1 /*
2 * Copyright (C) 2011-2013 Michael Niedermayer (michaelni@gmx.at)
3 *
4 * This file is part of libswresample
5 *
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.
10 *
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.
15 *
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
19 */
20
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"
27
28 #include <float.h>
29
30 #define ALIGN 32
31
swr_set_channel_mapping(struct SwrContext * s,const int * channel_map)32 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
33 if(!s || s->in_convert) // s needs to be allocated but not initialized
34 return AVERROR(EINVAL);
35 s->channel_map = channel_map;
36 return 0;
37 }
38
39 #if FF_API_OLD_CHANNEL_LAYOUT
40 FF_DISABLE_DEPRECATION_WARNINGS
swr_alloc_set_opts(struct SwrContext * s,int64_t out_ch_layout,enum AVSampleFormat out_sample_fmt,int out_sample_rate,int64_t in_ch_layout,enum AVSampleFormat in_sample_fmt,int in_sample_rate,int log_offset,void * log_ctx)41 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
42 int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
43 int64_t in_ch_layout, enum AVSampleFormat in_sample_fmt, int in_sample_rate,
44 int log_offset, void *log_ctx){
45 if(!s) s= swr_alloc();
46 if(!s) return NULL;
47
48 s->log_level_offset= log_offset;
49 s->log_ctx= log_ctx;
50
51 if (av_opt_set_int(s, "ocl", out_ch_layout, 0) < 0)
52 goto fail;
53
54 if (av_opt_set_int(s, "osf", out_sample_fmt, 0) < 0)
55 goto fail;
56
57 if (av_opt_set_int(s, "osr", out_sample_rate, 0) < 0)
58 goto fail;
59
60 if (av_opt_set_int(s, "icl", in_ch_layout, 0) < 0)
61 goto fail;
62
63 if (av_opt_set_int(s, "isf", in_sample_fmt, 0) < 0)
64 goto fail;
65
66 if (av_opt_set_int(s, "isr", in_sample_rate, 0) < 0)
67 goto fail;
68
69 if (av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> user_in_ch_layout), 0) < 0)
70 goto fail;
71
72 if (av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->user_out_ch_layout), 0) < 0)
73 goto fail;
74
75 av_opt_set_int(s, "uch", 0, 0);
76 return s;
77 fail:
78 av_log(s, AV_LOG_ERROR, "Failed to set option\n");
79 swr_free(&s);
80 return NULL;
81 }
82 FF_ENABLE_DEPRECATION_WARNINGS
83 #endif
84
swr_alloc_set_opts2(struct SwrContext ** ps,AVChannelLayout * out_ch_layout,enum AVSampleFormat out_sample_fmt,int out_sample_rate,AVChannelLayout * in_ch_layout,enum AVSampleFormat in_sample_fmt,int in_sample_rate,int log_offset,void * log_ctx)85 int swr_alloc_set_opts2(struct SwrContext **ps,
86 AVChannelLayout *out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
87 AVChannelLayout *in_ch_layout, enum AVSampleFormat in_sample_fmt, int in_sample_rate,
88 int log_offset, void *log_ctx) {
89 struct SwrContext *s = *ps;
90 int ret;
91
92 if (!s) s = swr_alloc();
93 if (!s) return AVERROR(ENOMEM);
94
95 *ps = s;
96
97 s->log_level_offset = log_offset;
98 s->log_ctx = log_ctx;
99
100 if ((ret = av_opt_set_chlayout(s, "ochl", out_ch_layout, 0)) < 0)
101 goto fail;
102
103 if ((ret = av_opt_set_int(s, "osf", out_sample_fmt, 0)) < 0)
104 goto fail;
105
106 if ((ret = av_opt_set_int(s, "osr", out_sample_rate, 0)) < 0)
107 goto fail;
108
109 if ((ret = av_opt_set_chlayout(s, "ichl", in_ch_layout, 0)) < 0)
110 goto fail;
111
112 if ((ret = av_opt_set_int(s, "isf", in_sample_fmt, 0)) < 0)
113 goto fail;
114
115 if ((ret = av_opt_set_int(s, "isr", in_sample_rate, 0)) < 0)
116 goto fail;
117
118 av_opt_set_int(s, "uch", 0, 0);
119
120 #if FF_API_OLD_CHANNEL_LAYOUT
121 // Clear old API values so they don't take precedence in swr_init()
122 av_opt_set_int(s, "icl", 0, 0);
123 av_opt_set_int(s, "ocl", 0, 0);
124 av_opt_set_int(s, "ich", 0, 0);
125 av_opt_set_int(s, "och", 0, 0);
126 #endif
127
128 return 0;
129 fail:
130 av_log(s, AV_LOG_ERROR, "Failed to set option\n");
131 swr_free(ps);
132 return ret;
133 }
134
set_audiodata_fmt(AudioData * a,enum AVSampleFormat fmt)135 static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
136 a->fmt = fmt;
137 a->bps = av_get_bytes_per_sample(fmt);
138 a->planar= av_sample_fmt_is_planar(fmt);
139 if (a->ch_count == 1)
140 a->planar = 1;
141 }
142
free_temp(AudioData * a)143 static void free_temp(AudioData *a){
144 av_free(a->data);
145 memset(a, 0, sizeof(*a));
146 }
147
clear_context(SwrContext * s)148 static void clear_context(SwrContext *s){
149 s->in_buffer_index= 0;
150 s->in_buffer_count= 0;
