]> git.sesse.net Git - ffmpeg/blob - libswresample/swresample.c
libswresample: support directly converting sampleformats and packed/planar in a singl...
[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
26 #define  C30DB  M_SQRT2
27 #define  C15DB  1.189207115
28 #define C__0DB  1.0
29 #define C_15DB  0.840896415
30 #define C_30DB  M_SQRT1_2
31 #define C_45DB  0.594603558
32 #define C_60DB  0.5
33
34
35 //TODO split options array out?
36 #define OFFSET(x) offsetof(SwrContext,x)
37 static const AVOption options[]={
38 {"ich",  "input channel count", OFFSET( in.ch_count   ), FF_OPT_TYPE_INT, {.dbl=2}, 1, SWR_CH_MAX, 0},
39 {"och", "output channel count", OFFSET(out.ch_count   ), FF_OPT_TYPE_INT, {.dbl=2}, 1, SWR_CH_MAX, 0},
40 {"isr",  "input sample rate"  , OFFSET( in_sample_rate), FF_OPT_TYPE_INT, {.dbl=48000}, 1, INT_MAX, 0},
41 {"osr", "output sample rate"  , OFFSET(out_sample_rate), FF_OPT_TYPE_INT, {.dbl=48000}, 1, INT_MAX, 0},
42 //{"ip" ,  "input planar"       , OFFSET( in.planar     ), FF_OPT_TYPE_INT, {.dbl=0},    0,       1, 0},
43 //{"op" , "output planar"       , OFFSET(out.planar     ), FF_OPT_TYPE_INT, {.dbl=0},    0,       1, 0},
44 {"isf",  "input sample format", OFFSET( in_sample_fmt ), FF_OPT_TYPE_INT, {.dbl=AV_SAMPLE_FMT_S16}, 0, AV_SAMPLE_FMT_NB-1+256, 0},
45 {"osf", "output sample format", OFFSET(out_sample_fmt ), FF_OPT_TYPE_INT, {.dbl=AV_SAMPLE_FMT_S16}, 0, AV_SAMPLE_FMT_NB-1+256, 0},
46 {"tsf", "internal sample format", OFFSET(int_sample_fmt ), FF_OPT_TYPE_INT, {.dbl=AV_SAMPLE_FMT_NONE}, -1, AV_SAMPLE_FMT_FLT, 0},
47 {"icl",  "input channel layout" , OFFSET( in_ch_layout), FF_OPT_TYPE_INT64, {.dbl=0}, 0, INT64_MAX, 0, "channel_layout"},
48 {"ocl",  "output channel layout", OFFSET(out_ch_layout), FF_OPT_TYPE_INT64, {.dbl=0}, 0, INT64_MAX, 0, "channel_layout"},
49 {"clev", "center mix level"     , OFFSET(clev)         , FF_OPT_TYPE_FLOAT, {.dbl=C_30DB}, 0, 4, 0},
50 {"slev", "sourround mix level"  , OFFSET(slev)         , FF_OPT_TYPE_FLOAT, {.dbl=C_30DB}, 0, 4, 0},
51 {"flags", NULL                  , OFFSET(flags)        , FF_OPT_TYPE_FLAGS, {.dbl=0}, 0,  UINT_MAX, 0, "flags"},
52 {"res", "force resampling", 0, FF_OPT_TYPE_CONST, {.dbl=SWR_FLAG_RESAMPLE}, INT_MIN, INT_MAX, 0, "flags"},
53
54 {0}
55 };
56
57 static const char* context_to_name(void* ptr) {
58     return "SWR";
59 }
60
61 static const AVClass av_class = { "SwrContext", context_to_name, options, LIBAVUTIL_VERSION_INT, OFFSET(log_level_offset), OFFSET(log_ctx) };
62
63 static int resample(SwrContext *s, AudioData *out_param, int out_count,
64                              const AudioData * in_param, int in_count);
65
66 SwrContext *swr_alloc(void){
67     SwrContext *s= av_mallocz(sizeof(SwrContext));
68     if(s){
69         s->av_class= &av_class;
70         av_opt_set_defaults2(s, 0, 0);
71     }
72     return s;
73 }
74
75 SwrContext *swr_alloc2(struct SwrContext *s, int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
76                        int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
77                        int log_offset, void *log_ctx){
78     if(!s) s= swr_alloc();
79     if(!s) return NULL;
80
81     s->log_level_offset= log_offset;
82     s->log_ctx= log_ctx;
83
84     av_set_int(s, "ocl", out_ch_layout);
85     av_set_int(s, "osf", out_sample_fmt);
86     av_set_int(s, "osr", out_sample_rate);
87     av_set_int(s, "icl", in_ch_layout);
88     av_set_int(s, "isf", in_sample_fmt);
89     av_set_int(s, "isr", in_sample_rate);
90
91     s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
92     s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
93     s->int_sample_fmt = AV_SAMPLE_FMT_S16;
94
95     return s;
96 }
97
98
99 static void free_temp(AudioData *a){
100     av_free(a->data);
101     memset(a, 0, sizeof(*a));
102 }
103
104 void swr_free(SwrContext **ss){
105     SwrContext *s= *ss;
106     if(s){
107         free_temp(&s->postin);
108         free_temp(&s->midbuf);
109         free_temp(&s->preout);
110         free_temp(&s->in_buffer);
111         swr_audio_convert_free(&s-> in_convert);
112         swr_audio_convert_free(&s->out_convert);
113         swr_audio_convert_free(&s->full_convert);
114         swr_resample_free(&s->resample);
115     }
116
117     av_freep(ss);
118 }
119
120 static int64_t guess_layout(int ch){
121     switch(ch){
122     case 1: return AV_CH_LAYOUT_MONO;
123     case 2: return AV_CH_LAYOUT_STEREO;
124     case 5: return AV_CH_LAYOUT_5POINT0;
125     case 6: return AV_CH_LAYOUT_5POINT1;
126     case 7: return AV_CH_LAYOUT_7POINT0;
127     case 8: return AV_CH_LAYOUT_7POINT1;
128     default: return 0;
129     }
130 }
131
132 int swr_init(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     swr_audio_convert_free(&s-> in_convert);
141     swr_audio_convert_free(&s->out_convert);
142     swr_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 = swr_resample_init(s->resample, s->out_sample_rate, s->in_sample_rate, 16, 10, 0, 0.8);
