]> git.sesse.net Git - ffmpeg/blob - libswresample/swresample.c
rematrix: add parameter to tune volume
[ffmpeg] / 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 {"isr",  "input sample rate"  , OFFSET( in_sample_rate), AV_OPT_TYPE_INT, {.dbl=48000}, 1, INT_MAX, 0},
42 {"osr", "output sample rate"  , OFFSET(out_sample_rate), AV_OPT_TYPE_INT, {.dbl=48000}, 1, INT_MAX, 0},
43 //{"ip" ,  "input planar"       , OFFSET( in.planar     ), AV_OPT_TYPE_INT, {.dbl=0},    0,       1, 0},
44 //{"op" , "output planar"       , OFFSET(out.planar     ), AV_OPT_TYPE_INT, {.dbl=0},    0,       1, 0},
45 {"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},
46 {"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},
47 {"tsf", "internal sample format", OFFSET(int_sample_fmt ), AV_OPT_TYPE_INT, {.dbl=AV_SAMPLE_FMT_NONE}, -1, AV_SAMPLE_FMT_FLT, 0},
48 {"icl",  "input channel layout" , OFFSET( in_ch_layout), AV_OPT_TYPE_INT64, {.dbl=0}, 0, INT64_MAX, 0, "channel_layout"},
49 {"ocl",  "output channel layout", OFFSET(out_ch_layout), AV_OPT_TYPE_INT64, {.dbl=0}, 0, INT64_MAX, 0, "channel_layout"},
50 {"clev", "center mix level"     , OFFSET(clev)         , AV_OPT_TYPE_FLOAT, {.dbl=C_30DB}, 0, 4, 0},
51 {"slev", "sourround mix level"  , OFFSET(slev)         , AV_OPT_TYPE_FLOAT, {.dbl=C_30DB}, 0, 4, 0},
52 {"rmvol", "rematrix volume"     , OFFSET(rematrix_volume), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, -1000, 1000, 0},
53 {"flags", NULL                  , OFFSET(flags)        , AV_OPT_TYPE_FLAGS, {.dbl=0}, 0,  UINT_MAX, 0, "flags"},
54 {"res", "force resampling", 0, AV_OPT_TYPE_CONST, {.dbl=SWR_FLAG_RESAMPLE}, INT_MIN, INT_MAX, 0, "flags"},
55
56 {0}
57 };
58
59 static const char* context_to_name(void* ptr) {
60     return "SWR";
61 }
62
63 static const AVClass av_class = { "SwrContext", context_to_name, options, LIBAVUTIL_VERSION_INT, OFFSET(log_level_offset), OFFSET(log_ctx) };
64
65 static int resample(SwrContext *s, AudioData *out_param, int out_count,
66                              const AudioData * in_param, int in_count);
67
68 SwrContext *swr_alloc(void){
69     SwrContext *s= av_mallocz(sizeof(SwrContext));
70     if(s){
71         s->av_class= &av_class;
72         av_opt_set_defaults2(s, 0, 0);
73     }
74     return s;
75 }
76
77 SwrContext *swr_alloc2(struct SwrContext *s, int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
78                        int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
79                        int log_offset, void *log_ctx){
80     if(!s) s= swr_alloc();
81     if(!s) return NULL;
82
83     s->log_level_offset= log_offset;
84     s->log_ctx= log_ctx;
85
86     av_set_int(s, "ocl", out_ch_layout);
87     av_set_int(s, "osf", out_sample_fmt);
88     av_set_int(s, "osr", out_sample_rate);
89     av_set_int(s, "icl", in_ch_layout);
90     av_set_int(s, "isf", in_sample_fmt);
91     av_set_int(s, "isr", in_sample_rate);
92
93     s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
94     s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
95     s->int_sample_fmt = AV_SAMPLE_FMT_S16;
96
97     return s;
98 }
99
100
101 static void free_temp(AudioData *a){
102     av_free(a->data);
103     memset(a, 0, sizeof(*a));
104 }
105
106 void swr_free(SwrContext **ss){
107     SwrContext *s= *ss;
108     if(s){
109         free_temp(&s->postin);
110         free_temp(&s->midbuf);
111         free_temp(&s->preout);
112         free_temp(&s->in_buffer);
113         swr_audio_convert_free(&s-> in_convert);
114         swr_audio_convert_free(&s->out_convert);
115         swr_audio_convert_free(&s->full_convert);
116         swr_resample_free(&s->resample);
117     }
118
119     av_freep(ss);
120 }
121
122 int swr_init(SwrContext *s){
123     s->in_buffer_index= 0;
124     s->in_buffer_count= 0;
125     s->resample_in_constraint= 0;
126     free_temp(&s->postin);
127     free_temp(&s->midbuf);
128     free_temp(&s->preout);
129     free_temp(&s->in_buffer);
130     swr_audio_convert_free(&s-> in_convert);
131     swr_audio_convert_free(&s->out_convert);
132     swr_audio_convert_free(&s->full_convert);
133
134     s-> in.planar= s-> in_sample_fmt >= 0x100;
