swr: set the default used_ch_count value to zero.
[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 = { "SwrContext", context_to_name, options, LIBAVUTIL_VERSION_INT, OFFSET(log_level_offset), OFFSET(log_ctx) };
65
66 static int resample(SwrContext *s, AudioData *out_param, int out_count,
67                              const AudioData * in_param, int in_count);
68
69 SwrContext *swr_alloc(void){
70     SwrContext *s= av_mallocz(sizeof(SwrContext));
71     if(s){
72         s->av_class= &av_class;
73         av_opt_set_defaults2(s, 0, 0);
74     }
75     return s;
76 }
77
78 SwrContext *swr_alloc2(struct SwrContext *s, int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
79                        int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
80                        const int *channel_map, int log_offset, void *log_ctx){
81     if(!s) s= swr_alloc();
82     if(!s) return NULL;
83
84     s->log_level_offset= log_offset;
85     s->log_ctx= log_ctx;
86
87     av_set_int(s, "ocl", out_ch_layout);
88     av_set_int(s, "osf", out_sample_fmt);
89     av_set_int(s, "osr", out_sample_rate);
90     av_set_int(s, "icl", in_ch_layout);
91     av_set_int(s, "isf", in_sample_fmt);
92     av_set_int(s, "isr", in_sample_rate);
93
94     s->channel_map = channel_map;
95     s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
96     s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
97     s->int_sample_fmt = AV_SAMPLE_FMT_S16;
98
99     return s;
100 }
101
102
103 static void free_temp(AudioData *a){
104     av_free(a->data);
105     memset(a, 0, sizeof(*a));
106 }
107
108 void swr_free(SwrContext **ss){
109     SwrContext *s= *ss;
110     if(s){
111         free_temp(&s->postin);
112         free_temp(&s->midbuf);
113         free_temp(&s->preout);
114         free_temp(&s->in_buffer);
115         swr_audio_convert_free(&s-> in_convert);
116         swr_audio_convert_free(&s->out_convert);
117         swr_audio_convert_free(&s->full_convert);
118         swr_resample_free(&s->resample);
119     }
120
121     av_freep(ss);
122 }
123
124 int swr_init(SwrContext *s){
125     s->in_buffer_index= 0;
126     s->in_buffer_count= 0;
127     s->resample_in_constraint= 0;
128     free_temp(&s->postin);
129     free_temp(&s->midbuf);
130     free_temp(&s->preout);
131     free_temp(&s->in_buffer);
132     swr_audio_convert_free(&s-> in_convert);
133     swr_audio_convert_free(&s->out_convert);
134     swr_audio_convert_free(&s->full_convert);
135
136     s-> in.planar= s-> in_sample_fmt >= 0x100;
137     s->out.planar= s->out_sample_fmt >= 0x100;
138     s-> in_sample_fmt &= 0xFF;
139     s->out_sample_fmt &= 0xFF;
140
141     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
142         av_log(s, AV_LOG_ERROR, "Requested sample format %s is invalid\n", av_get_sample_fmt_name(s->in_sample_fmt));
143         return AVERROR(EINVAL);
144     }
145     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
146         av_log(s, AV_LOG_ERROR, "Requested sample format %s is invalid\n", av_get_sample_fmt_name(s->out_sample_fmt));
147         return AVERROR(EINVAL);
148     }
149
150     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16
151         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLT){
152         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));
153         return AVERROR(EINVAL);
154     }
155
156     //FIXME should we allow/support using FLT on material that doesnt need it ?
