2 * Copyright (C) 2011-2013 Michael Niedermayer (michaelni@gmx.at)
4 * This file is part of libswresample
6 * libswresample is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * libswresample is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with libswresample; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "libavutil/opt.h"
22 #include "swresample_internal.h"
23 #include "audioconvert.h"
24 #include "libavutil/avassert.h"
25 #include "libavutil/channel_layout.h"
26 #include "libavutil/internal.h"
32 #include "libavutil/ffversion.h"
33 const char swr_ffversion[] = "FFmpeg version " FFMPEG_VERSION;
35 unsigned swresample_version(void)
37 av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
38 return LIBSWRESAMPLE_VERSION_INT;
41 const char *swresample_configuration(void)
43 return FFMPEG_CONFIGURATION;
46 const char *swresample_license(void)
48 #define LICENSE_PREFIX "libswresample license: "
49 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
52 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
53 if(!s || s->in_convert) // s needs to be allocated but not initialized
54 return AVERROR(EINVAL);
55 s->channel_map = channel_map;
59 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
60 int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
61 int64_t in_ch_layout, enum AVSampleFormat in_sample_fmt, int in_sample_rate,
62 int log_offset, void *log_ctx){
63 if(!s) s= swr_alloc();
66 s->log_level_offset= log_offset;
69 if (av_opt_set_int(s, "ocl", out_ch_layout, 0) < 0)
72 if (av_opt_set_int(s, "osf", out_sample_fmt, 0) < 0)
75 if (av_opt_set_int(s, "osr", out_sample_rate, 0) < 0)
78 if (av_opt_set_int(s, "icl", in_ch_layout, 0) < 0)
81 if (av_opt_set_int(s, "isf", in_sample_fmt, 0) < 0)
84 if (av_opt_set_int(s, "isr", in_sample_rate, 0) < 0)
87 if (av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_NONE, 0) < 0)
90 if (av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> user_in_ch_layout), 0) < 0)
93 if (av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->user_out_ch_layout), 0) < 0)
96 av_opt_set_int(s, "uch", 0, 0);
99 av_log(s, AV_LOG_ERROR, "Failed to set option\n");
104 static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
106 a->bps = av_get_bytes_per_sample(fmt);
107 a->planar= av_sample_fmt_is_planar(fmt);
108 if (a->ch_count == 1)
112 static void free_temp(AudioData *a){
114 memset(a, 0, sizeof(*a));
117 static void clear_context(SwrContext *s){
118 s->in_buffer_index= 0;
119 s->in_buffer_count= 0;
120 s->resample_in_constraint= 0;
121 memset(s->in.ch, 0, sizeof(s->in.ch));
122 memset(s->out.ch, 0, sizeof(s->out.ch));
123 free_temp(&s->postin);
124 free_temp(&s->midbuf);
125 free_temp(&s->preout);
126 free_temp(&s->in_buffer);
127 free_temp(&s->silence);
128 free_temp(&s->drop_temp);
129 free_temp(&s->dither.noise);
130 free_temp(&s->dither.temp);
131 swri_audio_convert_free(&s-> in_convert);
132 swri_audio_convert_free(&s->out_convert);
133 swri_audio_convert_free(&s->full_convert);
134 swri_rematrix_free(s);
136 s->delayed_samples_fixup = 0;
140 av_cold void swr_free(SwrContext **ss){
145 s->resampler->free(&s->resample);
151 av_cold void swr_close(SwrContext *s){
155 av_cold int swr_init(struct SwrContext *s){
157 char l1[1024], l2[1024];
161 if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
162 av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
163 return AVERROR(EINVAL);
165 if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
166 av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
167 return AVERROR(EINVAL);
170 s->out.ch_count = s-> user_out_ch_count;
171 s-> in.ch_count = s-> user_in_ch_count;
172 s->used_ch_count = s->user_used_ch_count;
174 s-> in_ch_layout = s-> user_in_ch_layout;
175 s->out_ch_layout = s->user_out_ch_layout;
177 s->int_sample_fmt= s->user_int_sample_fmt;
179 s->dither.method = s->user_dither_method;