135     s->out.planar= s->out_sample_fmt >= 0x100;
136     s-> in_sample_fmt &= 0xFF;
137     s->out_sample_fmt &= 0xFF;
138
139     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
140         av_log(s, AV_LOG_ERROR, "Requested sample format %s is invalid\n", av_get_sample_fmt_name(s->in_sample_fmt));
141         return AVERROR(EINVAL);
142     }
143     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
144         av_log(s, AV_LOG_ERROR, "Requested sample format %s is invalid\n", av_get_sample_fmt_name(s->out_sample_fmt));
145         return AVERROR(EINVAL);
146     }
147
148     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16
149         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLT){
150         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));
151         return AVERROR(EINVAL);
152     }
153
154     //FIXME should we allow/support using FLT on material that doesnt need it ?
155     if(s->in_sample_fmt <= AV_SAMPLE_FMT_S16 || s->int_sample_fmt==AV_SAMPLE_FMT_S16){
156         s->int_sample_fmt= AV_SAMPLE_FMT_S16;
157     }else
158         s->int_sample_fmt= AV_SAMPLE_FMT_FLT;
159
160
161     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
162         s->resample = swr_resample_init(s->resample, s->out_sample_rate, s->in_sample_rate, 16, 10, 0, 0.8);
163     }else
164         swr_resample_free(&s->resample);
165     if(s->int_sample_fmt != AV_SAMPLE_FMT_S16 && s->resample){
166         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16 currently\n"); //FIXME
167         return -1;
168     }
169
170     if(s-> in.ch_count && s-> in_ch_layout && s->in.ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
171         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than there actually is, ignoring layout\n");
172         s-> in_ch_layout= 0;
173     }
174
175     if(!s-> in_ch_layout)
176         s-> in_ch_layout= av_get_default_channel_layout(s->in.ch_count);
177     if(!s->out_ch_layout)
178         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
179
180     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0;
181
182 #define RSC 1 //FIXME finetune
183     if(!s-> in.ch_count)
184         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
185     if(!s->out.ch_count)
186         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
187
188 av_assert0(s-> in.ch_count);
189 av_assert0(s->out.ch_count);
190     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
191
192     s-> in.bps= av_get_bits_per_sample_fmt(s-> in_sample_fmt)/8;
193     s->int_bps= av_get_bits_per_sample_fmt(s->int_sample_fmt)/8;
194     s->out.bps= av_get_bits_per_sample_fmt(s->out_sample_fmt)/8;
195
196     if(!s->resample && !s->rematrix){
197         s->full_convert = swr_audio_convert_alloc(s->out_sample_fmt,
198                                                   s-> in_sample_fmt, s-> in.ch_count, 0);
199         return 0;
200     }
201
202     s->in_convert = swr_audio_convert_alloc(s->int_sample_fmt,
203                                             s-> in_sample_fmt, s-> in.ch_count, 0);
204     s->out_convert= swr_audio_convert_alloc(s->out_sample_fmt,
205                                             s->int_sample_fmt, s->out.ch_count, 0);
206
207
208     s->postin= s->in;
209     s->preout= s->out;
210     s->midbuf= s->in;
211     s->in_buffer= s->in;
212     if(!s->resample_first){
213         s->midbuf.ch_count= s->out.ch_count;
214         s->in_buffer.ch_count = s->out.ch_count;
215     }
216
217     s->in_buffer.bps = s->postin.bps = s->midbuf.bps = s->preout.bps =  s->int_bps;
218     s->in_buffer.planar = s->postin.planar = s->midbuf.planar = s->preout.planar =  1;
219
220
221     if(s->rematrix && swr_rematrix_init(s)<0)
222         return -1;
223
224     return 0;
225 }
226
227 static int realloc_audio(AudioData *a, int count){
228     int i, countb;
229     AudioData old;
230
231     if(a->count >= count)
232         return 0;
233
234     count*=2;
235
236     countb= FFALIGN(count*a->bps, 32);
237     old= *a;
238
239     av_assert0(a->planar);
240     av_assert0(a->bps);
241     av_assert0(a->ch_count);
242
243     a->data= av_malloc(countb*a->ch_count);