151 s->resample_in_constraint= 0;
152 memset(s->in.ch, 0, sizeof(s->in.ch));
153 memset(s->out.ch, 0, sizeof(s->out.ch));
154 free_temp(&s->postin);
155 free_temp(&s->midbuf);
156 free_temp(&s->preout);
157 free_temp(&s->in_buffer);
158 free_temp(&s->silence);
159 free_temp(&s->drop_temp);
160 free_temp(&s->dither.noise);
161 free_temp(&s->dither.temp);
162 av_channel_layout_uninit(&s->in_ch_layout);
163 av_channel_layout_uninit(&s->out_ch_layout);
164 swri_audio_convert_free(&s-> in_convert);
165 swri_audio_convert_free(&s->out_convert);
166 swri_audio_convert_free(&s->full_convert);
167 swri_rematrix_free(s);
168
169 s->delayed_samples_fixup = 0;
170 s->flushed = 0;
171 }
172
swr_free(SwrContext ** ss)173 av_cold void swr_free(SwrContext **ss){
174 SwrContext *s= *ss;
175 if(s){
176 clear_context(s);
177 av_channel_layout_uninit(&s->user_in_chlayout);
178 av_channel_layout_uninit(&s->user_out_chlayout);
179
180 if (s->resampler)
181 s->resampler->free(&s->resample);
182 }
183
184 av_freep(ss);
185 }
186
swr_close(SwrContext * s)187 av_cold void swr_close(SwrContext *s){
188 clear_context(s);
189 }
190
swr_init(struct SwrContext * s)191 av_cold int swr_init(struct SwrContext *s){
192 int ret;
193 char l1[1024], l2[1024];
194
195 clear_context(s);
196
197 if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
198 av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
199 return AVERROR(EINVAL);
200 }
201 if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
202 av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
203 return AVERROR(EINVAL);
204 }
205
206 if(s-> in_sample_rate <= 0){
207 av_log(s, AV_LOG_ERROR, "Requested input sample rate %d is invalid\n", s->in_sample_rate);
208 return AVERROR(EINVAL);
209 }
210 if(s->out_sample_rate <= 0){
211 av_log(s, AV_LOG_ERROR, "Requested output sample rate %d is invalid\n", s->out_sample_rate);
212 return AVERROR(EINVAL);
213 }
214 s->used_ch_count = s->user_used_ch_count;
215 #if FF_API_OLD_CHANNEL_LAYOUT
216 s->out.ch_count = s-> user_out_ch_count;
217 s-> in.ch_count = s-> user_in_ch_count;
218
219 // if the old/new fields are set inconsistently, prefer the old ones
220 if ((s->user_in_ch_count && s->user_in_ch_count != s->user_in_chlayout.nb_channels) ||
221 (s->user_in_ch_layout && (s->user_in_chlayout.order != AV_CHANNEL_ORDER_NATIVE ||
222 s->user_in_chlayout.u.mask != s->user_in_ch_layout))) {
223 av_channel_layout_uninit(&s->in_ch_layout);
224 if (s->user_in_ch_layout)
225 av_channel_layout_from_mask(&s->in_ch_layout, s->user_in_ch_layout);
226 else {
227 s->in_ch_layout.order = AV_CHANNEL_ORDER_UNSPEC;
228 s->in_ch_layout.nb_channels = s->user_in_ch_count;
229 }
230 } else
231 av_channel_layout_copy(&s->in_ch_layout, &s->user_in_chlayout);
232
233 if ((s->user_out_ch_count && s->user_out_ch_count != s->user_out_chlayout.nb_channels) ||
234 (s->user_out_ch_layout && (s->user_out_chlayout.order != AV_CHANNEL_ORDER_NATIVE ||
235 s->user_out_chlayout.u.mask != s->user_out_ch_layout))) {
236 av_channel_layout_uninit(&s->out_ch_layout);
237 if (s->user_out_ch_layout)
238 av_channel_layout_from_mask(&s->out_ch_layout, s->user_out_ch_layout);
239 else {
240 s->out_ch_layout.order = AV_CHANNEL_ORDER_UNSPEC;
241 s->out_ch_layout.nb_channels = s->user_out_ch_count;
242 }
243 } else
244 av_channel_layout_copy(&s->out_ch_layout, &s->user_out_chlayout);
245
246 if (!s->out.ch_count && !s->user_out_ch_layout)
247 s->out.ch_count = s->out_ch_layout.nb_channels;
248 if (!s-> in.ch_count && !s-> user_in_ch_layout)
249 s-> in.ch_count = s->in_ch_layout.nb_channels;
250 #else
251 s->out.ch_count = s-> user_out_chlayout.nb_channels;
252 s-> in.ch_count = s-> user_in_chlayout.nb_channels;
253
254 ret = av_channel_layout_copy(&s->in_ch_layout, &s->user_in_chlayout);
255 ret |= av_channel_layout_copy(&s->out_ch_layout, &s->user_out_chlayout);
256 if (ret < 0)
257 return ret;
258 #endif
259
260 s->int_sample_fmt= s->user_int_sample_fmt;
261
262 s->dither.method = s->user_dither_method;
263
264 if (!av_channel_layout_check(&s->in_ch_layout) || s->in_ch_layout.nb_channels > SWR_CH_MAX) {
265 av_channel_layout_describe(&s->in_ch_layout, l1, sizeof(l1));
266 av_log(s, AV_LOG_WARNING, "Input channel layout \"%s\" is invalid or unsupported.\n", l1);
267 av_channel_layout_uninit(&s->in_ch_layout);
268 }
269