173     }else
174         swr_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-> in_ch_layout)
181         s-> in_ch_layout= guess_layout(s->in.ch_count);
182     if(!s->out_ch_layout)
183         s->out_ch_layout= guess_layout(s->out.ch_count);
184
185     s->rematrix= s->out_ch_layout  !=s->in_ch_layout;
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->out.ch_count)
191         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
192
193 av_assert0(s-> in.ch_count);
194 av_assert0(s->out.ch_count);
195     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
196
197     s-> in.bps= av_get_bits_per_sample_fmt(s-> in_sample_fmt)/8;
198     s->int_bps= av_get_bits_per_sample_fmt(s->int_sample_fmt)/8;
199     s->out.bps= av_get_bits_per_sample_fmt(s->out_sample_fmt)/8;
200
201     if(!s->resample && !s->rematrix){
202         s->full_convert = swr_audio_convert_alloc(s->out_sample_fmt,
203                                                   s-> in_sample_fmt, s-> in.ch_count, 0);
204         return 0;
205     }
206
207     s->in_convert = swr_audio_convert_alloc(s->int_sample_fmt,
208                                             s-> in_sample_fmt, s-> in.ch_count, 0);
209     s->out_convert= swr_audio_convert_alloc(s->out_sample_fmt,
210                                             s->int_sample_fmt, s->out.ch_count, 0);
211
212
213     s->postin= s->in;
214     s->preout= s->out;
215     s->midbuf= s->in;
216     s->in_buffer= s->in;
217     if(!s->resample_first){
218         s->midbuf.ch_count= s->out.ch_count;
219         s->in_buffer.ch_count = s->out.ch_count;
220     }
221
222     s->in_buffer.bps = s->postin.bps = s->midbuf.bps = s->preout.bps =  s->int_bps;
223     s->in_buffer.planar = s->postin.planar = s->midbuf.planar = s->preout.planar =  1;
224
225
226     if(s->rematrix && swr_rematrix_init(s)<0)
227         return -1;
228
229     return 0;
230 }
231
232 static int realloc_audio(AudioData *a, int count){
233     int i, countb;
234     AudioData old;
235
236     if(a->count >= count)
237         return 0;
238
239     count*=2;
240
241     countb= FFALIGN(count*a->bps, 32);
242     old= *a;
243
244     av_assert0(a->planar);
245     av_assert0(a->bps);
246     av_assert0(a->ch_count);
247
248     a->data= av_malloc(countb*a->ch_count);
249     if(!a->data)
250         return AVERROR(ENOMEM);
251     for(i=0; i<a->ch_count; i++){
252         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
253         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
254     }
255     av_free(old.data);
256     a->count= count;
257
258     return 1;
259 }
260
261 static void copy(AudioData *out, AudioData *in,
262                  int count){
263     av_assert0(out->planar == in->planar);
264     av_assert0(out->bps == in->bps);
265     av_assert0(out->ch_count == in->ch_count);
266     if(out->planar){
267         int ch;
268         for(ch=0; ch<out->ch_count; ch++)
269             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
270     }else
271         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
272 }
273
274 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
275     int i;
276     if(out->planar){
277         for(i=0; i<out->ch_count; i++)
278             out->ch[i]= in_arg[i];
279     }else{
280         for(i=0; i<out->ch_count; i++)
281             out->ch[i]= in_arg[0] + i*out->bps;
282     }
283 }
284
285 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
286                          const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
287     AudioData *postin, *midbuf, *preout;
288     int ret, i/*, in_max*/;
289     AudioData * in= &s->in;
290     AudioData *out= &s->out;
291     AudioData preout_tmp, midbuf_tmp;
292
293     if(!s->resample){
294         if(in_count > out_count)
295             return -1;
296         out_count = in_count;
297     }
298
299     fill_audiodata(in ,  in_arg);
300     fill_audiodata(out, out_arg);
301
302     if(s->full_convert){
303         av_assert0(!s->resample);
304         swr_audio_convert(s->full_convert, out, in, in_count);
305         return out_count;
306     }
307
308 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
309 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
310
311     if((ret=realloc_audio(&s->postin, in_count))<0)
312         return ret;
313     if(s->resample_first){
314         av_assert0(s->midbuf.ch_count ==  s-> in.ch_count);
315         if((ret=realloc_audio(&s->midbuf, out_count))<0)
316             return ret;
317     }else{
318         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
319         if((ret=realloc_audio(&s->midbuf,  in_count))<0)
320             return ret;
321     }
322     if((ret=realloc_audio(&s->preout, out_count))<0)
323         return ret;
324
325     postin= &s->postin;
326
327     midbuf_tmp= s->midbuf;
