swr: allow flushing in multiple steps if the output is too small.
[ffmpeg.git] / libswresample / swresample.c
1 /*
2  * Copyright (C) 2011 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/audioconvert.h"
26
27 #define  C30DB  M_SQRT2
28 #define  C15DB  1.189207115
29 #define C__0DB  1.0
30 #define C_15DB  0.840896415
31 #define C_30DB  M_SQRT1_2
32 #define C_45DB  0.594603558
33 #define C_60DB  0.5
34
35
36 //TODO split options array out?
37 #define OFFSET(x) offsetof(SwrContext,x)
38 static const AVOption options[]={
39 {"ich",  "input channel count", OFFSET( in.ch_count   ), AV_OPT_TYPE_INT, {.dbl=2}, 0, SWR_CH_MAX, 0},
40 {"och", "output channel count", OFFSET(out.ch_count   ), AV_OPT_TYPE_INT, {.dbl=2}, 0, SWR_CH_MAX, 0},
41 {"uch",   "used channel count", OFFSET(used_ch_count  ), AV_OPT_TYPE_INT, {.dbl=0}, 0, SWR_CH_MAX, 0},
42 {"isr",  "input sample rate"  , OFFSET( in_sample_rate), AV_OPT_TYPE_INT, {.dbl=48000}, 1, INT_MAX, 0},
43 {"osr", "output sample rate"  , OFFSET(out_sample_rate), AV_OPT_TYPE_INT, {.dbl=48000}, 1, INT_MAX, 0},
44 //{"ip" ,  "input planar"       , OFFSET( in.planar     ), AV_OPT_TYPE_INT, {.dbl=0},    0,       1, 0},
45 //{"op" , "output planar"       , OFFSET(out.planar     ), AV_OPT_TYPE_INT, {.dbl=0},    0,       1, 0},
46 {"isf",  "input sample format", OFFSET( in_sample_fmt ), AV_OPT_TYPE_INT, {.dbl=AV_SAMPLE_FMT_S16}, 0, AV_SAMPLE_FMT_NB-1+256, 0},
47 {"osf", "output sample format", OFFSET(out_sample_fmt ), AV_OPT_TYPE_INT, {.dbl=AV_SAMPLE_FMT_S16}, 0, AV_SAMPLE_FMT_NB-1+256, 0},
48 {"tsf", "internal sample format", OFFSET(int_sample_fmt ), AV_OPT_TYPE_INT, {.dbl=AV_SAMPLE_FMT_NONE}, -1, AV_SAMPLE_FMT_FLT, 0},
49 {"icl",  "input channel layout" , OFFSET( in_ch_layout), AV_OPT_TYPE_INT64, {.dbl=0}, 0, INT64_MAX, 0, "channel_layout"},
50 {"ocl",  "output channel layout", OFFSET(out_ch_layout), AV_OPT_TYPE_INT64, {.dbl=0}, 0, INT64_MAX, 0, "channel_layout"},
51 {"clev", "center mix level"     , OFFSET(clev)         , AV_OPT_TYPE_FLOAT, {.dbl=C_30DB}, 0, 4, 0},
52 {"slev", "sourround mix level"  , OFFSET(slev)         , AV_OPT_TYPE_FLOAT, {.dbl=C_30DB}, 0, 4, 0},
53 {"rmvol", "rematrix volume"     , OFFSET(rematrix_volume), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, -1000, 1000, 0},
54 {"flags", NULL                  , OFFSET(flags)        , AV_OPT_TYPE_FLAGS, {.dbl=0}, 0,  UINT_MAX, 0, "flags"},
55 {"res", "force resampling", 0, AV_OPT_TYPE_CONST, {.dbl=SWR_FLAG_RESAMPLE}, INT_MIN, INT_MAX, 0, "flags"},
56
57 {0}
58 };
59
60 static const char* context_to_name(void* ptr) {
61     return "SWR";
62 }
63
64 static const AVClass av_class = {
65     .class_name                = "SwrContext",
66     .item_name                 = context_to_name,
67     .option                    = options,
68     .version                   = LIBAVUTIL_VERSION_INT,
69     .log_level_offset_offset   = OFFSET(log_level_offset),
70     .parent_log_context_offset = OFFSET(log_ctx),
71 };
72
73 unsigned swresample_version(void)
74 {
75     av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
76     return LIBSWRESAMPLE_VERSION_INT;
77 }
78
79 const char *swresample_configuration(void)
80 {
81     return FFMPEG_CONFIGURATION;
82 }
83
84 const char *swresample_license(void)
85 {
86 #define LICENSE_PREFIX "libswresample license: "
87     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
88 }
89
90 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
91     if(!s || s->in_convert) // s needs to be allocated but not initialized