270 if (!av_channel_layout_check(&s->out_ch_layout) || s->out_ch_layout.nb_channels > SWR_CH_MAX) {
271 av_channel_layout_describe(&s->out_ch_layout, l2, sizeof(l2));
272 av_log(s, AV_LOG_WARNING, "Output channel layout \"%s\" is invalid or unsupported.\n", l2);
273 av_channel_layout_uninit(&s->out_ch_layout);
274 }
275
276 switch(s->engine){
277 #if CONFIG_LIBSOXR
278 case SWR_ENGINE_SOXR: s->resampler = &swri_soxr_resampler; break;
279 #endif
280 case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
281 default:
282 av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
283 return AVERROR(EINVAL);
284 }
285
286 if(!s->used_ch_count)
287 s->used_ch_count= s->in.ch_count;
288
289 if (s->used_ch_count && s->in_ch_layout.order != AV_CHANNEL_ORDER_UNSPEC && s->used_ch_count != s->in_ch_layout.nb_channels) {
290 av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
291 av_channel_layout_uninit(&s->in_ch_layout);
292 }
293
294 if (!s->in_ch_layout.nb_channels || s->in_ch_layout.order == AV_CHANNEL_ORDER_UNSPEC)
295 av_channel_layout_default(&s->in_ch_layout, s->used_ch_count);
296 if (!s->out_ch_layout.nb_channels || s->out_ch_layout.order == AV_CHANNEL_ORDER_UNSPEC)
297 av_channel_layout_default(&s->out_ch_layout, s->out.ch_count);
298
299 s->rematrix = av_channel_layout_compare(&s->out_ch_layout, &s->in_ch_layout) ||
300 s->rematrix_volume!=1.0 ||
301 s->rematrix_custom;
302
303 if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
304 if( av_get_bytes_per_sample(s-> in_sample_fmt) <= 2
305 && av_get_bytes_per_sample(s->out_sample_fmt) <= 2){
306 s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
307 }else if( av_get_bytes_per_sample(s-> in_sample_fmt) <= 2
308 && !s->rematrix
309 && s->out_sample_rate==s->in_sample_rate
310 && !(s->flags & SWR_FLAG_RESAMPLE)){
311 s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
312 }else if( av_get_planar_sample_fmt(s-> in_sample_fmt) == AV_SAMPLE_FMT_S32P
313 && av_get_planar_sample_fmt(s->out_sample_fmt) == AV_SAMPLE_FMT_S32P
314 && !s->rematrix
315 && s->out_sample_rate == s->in_sample_rate
316 && !(s->flags & SWR_FLAG_RESAMPLE)
317 && s->engine != SWR_ENGINE_SOXR){
318 s->int_sample_fmt= AV_SAMPLE_FMT_S32P;
319 }else if(av_get_bytes_per_sample(s->in_sample_fmt) <= 4){
320 s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
321 }else{
322 s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
323 }
324 }
325 av_log(s, AV_LOG_DEBUG, "Using %s internally between filters\n", av_get_sample_fmt_name(s->int_sample_fmt));
326
327 if( s->int_sample_fmt != AV_SAMPLE_FMT_S16P
328 &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
329 &&s->int_sample_fmt != AV_SAMPLE_FMT_S64P
330 &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
331 &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
332 av_log(s, AV_LOG_ERROR, "Requested sample format %s is not supported internally, s16p/s32p/s64p/fltp/dblp are supported\n", av_get_sample_fmt_name(s->int_sample_fmt));
333 return AVERROR(EINVAL);
334 }
335
336 set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
337 set_audiodata_fmt(&s->out, s->out_sample_fmt);
338
339 if (s->firstpts_in_samples != AV_NOPTS_VALUE) {
340 if (!s->async && s->min_compensation >= FLT_MAX/2)
341 s->async = 1;
342 s->firstpts =
343 s->outpts = s->firstpts_in_samples * s->out_sample_rate;
344 } else
345 s->firstpts = AV_NOPTS_VALUE;
346
347 if (s->async) {
348 if (s->min_compensation >= FLT_MAX/2)
349 s->min_compensation = 0.001;
350 if (s->async > 1.0001) {
351 s->max_soft_compensation = s->async / (double) s->in_sample_rate;
352 }
353 }
354
355 if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
356 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);
357 if (!s->resample) {
358 av_log(s, AV_LOG_ERROR, "Failed to initialize resampler\n");
359 return AVERROR(ENOMEM);
360 }
361 }else
362 s->resampler->free(&s->resample);
363 if( s->int_sample_fmt != AV_SAMPLE_FMT_S16P
364 && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
365 && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
366 && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
367 && s->resample){
368 av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16p/s32p/fltp/dblp\n");
369 ret = AVERROR(EINVAL);
370 goto fail;
371 }
372
373 #define RSC 1 //FIXME finetune
374 if(!s-> in.ch_count)
375 s-> in.ch_count = s->in_ch_layout.nb_channels;
376 if(!s->used_ch_count)