328     midbuf= &midbuf_tmp;
329     preout_tmp= s->preout;
330     preout= &preout_tmp;
331
332     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar)
333         postin= in;
334
335     if(s->resample_first ? !s->resample : !s->rematrix)
336         midbuf= postin;
337
338     if(s->resample_first ? !s->rematrix : !s->resample)
339         preout= midbuf;
340
341     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
342         if(preout==in){
343             out_count= FFMIN(out_count, in_count); //TODO check at teh end if this is needed or redundant
344             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
345             copy(out, in, out_count);
346             return out_count;
347         }
348         else if(preout==postin) preout= midbuf= postin= out;
349         else if(preout==midbuf) preout= midbuf= out;
350         else                    preout= out;
351     }
352
353     if(in != postin){
354         swr_audio_convert(s->in_convert, postin, in, in_count);
355     }
356
357     if(s->resample_first){
358         if(postin != midbuf)
359             out_count= resample(s, midbuf, out_count, postin, in_count);
360         if(midbuf != preout)
361             swr_rematrix(s, preout, midbuf, out_count, preout==out);
362     }else{
363         if(postin != midbuf)
364             swr_rematrix(s, midbuf, postin, in_count, midbuf==out);
365         if(midbuf != preout)
366             out_count= resample(s, preout, out_count, midbuf, in_count);
367     }
368
369     if(preout != out){
370 //FIXME packed doesnt need more than 1 chan here!
371         swr_audio_convert(s->out_convert, out, preout, out_count);
372     }
373     return out_count;
374 }
375
376 /**
377  *
378  * out may be equal in.
379  */
380 static void buf_set(AudioData *out, AudioData *in, int count){
381     if(in->planar){
382         int ch;
383         for(ch=0; ch<out->ch_count; ch++)
384             out->ch[ch]= in->ch[ch] + count*out->bps;
385     }else
386         out->ch[0]= in->ch[0] + count*out->ch_count*out->bps;
387 }
388
389 /**
390  *
391  * @return number of samples output per channel
392  */
393 static int resample(SwrContext *s, AudioData *out_param, int out_count,
394                              const AudioData * in_param, int in_count){
395     AudioData in, out, tmp;
396     int ret_sum=0;
397     int border=0;
398
399     tmp=out=*out_param;
400     in =  *in_param;
401
402     do{
403         int ret, size, consumed;
404         if(!s->resample_in_constraint && s->in_buffer_count){
405             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
406             ret= swr_multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
407             out_count -= ret;
408             ret_sum += ret;
409             buf_set(&out, &out, ret);
410             s->in_buffer_count -= consumed;
411             s->in_buffer_index += consumed;
412
413             if(!in_count)
414                 break;
415             if(s->in_buffer_count <= border){
416                 buf_set(&in, &in, -s->in_buffer_count);
417                 in_count += s->in_buffer_count;
418                 s->in_buffer_count=0;
419                 s->in_buffer_index=0;
420                 border = 0;
421             }
422         }
423
424         if(in_count && !s->in_buffer_count){
425             s->in_buffer_index=0;
426             ret= swr_multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
427             out_count -= ret;
428             ret_sum += ret;
429             buf_set(&out, &out, ret);
430             in_count -= consumed;
431             buf_set(&in, &in, consumed);
432         }
433
434         //TODO is this check sane considering the advanced copy avoidance below
435         size= s->in_buffer_index + s->in_buffer_count + in_count;
436         if(   size > s->in_buffer.count
437            && s->in_buffer_count + in_count <= s->in_buffer_index){
438             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
439             copy(&s->in_buffer, &tmp, s->in_buffer_count);
440             s->in_buffer_index=0;
441         }else
442             if((ret=realloc_audio(&s->in_buffer, size)) < 0)
443                 return ret;
444
445         if(in_count){
446             int count= in_count;
447             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
448
449             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
450             copy(&tmp, &in, /*in_*/count);
451             s->in_buffer_count += count;
452             in_count -= count;
453             border += count;
454             buf_set(&in, &in, count);
455             s->resample_in_constraint= 0;
456             if(s->in_buffer_count != count || in_count)
457                 continue;
458         }
459         break;
460     }while(1);
461
462     s->resample_in_constraint= !!out_count;
463
464     return ret_sum;
465 }