181 if(av_get_channel_layout_nb_channels(s-> in_ch_layout) > SWR_CH_MAX) {
182 av_log(s, AV_LOG_WARNING, "Input channel layout 0x%"PRIx64" is invalid or unsupported.\n", s-> in_ch_layout);
186 if(av_get_channel_layout_nb_channels(s->out_ch_layout) > SWR_CH_MAX) {
187 av_log(s, AV_LOG_WARNING, "Output channel layout 0x%"PRIx64" is invalid or unsupported.\n", s->out_ch_layout);
188 s->out_ch_layout = 0;
193 case SWR_ENGINE_SOXR: s->resampler = &swri_soxr_resampler; break;
195 case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
197 av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
198 return AVERROR(EINVAL);
201 if(!s->used_ch_count)
202 s->used_ch_count= s->in.ch_count;
204 if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
205 av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
209 if(!s-> in_ch_layout)
210 s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
211 if(!s->out_ch_layout)
212 s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
214 s->rematrix= s->out_ch_layout !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
217 if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
218 if( av_get_bytes_per_sample(s-> in_sample_fmt) <= 2
219 && av_get_bytes_per_sample(s->out_sample_fmt) <= 2){
220 s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
221 }else if( av_get_bytes_per_sample(s-> in_sample_fmt) <= 2
223 && s->out_sample_rate==s->in_sample_rate
224 && !(s->flags & SWR_FLAG_RESAMPLE)){
225 s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
226 }else if( av_get_planar_sample_fmt(s-> in_sample_fmt) == AV_SAMPLE_FMT_S32P
227 && av_get_planar_sample_fmt(s->out_sample_fmt) == AV_SAMPLE_FMT_S32P
229 && s->engine != SWR_ENGINE_SOXR){
230 s->int_sample_fmt= AV_SAMPLE_FMT_S32P;
231 }else if(av_get_bytes_per_sample(s->in_sample_fmt) <= 4){
232 s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
234 s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
237 av_log(s, AV_LOG_DEBUG, "Using %s internally between filters\n", av_get_sample_fmt_name(s->int_sample_fmt));
239 if( s->int_sample_fmt != AV_SAMPLE_FMT_S16P
240 &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
241 &&s->int_sample_fmt != AV_SAMPLE_FMT_S64P
242 &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
243 &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
244 av_log(s, AV_LOG_ERROR, "Requested sample format %s is not supported internally, S16/S32/S64/FLT/DBL is supported\n", av_get_sample_fmt_name(s->int_sample_fmt));
245 return AVERROR(EINVAL);
248 set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
249 set_audiodata_fmt(&s->out, s->out_sample_fmt);
251 if (s->firstpts_in_samples != AV_NOPTS_VALUE) {
252 if (!s->async && s->min_compensation >= FLT_MAX/2)
255 s->outpts = s->firstpts_in_samples * s->out_sample_rate;
257 s->firstpts = AV_NOPTS_VALUE;
260 if (s->min_compensation >= FLT_MAX/2)
261 s->min_compensation = 0.001;
262 if (s->async > 1.0001) {
263 s->max_soft_compensation = s->async / (double) s->in_sample_rate;
267 if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
268 s->resample = s->resampler->init(s->resample, s->out_sample_rate, s->in_sample_rate, s->filter_size, s->phase_shift, s->linear_interp, s->cutoff, s->int_sample_fmt, s->filter_type, s->kaiser_beta, s->precision, s->cheby, s->exact_rational);
270 av_log(s, AV_LOG_ERROR, "Failed to initialize resampler\n");
271 return AVERROR(ENOMEM);
274 s->resampler->free(&s->resample);
275 if( s->int_sample_fmt != AV_SAMPLE_FMT_S16P
276 && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
277 && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
278 && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
280 av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
281 ret = AVERROR(EINVAL);
285 #define RSC 1 //FIXME finetune
287 s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
288 if(!s->used_ch_count)
