swr: change minimum I/O channel count 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}, 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 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
74     if(!s || s->in_convert) // s needs to be allocated but not initialized
75         return AVERROR(EINVAL);
76     s->channel_map = channel_map;
77     return 0;
78 }
79
80 struct SwrContext *swr_alloc(void){
81     SwrContext *s= av_mallocz(sizeof(SwrContext));
82     if(s){
83         s->av_class= &av_class;
84         av_opt_set_defaults(s);
85     }
86     return s;
87 }
88
89 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
90                                       int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
91                                       int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
92                                       int log_offset, void *log_ctx){
93     if(!s) s= swr_alloc();
94     if(!s) return NULL;
95
96     s->log_level_offset= log_offset;
97     s->log_ctx= log_ctx;
98
99     av_opt_set_int(s, "ocl", out_ch_layout,   0);
100     av_opt_set_int(s, "osf", out_sample_fmt,  0);
101     av_opt_set_int(s, "osr", out_sample_rate, 0);
102     av_opt_set_int(s, "icl", in_ch_layout,    0);
103     av_opt_set_int(s, "isf", in_sample_fmt,   0);
104     av_opt_set_int(s, "isr", in_sample_rate,  0);
105     av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_S16, 0);
106     av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> in_ch_layout), 0);
107     av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->out_ch_layout), 0);
108     return s;
109 }
110
111
112 static void free_temp(AudioData *a){
113     av_free(a->data);
114     memset(a, 0, sizeof(*a));
115 }
116
117 void swr_free(SwrContext **ss){
118     SwrContext *s= *ss;
119     if(s){
120         free_temp(&s->postin);
121         free_temp(&s->midbuf);
122         free_temp(&s->preout);
123         free_temp(&s->in_buffer);
124         swri_audio_convert_free(&s-> in_convert);
125         swri_audio_convert_free(&s->out_convert);
126         swri_audio_convert_free(&s->full_convert);
127         swri_resample_free(&s->resample);
128     }
129
130     av_freep(ss);
131 }
132
133 int swr_init(struct SwrContext *s){
134     s->in_buffer_index= 0;
135     s->in_buffer_count= 0;
136     s->resample_in_constraint= 0;
137     free_temp(&s->postin);
138     free_temp(&s->midbuf);
139     free_temp(&s->preout);
140     free_temp(&s->in_buffer);
141     swri_audio_convert_free(&s-> in_convert);
142     swri_audio_convert_free(&s->out_convert);
143     swri_audio_convert_free(&s->full_convert);
144
145     s-> in.planar= s-> in_sample_fmt >= 0x100;
146     s->out.planar= s->out_sample_fmt >= 0x100;
147     s-> in_sample_fmt &= 0xFF;
148     s->out_sample_fmt &= 0xFF;
149
150     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
151         av_log(s, AV_LOG_ERROR, "Requested sample format %s is invalid\n", av_get_sample_fmt_name(s->in_sample_fmt));
152         return AVERROR(EINVAL);
153     }
154     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
155         av_log(s, AV_LOG_ERROR, "Requested sample format %s is invalid\n", av_get_sample_fmt_name(s->out_sample_fmt));
156         return AVERROR(EINVAL);
157     }
158
159     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16
160         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLT){
161         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));
162         return AVERROR(EINVAL);
163     }
164
165     //FIXME should we allow/support using FLT on material that doesnt need it ?
