6c6482a19fe9819f12f929146e58cb9eab27b23b
[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}, 1, SWR_CH_MAX, 0},
40 {"och", "output channel count", OFFSET(out.ch_count   ), AV_OPT_TYPE_INT, {.dbl=2}, 1, 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 static int resample(SwrContext *s, AudioData *out_param, int out_count,
74                              const AudioData * in_param, int in_count);
75
76 struct SwrContext *swr_alloc(void){
77     SwrContext *s= av_mallocz(sizeof(SwrContext));
78     if(s){
79         s->av_class= &av_class;
80         av_opt_set_defaults(s);
81     }
82     return s;
83 }
84
85 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
86                                       int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
87                                       int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
88                                       const int *channel_map, int log_offset, void *log_ctx){
89     if(!s) s= swr_alloc();
90     if(!s) return NULL;
91
92     s->log_level_offset= log_offset;
93     s->log_ctx= log_ctx;
94
95     av_opt_set_int(s, "ocl", out_ch_layout,   0);
96     av_opt_set_int(s, "osf", out_sample_fmt,  0);
97     av_opt_set_int(s, "osr", out_sample_rate, 0);
98     av_opt_set_int(s, "icl", in_ch_layout,    0);
99     av_opt_set_int(s, "isf", in_sample_fmt,   0);
100     av_opt_set_int(s, "isr", in_sample_rate,  0);
101
102     s->channel_map = channel_map;
103     s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
104     s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
105     s->int_sample_fmt = AV_SAMPLE_FMT_S16;
106
107     return s;
108 }
109
110
111 static void free_temp(AudioData *a){
112     av_free(a->data);
113     memset(a, 0, sizeof(*a));
114 }
115
116 void swr_free(SwrContext **ss){
117     SwrContext *s= *ss;
118     if(s){
119         free_temp(&s->postin);
120         free_temp(&s->midbuf);
121         free_temp(&s->preout);
122         free_temp(&s->in_buffer);
123         swri_audio_convert_free(&s-> in_convert);
124         swri_audio_convert_free(&s->out_convert);
125         swri_audio_convert_free(&s->full_convert);
126         swri_resample_free(&s->resample);
127     }
128
129     av_freep(ss);
130 }
131
132 int swr_init(struct SwrContext *s){
133     s->in_buffer_index= 0;
134     s->in_buffer_count= 0;
135     s->resample_in_constraint= 0;
136     free_temp(&s->postin);
137     free_temp(&s->midbuf);
138     free_temp(&s->preout);
139     free_temp(&s->in_buffer);
140     swri_audio_convert_free(&s-> in_convert);
141     swri_audio_convert_free(&s->out_convert);
142     swri_audio_convert_free(&s->full_convert);
143
144     s-> in.planar= s-> in_sample_fmt >= 0x100;
145     s->out.planar= s->out_sample_fmt >= 0x100;
146     s-> in_sample_fmt &= 0xFF;
147     s->out_sample_fmt &= 0xFF;
148
149     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
150         av_log(s, AV_LOG_ERROR, "Requested sample format %s is invalid\n", av_get_sample_fmt_name(s->in_sample_fmt));
151         return AVERROR(EINVAL);
152     }
153     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
154         av_log(s, AV_LOG_ERROR, "Requested sample format %s is invalid\n", av_get_sample_fmt_name(s->out_sample_fmt));
155         return AVERROR(EINVAL);
156     }
157
158     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16
159         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLT){
160         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));
161         return AVERROR(EINVAL);
162     }
163
164     //FIXME should we allow/support using FLT on material that doesnt need it ?