289 s->used_ch_count= s->in.ch_count;
291 s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
293 if(!s-> in.ch_count){
294 av_assert0(!s->in_ch_layout);
295 av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
296 ret = AVERROR(EINVAL);
300 av_get_channel_layout_string(l1, sizeof(l1), s-> in.ch_count, s-> in_ch_layout);
301 av_get_channel_layout_string(l2, sizeof(l2), s->out.ch_count, s->out_ch_layout);
302 if (s->out_ch_layout && s->out.ch_count != av_get_channel_layout_nb_channels(s->out_ch_layout)) {
303 av_log(s, AV_LOG_ERROR, "Output channel layout %s mismatches specified channel count %d\n", l2, s->out.ch_count);
304 ret = AVERROR(EINVAL);
307 if (s->in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s->in_ch_layout)) {
308 av_log(s, AV_LOG_ERROR, "Input channel layout %s mismatches specified channel count %d\n", l1, s->used_ch_count);
309 ret = AVERROR(EINVAL);
313 if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
314 av_log(s, AV_LOG_ERROR, "Rematrix is needed between %s and %s "
315 "but there is not enough information to do it\n", l1, l2);
316 ret = AVERROR(EINVAL);
320 av_assert0(s->used_ch_count);
321 av_assert0(s->out.ch_count);
322 s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
326 s->drop_temp= s->out;
328 if ((ret = swri_dither_init(s, s->out_sample_fmt, s->int_sample_fmt)) < 0)
331 if(!s->resample && !s->rematrix && !s->channel_map && !s->dither.method){
332 s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
333 s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
337 s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
338 s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
339 s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
340 s->int_sample_fmt, s->out.ch_count, NULL, 0);
342 if (!s->in_convert || !s->out_convert) {
343 ret = AVERROR(ENOMEM);
353 s->midbuf.ch_count= s->used_ch_count;
355 s->in_buffer.ch_count= s->used_ch_count;
357 if(!s->resample_first){
358 s->midbuf.ch_count= s->out.ch_count;
360 s->in_buffer.ch_count = s->out.ch_count;
363 set_audiodata_fmt(&s->postin, s->int_sample_fmt);
364 set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
365 set_audiodata_fmt(&s->preout, s->int_sample_fmt);
368 set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
371 av_assert0(!s->preout.count);
372 s->dither.noise = s->preout;
373 s->dither.temp = s->preout;
374 if (s->dither.method > SWR_DITHER_NS) {
375 s->dither.noise.bps = 4;
376 s->dither.noise.fmt = AV_SAMPLE_FMT_FLTP;
377 s->dither.noise_scale = 1;
380 if(s->rematrix || s->dither.method) {
381 ret = swri_rematrix_init(s);
393 int swri_realloc_audio(AudioData *a, int count){
397 if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
398 return AVERROR(EINVAL);
400 if(a->count >= count)
405 countb= FFALIGN(count*a->bps, ALIGN);
409 av_assert0(a->ch_count);
411 a->data= av_mallocz_array(countb, a->ch_count);
413 return AVERROR(ENOMEM);
414 for(i=0; i<a->ch_count; i++){
415 a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
416 if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
418 if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
425 static void copy(AudioData *out, AudioData *in,
427 av_assert0(out->planar == in->planar);
428 av_assert0(out->bps == in->bps);
429 av_assert0(out->ch_count == in->ch_count);
432 for(ch=0; ch<out->ch_count; ch++)
433 memcpy(out->ch[ch], in->ch[ch], count*out->bps);
435 memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
438 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
441 memset(out->ch, 0, sizeof(out->ch));
442 }else if(out->planar){
443 for(i=0; i<out->ch_count; i++)
444 out->ch[i]= in_arg[i];
446 for(i=0; i<out->ch_count; i++)
447 out->ch[i]= in_arg[0] + i*out->bps;
451 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
454 for(i=0; i<out->ch_count; i++)
455 in_arg[i]= out->ch[i];
457 in_arg[0]= out->ch[0];
463 * out may be equal in.
