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"
33 typedef struct ResampleContext {
34 const AVClass *av_class;
43 int compensation_distance;
47 enum SwrFilterType filter_type;
50 enum AVSampleFormat format;
56 * 0th order modified bessel function of the first kind.
58 static double bessel(double x){
63 static const double inv[100]={
64 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),
65 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),
66 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),
67 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),
68 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),
69 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),
70 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),
71 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),
72 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),
73 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)
77 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 filter_type filter type
91 * @param kaiser_beta kaiser window beta
92 * @return 0 on success, negative on error
94 static int build_filter(ResampleContext *c, void *filter, double factor, int tap_count, int alloc, int phase_count, int scale,
95 int filter_type, int kaiser_beta){
98 double *tab = av_malloc_array(tap_count, sizeof(*tab));
99 const int center= (tap_count-1)/2;
102 return AVERROR(ENOMEM);
104 /* if upsampling, only need to interpolate, no filter */
108 for(ph=0;ph<phase_count;ph++) {
110 for(i=0;i<tap_count;i++) {
111 x = M_PI * ((double)(i - center) - (double)ph / phase_count) * factor;
115 case SWR_FILTER_TYPE_CUBIC:{
116 const float d= -0.5; //first order derivative = -0.5
117 x = fabs(((double)(i - center) - (double)ph / phase_count) * factor);
118 if(x<1.0) y= 1 - 3*x*x + 2*x*x*x + d*( -x*x + x*x*x);
119 else y= d*(-4 + 8*x - 5*x*x + x*x*x);
121 case SWR_FILTER_TYPE_BLACKMAN_NUTTALL:
122 w = 2.0*x / (factor*tap_count) + M_PI;
123 y *= 0.3635819 - 0.4891775 * cos(w) + 0.1365995 * cos(2*w) - 0.0106411 * cos(3*w);
125 case SWR_FILTER_TYPE_KAISER:
126 w = 2.0*x / (factor*tap_count*M_PI);
127 y *= bessel(kaiser_beta*sqrt(FFMAX(1-w*w, 0)));
137 /* normalize so that an uniform color remains the same */
139 case AV_SAMPLE_FMT_S16P:
140 for(i=0;i<tap_count;i++)
141 ((int16_t*)filter)[ph * alloc + i] = av_clip(lrintf(tab[i] * scale / norm), INT16_MIN, INT16_MAX);
143 case AV_SAMPLE_FMT_S32P:
144 for(i=0;i<tap_count;i++)
145 ((int32_t*)filter)[ph * alloc + i] = av_clipl_int32(llrint(tab[i] * scale / norm));
147 case AV_SAMPLE_FMT_FLTP:
148 for(i=0;i<tap_count;i++)
149 ((float*)filter)[ph * alloc + i] = tab[i] * scale / norm;
151 case AV_SAMPLE_FMT_DBLP:
152 for(i=0;i<tap_count;i++)
153 ((double*)filter)[ph * alloc + i] = tab[i] * scale / norm;
161 double sine[LEN + tap_count];
162 double filtered[LEN];
163 double maxff=-2, minff=2, maxsf=-2, minsf=2;
164 for(i=0; i<LEN; i++){
165 double ss=0, sf=0, ff=0;
166 for(j=0; j<LEN+tap_count; j++)
167 sine[j]= cos(i*j*M_PI/LEN);
168 for(j=0; j<LEN; j++){
171 for(k=0; k<tap_count; k++)
172 sum += filter[ph * tap_count + k] * sine[k+j];
173 filtered[j]= sum / (1<<FILTER_SHIFT);
174 ss+= sine[j + center] * sine[j + center];
175 ff+= filtered[j] * filtered[j];
176 sf+= sine[j + center] * filtered[j];
181 maxff= FFMAX(maxff, ff);
182 minff= FFMIN(minff, ff);
183 maxsf= FFMAX(maxsf, sf);
184 minsf= FFMIN(minsf, sf);
186 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);
198 static ResampleContext *resample_init(ResampleContext *c, int out_rate, int in_rate, int filter_size, int phase_shift, int linear,
199 double cutoff0, enum AVSampleFormat format, enum SwrFilterType filter_type, int kaiser_beta,
200 double precision, int cheby){
201 double cutoff = cutoff0? cutoff0 : 0.97;
202 double factor= FFMIN(out_rate * cutoff / in_rate, 1.0);
203 int phase_count= 1<<phase_shift;
205 if (!c || c->phase_shift != phase_shift || c->linear!=linear || c->factor != factor
206 || c->filter_length != FFMAX((int)ceil(filter_size/factor), 1) || c->format != format
207 || c->filter_type != filter_type || c->kaiser_beta != kaiser_beta) {
208 c = av_mallocz(sizeof(*c));
214 c->felem_size= av_get_bytes_per_sample(c->format);
217 case AV_SAMPLE_FMT_S16P:
218 c->filter_shift = 15;
220 case AV_SAMPLE_FMT_S32P:
221 c->filter_shift = 30;
223 case AV_SAMPLE_FMT_FLTP:
224 case AV_SAMPLE_FMT_DBLP:
228 av_log(NULL, AV_LOG_ERROR, "Unsupported sample format\n");
232 if (filter_size/factor > INT32_MAX/256) {
233 av_log(NULL, AV_LOG_ERROR, "Filter length too large\n");
237 c->phase_shift = phase_shift;
