swr: use native planar sample formats.
[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     av_opt_set_int(s, "uch", 0, 0);
109     return s;
110 }
111
112
113 static void free_temp(AudioData *a){
114     av_free(a->data);
115     memset(a, 0, sizeof(*a));
116 }
117
118 void swr_free(SwrContext **ss){
119     SwrContext *s= *ss;
120     if(s){
121         free_temp(&s->postin);
122         free_temp(&s->midbuf);
123         free_temp(&s->preout);
124         free_temp(&s->in_buffer);
125         swri_audio_convert_free(&s-> in_convert);
126         swri_audio_convert_free(&s->out_convert);
127         swri_audio_convert_free(&s->full_convert);
128         swri_resample_free(&s->resample);
129     }
130
131     av_freep(ss);
132 }
133
134 int swr_init(struct SwrContext *s){
135     s->in_buffer_index= 0;
136     s->in_buffer_count= 0;
137     s->resample_in_constraint= 0;
138     free_temp(&s->postin);
139     free_temp(&s->midbuf);
140     free_temp(&s->preout);
141     free_temp(&s->in_buffer);
142     swri_audio_convert_free(&s-> in_convert);
143     swri_audio_convert_free(&s->out_convert);
144     swri_audio_convert_free(&s->full_convert);
145
146     s-> in.planar= av_sample_fmt_is_planar(s-> in_sample_fmt);
147     s->out.planar= av_sample_fmt_is_planar(s->out_sample_fmt);
148     s-> in_sample_fmt= av_get_alt_sample_fmt(s-> in_sample_fmt, 0);
149     s->out_sample_fmt= av_get_alt_sample_fmt(s->out_sample_fmt, 0);
150
151     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
152         av_log(s, AV_LOG_ERROR, "Requested sample format %s is invalid\n", av_get_sample_fmt_name(s->in_sample_fmt));
153         return AVERROR(EINVAL);
154     }
155     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
156         av_log(s, AV_LOG_ERROR, "Requested sample format %s is invalid\n", av_get_sample_fmt_name(s->out_sample_fmt));
157         return AVERROR(EINVAL);
158     }
159
160     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16
161         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLT){
162         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));
163         return AVERROR(EINVAL);
164     }
165
166     //FIXME should we allow/support using FLT on material that doesnt need it ?
167     if(s->in_sample_fmt <= AV_SAMPLE_FMT_S16 || s->int_sample_fmt==AV_SAMPLE_FMT_S16){
168         s->int_sample_fmt= AV_SAMPLE_FMT_S16;
169     }else
170         s->int_sample_fmt= AV_SAMPLE_FMT_FLT;
171
172
173     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
174         s->resample = swri_resample_init(s->resample, s->out_sample_rate, s->in_sample_rate, 16, 10, 0, 0.8);
175     }else
176         swri_resample_free(&s->resample);
177     if(s->int_sample_fmt != AV_SAMPLE_FMT_S16 && s->resample){
178         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16 currently\n"); //FIXME
179         return -1;
180     }
181
182     if(!s->used_ch_count)
183         s->used_ch_count= s->in.ch_count;
184
185     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
186         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
187         s-> in_ch_layout= 0;
188     }
189
190     if(!s-> in_ch_layout)
191         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
192     if(!s->out_ch_layout)
193         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
194
195     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0;
196
197 #define RSC 1 //FIXME finetune
198     if(!s-> in.ch_count)
199         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
200     if(!s->used_ch_count)
201         s->used_ch_count= s->in.ch_count;
202     if(!s->out.ch_count)
203         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
204
205 av_assert0(s-> in.ch_count);
206 av_assert0(s->used_ch_count);
207 av_assert0(s->out.ch_count);
208     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
209
210     s-> in.bps= av_get_bytes_per_sample(s-> in_sample_fmt);
211     s->int_bps= av_get_bytes_per_sample(s->int_sample_fmt);
212     s->out.bps= av_get_bytes_per_sample(s->out_sample_fmt);
213
214     if(!s->resample && !s->rematrix && !s->channel_map){
215         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
216                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
217         return 0;
218     }
219
220     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
221                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
222     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
223                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
224
225
226     s->postin= s->in;
227     s->preout= s->out;
228     s->midbuf= s->in;