92         return AVERROR(EINVAL);
93     s->channel_map = channel_map;
94     return 0;
95 }
96
97 struct SwrContext *swr_alloc(void){
98     SwrContext *s= av_mallocz(sizeof(SwrContext));
99     if(s){
100         s->av_class= &av_class;
101         av_opt_set_defaults(s);
102     }
103     return s;
104 }
105
106 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
107                                       int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
108                                       int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
109                                       int log_offset, void *log_ctx){
110     if(!s) s= swr_alloc();
111     if(!s) return NULL;
112
113     s->log_level_offset= log_offset;
114     s->log_ctx= log_ctx;
115
116     av_opt_set_int(s, "ocl", out_ch_layout,   0);
117     av_opt_set_int(s, "osf", out_sample_fmt,  0);
118     av_opt_set_int(s, "osr", out_sample_rate, 0);
119     av_opt_set_int(s, "icl", in_ch_layout,    0);
120     av_opt_set_int(s, "isf", in_sample_fmt,   0);
121     av_opt_set_int(s, "isr", in_sample_rate,  0);
122     av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_S16, 0);
123     av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> in_ch_layout), 0);
124     av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->out_ch_layout), 0);
125     av_opt_set_int(s, "uch", 0, 0);
126     return s;
127 }
128
129
130 static void free_temp(AudioData *a){
131     av_free(a->data);
132     memset(a, 0, sizeof(*a));
133 }
134
135 void swr_free(SwrContext **ss){
136     SwrContext *s= *ss;
137     if(s){
138         free_temp(&s->postin);
139         free_temp(&s->midbuf);
140         free_temp(&s->preout);
141         free_temp(&s->in_buffer);
142         swri_audio_convert_free(&s-> in_convert);
143         swri_audio_convert_free(&s->out_convert);
144         swri_audio_convert_free(&s->full_convert);
145         swri_resample_free(&s->resample);
146     }
147
148     av_freep(ss);
149 }
150
151 int swr_init(struct SwrContext *s){
152     s->in_buffer_index= 0;
153     s->in_buffer_count= 0;
154     s->resample_in_constraint= 0;
155     free_temp(&s->postin);
156     free_temp(&s->midbuf);
157     free_temp(&s->preout);
158     free_temp(&s->in_buffer);
159     swri_audio_convert_free(&s-> in_convert);
160     swri_audio_convert_free(&s->out_convert);
161     swri_audio_convert_free(&s->full_convert);
162
163     s-> in.planar= av_sample_fmt_is_planar(s-> in_sample_fmt);
164     s->out.planar= av_sample_fmt_is_planar(s->out_sample_fmt);
165     s-> in_sample_fmt= av_get_alt_sample_fmt(s-> in_sample_fmt, 0);
166     s->out_sample_fmt= av_get_alt_sample_fmt(s->out_sample_fmt, 0);
167
168     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
169         av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
170         return AVERROR(EINVAL);
171     }
172     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
173         av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
174         return AVERROR(EINVAL);
175     }
176
177     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16
178         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLT){
179         av_log(s, AV_LOG_ERROR, "Requested sample format %s is not supported internally, only float & S16 is supported\n", av_get_sample_fmt_name(s->int_sample_fmt));
180         return AVERROR(EINVAL);
181     }
182
183     //FIXME should we allow/support using FLT on material that doesnt need it ?