238 c->phase_mask = phase_count - 1;
241 c->filter_length = FFMAX((int)ceil(filter_size/factor), 1);
242 c->filter_alloc = FFALIGN(c->filter_length, 8);
243 c->filter_bank = av_calloc(c->filter_alloc, (phase_count+1)*c->felem_size);
244 c->filter_type = filter_type;
245 c->kaiser_beta = kaiser_beta;
248 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))
250 memcpy(c->filter_bank + (c->filter_alloc*phase_count+1)*c->felem_size, c->filter_bank, (c->filter_alloc-1)*c->felem_size);
251 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);
254 c->compensation_distance= 0;
255 if(!av_reduce(&c->src_incr, &c->dst_incr, out_rate, in_rate * (int64_t)phase_count, INT32_MAX/2))
257 c->ideal_dst_incr= c->dst_incr;
259 c->index= -phase_count*((c->filter_length-1)/2);
264 av_freep(&c->filter_bank);
269 static void resample_free(ResampleContext **c){
272 av_freep(&(*c)->filter_bank);
276 static int set_compensation(ResampleContext *c, int sample_delta, int compensation_distance){
277 c->compensation_distance= compensation_distance;
278 if (compensation_distance)
279 c->dst_incr = c->ideal_dst_incr - c->ideal_dst_incr * (int64_t)sample_delta / compensation_distance;
281 c->dst_incr = c->ideal_dst_incr;
285 #define TEMPLATE_RESAMPLE_S16
286 #include "resample_template.c"
287 #undef TEMPLATE_RESAMPLE_S16
289 #define TEMPLATE_RESAMPLE_S32
290 #include "resample_template.c"
291 #undef TEMPLATE_RESAMPLE_S32
293 #define TEMPLATE_RESAMPLE_FLT
294 #include "resample_template.c"
295 #undef TEMPLATE_RESAMPLE_FLT
297 #define TEMPLATE_RESAMPLE_DBL
298 #include "resample_template.c"
299 #undef TEMPLATE_RESAMPLE_DBL
301 // XXX FIXME the whole C loop should be written in asm so this x86 specific code here isnt needed
302 #if HAVE_MMXEXT_INLINE
304 #include "x86/resample_mmx.h"
306 #define TEMPLATE_RESAMPLE_S16_MMX2
307 #include "resample_template.c"
308 #undef TEMPLATE_RESAMPLE_S16_MMX2
311 #define TEMPLATE_RESAMPLE_FLT_SSE
312 #include "resample_template.c"
313 #undef TEMPLATE_RESAMPLE_FLT_SSE
317 #define TEMPLATE_RESAMPLE_S16_SSE2
318 #include "resample_template.c"
319 #undef TEMPLATE_RESAMPLE_S16_SSE2
321 #define TEMPLATE_RESAMPLE_DBL_SSE2
322 #include "resample_template.c"
323 #undef TEMPLATE_RESAMPLE_DBL_SSE2
327 #define TEMPLATE_RESAMPLE_FLT_AVX
328 #include "resample_template.c"
329 #undef TEMPLATE_RESAMPLE_FLT_AVX
332 #endif // HAVE_MMXEXT_INLINE
334 static int multiple_resample(ResampleContext *c, AudioData *dst, int dst_size, AudioData *src, int src_size, int *consumed){
336 int av_unused mm_flags = av_get_cpu_flags();
339 for(i=0; i<dst->ch_count; i++){
340 #if HAVE_MMXEXT_INLINE
342 if(c->format == AV_SAMPLE_FMT_S16P && (mm_flags&AV_CPU_FLAG_SSE2)) ret= swri_resample_int16_sse2 (c, (int16_t*)dst->ch[i], (const int16_t*)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
345 if(c->format == AV_SAMPLE_FMT_S16P && (mm_flags&AV_CPU_FLAG_MMX2 )){
346 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);
350 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);
351 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);
353 else if(c->format == AV_SAMPLE_FMT_FLTP && (mm_flags&AV_CPU_FLAG_AVX))
354 ret= swri_resample_float_avx (c, (float*)dst->ch[i], (const float*)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
357 else if(c->format == AV_SAMPLE_FMT_FLTP && (mm_flags&AV_CPU_FLAG_SSE))
358 ret= swri_resample_float_sse (c, (float*)dst->ch[i], (const float*)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
360 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);
362 else if(c->format == AV_SAMPLE_FMT_DBLP && (mm_flags&AV_CPU_FLAG_SSE2))
363 ret= swri_resample_double_sse2(c,(double *)dst->ch[i], (const double *)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
365 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);
372 static int64_t get_delay(struct SwrContext *s, int64_t base){
373 ResampleContext *c = s->resample;
374 int64_t num = s->in_buffer_count - (c->filter_length-1)/2;
375 num <<= c->phase_shift;
379 return av_rescale(num, base, s->in_sample_rate*(int64_t)c->src_incr << c->phase_shift);
382 static int resample_flush(struct SwrContext *s) {
383 AudioData *a= &s->in_buffer;
385 if((ret = swri_realloc_audio(a, s->in_buffer_index + 2*s->in_buffer_count)) < 0)
387 av_assert0(a->planar);
388 for(i=0; i<a->ch_count; i++){
389 for(j=0; j<s->in_buffer_count; j++){
390 memcpy(a->ch[i] + (s->in_buffer_index+s->in_buffer_count+j )*a->bps,
391 a->ch[i] + (s->in_buffer_index+s->in_buffer_count-j-1)*a->bps, a->bps);
394 s->in_buffer_count += (s->in_buffer_count+1)/2;
398 struct Resampler const swri_resampler={