465 static void buf_set(AudioData *out, AudioData *in, int count){
468 for(ch=0; ch<out->ch_count; ch++)
469 out->ch[ch]= in->ch[ch] + count*out->bps;
471 for(ch=out->ch_count-1; ch>=0; ch--)
472 out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
478 * @return number of samples output per channel
480 static int resample(SwrContext *s, AudioData *out_param, int out_count,
481 const AudioData * in_param, int in_count){
482 AudioData in, out, tmp;
485 int padless = ARCH_X86 && s->engine == SWR_ENGINE_SWR ? 7 : 0;
487 av_assert1(s->in_buffer.ch_count == in_param->ch_count);
488 av_assert1(s->in_buffer.planar == in_param->planar);
489 av_assert1(s->in_buffer.fmt == in_param->fmt);
494 border = s->resampler->invert_initial_buffer(s->resample, &s->in_buffer,
495 &in, in_count, &s->in_buffer_index, &s->in_buffer_count);
496 if (border == INT_MAX) {
498 } else if (border < 0) {
501 buf_set(&in, &in, border);
503 s->resample_in_constraint = 0;
507 int ret, size, consumed;
508 if(!s->resample_in_constraint && s->in_buffer_count){
509 buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
510 ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
513 buf_set(&out, &out, ret);
514 s->in_buffer_count -= consumed;
515 s->in_buffer_index += consumed;
519 if(s->in_buffer_count <= border){
520 buf_set(&in, &in, -s->in_buffer_count);
521 in_count += s->in_buffer_count;
522 s->in_buffer_count=0;
523 s->in_buffer_index=0;
528 if((s->flushed || in_count > padless) && !s->in_buffer_count){
529 s->in_buffer_index=0;
530 ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, FFMAX(in_count-padless, 0), &consumed);
533 buf_set(&out, &out, ret);
534 in_count -= consumed;
535 buf_set(&in, &in, consumed);
538 //TODO is this check sane considering the advanced copy avoidance below
539 size= s->in_buffer_index + s->in_buffer_count + in_count;
540 if( size > s->in_buffer.count
541 && s->in_buffer_count + in_count <= s->in_buffer_index){
542 buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
543 copy(&s->in_buffer, &tmp, s->in_buffer_count);
544 s->in_buffer_index=0;
546 if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
551 if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
553 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
554 copy(&tmp, &in, /*in_*/count);
555 s->in_buffer_count += count;
558 buf_set(&in, &in, count);
559 s->resample_in_constraint= 0;
560 if(s->in_buffer_count != count || in_count)
570 s->resample_in_constraint= !!out_count;
575 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
576 AudioData *in , int in_count){
577 AudioData *postin, *midbuf, *preout;
579 AudioData preout_tmp, midbuf_tmp;
582 av_assert0(!s->resample);
583 swri_audio_convert(s->full_convert, out, in, in_count);
587 // in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
588 // in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
590 if((ret=swri_realloc_audio(&s->postin, in_count))<0)
592 if(s->resample_first){
593 av_assert0(s->midbuf.ch_count == s->used_ch_count);
594 if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
597 av_assert0(s->midbuf.ch_count == s->out.ch_count);
598 if((ret=swri_realloc_audio(&s->midbuf, in_count))<0)
601 if((ret=swri_realloc_audio(&s->preout, out_count))<0)
606 midbuf_tmp= s->midbuf;
608 preout_tmp= s->preout;
611 if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
614 if(s->resample_first ? !s->resample : !s->rematrix)
617 if(s->resample_first ? !s->rematrix : !s->resample)
620 if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar
621 && !(s->out_sample_fmt==AV_SAMPLE_FMT_S32P && (s->dither.output_sample_bits&31))){
623 out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
624 av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
625 copy(out, in, out_count);
628 else if(preout==postin) preout= midbuf= postin= out;
629 else if(preout==midbuf) preout= midbuf= out;
634 swri_audio_convert(s->in_convert, postin, in, in_count);
637 if(s->resample_first){
639 out_count= resample(s, midbuf, out_count, postin, in_count);