377 s->used_ch_count= s->in.ch_count;
378 if(!s->out.ch_count)
379 s->out.ch_count = s->out_ch_layout.nb_channels;
380
381 if(!s-> in.ch_count){
382 av_assert0(s->in_ch_layout.order == AV_CHANNEL_ORDER_UNSPEC);
383 av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
384 ret = AVERROR(EINVAL);
385 goto fail;
386 }
387
388 av_channel_layout_describe(&s->out_ch_layout, l2, sizeof(l2));
389 #if FF_API_OLD_CHANNEL_LAYOUT
390 if (s->out_ch_layout.order != AV_CHANNEL_ORDER_UNSPEC && s->out.ch_count != s->out_ch_layout.nb_channels) {
391 av_log(s, AV_LOG_ERROR, "Output channel layout %s mismatches specified channel count %d\n", l2, s->out.ch_count);
392 ret = AVERROR(EINVAL);
393 goto fail;
394 }
395 #endif
396 av_channel_layout_describe(&s->in_ch_layout, l1, sizeof(l1));
397 if (s->in_ch_layout.order != AV_CHANNEL_ORDER_UNSPEC && s->used_ch_count != s->in_ch_layout.nb_channels) {
398 av_log(s, AV_LOG_ERROR, "Input channel layout %s mismatches specified channel count %d\n", l1, s->used_ch_count);
399 ret = AVERROR(EINVAL);
400 goto fail;
401 }
402
403 if (( s->out_ch_layout.order == AV_CHANNEL_ORDER_UNSPEC
404 || s-> in_ch_layout.order == AV_CHANNEL_ORDER_UNSPEC) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
405 av_log(s, AV_LOG_ERROR, "Rematrix is needed between %s and %s "
406 "but there is not enough information to do it\n", l1, l2);
407 ret = AVERROR(EINVAL);
408 goto fail;
409 }
410
411 av_assert0(s->used_ch_count);
412 av_assert0(s->out.ch_count);
413 s->resample_first= RSC*s->out.ch_count/s->used_ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
414
415 s->in_buffer= s->in;
416 s->silence = s->in;
417 s->drop_temp= s->out;
418
419 if ((ret = swri_dither_init(s, s->out_sample_fmt, s->int_sample_fmt)) < 0)
420 goto fail;
421
422 if(!s->resample && !s->rematrix && !s->channel_map && !s->dither.method){
423 s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
424 s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
425 return 0;
426 }
427
428 s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
429 s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
430 s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
431 s->int_sample_fmt, s->out.ch_count, NULL, 0);
432
433 if (!s->in_convert || !s->out_convert) {
434 ret = AVERROR(ENOMEM);
435 goto fail;
436 }
437
438 s->postin= s->in;
439 s->preout= s->out;
440 s->midbuf= s->in;
441
442 if(s->channel_map){
443 s->postin.ch_count=
444 s->midbuf.ch_count= s->used_ch_count;
445 if(s->resample)
446 s->in_buffer.ch_count= s->used_ch_count;
447 }
448 if(!s->resample_first){
449 s->midbuf.ch_count= s->out.ch_count;
450 if(s->resample)
451 s->in_buffer.ch_count = s->out.ch_count;
452 }
453
454 set_audiodata_fmt(&s->postin, s->int_sample_fmt);
455 set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
456 set_audiodata_fmt(&s->preout, s->int_sample_fmt);
457
458 if(s->resample){
459 set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
460 }
461
462 av_assert0(!s->preout.count);
463 s->dither.noise = s->preout;
464 s->dither.temp = s->preout;
465 if (s->dither.method > SWR_DITHER_NS) {
466 s->dither.noise.bps = 4;
467 s->dither.noise.fmt = AV_SAMPLE_FMT_FLTP;
468 s->dither.noise_scale = 1;
469 }
470
471 if(s->rematrix || s->dither.method) {
472 ret = swri_rematrix_init(s);
473 if (ret < 0)
474 goto fail;
475 }
476
477 return 0;
478 fail:
479 swr_close(s);
480 return ret;
481
482 }
483
swri_realloc_audio(AudioData * a,int count)484 int swri_realloc_audio(AudioData *a, int count){
485 int i, countb;
486 AudioData old;
487
488 if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
489 return AVERROR(EINVAL);
490
491 if(a->count >= count)
492 return 0;
493
494 count*=2;
495
496 countb= FFALIGN(count*a->bps, ALIGN);
497 old= *a;
498
499 av_assert0(a->bps);
500 av_assert0(a->ch_count);
501
502 a->data = av_calloc(countb, a->ch_count);
503 if(!a->data)
504 return AVERROR(ENOMEM);
505 for(i=0; i<a->ch_count; i++){
506 a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
507 if(a->count && a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
508 }
509 if(a->count && !a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
510 av_freep(&old.data);
511 a->count= count;
512
513 return 1;
514 }
515
copy(AudioData * out,AudioData * in,int count)516 static void copy(AudioData *out, AudioData *in,
517 int count){
518 av_assert0(out->planar == in->planar);
