remove dithering of filter coefficients, improves precision by 1-2 bits and
[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
51a88020 33#if 1
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
40#else
f25ba8b3 41#define FILTER_SHIFT 22
51a88020
MN
42
43#define FELEM int32_t
44#define FELEM2 int64_t
45#define FELEM_MAX INT32_MAX
46#define FELEM_MIN INT32_MIN
47#endif
48
49
aaaf1635 50typedef struct AVResampleContext{
51a88020 51 FELEM *filter_bank;
aaaf1635
MN
52 int filter_length;
53 int ideal_dst_incr;
54 int dst_incr;
55 int index;
56 int frac;
57 int src_incr;
58 int compensation_distance;
ed861c6b
MN
59 int phase_shift;
60 int phase_mask;
61 int linear;
aaaf1635
MN
62}AVResampleContext;
63
64/**
65 * 0th order modified bessel function of the first kind.
66 */
7b49ce2e 67static double bessel(double x){
aaaf1635
MN
68 double v=1;
69 double t=1;
70 int i;
115329f1 71
aaaf1635
MN
72 for(i=1; i<50; i++){
73 t *= i;
74 v += pow(x*x/4, i)/(t*t);
75 }
76 return v;
77}
78
79/**
80 * builds a polyphase filterbank.
81 * @param factor resampling factor
82 * @param scale wanted sum of coefficients for each filter
83 * @param type 0->cubic, 1->blackman nuttall windowed sinc, 2->kaiser windowed sinc beta=16
84 */
51a88020 85void av_build_filter(FELEM *filter, double factor, int tap_count, int phase_count, int scale, int type){
aaaf1635
MN
86 int ph, i, v;
87 double x, y, w, tab[tap_count];
88 const int center= (tap_count-1)/2;
89
90 /* if upsampling, only need to interpolate, no filter */
91 if (factor > 1.0)
92 factor = 1.0;
93
94 for(ph=0;ph<phase_count;ph++) {
95 double norm = 0;
aaaf1635
MN
96 for(i=0;i<tap_count;i++) {
97 x = M_PI * ((double)(i - center) - (double)ph / phase_count) * factor;
98 if (x == 0) y = 1.0;
99 else y = sin(x) / x;
100 switch(type){
101 case 0:{
102 const float d= -0.5; //first order derivative = -0.5
103 x = fabs(((double)(i - center) - (double)ph / phase_count) * factor);
104 if(x<1.0) y= 1 - 3*x*x + 2*x*x*x + d*( -x*x + x*x*x);
105 else y= d*(-4 + 8*x - 5*x*x + x*x*x);
106 break;}
107 case 1:
108 w = 2.0*x / (factor*tap_count) + M_PI;
109 y *= 0.3635819 - 0.4891775 * cos(w) + 0.1365995 * cos(2*w) - 0.0106411 * cos(3*w);
110 break;
111 case 2:
112 w = 2.0*x / (factor*tap_count*M_PI);
08f7073a 113 y *= bessel(16*sqrt(FFMAX(1-w*w, 0)));
aaaf1635
MN
114 break;
115 }
116
117 tab[i] = y;
118 norm += y;
119 }
120
121 /* normalize so that an uniform color remains the same */
122 for(i=0;i<tap_count;i++) {
66a148a1 123 v = av_clip(lrintf(tab[i] * scale / norm), FELEM_MIN, FELEM_MAX);
aaaf1635 124 filter[ph * tap_count + i] = v;
aaaf1635
MN
125 }
126 }
127}
128
129/**
130 * initalizes a audio resampler.
131 * note, if either rate is not a integer then simply scale both rates up so they are
132 */
6e225de2 133AVResampleContext *av_resample_init(int out_rate, int in_rate, int filter_size, int phase_shift, int linear, double cutoff){
aaaf1635 134 AVResampleContext *c= av_mallocz(sizeof(AVResampleContext));
6e225de2 135 double factor= FFMIN(out_rate * cutoff / in_rate, 1.0);
ed861c6b 136 int phase_count= 1<<phase_shift;
115329f1 137
ed861c6b
MN
138 c->phase_shift= phase_shift;
139 c->phase_mask= phase_count-1;
140 c->linear= linear;
aaaf1635 141
f0ff20a1 142 c->filter_length= FFMAX((int)ceil(filter_size/factor), 1);
ed861c6b
MN
143 c->filter_bank= av_mallocz(c->filter_length*(phase_count+1)*sizeof(FELEM));
144 av_build_filter(c->filter_bank, factor, c->filter_length, phase_count, 1<<FILTER_SHIFT, 1);
145 memcpy(&c->filter_bank[c->filter_length*phase_count+1], c->filter_bank, (c->filter_length-1)*sizeof(FELEM));
146 c->filter_bank[c->filter_length*phase_count]= c->filter_bank[c->filter_length - 1];
aaaf1635
MN
147
148 c->src_incr= out_rate;
ed861c6b
MN
149 c->ideal_dst_incr= c->dst_incr= in_rate * phase_count;
150 c->index= -phase_count*((c->filter_length-1)/2);
aaaf1635
MN
151
152 return c;
153}
154
155void av_resample_close(AVResampleContext *c){
156 av_freep(&c->filter_bank);
157 av_freep(&c);
158}
159
788d7a8c
MN
160/**
161 * Compensates samplerate/timestamp drift. The compensation is done by changing
162 * the resampler parameters, so no audible clicks or similar distortions ocur
163 * @param compensation_distance distance in output samples over which the compensation should be performed
164 * @param sample_delta number of output samples which should be output less
165 *
166 * example: av_resample_compensate(c, 10, 500)
167 * here instead of 510 samples only 500 samples would be output
168 *
115329f1 169 * note, due to rounding the actual compensation might be slightly different,
788d7a8c
MN
170 * especially if the compensation_distance is large and the in_rate used during init is small
171 */
aaaf1635 172void av_resample_compensate(AVResampleContext *c, int sample_delta, int compensation_distance){
08f7073a 173// sample_delta += (c->ideal_dst_incr - c->dst_incr)*(int64_t)c->compensation_distance / c->ideal_dst_incr;
aaaf1635 174 c->compensation_distance= compensation_distance;
08f7073a 175 c->dst_incr = c->ideal_dst_incr - c->ideal_dst_incr * (int64_t)sample_delta / compensation_distance;
aaaf1635
MN
176}
177
178/**
179 * resamples.