641 swri_rematrix(s, preout, midbuf, out_count, preout==out);
644 swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
646 out_count= resample(s, preout, out_count, midbuf, in_count);
649 if(preout != out && out_count){
650 AudioData *conv_src = preout;
651 if(s->dither.method){
653 int dither_count= FFMAX(out_count, 1<<16);
656 conv_src = &s->dither.temp;
657 if((ret=swri_realloc_audio(&s->dither.temp, dither_count))<0)
661 if((ret=swri_realloc_audio(&s->dither.noise, dither_count))<0)
664 for(ch=0; ch<s->dither.noise.ch_count; ch++)
665 if((ret=swri_get_dither(s, s->dither.noise.ch[ch], s->dither.noise.count, (12345678913579ULL*ch + 3141592) % 2718281828U, s->dither.noise.fmt))<0)
667 av_assert0(s->dither.noise.ch_count == preout->ch_count);
669 if(s->dither.noise_pos + out_count > s->dither.noise.count)
670 s->dither.noise_pos = 0;
672 if (s->dither.method < SWR_DITHER_NS){
673 if (s->mix_2_1_simd) {
674 int len1= out_count&~15;
675 int off = len1 * preout->bps;
678 for(ch=0; ch<preout->ch_count; ch++)
679 s->mix_2_1_simd(conv_src->ch[ch], preout->ch[ch], s->dither.noise.ch[ch] + s->dither.noise.bps * s->dither.noise_pos, s->native_simd_one, 0, 0, len1);
680 if(out_count != len1)
681 for(ch=0; ch<preout->ch_count; ch++)
682 s->mix_2_1_f(conv_src->ch[ch] + off, preout->ch[ch] + off, s->dither.noise.ch[ch] + s->dither.noise.bps * s->dither.noise_pos + off + len1, s->native_one, 0, 0, out_count - len1);
684 for(ch=0; ch<preout->ch_count; ch++)
685 s->mix_2_1_f(conv_src->ch[ch], preout->ch[ch], s->dither.noise.ch[ch] + s->dither.noise.bps * s->dither.noise_pos, s->native_one, 0, 0, out_count);
688 switch(s->int_sample_fmt) {
689 case AV_SAMPLE_FMT_S16P :swri_noise_shaping_int16(s, conv_src, preout, &s->dither.noise, out_count); break;
690 case AV_SAMPLE_FMT_S32P :swri_noise_shaping_int32(s, conv_src, preout, &s->dither.noise, out_count); break;
691 case AV_SAMPLE_FMT_FLTP :swri_noise_shaping_float(s, conv_src, preout, &s->dither.noise, out_count); break;
692 case AV_SAMPLE_FMT_DBLP :swri_noise_shaping_double(s,conv_src, preout, &s->dither.noise, out_count); break;
695 s->dither.noise_pos += out_count;
697 //FIXME packed doesn't need more than 1 chan here!
698 swri_audio_convert(s->out_convert, out, conv_src, out_count);
703 int swr_is_initialized(struct SwrContext *s) {
704 return !!s->in_buffer.ch_count;
707 int attribute_align_arg swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
708 const uint8_t *in_arg [SWR_CH_MAX], int in_count){
709 AudioData * in= &s->in;
710 AudioData *out= &s->out;
711 int av_unused max_output;
713 if (!swr_is_initialized(s)) {
714 av_log(s, AV_LOG_ERROR, "Context has not been initialized\n");
715 return AVERROR(EINVAL);
717 #if defined(ASSERT_LEVEL) && ASSERT_LEVEL >1
718 max_output = swr_get_out_samples(s, in_count);
721 while(s->drop_output > 0){
723 uint8_t *tmp_arg[SWR_CH_MAX];
724 #define MAX_DROP_STEP 16384
725 if((ret=swri_realloc_audio(&s->drop_temp, FFMIN(s->drop_output, MAX_DROP_STEP)))<0)
728 reversefill_audiodata(&s->drop_temp, tmp_arg);
729 s->drop_output *= -1; //FIXME find a less hackish solution
730 ret = swr_convert(s, tmp_arg, FFMIN(-s->drop_output, MAX_DROP_STEP), in_arg, in_count); //FIXME optimize but this is as good as never called so maybe it doesn't matter
731 s->drop_output *= -1;
734 s->drop_output -= ret;
735 if (!s->drop_output && !out_arg)
740 av_assert0(s->drop_output);
747 s->resampler->flush(s);
748 s->resample_in_constraint = 0;
750 }else if(!s->in_buffer_count){
754 fill_audiodata(in , (void*)in_arg);
756 fill_audiodata(out, out_arg);
759 int ret = swr_convert_internal(s, out, out_count, in, in_count);
760 if(ret>0 && !s->drop_output)
761 s->outpts += ret * (int64_t)s->in_sample_rate;
763 av_assert2(max_output < 0 || ret < 0 || ret <= max_output);
770 size = FFMIN(out_count, s->in_buffer_count);
772 buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
773 ret= swr_convert_internal(s, out, size, &tmp, size);
777 s->in_buffer_count -= ret;
778 s->in_buffer_index += ret;
779 buf_set(out, out, ret);