519 av_assert0(out->bps == in->bps);
520 av_assert0(out->ch_count == in->ch_count);
521 if(out->planar){
522 int ch;
523 for(ch=0; ch<out->ch_count; ch++)
524 memcpy(out->ch[ch], in->ch[ch], count*out->bps);
525 }else
526 memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
527 }
528
fill_audiodata(AudioData * out,uint8_t * in_arg[SWR_CH_MAX])529 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
530 int i;
531 if(!in_arg){
532 memset(out->ch, 0, sizeof(out->ch));
533 }else if(out->planar){
534 for(i=0; i<out->ch_count; i++)
535 out->ch[i]= in_arg[i];
536 }else{
537 for(i=0; i<out->ch_count; i++)
538 out->ch[i]= in_arg[0] + i*out->bps;
539 }
540 }
541
reversefill_audiodata(AudioData * out,uint8_t * in_arg[SWR_CH_MAX])542 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
543 int i;
544 if(out->planar){
545 for(i=0; i<out->ch_count; i++)
546 in_arg[i]= out->ch[i];
547 }else{
548 in_arg[0]= out->ch[0];
549 }
550 }
551
552 /**
553 *
554 * out may be equal in.
555 */
buf_set(AudioData * out,AudioData * in,int count)556 static void buf_set(AudioData *out, AudioData *in, int count){
557 int ch;
558 if(in->planar){
559 for(ch=0; ch<out->ch_count; ch++)
560 out->ch[ch]= in->ch[ch] + count*out->bps;
561 }else{
562 for(ch=out->ch_count-1; ch>=0; ch--)
563 out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
564 }
565 }
566
567 /**
568 *
569 * @return number of samples output per channel
570 */
resample(SwrContext * s,AudioData * out_param,int out_count,const AudioData * in_param,int in_count)571 static int resample(SwrContext *s, AudioData *out_param, int out_count,
572 const AudioData * in_param, int in_count){
573 AudioData in, out, tmp;
574 int ret_sum=0;
575 int border=0;
576 int padless = ARCH_X86 && s->engine == SWR_ENGINE_SWR ? 7 : 0;
577
578 av_assert1(s->in_buffer.ch_count == in_param->ch_count);
579 av_assert1(s->in_buffer.planar == in_param->planar);
580 av_assert1(s->in_buffer.fmt == in_param->fmt);
581
582 tmp=out=*out_param;
583 in = *in_param;
584
585 border = s->resampler->invert_initial_buffer(s->resample, &s->in_buffer,
586 &in, in_count, &s->in_buffer_index, &s->in_buffer_count);
587 if (border == INT_MAX) {
588 return 0;
589 } else if (border < 0) {
590 return border;
591 } else if (border) {
592 buf_set(&in, &in, border);
593 in_count -= border;
594 s->resample_in_constraint = 0;
595 }
596
597 do{
598 int ret, size, consumed;
599 if(!s->resample_in_constraint && s->in_buffer_count){
600 buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
601 ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
602 out_count -= ret;
603 ret_sum += ret;
604 buf_set(&out, &out, ret);
605 s->in_buffer_count -= consumed;
606 s->in_buffer_index += consumed;
607
608 if(!in_count)
609 break;
610 if(s->in_buffer_count <= border){
611 buf_set(&in, &in, -s->in_buffer_count);
612 in_count += s->in_buffer_count;
613 s->in_buffer_count=0;
614 s->in_buffer_index=0;
615 border = 0;
616 }
617 }
618
619 if((s->flushed || in_count > padless) && !s->in_buffer_count){
620 s->in_buffer_index=0;
621 ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, FFMAX(in_count-padless, 0), &consumed);
622 out_count -= ret;
623 ret_sum += ret;
624 buf_set(&out, &out, ret);
625 in_count -= consumed;
626 buf_set(&in, &in, consumed);
627 }
628
629 //TODO is this check sane considering the advanced copy avoidance below
630 size= s->in_buffer_index + s->in_buffer_count + in_count;
631 if( size > s->in_buffer.count
632 && s->in_buffer_count + in_count <= s->in_buffer_index){
633 buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
634 copy(&s->in_buffer, &tmp, s->in_buffer_count);
635 s->in_buffer_index=0;
636 }else
637 if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
638 return ret;
639
640 if(in_count){
641 int count= in_count;
642 if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
643
644 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
645 copy(&tmp, &in, /*in_*/count);
646 s->in_buffer_count += count;
647 in_count -= count;
648 border += count;
649 buf_set(&in, &in, count);
650 s->resample_in_constraint= 0;
651 if(s->in_buffer_count != count || in_count)
652 continue;
653 if (padless) {
654 padless = 0;
655 continue;
656 }
657 }
658 break;
659 }while(1);
660
661 s->resample_in_constraint= !!out_count;
662
663 return ret_sum;
664 }
665