180 * @param src an array of unconsumed samples
181 * @param consumed the number of samples of src which have been consumed are returned here
182 * @param src_size the number of unconsumed samples available
183 * @param dst_size the amount of space in samples available in dst
184 * @param update_ctx if this is 0 then the context wont be modified, that way several channels can be resampled with the same context
185 * @return the number of samples written in dst or -1 if an error occured
186 */
187int av_resample(AVResampleContext *c, short *dst, short *src, int *consumed, int src_size, int dst_size, int update_ctx){
188 int dst_index, i;
189 int index= c->index;
190 int frac= c->frac;
191 int dst_incr_frac= c->dst_incr % c->src_incr;
192 int dst_incr= c->dst_incr / c->src_incr;
80e85288 193 int compensation_distance= c->compensation_distance;
53f0090d 194
6cb5dcb3 195 if(compensation_distance == 0 && c->filter_length == 1 && c->phase_shift==0){
53f0090d
MN
196 int64_t index2= ((int64_t)index)<<32;
197 int64_t incr= (1LL<<32) * c->dst_incr / c->src_incr;
198 dst_size= FFMIN(dst_size, (src_size-1-index) * (int64_t)c->src_incr / c->dst_incr);
115329f1 199
6cb5dcb3 200 for(dst_index=0; dst_index < dst_size; dst_index++){
53f0090d
MN
201 dst[dst_index] = src[index2>>32];
202 index2 += incr;
6cb5dcb3 203 }
53f0090d
MN
204 frac += dst_index * dst_incr_frac;
205 index += dst_index * dst_incr;
206 index += frac / c->src_incr;
207 frac %= c->src_incr;
6cb5dcb3 208 }else{
aaaf1635 209 for(dst_index=0; dst_index < dst_size; dst_index++){
ed861c6b
MN
210 FELEM *filter= c->filter_bank + c->filter_length*(index & c->phase_mask);
211 int sample_index= index >> c->phase_shift;
51a88020 212 FELEM2 val=0;
115329f1 213
aaaf1635
MN
214 if(sample_index < 0){
215 for(i=0; i<c->filter_length; i++)
c26abfa5 216 val += src[FFABS(sample_index + i) % src_size] * filter[i];
aaaf1635
MN
217 }else if(sample_index + c->filter_length > src_size){
218 break;
ed861c6b 219 }else if(c->linear){
aaaf1635 220 int64_t v=0;
f25ba8b3 221 int sub_phase= (frac<<8) / c->src_incr;
aaaf1635 222 for(i=0; i<c->filter_length; i++){
f25ba8b3 223 int64_t coeff= filter[i]*(256 - sub_phase) + filter[i + c->filter_length]*sub_phase;
aaaf1635
MN
224 v += src[sample_index + i] * coeff;
225 }
f25ba8b3 226 val= v>>8;
ed861c6b 227 }else{
aaaf1635 228 for(i=0; i<c->filter_length; i++){
51a88020 229 val += src[sample_index + i] * (FELEM2)filter[i];
aaaf1635 230 }
aaaf1635
MN
231 }
232
233 val = (val + (1<<(FILTER_SHIFT-1)))>>FILTER_SHIFT;
234 dst[dst_index] = (unsigned)(val + 32768) > 65535 ? (val>>31) ^ 32767 : val;
235
236 frac += dst_incr_frac;
237 index += dst_incr;
238 if(frac >= c->src_incr){
239 frac -= c->src_incr;
240 index++;
241 }
80e85288
MN
242
243 if(dst_index + 1 == compensation_distance){
244 compensation_distance= 0;
245 dst_incr_frac= c->ideal_dst_incr % c->src_incr;
246 dst_incr= c->ideal_dst_incr / c->src_incr;
247 }
aaaf1635 248 }
6cb5dcb3 249 }
ed861c6b 250 *consumed= FFMAX(index, 0) >> c->phase_shift;
4e255822 251 if(index>=0) index &= c->phase_mask;
b9d2085b 252
80e85288
MN
253 if(compensation_distance){
254 compensation_distance -= dst_index;
255 assert(compensation_distance > 0);
256 }
aaaf1635 257 if(update_ctx){
aaaf1635 258 c->frac= frac;
b9d2085b 259 c->index= index;
80e85288
MN
260 c->dst_incr= dst_incr_frac + c->src_incr*dst_incr;
261 c->compensation_distance= compensation_distance;
aaaf1635 262 }
115329f1 263#if 0
08f7073a
MN
264 if(update_ctx && !c->compensation_distance){
265#undef rand
266 av_resample_compensate(c, rand() % (8000*2) - 8000, 8000*2);
267av_log(NULL, AV_LOG_DEBUG, "%d %d %d\n", c->dst_incr, c->ideal_dst_incr, c->compensation_distance);
268 }
269#endif
115329f1 270
aaaf1635
MN
271 return dst_index;
272}