166     if(s->in_sample_fmt <= AV_SAMPLE_FMT_S16 || s->int_sample_fmt==AV_SAMPLE_FMT_S16){
167         s->int_sample_fmt= AV_SAMPLE_FMT_S16;
168     }else
169         s->int_sample_fmt= AV_SAMPLE_FMT_FLT;
170
171
172     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
173         s->resample = swri_resample_init(s->resample, s->out_sample_rate, s->in_sample_rate, 16, 10, 0, 0.8);
174     }else
175         swri_resample_free(&s->resample);
176     if(s->int_sample_fmt != AV_SAMPLE_FMT_S16 && s->resample){
177         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16 currently\n"); //FIXME
178         return -1;
179     }
180
181     if(!s->used_ch_count)
182         s->used_ch_count= s->in.ch_count;
183
184     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
185         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
186         s-> in_ch_layout= 0;
187     }
188
189     if(!s-> in_ch_layout)
190         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
191     if(!s->out_ch_layout)
192         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
193
194     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0;
195
196 #define RSC 1 //FIXME finetune
197     if(!s-> in.ch_count)
198         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
199     if(!s->used_ch_count)
200         s->used_ch_count= s->in.ch_count;
201     if(!s->out.ch_count)
202         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
203
204 av_assert0(s-> in.ch_count);
205 av_assert0(s->used_ch_count);
206 av_assert0(s->out.ch_count);
207     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
208
209     s-> in.bps= av_get_bytes_per_sample(s-> in_sample_fmt);
210     s->int_bps= av_get_bytes_per_sample(s->int_sample_fmt);
211     s->out.bps= av_get_bytes_per_sample(s->out_sample_fmt);
212
213     if(!s->resample && !s->rematrix && !s->channel_map){
214         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
215                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
216         return 0;
217     }
218
219     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
220                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
221     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
222                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
223
224
225     s->postin= s->in;
226     s->preout= s->out;
227     s->midbuf= s->in;
228     s->in_buffer= s->in;
229     if(s->channel_map){
230         s->postin.ch_count=
231         s->midbuf.ch_count=
232         s->in_buffer.ch_count= s->used_ch_count;
233     }
234     if(!s->resample_first){
235         s->midbuf.ch_count= s->out.ch_count;
236         s->in_buffer.ch_count = s->out.ch_count;
237     }
238
239     s->in_buffer.bps = s->postin.bps = s->midbuf.bps = s->preout.bps =  s->int_bps;
240     s->in_buffer.planar = s->postin.planar = s->midbuf.planar = s->preout.planar =  1;
241
242
243     if(s->rematrix)
244         return swri_rematrix_init(s);
245
246     return 0;
247 }
248
249 static int realloc_audio(AudioData *a, int count){
250     int i, countb;
251     AudioData old;
252
253     if(a->count >= count)
254         return 0;
255
256     count*=2;
257
258     countb= FFALIGN(count*a->bps, 32);
259     old= *a;
260
261     av_assert0(a->planar);
262     av_assert0(a->bps);
263     av_assert0(a->ch_count);
264
265     a->data= av_malloc(countb*a->ch_count);
266     if(!a->data)
267         return AVERROR(ENOMEM);
268     for(i=0; i<a->ch_count; i++){
269         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
270         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
271     }
272     av_free(old.data);
273     a->count= count;
274
275     return 1;
276 }
277
278 static void copy(AudioData *out, AudioData *in,
279                  int count){
280     av_assert0(out->planar == in->planar);
281     av_assert0(out->bps == in->bps);
282     av_assert0(out->ch_count == in->ch_count);
283     if(out->planar){
284         int ch;
285         for(ch=0; ch<out->ch_count; ch++)
286             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
287     }else
288         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
289 }
290
291 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
292     int i;
293     if(out->planar){
294         for(i=0; i<out->ch_count; i++)
295             out->ch[i]= in_arg[i];
296     }else{
297         for(i=0; i<out->ch_count; i++)
298             out->ch[i]= in_arg[0] + i*out->bps;
299     }
300 }
301
302 /**
303  *
304  * out may be equal in.
305  */
306 static void buf_set(AudioData *out, AudioData *in, int count){
307     if(in->planar){
308         int ch;
309         for(ch=0; ch<out->ch_count; ch++)
310             out->ch[ch]= in->ch[ch] + count*out->bps;
311     }else
312         out->ch[0]= in->ch[0] + count*out->ch_count*out->bps;
313 }
314
315 /**
316  *
317  * @return number of samples output per channel
318  */
319 static int resample(SwrContext *s, AudioData *out_param, int out_count,
320                              const AudioData * in_param, int in_count){
321     AudioData in, out, tmp;
322     int ret_sum=0;
323     int border=0;
324
325     tmp=out=*out_param;
326     in =  *in_param;
327
328     do{
329         int ret, size, consumed;
330         if(!s->resample_in_constraint && s->in_buffer_count){
331             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
332             ret= swri_multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
333             out_count -= ret;
334             ret_sum += ret;
335             buf_set(&out, &out, ret);
336             s->in_buffer_count -= consumed;
337             s->in_buffer_index += consumed;
338
339             if(!in_count)
340                 break;
341             if(s->in_buffer_count <= border){
342                 buf_set(&in, &in, -s->in_buffer_count);