244     if(!a->data)
245         return AVERROR(ENOMEM);
246     for(i=0; i<a->ch_count; i++){
247         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
248         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
249     }
250     av_free(old.data);
251     a->count= count;
252
253     return 1;
254 }
255
256 static void copy(AudioData *out, AudioData *in,
257                  int count){
258     av_assert0(out->planar == in->planar);
259     av_assert0(out->bps == in->bps);
260     av_assert0(out->ch_count == in->ch_count);
261     if(out->planar){
262         int ch;
263         for(ch=0; ch<out->ch_count; ch++)
264             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
265     }else
266         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
267 }
268
269 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
270     int i;
271     if(out->planar){
272         for(i=0; i<out->ch_count; i++)
273             out->ch[i]= in_arg[i];
274     }else{
275         for(i=0; i<out->ch_count; i++)
276             out->ch[i]= in_arg[0] + i*out->bps;
277     }
278 }
279
280 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
281                          const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
282     AudioData *postin, *midbuf, *preout;
283     int ret/*, in_max*/;
284     AudioData * in= &s->in;
285     AudioData *out= &s->out;
286     AudioData preout_tmp, midbuf_tmp;
287
288     if(!s->resample){
289         if(in_count > out_count)
290             return -1;
291         out_count = in_count;
292     }
293
294     fill_audiodata(in ,  (void*)in_arg);
295     fill_audiodata(out, out_arg);
296
297     if(s->full_convert){
298         av_assert0(!s->resample);
299         swr_audio_convert(s->full_convert, out, in, in_count);
300         return out_count;
301     }
302
303 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
304 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
305
306     if((ret=realloc_audio(&s->postin, in_count))<0)
307         return ret;
308     if(s->resample_first){
309         av_assert0(s->midbuf.ch_count ==  s-> in.ch_count);
310         if((ret=realloc_audio(&s->midbuf, out_count))<0)
311             return ret;
312     }else{
313         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
314         if((ret=realloc_audio(&s->midbuf,  in_count))<0)
315             return ret;
316     }
317     if((ret=realloc_audio(&s->preout, out_count))<0)
318         return ret;
319
320     postin= &s->postin;
321
322     midbuf_tmp= s->midbuf;
323     midbuf= &midbuf_tmp;
324     preout_tmp= s->preout;
325     preout= &preout_tmp;
326
327     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar)
328         postin= in;
329
330     if(s->resample_first ? !s->resample : !s->rematrix)
331         midbuf= postin;
332
333     if(s->resample_first ? !s->rematrix : !s->resample)
334         preout= midbuf;
335
336     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
337         if(preout==in){
338             out_count= FFMIN(out_count, in_count); //TODO check at teh end if this is needed or redundant
339             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
340             copy(out, in, out_count);
341             return out_count;
342         }
343         else if(preout==postin) preout= midbuf= postin= out;
344         else if(preout==midbuf) preout= midbuf= out;
345         else                    preout= out;
346     }
347
348     if(in != postin){
349         swr_audio_convert(s->in_convert, postin, in, in_count);
350     }
351
352     if(s->resample_first){
353         if(postin != midbuf)
354             out_count= resample(s, midbuf, out_count, postin, in_count);
355         if(midbuf != preout)
356             swr_rematrix(s, preout, midbuf, out_count, preout==out);
357     }else{
358         if(postin != midbuf)
359             swr_rematrix(s, midbuf, postin, in_count, midbuf==out);
360         if(midbuf != preout)
361             out_count= resample(s, preout, out_count, midbuf, in_count);
362     }
363
364     if(preout != out){
365 //FIXME packed doesnt need more than 1 chan here!