781 if(!s->in_buffer_count)
782 s->in_buffer_index = 0;
786 size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
788 if(in_count > out_count) { //FIXME move after swr_convert_internal
789 if( size > s->in_buffer.count
790 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
791 buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
792 copy(&s->in_buffer, &tmp, s->in_buffer_count);
793 s->in_buffer_index=0;
795 if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
800 size = FFMIN(in_count, out_count);
801 ret= swr_convert_internal(s, out, size, in, size);
804 buf_set(in, in, ret);
809 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
810 copy(&tmp, in, in_count);
811 s->in_buffer_count += in_count;
814 if(ret2>0 && !s->drop_output)
815 s->outpts += ret2 * (int64_t)s->in_sample_rate;
816 av_assert2(max_output < 0 || ret2 < 0 || ret2 <= max_output);
821 int swr_drop_output(struct SwrContext *s, int count){
822 const uint8_t *tmp_arg[SWR_CH_MAX];
823 s->drop_output += count;
825 if(s->drop_output <= 0)
828 av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
829 return swr_convert(s, NULL, s->drop_output, tmp_arg, 0);
832 int swr_inject_silence(struct SwrContext *s, int count){
834 uint8_t *tmp_arg[SWR_CH_MAX];
839 #define MAX_SILENCE_STEP 16384
840 while (count > MAX_SILENCE_STEP) {
841 if ((ret = swr_inject_silence(s, MAX_SILENCE_STEP)) < 0)
843 count -= MAX_SILENCE_STEP;
846 if((ret=swri_realloc_audio(&s->silence, count))<0)
849 if(s->silence.planar) for(i=0; i<s->silence.ch_count; i++) {
850 memset(s->silence.ch[i], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps);
852 memset(s->silence.ch[0], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps*s->silence.ch_count);
854 reversefill_audiodata(&s->silence, tmp_arg);
855 av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
856 ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
860 int64_t swr_get_delay(struct SwrContext *s, int64_t base){
861 if (s->resampler && s->resample){
862 return s->resampler->get_delay(s, base);
864 return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
868 int swr_get_out_samples(struct SwrContext *s, int in_samples)
873 return AVERROR(EINVAL);
875 if (s->resampler && s->resample) {
876 if (!s->resampler->get_out_samples)
877 return AVERROR(ENOSYS);
878 out_samples = s->resampler->get_out_samples(s, in_samples);
880 out_samples = s->in_buffer_count + in_samples;
881 av_assert0(s->out_sample_rate == s->in_sample_rate);
884 if (out_samples > INT_MAX)
885 return AVERROR(EINVAL);
890 int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
893 if (!s || compensation_distance < 0)
894 return AVERROR(EINVAL);
895 if (!compensation_distance && sample_delta)
896 return AVERROR(EINVAL);
898 s->flags |= SWR_FLAG_RESAMPLE;
903 if (!s->resampler->set_compensation){
904 return AVERROR(EINVAL);
906 return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
910 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
914 if (s->firstpts == AV_NOPTS_VALUE)
915 s->outpts = s->firstpts = pts;
917 if(s->min_compensation >= FLT_MAX) {
918 return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
920 int64_t delta = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate) - s->outpts + s->drop_output*(int64_t)s->in_sample_rate;
921 double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
923 if(fabs(fdelta) > s->min_compensation) {
924 if(s->outpts == s->firstpts || fabs(fdelta) > s->min_hard_compensation){
926 if(delta > 0) ret = swr_inject_silence(s, delta / s->out_sample_rate);
927 else ret = swr_drop_output (s, -delta / s-> in_sample_rate);
929 av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
931 } else if(s->soft_compensation_duration && s->max_soft_compensation) {
932 int duration = s->out_sample_rate * s->soft_compensation_duration;
933 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
934 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
935 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
936 swr_set_compensation(s, comp, duration);