229     s->in_buffer= s->in;
230     if(s->channel_map){
231         s->postin.ch_count=
232         s->midbuf.ch_count=
233         s->in_buffer.ch_count= s->used_ch_count;
234     }
235     if(!s->resample_first){
236         s->midbuf.ch_count= s->out.ch_count;
237         s->in_buffer.ch_count = s->out.ch_count;
238     }
239
240     s->in_buffer.bps = s->postin.bps = s->midbuf.bps = s->preout.bps =  s->int_bps;
241     s->in_buffer.planar = s->postin.planar = s->midbuf.planar = s->preout.planar =  1;
242
243
244     if(s->rematrix)
245         return swri_rematrix_init(s);
246
247     return 0;
248 }
249
250 static int realloc_audio(AudioData *a, int count){
251     int i, countb;
252     AudioData old;
253
254     if(a->count >= count)
255         return 0;
256
257     count*=2;
258
259     countb= FFALIGN(count*a->bps, 32);
260     old= *a;
261
262     av_assert0(a->planar);
263     av_assert0(a->bps);
264     av_assert0(a->ch_count);
265
266     a->data= av_malloc(countb*a->ch_count);
267     if(!a->data)
268         return AVERROR(ENOMEM);
269     for(i=0; i<a->ch_count; i++){
270         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
271         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
272     }
273     av_free(old.data);
274     a->count= count;
275
276     return 1;
277 }
278
279 static void copy(AudioData *out, AudioData *in,
280                  int count){
281     av_assert0(out->planar == in->planar);
282     av_assert0(out->bps == in->bps);
283     av_assert0(out->ch_count == in->ch_count);
284     if(out->planar){
285         int ch;
286         for(ch=0; ch<out->ch_count; ch++)
287             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
288     }else
289         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
290 }
291
292 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
293     int i;
294     if(out->planar){
295         for(i=0; i<out->ch_count; i++)
296             out->ch[i]= in_arg[i];
297     }else{
298         for(i=0; i<out->ch_count; i++)
299             out->ch[i]= in_arg[0] + i*out->bps;
300     }
301 }
302
303 /**
304  *
305  * out may be equal in.
306  */
307 static void buf_set(AudioData *out, AudioData *in, int count){
308     if(in->planar){
309         int ch;
310         for(ch=0; ch<out->ch_count; ch++)
311             out->ch[ch]= in->ch[ch] + count*out->bps;
312     }else
313         out->ch[0]= in->ch[0] + count*out->ch_count*out->bps;
314 }
315
316 /**
317  *
318  * @return number of samples output per channel
319  */
320 static int resample(SwrContext *s, AudioData *out_param, int out_count,
321                              const AudioData * in_param, int in_count){
322     AudioData in, out, tmp;
323     int ret_sum=0;
324     int border=0;
325
326     tmp=out=*out_param;
327     in =  *in_param;
328
329     do{
330         int ret, size, consumed;
331         if(!s->resample_in_constraint && s->in_buffer_count){
332             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
333             ret= swri_multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
334             out_count -= ret;
335             ret_sum += ret;
336             buf_set(&out, &out, ret);
337             s->in_buffer_count -= consumed;
338             s->in_buffer_index += consumed;
339
340             if(!in_count)
341                 break;
342             if(s->in_buffer_count <= border){
343                 buf_set(&in, &in, -s->in_buffer_count);
344                 in_count += s->in_buffer_count;
345                 s->in_buffer_count=0;
346                 s->in_buffer_index=0;
347                 border = 0;
348             }
349         }
350
351         if(in_count && !s->in_buffer_count){
352             s->in_buffer_index=0;
353             ret= swri_multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
354             out_count -= ret;
355             ret_sum += ret;
356             buf_set(&out, &out, ret);
357             in_count -= consumed;
358             buf_set(&in, &in, consumed);
359         }
360
361         //TODO is this check sane considering the advanced copy avoidance below
362         size= s->in_buffer_index + s->in_buffer_count + in_count;
363         if(   size > s->in_buffer.count
364            && s->in_buffer_count + in_count <= s->in_buffer_index){
365             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
366             copy(&s->in_buffer, &tmp, s->in_buffer_count);
367             s->in_buffer_index=0;
368         }else
369             if((ret=realloc_audio(&s->in_buffer, size)) < 0)
370                 return ret;
371
372         if(in_count){
373             int count= in_count;