165     if(s->in_sample_fmt <= AV_SAMPLE_FMT_S16 || s->int_sample_fmt==AV_SAMPLE_FMT_S16){
166         s->int_sample_fmt= AV_SAMPLE_FMT_S16;
167     }else
168         s->int_sample_fmt= AV_SAMPLE_FMT_FLT;
169
170
171     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
172         s->resample = swri_resample_init(s->resample, s->out_sample_rate, s->in_sample_rate, 16, 10, 0, 0.8);
173     }else
174         swri_resample_free(&s->resample);
175     if(s->int_sample_fmt != AV_SAMPLE_FMT_S16 && s->resample){
176         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16 currently\n"); //FIXME
177         return -1;
178     }
179
180     if(!s->used_ch_count)
181         s->used_ch_count= s->in.ch_count;
182
183     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
184         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
185         s-> in_ch_layout= 0;
186     }
187
188     if(!s-> in_ch_layout)
189         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
190     if(!s->out_ch_layout)
191         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
192
193     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0;
194
195 #define RSC 1 //FIXME finetune
196     if(!s-> in.ch_count)
197         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
198     if(!s->used_ch_count)
199         s->used_ch_count= s->in.ch_count;
200     if(!s->out.ch_count)
201         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
202
203 av_assert0(s-> in.ch_count);
204 av_assert0(s->used_ch_count);
205 av_assert0(s->out.ch_count);
206     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
207
208     s-> in.bps= av_get_bytes_per_sample(s-> in_sample_fmt);
209     s->int_bps= av_get_bytes_per_sample(s->int_sample_fmt);
210     s->out.bps= av_get_bytes_per_sample(s->out_sample_fmt);
211
212     if(!s->resample && !s->rematrix && !s->channel_map){
213         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
214                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
215         return 0;
216     }
217
218     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
219                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
220     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
221                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
222
223
224     s->postin= s->in;
225     s->preout= s->out;
226     s->midbuf= s->in;
227     s->in_buffer= s->in;
228     if(s->channel_map){
229         s->postin.ch_count=
230         s->midbuf.ch_count=
231         s->in_buffer.ch_count= s->used_ch_count;
232     }
233     if(!s->resample_first){
234         s->midbuf.ch_count= s->out.ch_count;
235         s->in_buffer.ch_count = s->out.ch_count;
236     }
237
238     s->in_buffer.bps = s->postin.bps = s->midbuf.bps = s->preout.bps =  s->int_bps;
239     s->in_buffer.planar = s->postin.planar = s->midbuf.planar = s->preout.planar =  1;
240
241
242     if(s->rematrix)
243         return swri_rematrix_init(s);
244
245     return 0;
246 }
247
248 static int realloc_audio(AudioData *a, int count){
249     int i, countb;
250     AudioData old;
251
252     if(a->count >= count)
253         return 0;
254
255     count*=2;
256
257     countb= FFALIGN(count*a->bps, 32);
258     old= *a;
259
260     av_assert0(a->planar);
261     av_assert0(a->bps);
262     av_assert0(a->ch_count);
263
264     a->data= av_malloc(countb*a->ch_count);
265     if(!a->data)
266         return AVERROR(ENOMEM);
267     for(i=0; i<a->ch_count; i++){
268         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
269         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
270     }
271     av_free(old.data);
272     a->count= count;
273
274     return 1;
275 }
276
277 static void copy(AudioData *out, AudioData *in,
278                  int count){
279     av_assert0(out->planar == in->planar);
280     av_assert0(out->bps == in->bps);
281     av_assert0(out->ch_count == in->ch_count);
282     if(out->planar){
283         int ch;
284         for(ch=0; ch<out->ch_count; ch++)
285             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
286     }else
287         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
288 }
289
290 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
291     int i;
292     if(out->planar){
293         for(i=0; i<out->ch_count; i++)
294             out->ch[i]= in_arg[i];
295     }else{
296         for(i=0; i<out->ch_count; i++)
297             out->ch[i]= in_arg[0] + i*out->bps;
298     }
299 }
300
301 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
302                          const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
303     AudioData *postin, *midbuf, *preout;
304     int ret/*, in_max*/;
305     AudioData * in= &s->in;
306     AudioData *out= &s->out;
307     AudioData preout_tmp, midbuf_tmp;
308
309     if(!s->resample){
310         if(in_count > out_count)
311             return -1;
312         out_count = in_count;
313     }
314
315     if(!in_arg){
316         if(s->in_buffer_count){
317             AudioData *a= &s->in_buffer;
318             int i, j, ret;
319             if((ret=realloc_audio(a, s->in_buffer_index + 2*s->in_buffer_count)) < 0)
320                 return ret;
321             av_assert0(a->planar);
322             for(i=0; i<a->ch_count; i++){
323                 for(j=0; j<s->in_buffer_count; j++){
324                     memcpy(a->ch[i] + (s->in_buffer_index+s->in_buffer_count+j  )*a->bps,
325                            a->ch[i] + (s->in_buffer_index+s->in_buffer_count-j-1)*a->bps, a->bps);
326                 }
327             }
328             s->in_buffer_count += (s->in_buffer_count+1)/2;
329             s->resample_in_constraint = 0;
330         }else{
331             return 0;
332         }
333     }else
334         fill_audiodata(in ,  (void*)in_arg);