swr_convert_internal(struct SwrContext * s,AudioData * out,int out_count,AudioData * in,int in_count)666 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
667 AudioData *in , int in_count){
668 AudioData *postin, *midbuf, *preout;
669 int ret/*, in_max*/;
670 AudioData preout_tmp, midbuf_tmp;
671
672 if(s->full_convert){
673 av_assert0(!s->resample);
674 swri_audio_convert(s->full_convert, out, in, in_count);
675 return out_count;
676 }
677
678 // in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
679 // in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
680
681 if((ret=swri_realloc_audio(&s->postin, in_count))<0)
682 return ret;
683 if(s->resample_first){
684 av_assert0(s->midbuf.ch_count == s->used_ch_count);
685 if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
686 return ret;
687 }else{
688 av_assert0(s->midbuf.ch_count == s->out.ch_count);
689 if((ret=swri_realloc_audio(&s->midbuf, in_count))<0)
690 return ret;
691 }
692 if((ret=swri_realloc_audio(&s->preout, out_count))<0)
693 return ret;
694
695 postin= &s->postin;
696
697 midbuf_tmp= s->midbuf;
698 midbuf= &midbuf_tmp;
699 preout_tmp= s->preout;
700 preout= &preout_tmp;
701
702 if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
703 postin= in;
704
705 if(s->resample_first ? !s->resample : !s->rematrix)
706 midbuf= postin;
707
708 if(s->resample_first ? !s->rematrix : !s->resample)
709 preout= midbuf;
710
711 if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar
712 && !(s->out_sample_fmt==AV_SAMPLE_FMT_S32P && (s->dither.output_sample_bits&31))){
713 if(preout==in){
714 out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
715 av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
716 copy(out, in, out_count);
717 return out_count;
718 }
719 else if(preout==postin) preout= midbuf= postin= out;
720 else if(preout==midbuf) preout= midbuf= out;
721 else preout= out;
722 }
723
724 if(in != postin){
725 swri_audio_convert(s->in_convert, postin, in, in_count);
726 }
727
728 if(s->resample_first){
729 if(postin != midbuf)
730 if ((out_count = resample(s, midbuf, out_count, postin, in_count)) < 0)
731 return out_count;
732 if(midbuf != preout)
733 swri_rematrix(s, preout, midbuf, out_count, preout==out);
734 }else{
735 if(postin != midbuf)
736 swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
737 if(midbuf != preout)
738 if ((out_count = resample(s, preout, out_count, midbuf, in_count)) < 0)
739 return out_count;
740 }
741
742 if(preout != out && out_count){
743 AudioData *conv_src = preout;
744 if(s->dither.method){
745 int ch;
746 int dither_count= FFMAX(out_count, 1<<16);
747
748 if (preout == in) {
749 conv_src = &s->dither.temp;
750 if((ret=swri_realloc_audio(&s->dither.temp, dither_count))<0)
751 return ret;
752 }
753
754 if((ret=swri_realloc_audio(&s->dither.noise, dither_count))<0)
755 return ret;
756 if(ret)
757 for(ch=0; ch<s->dither.noise.ch_count; ch++)
758 if((ret=swri_get_dither(s, s->dither.noise.ch[ch], s->dither.noise.count, (12345678913579ULL*ch + 3141592) % 2718281828U, s->dither.noise.fmt))<0)
759 return ret;
760 av_assert0(s->dither.noise.ch_count == preout->ch_count);
761
762 if(s->dither.noise_pos + out_count > s->dither.noise.count)
763 s->dither.noise_pos = 0;
764
765 if (s->dither.method < SWR_DITHER_NS){
766 if (s->mix_2_1_simd) {
767 int len1= out_count&~15;
768 int off = len1 * preout->bps;
769
770 if(len1)
771 for(ch=0; ch<preout->ch_count; ch++)
772 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);
773 if(out_count != len1)
774 for(ch=0; ch<preout->ch_count; ch++)
775 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, s->native_one, 0, 0, out_count - len1);
776 } else {
777 for(ch=0; ch<preout->ch_count; ch++)
778 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);
779 }
780 } else {
781 switch(s->int_sample_fmt) {
782 case AV_SAMPLE_FMT_S16P :swri_noise_shaping_int16(s, conv_src, preout, &s->dither.noise, out_count); break;
783 case AV_SAMPLE_FMT_S32P :swri_noise_shaping_int32(s, conv_src, preout, &s->dither.noise, out_count); break;
784 case AV_SAMPLE_FMT_FLTP :swri_noise_shaping_float(s, conv_src, preout, &s->dither.noise, out_count); break;
785 case AV_SAMPLE_FMT_DBLP :swri_noise_shaping_double(s,conv_src, preout, &s->dither.noise, out_count); break;
786 }
787 }
788 s->dither.noise_pos += out_count;
789 }
790 //FIXME packed doesn't need more than 1 chan here!