184     if(s->in_sample_fmt <= AV_SAMPLE_FMT_S16 || s->int_sample_fmt==AV_SAMPLE_FMT_S16){
185         s->int_sample_fmt= AV_SAMPLE_FMT_S16;
186     }else
187         s->int_sample_fmt= AV_SAMPLE_FMT_FLT;
188
189
190     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
191         s->resample = swri_resample_init(s->resample, s->out_sample_rate, s->in_sample_rate, 16, 10, 0, 0.8);
192     }else
193         swri_resample_free(&s->resample);
194     if(s->int_sample_fmt != AV_SAMPLE_FMT_S16 && s->resample){
195         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16 currently\n"); //FIXME
196         return -1;
197     }
198
199     if(!s->used_ch_count)
200         s->used_ch_count= s->in.ch_count;
201
202     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
203         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
204         s-> in_ch_layout= 0;
205     }
206
207     if(!s-> in_ch_layout)
208         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
209     if(!s->out_ch_layout)
210         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
211
212     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
213                  s->rematrix_custom;
214
215 #define RSC 1 //FIXME finetune
216     if(!s-> in.ch_count)
217         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
218     if(!s->used_ch_count)
219         s->used_ch_count= s->in.ch_count;
220     if(!s->out.ch_count)
221         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
222
223     if(!s-> in.ch_count){
224         av_assert0(!s->in_ch_layout);
225         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
226         return -1;
227     }
228
229     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
230         av_log(s, AV_LOG_ERROR, "Rematrix is needed but there is not enough information to do it\n");
231         return -1;
232     }
233
234 av_assert0(s->used_ch_count);
235 av_assert0(s->out.ch_count);
236     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
237
238     s-> in.bps= av_get_bytes_per_sample(s-> in_sample_fmt);
239     s->int_bps= av_get_bytes_per_sample(s->int_sample_fmt);
240     s->out.bps= av_get_bytes_per_sample(s->out_sample_fmt);
241
242     if(!s->resample && !s->rematrix && !s->channel_map){
243         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
244                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
245         return 0;
246     }
247
248     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
249                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
250     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
251                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
252
253
254     s->postin= s->in;
255     s->preout= s->out;
256     s->midbuf= s->in;
257     s->in_buffer= s->in;
258     if(s->channel_map){
259         s->postin.ch_count=
260         s->midbuf.ch_count=
261         s->in_buffer.ch_count= s->used_ch_count;
262     }
263     if(!s->resample_first){
264         s->midbuf.ch_count= s->out.ch_count;
265         s->in_buffer.ch_count = s->out.ch_count;
266     }
267
268     s->in_buffer.bps = s->postin.bps = s->midbuf.bps = s->preout.bps =  s->int_bps;
269     s->in_buffer.planar = s->postin.planar = s->midbuf.planar = s->preout.planar =  1;
270
271
272     if(s->rematrix)
273         return swri_rematrix_init(s);
274
275     return 0;
276 }
277
278 static int realloc_audio(AudioData *a, int count){
279     int i, countb;
280     AudioData old;
281
282     if(a->count >= count)
283         return 0;
284
285     count*=2;
286
287     countb= FFALIGN(count*a->bps, 32);
288     old= *a;
289
290     av_assert0(a->planar);
291     av_assert0(a->bps);
292     av_assert0(a->ch_count);
293
294     a->data= av_malloc(countb*a->ch_count);
295     if(!a->data)
296         return AVERROR(ENOMEM);
297     for(i=0; i<a->ch_count; i++){
298         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
299         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
300     }
301     av_free(old.data);
302     a->count= count;
303
304     return 1;
305 }
306
307 static void copy(AudioData *out, AudioData *in,
308                  int count){
309     av_assert0(out->planar == in->planar);
310     av_assert0(out->bps == in->bps);
311     av_assert0(out->ch_count == in->ch_count);
312     if(out->planar){
313         int ch;
314         for(ch=0; ch<out->ch_count; ch++)
315             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
316     }else
317         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
318 }
319
320 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
321     int i;
322     if(out->planar){
323         for(i=0; i<out->ch_count; i++)
324             out->ch[i]= in_arg[i];
325     }else{
326         for(i=0; i<out->ch_count; i++)
327             out->ch[i]= in_arg[0] + i*out->bps;
328     }
329 }
330
331 /**
332  *
333  * out may be equal in.
334  */
335 static void buf_set(AudioData *out, AudioData *in, int count){
336     if(in->planar){
337         int ch;
338         for(ch=0; ch<out->ch_count; ch++)
339             out->ch[ch]= in->ch[ch] + count*out->bps;
340     }else
341         out->ch[0]= in->ch[0] + count*out->ch_count*out->bps;
342 }
343
344 /**
345  *
346  * @return number of samples output per channel
347  */
348 static int resample(SwrContext *s, AudioData *out_param, int out_count,
349                              const AudioData * in_param, int in_count){
350     AudioData in, out, tmp;
351     int ret_sum=0;
352     int border=0;
353
354     tmp=out=*out_param;
355     in =  *in_param;
356
357     do{
358         int ret, size, consumed;
359         if(!s->resample_in_constraint && s->in_buffer_count){
360             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
361             ret= swri_multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
362             out_count -= ret;
363             ret_sum += ret;
364             buf_set(&out, &out, ret);
365             s->in_buffer_count -= consumed;
366             s->in_buffer_index += consumed;
367
