5dcb3 238 for(dst_index=0; dst_index < dst_size; dst_index++){
53f0090d
MN
239 dst[dst_index] = src[index2>>32];
240 index2 += incr;
6cb5dcb3 241 }
53f0090d
MN
242 frac += dst_index * dst_incr_frac;
243 index += dst_index * dst_incr;
244 index += frac / c->src_incr;
245 frac %= c->src_incr;
6cb5dcb3 246 }else{
aaaf1635 247 for(dst_index=0; dst_index < dst_size; dst_index++){
ed861c6b
MN
248 FELEM *filter= c->filter_bank + c->filter_length*(index & c->phase_mask);
249 int sample_index= index >> c->phase_shift;
51a88020 250 FELEM2 val=0;
115329f1 251
aaaf1635
MN
252 if(sample_index < 0){
253 for(i=0; i<c->filter_length; i++)
c26abfa5 254 val += src[FFABS(sample_index + i) % src_size] * filter[i];
aaaf1635
MN
255 }else if(sample_index + c->filter_length > src_size){
256 break;
ed861c6b 257 }else if(c->linear){
aaaf1635 258 int64_t v=0;
f25ba8b3 259 int sub_phase= (frac<<8) / c->src_incr;
aaaf1635 260 for(i=0; i<c->filter_length; i++){
f25ba8b3 261 int64_t coeff= filter[i]*(256 - sub_phase) + filter[i + c->filter_length]*sub_phase;
aaaf1635
MN
262 v += src[sample_index + i] * coeff;
263 }
f25ba8b3 264 val= v>>8;
ed861c6b 265 }else{
aaaf1635 266 for(i=0; i<c->filter_length; i++){
51a88020 267 val += src[sample_index + i] * (FELEM2)filter[i];
aaaf1635 268 }
aaaf1635
MN
269 }
270
271 val = (val + (1<<(FILTER_SHIFT-1)))>>FILTER_SHIFT;
272 dst[dst_index] = (unsigned)(val + 32768) > 65535 ? (val>>31) ^ 32767 : val;
273
274 frac += dst_incr_frac;
275 index += dst_incr;
276 if(frac >= c->src_incr){
277 frac -= c->src_incr;
278 index++;
279 }
80e85288
MN
280
281 if(dst_index + 1 == compensation_distance){
282 compensation_distance= 0;
283 dst_incr_frac= c->ideal_dst_incr % c->src_incr;
284 dst_incr= c->ideal_dst_incr / c->src_incr;
285 }
aaaf1635 286 }
6cb5dcb3 287 }
ed861c6b 288 *consumed= FFMAX(index, 0) >> c->phase_shift;
4e255822 289 if(index>=0) index &= c->phase_mask;
b9d2085b 290
80e85288
MN
291 if(compensation_distance){
292 compensation_distance -= dst_index;
293 assert(compensation_distance > 0);
294 }
aaaf1635 295 if(update_ctx){
aaaf1635 296 c->frac= frac;
b9d2085b 297 c->index= index;
80e85288
MN
298 c->dst_incr= dst_incr_frac + c->src_incr*dst_incr;
299 c->compensation_distance= compensation_distance;
aaaf1635 300 }
115329f1 301#if 0
08f7073a
MN
302 if(update_ctx && !c->compensation_distance){
303#undef rand
304 av_resample_compensate(c, rand() % (8000*2) - 8000, 8000*2);
305av_log(NULL, AV_LOG_DEBUG, "%d %d %d\n", c->dst_incr, c->ideal_dst_incr, c->compensation_distance);
306 }
307#endif
115329f1 308
aaaf1635
MN
309 return dst_index;
310}