791 swri_audio_convert(s->out_convert, out, conv_src, out_count);
792 }
793 return out_count;
794 }
795
swr_is_initialized(struct SwrContext * s)796 int swr_is_initialized(struct SwrContext *s) {
797 return !!s->in_buffer.ch_count;
798 }
799
swr_convert(struct SwrContext * s,uint8_t ** out_arg,int out_count,const uint8_t ** in_arg,int in_count)800 int attribute_align_arg swr_convert(struct SwrContext *s,
801 uint8_t **out_arg, int out_count,
802 const uint8_t **in_arg, int in_count)
803 {
804 AudioData * in= &s->in;
805 AudioData *out= &s->out;
806 int av_unused max_output;
807
808 if (!swr_is_initialized(s)) {
809 av_log(s, AV_LOG_ERROR, "Context has not been initialized\n");
810 return AVERROR(EINVAL);
811 }
812 #if defined(ASSERT_LEVEL) && ASSERT_LEVEL >1
813 max_output = swr_get_out_samples(s, in_count);
814 #endif
815
816 while(s->drop_output > 0){
817 int ret;
818 uint8_t *tmp_arg[SWR_CH_MAX];
819 #define MAX_DROP_STEP 16384
820 if((ret=swri_realloc_audio(&s->drop_temp, FFMIN(s->drop_output, MAX_DROP_STEP)))<0)
821 return ret;
822
823 reversefill_audiodata(&s->drop_temp, tmp_arg);
824 s->drop_output *= -1; //FIXME find a less hackish solution
825 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
826 s->drop_output *= -1;
827 in_count = 0;
828 if(ret>0) {
829 s->drop_output -= ret;
830 if (!s->drop_output && !out_arg)
831 return 0;
832 continue;
833 }
834
835 av_assert0(s->drop_output);
836 return 0;
837 }
838
839 if(!in_arg){
840 if(s->resample){
841 if (!s->flushed)
842 s->resampler->flush(s);
843 s->resample_in_constraint = 0;
844 s->flushed = 1;
845 }else if(!s->in_buffer_count){
846 return 0;
847 }
848 }else
849 fill_audiodata(in , (void*)in_arg);
850
851 fill_audiodata(out, out_arg);
852
853 if(s->resample){
854 int ret = swr_convert_internal(s, out, out_count, in, in_count);
855 if(ret>0 && !s->drop_output)
856 s->outpts += ret * (int64_t)s->in_sample_rate;
857
858 av_assert2(max_output < 0 || ret <= max_output);
859
860 return ret;
861 }else{
862 AudioData tmp= *in;
863 int ret2=0;
864 int ret, size;
865 size = FFMIN(out_count, s->in_buffer_count);
866 if(size){
867 buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
868 ret= swr_convert_internal(s, out, size, &tmp, size);
869 if(ret<0)
870 return ret;
871 ret2= ret;
872 s->in_buffer_count -= ret;
873 s->in_buffer_index += ret;
874 buf_set(out, out, ret);
875 out_count -= ret;
876 if(!s->in_buffer_count)
877 s->in_buffer_index = 0;
878 }
879
880 if(in_count){
881 size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
882
883 if(in_count > out_count) { //FIXME move after swr_convert_internal
884 if( size > s->in_buffer.count
885 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
886 buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
887 copy(&s->in_buffer, &tmp, s->in_buffer_count);
888 s->in_buffer_index=0;
889 }else
890 if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
891 return ret;
892 }
893
894 if(out_count){
895 size = FFMIN(in_count, out_count);
896 ret= swr_convert_internal(s, out, size, in, size);
897 if(ret<0)
898 return ret;
899 buf_set(in, in, ret);
900 in_count -= ret;
901 ret2 += ret;
902 }
903 if(in_count){
904 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
905 copy(&tmp, in, in_count);
906 s->in_buffer_count += in_count;
907 }
908 }
909 if(ret2>0 && !s->drop_output)
910 s->outpts += ret2 * (int64_t)s->in_sample_rate;
911 av_assert2(max_output < 0 || ret2 < 0 || ret2 <= max_output);
912 return ret2;
913 }
914 }
915
swr_drop_output(struct SwrContext * s,int count)916 int swr_drop_output(struct SwrContext *s, int count){
917 const uint8_t *tmp_arg[SWR_CH_MAX];
918 s->drop_output += count;
919
920 if(s->drop_output <= 0)
921 return 0;
922
923 av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