157     if(s->in_sample_fmt <= AV_SAMPLE_FMT_S16 || s->int_sample_fmt==AV_SAMPLE_FMT_S16){
158         s->int_sample_fmt= AV_SAMPLE_FMT_S16;
159     }else
160         s->int_sample_fmt= AV_SAMPLE_FMT_FLT;
161
162
163     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
164         s->resample = swr_resample_init(s->resample, s->out_sample_rate, s->in_sample_rate, 16, 10, 0, 0.8);
165     }else
166         swr_resample_free(&s->resample);
167     if(s->int_sample_fmt != AV_SAMPLE_FMT_S16 && s->resample){
168         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16 currently\n"); //FIXME
169         return -1;
170     }
171
172     if(!s->used_ch_count)
173         s->used_ch_count= s->in.ch_count;
174
175     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
176         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
177         s-> in_ch_layout= 0;
178     }
179
180     if(!s-> in_ch_layout)
181         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
182     if(!s->out_ch_layout)
183         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
184
185     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0;
186
187 #define RSC 1 //FIXME finetune
188     if(!s-> in.ch_count)
189         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
190     if(!s->used_ch_count)
191         s->used_ch_count= s->in.ch_count;
192     if(!s->out.ch_count)
193         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
194
195 av_assert0(s-> in.ch_count);
196 av_assert0(s->used_ch_count);
197 av_assert0(s->out.ch_count);
198     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
199
200     s-> in.bps= av_get_bits_per_sample_fmt(s-> in_sample_fmt)/8;
201     s->int_bps= av_get_bits_per_sample_fmt(s->int_sample_fmt)/8;
202     s->out.bps= av_get_bits_per_sample_fmt(s->out_sample_fmt)/8;
203
204     if(!s->resample && !s->rematrix && !s->channel_map){
205         s->full_convert = swr_audio_convert_alloc(s->out_sample_fmt,
206                                                   s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
207         return 0;
208     }
209
210     s->in_convert = swr_audio_convert_alloc(s->int_sample_fmt,
211                                             s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
212     s->out_convert= swr_audio_convert_alloc(s->out_sample_fmt,
213                                             s->int_sample_fmt, s->out.ch_count, NULL, 0);
214
215
216     s->postin= s->in;
217     s->preout= s->out;
218     s->midbuf= s->in;
219     s->in_buffer= s->in;
220     if(s->channel_map){
221         s->postin.ch_count=
222         s->midbuf.ch_count=
223         s->in_buffer.ch_count= s->used_ch_count;
224     }
225     if(!s->resample_first){
226         s->midbuf.ch_count= s->out.ch_count;
227         s->in_buffer.ch_count = s->out.ch_count;
228     }
229
230     s->in_buffer.bps = s->postin.bps = s->midbuf.bps = s->preout.bps =  s->int_bps;
231     s->in_buffer.planar = s->postin.planar = s->midbuf.planar = s->preout.planar =  1;
232
233
234     if(s->rematrix && swr_rematrix_init(s)<0)
235         return -1;
236
237     return 0;
238 }
239
240 static int realloc_audio(AudioData *a, int count){
241     int i, countb;
242     AudioData old;
243
244     if(a->count >= count)
245         return 0;
246
247     count*=2;
248
249     countb= FFALIGN(count*a->bps, 32);
250     old= *a;
251
252     av_assert0(a->planar);
253     av_assert0(a->bps);
254     av_assert0(a->ch_count);
255
256     a->data= av_malloc(countb*a->ch_count);
257     if(!a->data)
258         return AVERROR(ENOMEM);
259     for(i=0; i<a->ch_count; i++){
260         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
261         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
262     }
263     av_free(old.data);
264     a->count= count;
265
266     return 1;
267 }
268
269 static void copy(AudioData *out, AudioData *in,
270                  int count){
271     av_assert0(out->planar == in->planar);
272     av_assert0(out->bps == in->bps);
273     av_assert0(out->ch_count == in->ch_count);
274     if(out->planar){
275         int ch;
276         for(ch=0; ch<out->ch_count; ch++)
277             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
278     }else
279         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
280 }
281
282 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
283     int i;
284     if(out->planar){
285         for(i=0; i<out->ch_count; i++)
286             out->ch[i]= in_arg[i];
287     }else{
288         for(i=0; i<out->ch_count; i++)
289             out->ch[i]= in_arg[0] + i*out->bps;
290     }
291 }
292
293 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
294                          const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
295     AudioData *postin, *midbuf, *preout;
296     int ret/*, in_max*/;
297     AudioData * in= &s->in;
298     AudioData *out= &s->out;
299     AudioData preout_tmp, midbuf_tmp;
300
301     if(!s->resample){
302         if(in_count > out_count)
303             return -1;
304         out_count = in_count;
305     }
306
307     if(!in_arg){
308         if(s->in_buffer_count){
309             AudioData *a= &s->in_buffer;
310             int i, j, ret;
311             if((ret=realloc_audio(a, s->in_buffer_index + 2*s->in_buffer_count)) < 0)
312                 return ret;
313             av_assert0(a->planar);
314             for(i=0; i<a->ch_count; i++){
315                 for(j=0; j<s->in_buffer_count; j++){
316                     memcpy(a->ch[i] + (s->in_buffer_index+s->in_buffer_count+j  )*a->bps,
317                            a->ch[i] + (s->in_buffer_index+s->in_buffer_count-j-1)*a->bps, a->bps);
318                 }
319             }
320             s->in_buffer_count += (s->in_buffer_count+1)/2;
321             s->resample_in_constraint = 0;
322         }else{
323             return 0;
324         }
325     }else