366         swr_audio_convert(s->out_convert, out, preout, out_count);
367     }
368     return out_count;
369 }
370
371 /**
372  *
373  * out may be equal in.
374  */
375 static void buf_set(AudioData *out, AudioData *in, int count){
376     if(in->planar){
377         int ch;
378         for(ch=0; ch<out->ch_count; ch++)
379             out->ch[ch]= in->ch[ch] + count*out->bps;
380     }else
381         out->ch[0]= in->ch[0] + count*out->ch_count*out->bps;
382 }
383
384 /**
385  *
386  * @return number of samples output per channel
387  */
388 static int resample(SwrContext *s, AudioData *out_param, int out_count,
389                              const AudioData * in_param, int in_count){
390     AudioData in, out, tmp;
391     int ret_sum=0;
392     int border=0;
393
394     tmp=out=*out_param;
395     in =  *in_param;
396
397     do{
398         int ret, size, consumed;
399         if(!s->resample_in_constraint && s->in_buffer_count){
400             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
401             ret= swr_multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
402             out_count -= ret;
403             ret_sum += ret;
404             buf_set(&out, &out, ret);
405             s->in_buffer_count -= consumed;
406             s->in_buffer_index += consumed;
407
408             if(!in_count)
409                 break;
410             if(s->in_buffer_count <= border){
411                 buf_set(&in, &in, -s->in_buffer_count);
412                 in_count += s->in_buffer_count;
413                 s->in_buffer_count=0;
414                 s->in_buffer_index=0;
415                 border = 0;
416             }
417         }
418
419         if(in_count && !s->in_buffer_count){
420             s->in_buffer_index=0;
421             ret= swr_multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
422             out_count -= ret;
423             ret_sum += ret;
424             buf_set(&out, &out, ret);
425             in_count -= consumed;
426             buf_set(&in, &in, consumed);
427         }
428
429         //TODO is this check sane considering the advanced copy avoidance below
430         size= s->in_buffer_index + s->in_buffer_count + in_count;
431         if(   size > s->in_buffer.count
432            && s->in_buffer_count + in_count <= s->in_buffer_index){
433             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
434             copy(&s->in_buffer, &tmp, s->in_buffer_count);
435             s->in_buffer_index=0;
436         }else
437             if((ret=realloc_audio(&s->in_buffer, size)) < 0)
438                 return ret;
439
440         if(in_count){
441             int count= in_count;
442             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
443
444             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
445             copy(&tmp, &in, /*in_*/count);
446             s->in_buffer_count += count;
447             in_count -= count;
448             border += count;
449             buf_set(&in, &in, count);
450             s->resample_in_constraint= 0;
451             if(s->in_buffer_count != count || in_count)
452                 continue;
453         }
454         break;
455     }while(1);
456
457     s->resample_in_constraint= !!out_count;
458
459     return ret_sum;
460 }