374             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
375
376             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
377             copy(&tmp, &in, /*in_*/count);
378             s->in_buffer_count += count;
379             in_count -= count;
380             border += count;
381             buf_set(&in, &in, count);
382             s->resample_in_constraint= 0;
383             if(s->in_buffer_count != count || in_count)
384                 continue;
385         }
386         break;
387     }while(1);
388
389     s->resample_in_constraint= !!out_count;
390
391     return ret_sum;
392 }
393
394 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
395                          const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
396     AudioData *postin, *midbuf, *preout;
397     int ret/*, in_max*/;
398     AudioData * in= &s->in;
399     AudioData *out= &s->out;
400     AudioData preout_tmp, midbuf_tmp;
401
402     if(!s->resample){
403         if(in_count > out_count)
404             return -1;
405         out_count = in_count;
406     }
407
408     if(!in_arg){
409         if(s->in_buffer_count){
410             AudioData *a= &s->in_buffer;
411             int i, j, ret;
412             if((ret=realloc_audio(a, s->in_buffer_index + 2*s->in_buffer_count)) < 0)
413                 return ret;
414             av_assert0(a->planar);
415             for(i=0; i<a->ch_count; i++){
416                 for(j=0; j<s->in_buffer_count; j++){
417                     memcpy(a->ch[i] + (s->in_buffer_index+s->in_buffer_count+j  )*a->bps,
418                            a->ch[i] + (s->in_buffer_index+s->in_buffer_count-j-1)*a->bps, a->bps);
419                 }
420             }
421             s->in_buffer_count += (s->in_buffer_count+1)/2;
422             s->resample_in_constraint = 0;
423         }else{
424             return 0;
425         }
426     }else
427         fill_audiodata(in ,  (void*)in_arg);
428     fill_audiodata(out, out_arg);
429
430     if(s->full_convert){
431         av_assert0(!s->resample);
432         swri_audio_convert(s->full_convert, out, in, in_count);
433         return out_count;
434     }
435
436 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
437 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
438
439     if((ret=realloc_audio(&s->postin, in_count))<0)
440         return ret;
441     if(s->resample_first){
442         av_assert0(s->midbuf.ch_count == s->used_ch_count);
443         if((ret=realloc_audio(&s->midbuf, out_count))<0)
444             return ret;
445     }else{
446         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
447         if((ret=realloc_audio(&s->midbuf,  in_count))<0)
448             return ret;
449     }
450     if((ret=realloc_audio(&s->preout, out_count))<0)
451         return ret;
452
453     postin= &s->postin;
454
455     midbuf_tmp= s->midbuf;
456     midbuf= &midbuf_tmp;
457     preout_tmp= s->preout;
458     preout= &preout_tmp;
459
460     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar)
461         postin= in;
462
463     if(s->resample_first ? !s->resample : !s->rematrix)
464         midbuf= postin;
465
466     if(s->resample_first ? !s->rematrix : !s->resample)
467         preout= midbuf;
468
469     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
470         if(preout==in){
471             out_count= FFMIN(out_count, in_count); //TODO check at teh end if this is needed or redundant
472             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
473             copy(out, in, out_count);
474             return out_count;
475         }
476         else if(preout==postin) preout= midbuf= postin= out;
477         else if(preout==midbuf) preout= midbuf= out;
478         else                    preout= out;
479     }
480
481     if(in != postin){
482         swri_audio_convert(s->in_convert, postin, in, in_count);
483     }
484
485     if(s->resample_first){
486         if(postin != midbuf)
487             out_count= resample(s, midbuf, out_count, postin, in_count);
488         if(midbuf != preout)
489             swri_rematrix(s, preout, midbuf, out_count, preout==out);
490     }else{
491         if(postin != midbuf)
492             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
493         if(midbuf != preout)
494             out_count= resample(s, preout, out_count, midbuf, in_count);
495     }
496
497     if(preout != out){
498 //FIXME packed doesnt need more than 1 chan here!
499         swri_audio_convert(s->out_convert, out, preout, out_count);
500     }
501     if(!in_arg)
502         s->in_buffer_count = 0;
503     return out_count;
504 }
505