326         fill_audiodata(in ,  (void*)in_arg);
327     fill_audiodata(out, out_arg);
328
329     if(s->full_convert){
330         av_assert0(!s->resample);
331         swr_audio_convert(s->full_convert, out, in, in_count);
332         return out_count;
333     }
334
335 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
336 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
337
338     if((ret=realloc_audio(&s->postin, in_count))<0)
339         return ret;
340     if(s->resample_first){
341         av_assert0(s->midbuf.ch_count == s->used_ch_count);
342         if((ret=realloc_audio(&s->midbuf, out_count))<0)
343             return ret;
344     }else{
345         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
346         if((ret=realloc_audio(&s->midbuf,  in_count))<0)
347             return ret;
348     }
349     if((ret=realloc_audio(&s->preout, out_count))<0)
350         return ret;
351
352     postin= &s->postin;
353
354     midbuf_tmp= s->midbuf;
355     midbuf= &midbuf_tmp;
356     preout_tmp= s->preout;
357     preout= &preout_tmp;
358
359     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar)
360         postin= in;
361
362     if(s->resample_first ? !s->resample : !s->rematrix)
363         midbuf= postin;
364
365     if(s->resample_first ? !s->rematrix : !s->resample)
366         preout= midbuf;
367
368     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
369         if(preout==in){
370             out_count= FFMIN(out_count, in_count); //TODO check at teh end if this is needed or redundant
371             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
372             copy(out, in, out_count);
373             return out_count;
374         }
375         else if(preout==postin) preout= midbuf= postin= out;
376         else if(preout==midbuf) preout= midbuf= out;
377         else                    preout= out;
378     }
379
380     if(in != postin){
381         swr_audio_convert(s->in_convert, postin, in, in_count);
382     }
383
384     if(s->resample_first){
385         if(postin != midbuf)
386             out_count= resample(s, midbuf, out_count, postin, in_count);
387         if(midbuf != preout)
388             swr_rematrix(s, preout, midbuf, out_count, preout==out);
389     }else{
390         if(postin != midbuf)
391             swr_rematrix(s, midbuf, postin, in_count, midbuf==out);
392         if(midbuf != preout)
393             out_count= resample(s, preout, out_count, midbuf, in_count);
394     }
395
396     if(preout != out){
397 //FIXME packed doesnt need more than 1 chan here!
398         swr_audio_convert(s->out_convert, out, preout, out_count);
399     }
400     if(!in_arg)
401         s->in_buffer_count = 0;
402     return out_count;
403 }
404
405 /**
406  *
407  * out may be equal in.
408  */
409 static void buf_set(AudioData *out, AudioData *in, int count){
410     if(in->planar){
411         int ch;
412         for(ch=0; ch<out->ch_count; ch++)
413             out->ch[ch]= in->ch[ch] + count*out->bps;
414     }else
415         out->ch[0]= in->ch[0] + count*out->ch_count*out->bps;
416 }
417
418 /**
419  *
420  * @return number of samples output per channel
421  */
422 static int resample(SwrContext *s, AudioData *out_param, int out_count,
423                              const AudioData * in_param, int in_count){
424     AudioData in, out, tmp;
425     int ret_sum=0;
426     int border=0;
427
428     tmp=out=*out_param;
429     in =  *in_param;
430
431     do{
432         int ret, size, consumed;
433         if(!s->resample_in_constraint && s->in_buffer_count){
434             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
435             ret= swr_multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
436             out_count -= ret;
437             ret_sum += ret;
438             buf_set(&out, &out, ret);
439             s->in_buffer_count -= consumed;
440             s->in_buffer_index += consumed;
441
442             if(!in_count)
443                 break;
444             if(s->in_buffer_count <= border){
445                 buf_set(&in, &in, -s->in_buffer_count);
446                 in_count += s->in_buffer_count;
447                 s->in_buffer_count=0;
448                 s->in_buffer_index=0;
449                 border = 0;
450             }
451         }
452
453         if(in_count && !s->in_buffer_count){
454             s->in_buffer_index=0;
455             ret= swr_multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
456             out_count -= ret;
457             ret_sum += ret;
458             buf_set(&out, &out, ret);
459             in_count -= consumed;
460             buf_set(&in, &in, consumed);
461         }
462
463         //TODO is this check sane considering the advanced copy avoidance below
464         size= s->in_buffer_index + s->in_buffer_count + in_count;
465         if(   size > s->in_buffer.count
466            && s->in_buffer_count + in_count <= s->in_buffer_index){
467             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
468             copy(&s->in_buffer, &tmp, s->in_buffer_count);
469             s->in_buffer_index=0;
470         }else
471             if((ret=realloc_audio(&s->in_buffer, size)) < 0)
472                 return ret;
473
474         if(in_count){
475             int count= in_count;
476             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
477
478             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
479             copy(&tmp, &in, /*in_*/count);
480             s->in_buffer_count += count;
481             in_count -= count;
482             border += count;
483             buf_set(&in, &in, count);
484             s->resample_in_constraint= 0;
485             if(s->in_buffer_count != count || in_count)
486                 continue;
487         }
488         break;
489     }while(1);
490
491     s->resample_in_constraint= !!out_count;
492
493     return ret_sum;
494 }