368             if(!in_count)
369                 break;
370             if(s->in_buffer_count <= border){
371                 buf_set(&in, &in, -s->in_buffer_count);
372                 in_count += s->in_buffer_count;
373                 s->in_buffer_count=0;
374                 s->in_buffer_index=0;
375                 border = 0;
376             }
377         }
378
379         if(in_count && !s->in_buffer_count){
380             s->in_buffer_index=0;
381             ret= swri_multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
382             out_count -= ret;
383             ret_sum += ret;
384             buf_set(&out, &out, ret);
385             in_count -= consumed;
386             buf_set(&in, &in, consumed);
387         }
388
389         //TODO is this check sane considering the advanced copy avoidance below
390         size= s->in_buffer_index + s->in_buffer_count + in_count;
391         if(   size > s->in_buffer.count
392            && s->in_buffer_count + in_count <= s->in_buffer_index){
393             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
394             copy(&s->in_buffer, &tmp, s->in_buffer_count);
395             s->in_buffer_index=0;
396         }else
397             if((ret=realloc_audio(&s->in_buffer, size)) < 0)
398                 return ret;
399
400         if(in_count){
401             int count= in_count;
402             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
403
404             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
405             copy(&tmp, &in, /*in_*/count);
406             s->in_buffer_count += count;
407             in_count -= count;
408             border += count;
409             buf_set(&in, &in, count);
410             s->resample_in_constraint= 0;
411             if(s->in_buffer_count != count || in_count)
412                 continue;
413         }
414         break;
415     }while(1);
416
417     s->resample_in_constraint= !!out_count;
418
419     return ret_sum;
420 }
421
422 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
423                          const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
424     AudioData *postin, *midbuf, *preout;
425     int ret/*, in_max*/;
426     AudioData * in= &s->in;
427     AudioData *out= &s->out;
428     AudioData preout_tmp, midbuf_tmp;
429
430     if(!s->resample){
431         if(in_count > out_count)
432             return -1;
433         out_count = in_count;
434     }
435
436     if(!in_arg){
437         if(s->in_buffer_count){
438             if (!s->flushed) {
439                 AudioData *a= &s->in_buffer;
440                 int i, j, ret;
441                 if((ret=realloc_audio(a, s->in_buffer_index + 2*s->in_buffer_count)) < 0)
442                     return ret;
443                 av_assert0(a->planar);
444                 for(i=0; i<a->ch_count; i++){
445                     for(j=0; j<s->in_buffer_count; j++){
446                         memcpy(a->ch[i] + (s->in_buffer_index+s->in_buffer_count+j  )*a->bps,
447                             a->ch[i] + (s->in_buffer_index+s->in_buffer_count-j-1)*a->bps, a->bps);
448                     }
449                 }
450                 s->in_buffer_count += (s->in_buffer_count+1)/2;
451                 s->resample_in_constraint = 0;
452                 s->flushed = 1;
453             }
454         }else{
455             return 0;
456         }
457     }else
458         fill_audiodata(in ,  (void*)in_arg);
459     fill_audiodata(out, out_arg);
460
461     if(s->full_convert){
462         av_assert0(!s->resample);
463         swri_audio_convert(s->full_convert, out, in, in_count);
464         return out_count;
465     }
466
467 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
468 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
469
470     if((ret=realloc_audio(&s->postin, in_count))<0)
471         return ret;
472     if(s->resample_first){
473         av_assert0(s->midbuf.ch_count == s->used_ch_count);
474         if((ret=realloc_audio(&s->midbuf, out_count))<0)
475             return ret;
476     }else{
477         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
478         if((ret=realloc_audio(&s->midbuf,  in_count))<0)
479             return ret;
480     }
481     if((ret=realloc_audio(&s->preout, out_count))<0)
482         return ret;
483
484     postin= &s->postin;
485
486     midbuf_tmp= s->midbuf;
487     midbuf= &midbuf_tmp;
488     preout_tmp= s->preout;
489     preout= &preout_tmp;
490
491     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar)
492         postin= in;
493
494     if(s->resample_first ? !s->resample : !s->rematrix)
495         midbuf= postin;
496
497     if(s->resample_first ? !s->rematrix : !s->resample)
498         preout= midbuf;
499
500     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
501         if(preout==in){
502             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
503             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
504             copy(out, in, out_count);
505             return out_count;
506         }
507         else if(preout==postin) preout= midbuf= postin= out;
508         else if(preout==midbuf) preout= midbuf= out;
509         else                    preout= out;
510     }
511
512     if(in != postin){
513         swri_audio_convert(s->in_convert, postin, in, in_count);
514     }
515
516     if(s->resample_first){
517         if(postin != midbuf)
518             out_count= resample(s, midbuf, out_count, postin, in_count);
519         if(midbuf != preout)
520             swri_rematrix(s, preout, midbuf, out_count, preout==out);
521     }else{
522         if(postin != midbuf)
523             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
524         if(midbuf != preout)
525             out_count= resample(s, preout, out_count, midbuf, in_count);
526     }
527
528     if(preout != out){
529 //FIXME packed doesnt need more than 1 chan here!
530         swri_audio_convert(s->out_convert, out, preout, out_count);
531     }
532     if(!in_arg)
533         s->in_buffer_count = 0;
534     return out_count;
535 }
536