343                 in_count += s->in_buffer_count;
344                 s->in_buffer_count=0;
345                 s->in_buffer_index=0;
346                 border = 0;
347             }
348         }
349
350         if(in_count && !s->in_buffer_count){
351             s->in_buffer_index=0;
352             ret= swri_multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
353             out_count -= ret;
354             ret_sum += ret;
355             buf_set(&out, &out, ret);
356             in_count -= consumed;
357             buf_set(&in, &in, consumed);
358         }
359
360         //TODO is this check sane considering the advanced copy avoidance below
361         size= s->in_buffer_index + s->in_buffer_count + in_count;
362         if(   size > s->in_buffer.count
363            && s->in_buffer_count + in_count <= s->in_buffer_index){
364             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
365             copy(&s->in_buffer, &tmp, s->in_buffer_count);
366             s->in_buffer_index=0;
367         }else
368             if((ret=realloc_audio(&s->in_buffer, size)) < 0)
369                 return ret;
370
371         if(in_count){
372             int count= in_count;
373             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
374
375             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
376             copy(&tmp, &in, /*in_*/count);
377             s->in_buffer_count += count;
378             in_count -= count;
379             border += count;
380             buf_set(&in, &in, count);
381             s->resample_in_constraint= 0;
382             if(s->in_buffer_count != count || in_count)
383                 continue;
384         }
385         break;
386     }while(1);
387
388     s->resample_in_constraint= !!out_count;
389
390     return ret_sum;
391 }
392
393 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
394                          const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
395     AudioData *postin, *midbuf, *preout;
396     int ret/*, in_max*/;
397     AudioData * in= &s->in;
398     AudioData *out= &s->out;
399     AudioData preout_tmp, midbuf_tmp;
400
401     if(!s->resample){
402         if(in_count > out_count)
403             return -1;
404         out_count = in_count;
405     }
406
407     if(!in_arg){
408         if(s->in_buffer_count){
409             AudioData *a= &s->in_buffer;
410             int i, j, ret;
411             if((ret=realloc_audio(a, s->in_buffer_index + 2*s->in_buffer_count)) < 0)
412                 return ret;
413             av_assert0(a->planar);
414             for(i=0; i<a->ch_count; i++){
415                 for(j=0; j<s->in_buffer_count; j++){
416                     memcpy(a->ch[i] + (s->in_buffer_index+s->in_buffer_count+j  )*a->bps,
417                            a->ch[i] + (s->in_buffer_index+s->in_buffer_count-j-1)*a->bps, a->bps);
418                 }
419             }
420             s->in_buffer_count += (s->in_buffer_count+1)/2;
421             s->resample_in_constraint = 0;
422         }else{
423             return 0;
424         }
425     }else
426         fill_audiodata(in ,  (void*)in_arg);
427     fill_audiodata(out, out_arg);
428
429     if(s->full_convert){
430         av_assert0(!s->resample);
431         swri_audio_convert(s->full_convert, out, in, in_count);
432         return out_count;
433     }
434
435 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
436 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
437
438     if((ret=realloc_audio(&s->postin, in_count))<0)
439         return ret;
440     if(s->resample_first){
441         av_assert0(s->midbuf.ch_count == s->used_ch_count);
442         if((ret=realloc_audio(&s->midbuf, out_count))<0)
443             return ret;
444     }else{
445         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
446         if((ret=realloc_audio(&s->midbuf,  in_count))<0)
447             return ret;
448     }
449     if((ret=realloc_audio(&s->preout, out_count))<0)
450         return ret;
451
452     postin= &s->postin;
453
454     midbuf_tmp= s->midbuf;
455     midbuf= &midbuf_tmp;
456     preout_tmp= s->preout;
457     preout= &preout_tmp;
458
459     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar)
460         postin= in;
461
462     if(s->resample_first ? !s->resample : !s->rematrix)
463         midbuf= postin;
464
465     if(s->resample_first ? !s->rematrix : !s->resample)
466         preout= midbuf;
467
468     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
469         if(preout==in){
470             out_count= FFMIN(out_count, in_count); //TODO check at teh end if this is needed or redundant
471             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
472             copy(out, in, out_count);
473             return out_count;
474         }
475         else if(preout==postin) preout= midbuf= postin= out;
476         else if(preout==midbuf) preout= midbuf= out;
477         else                    preout= out;
478     }
479
480     if(in != postin){
481         swri_audio_convert(s->in_convert, postin, in, in_count);
482     }
483
484     if(s->resample_first){
485         if(postin != midbuf)
486             out_count= resample(s, midbuf, out_count, postin, in_count);
487         if(midbuf != preout)
488             swri_rematrix(s, preout, midbuf, out_count, preout==out);
489     }else{
490         if(postin != midbuf)
491             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
492         if(midbuf != preout)
493             out_count= resample(s, preout, out_count, midbuf, in_count);
494     }
495
496     if(preout != out){
497 //FIXME packed doesnt need more than 1 chan here!
498         swri_audio_convert(s->out_convert, out, preout, out_count);
499     }
500     if(!in_arg)
501         s->in_buffer_count = 0;
502     return out_count;
503 }
504