924 return swr_convert(s, NULL, s->drop_output, tmp_arg, 0);
925 }
926
swr_inject_silence(struct SwrContext * s,int count)927 int swr_inject_silence(struct SwrContext *s, int count){
928 int ret, i;
929 uint8_t *tmp_arg[SWR_CH_MAX];
930
931 if(count <= 0)
932 return 0;
933
934 #define MAX_SILENCE_STEP 16384
935 while (count > MAX_SILENCE_STEP) {
936 if ((ret = swr_inject_silence(s, MAX_SILENCE_STEP)) < 0)
937 return ret;
938 count -= MAX_SILENCE_STEP;
939 }
940
941 if((ret=swri_realloc_audio(&s->silence, count))<0)
942 return ret;
943
944 if(s->silence.planar) for(i=0; i<s->silence.ch_count; i++) {
945 memset(s->silence.ch[i], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps);
946 } else
947 memset(s->silence.ch[0], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps*s->silence.ch_count);
948
949 reversefill_audiodata(&s->silence, tmp_arg);
950 av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
951 ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
952 return ret;
953 }
954
swr_get_delay(struct SwrContext * s,int64_t base)955 int64_t swr_get_delay(struct SwrContext *s, int64_t base){
956 if (s->resampler && s->resample){
957 return s->resampler->get_delay(s, base);
958 }else{
959 return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
960 }
961 }
962
swr_get_out_samples(struct SwrContext * s,int in_samples)963 int swr_get_out_samples(struct SwrContext *s, int in_samples)
964 {
965 int64_t out_samples;
966
967 if (in_samples < 0)
968 return AVERROR(EINVAL);
969
970 if (s->resampler && s->resample) {
971 if (!s->resampler->get_out_samples)
972 return AVERROR(ENOSYS);
973 out_samples = s->resampler->get_out_samples(s, in_samples);
974 } else {
975 out_samples = s->in_buffer_count + in_samples;
976 av_assert0(s->out_sample_rate == s->in_sample_rate);
977 }
978
979 if (out_samples > INT_MAX)
980 return AVERROR(EINVAL);
981
982 return out_samples;
983 }
984
swr_set_compensation(struct SwrContext * s,int sample_delta,int compensation_distance)985 int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
986 int ret;
987
988 if (!s || compensation_distance < 0)
989 return AVERROR(EINVAL);
990 if (!compensation_distance && sample_delta)
991 return AVERROR(EINVAL);
992 if (!s->resample) {
993 s->flags |= SWR_FLAG_RESAMPLE;
994 ret = swr_init(s);
995 if (ret < 0)
996 return ret;
997 }
998 if (!s->resampler->set_compensation){
999 return AVERROR(EINVAL);
1000 }else{
1001 return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
1002 }
1003 }
1004
swr_next_pts(struct SwrContext * s,int64_t pts)1005 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
1006 if(pts == INT64_MIN)
1007 return s->outpts;
1008
1009 if (s->firstpts == AV_NOPTS_VALUE)
1010 s->outpts = s->firstpts = pts;
1011
1012 if(s->min_compensation >= FLT_MAX) {
1013 return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
1014 } else {
1015 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;
1016 double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
1017
1018 if(fabs(fdelta) > s->min_compensation) {
1019 if(s->outpts == s->firstpts || fabs(fdelta) > s->min_hard_compensation){
1020 int ret;
1021 if(delta > 0) ret = swr_inject_silence(s, delta / s->out_sample_rate);
1022 else ret = swr_drop_output (s, -delta / s-> in_sample_rate);
1023 if(ret<0){
1024 av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
1025 }
1026 } else if(s->soft_compensation_duration && s->max_soft_compensation) {
1027 int duration = s->out_sample_rate * s->soft_compensation_duration;
1028 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
1029 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
1030 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
1031 swr_set_compensation(s, comp, duration);
1032 }
1033 }
1034
1035 return s->outpts;
1036 }
1037 }
1038