335     fill_audiodata(out, out_arg);
336
337     if(s->full_convert){
338         av_assert0(!s->resample);
339         swri_audio_convert(s->full_convert, out, in, in_count);
340         return out_count;
341     }
342
343 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
344 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
345
346     if((ret=realloc_audio(&s->postin, in_count))<0)
347         return ret;
348     if(s->resample_first){
349         av_assert0(s->midbuf.ch_count == s->used_ch_count);
350         if((ret=realloc_audio(&s->midbuf, out_count))<0)
351             return ret;
352     }else{
353         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
354         if((ret=realloc_audio(&s->midbuf,  in_count))<0)
355             return ret;
356     }
357     if((ret=realloc_audio(&s->preout, out_count))<0)
358         return ret;
359
360     postin= &s->postin;
361
362     midbuf_tmp= s->midbuf;
363     midbuf= &midbuf_tmp;
364     preout_tmp= s->preout;
365     preout= &preout_tmp;
366
367     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar)
368         postin= in;
369
370     if(s->resample_first ? !s->resample : !s->rematrix)
371         midbuf= postin;
372
373     if(s->resample_first ? !s->rematrix : !s->resample)
374         preout= midbuf;
375
376     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
377         if(preout==in){
378             out_count= FFMIN(out_count, in_count); //TODO check at teh end if this is needed or redundant
379             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
380             copy(out, in, out_count);
381             return out_count;
382         }
383         else if(preout==postin) preout= midbuf= postin= out;
384         else if(preout==midbuf) preout= midbuf= out;
385         else                    preout= out;
386     }
387
388     if(in != postin){
389         swri_audio_convert(s->in_convert, postin, in, in_count);
390     }
391
392     if(s->resample_first){
393         if(postin != midbuf)
394             out_count= resample(s, midbuf, out_count, postin, in_count);
395         if(midbuf != preout)
396             swri_rematrix(s, preout, midbuf, out_count, preout==out);
397     }else{
398         if(postin != midbuf)
399             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
400         if(midbuf != preout)
401             out_count= resample(s, preout, out_count, midbuf, in_count);
402     }
403
404     if(preout != out){
405 //FIXME packed doesnt need more than 1 chan here!
406         swri_audio_convert(s->out_convert, out, preout, out_count);
407     }
408     if(!in_arg)
409         s->in_buffer_count = 0;
410     return out_count;
411 }
412
413 /**
414  *
415  * out may be equal in.
416  */
417 static void buf_set(AudioData *out, AudioData *in, int count){
418     if(in->planar){
419         int ch;
420         for(ch=0; ch<out->ch_count; ch++)
421             out->ch[ch]= in->ch[ch] + count*out->bps;
422     }else
423         out->ch[0]= in->ch[0] + count*out->ch_count*out->bps;
424 }
425
426 /**
427  *
428  * @return number of samples output per channel
429  */
430 static int resample(SwrContext *s, AudioData *out_param, int out_count,
431                              const AudioData * in_param, int in_count){
432     AudioData in, out, tmp;
433     int ret_sum=0;
434     int border=0;
435
436     tmp=out=*out_param;
437     in =  *in_param;
438
439     do{
440         int ret, size, consumed;
441         if(!s->resample_in_constraint && s->in_buffer_count){
442             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
443             ret= swri_multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
444             out_count -= ret;
445             ret_sum += ret;
446             buf_set(&out, &out, ret);
447             s->in_buffer_count -= consumed;
448             s->in_buffer_index += consumed;
449
450             if(!in_count)
451                 break;
452             if(s->in_buffer_count <= border){
453                 buf_set(&in, &in, -s->in_buffer_count);
454                 in_count += s->in_buffer_count;
455                 s->in_buffer_count=0;
456                 s->in_buffer_index=0;
457                 border = 0;
458             }
459         }
460
461         if(in_count && !s->in_buffer_count){
462             s->in_buffer_index=0;
463             ret= swri_multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
464             out_count -= ret;
465             ret_sum += ret;
466             buf_set(&out, &out, ret);
467             in_count -= consumed;
468             buf_set(&in, &in, consumed);
469         }
470
471         //TODO is this check sane considering the advanced copy avoidance below
472         size= s->in_buffer_index + s->in_buffer_count + in_count;
473         if(   size > s->in_buffer.count
474            && s->in_buffer_count + in_count <= s->in_buffer_index){
475             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
476             copy(&s->in_buffer, &tmp, s->in_buffer_count);
477             s->in_buffer_index=0;
478         }else
479             if((ret=realloc_audio(&s->in_buffer, size)) < 0)
480                 return ret;
481
482         if(in_count){
483             int count= in_count;
484             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
485
486             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
487             copy(&tmp, &in, /*in_*/count);
488             s->in_buffer_count += count;
489             in_count -= count;
490             border += count;
491             buf_set(&in, &in, count);
492             s->resample_in_constraint= 0;
493             if(s->in_buffer_count != count || in_count)
494                 continue;
495         }
496         break;
497     }while(1);
498
499     s->resample_in_constraint= !!out_count;
500
501     return ret_sum;
502 }