git.sesse.net Git - ffmpeg/blob - 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 = {
  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 static int resample(SwrContext *s, AudioData *out_param, int out_count,
  74                              const AudioData * in_param, int in_count);
  75
  76 SwrContext *swr_alloc(void){
  77     SwrContext *s= av_mallocz(sizeof(SwrContext));
  78     if(s){
  79         s->av_class= &av_class;
  80         av_opt_set_defaults(s);
  81     }
  82     return s;
  83 }
  84
  85 SwrContext *swr_alloc2(struct SwrContext *s, int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
  86                        int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
  87                        const int *channel_map, int log_offset, void *log_ctx){
  88     if(!s) s= swr_alloc();
  89     if(!s) return NULL;
  90
  91     s->log_level_offset= log_offset;
  92     s->log_ctx= log_ctx;
  93
  94     av_opt_set_int(s, "ocl", out_ch_layout,   0);
  95     av_opt_set_int(s, "osf", out_sample_fmt,  0);
  96     av_opt_set_int(s, "osr", out_sample_rate, 0);
  97     av_opt_set_int(s, "icl", in_ch_layout,    0);
  98     av_opt_set_int(s, "isf", in_sample_fmt,   0);
  99     av_opt_set_int(s, "isr", in_sample_rate,  0);
 100
 101     s->channel_map = channel_map;
 102     s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
 103     s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
 104     s->int_sample_fmt = AV_SAMPLE_FMT_S16;
 105
 106     return s;
 107 }
 108
 109
 110 static void free_temp(AudioData *a){
 111     av_free(a->data);
 112     memset(a, 0, sizeof(*a));
 113 }
 114
 115 void swr_free(SwrContext **ss){
 116     SwrContext *s= *ss;
 117     if(s){
 118         free_temp(&s->postin);
 119         free_temp(&s->midbuf);
 120         free_temp(&s->preout);
 121         free_temp(&s->in_buffer);
 122         swri_audio_convert_free(&s-> in_convert);
 123         swri_audio_convert_free(&s->out_convert);
 124         swri_audio_convert_free(&s->full_convert);
 125         swr_resample_free(&s->resample);
 126     }
 127
 128     av_freep(ss);
 129 }
 130
 131 int swr_init(SwrContext *s){
 132     s->in_buffer_index= 0;
 133     s->in_buffer_count= 0;
 134     s->resample_in_constraint= 0;
 135     free_temp(&s->postin);
 136     free_temp(&s->midbuf);
 137     free_temp(&s->preout);
 138     free_temp(&s->in_buffer);
 139     swri_audio_convert_free(&s-> in_convert);
 140     swri_audio_convert_free(&s->out_convert);
 141     swri_audio_convert_free(&s->full_convert);
 142
 143     s-> in.planar= s-> in_sample_fmt >= 0x100;
 144     s->out.planar= s->out_sample_fmt >= 0x100;
 145     s-> in_sample_fmt &= 0xFF;
 146     s->out_sample_fmt &= 0xFF;
 147
 148     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
 149         av_log(s, AV_LOG_ERROR, "Requested sample format %s is invalid\n", av_get_sample_fmt_name(s->in_sample_fmt));
 150         return AVERROR(EINVAL);
 151     }
 152     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
 153         av_log(s, AV_LOG_ERROR, "Requested sample format %s is invalid\n", av_get_sample_fmt_name(s->out_sample_fmt));
 154         return AVERROR(EINVAL);
 155     }
 156
 157     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16
 158         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLT){
 159         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));
 160         return AVERROR(EINVAL);
 161     }
 162
 163     //FIXME should we allow/support using FLT on material that doesnt need it ?
 164     if(s->in_sample_fmt <= AV_SAMPLE_FMT_S16 || s->int_sample_fmt==AV_SAMPLE_FMT_S16){
 165         s->int_sample_fmt= AV_SAMPLE_FMT_S16;
 166     }else
 167         s->int_sample_fmt= AV_SAMPLE_FMT_FLT;
 168
 169
 170     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
 171         s->resample = swr_resample_init(s->resample, s->out_sample_rate, s->in_sample_rate, 16, 10, 0, 0.8);
 172     }else
 173         swr_resample_free(&s->resample);
 174     if(s->int_sample_fmt != AV_SAMPLE_FMT_S16 && s->resample){
 175         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16 currently\n"); //FIXME
 176         return -1;
 177     }
 178
 179     if(!s->used_ch_count)
 180         s->used_ch_count= s->in.ch_count;
 181
 182     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
 183         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
 184         s-> in_ch_layout= 0;
 185     }
 186
 187     if(!s-> in_ch_layout)
 188         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
 189     if(!s->out_ch_layout)
 190         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
 191
 192     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0;
 193
 194 #define RSC 1 //FIXME finetune
 195     if(!s-> in.ch_count)
 196         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
 197     if(!s->used_ch_count)
 198         s->used_ch_count= s->in.ch_count;
 199     if(!s->out.ch_count)
 200         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
 201
 202 av_assert0(s-> in.ch_count);
 203 av_assert0(s->used_ch_count);
 204 av_assert0(s->out.ch_count);
 205     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
 206
 207     s-> in.bps= av_get_bytes_per_sample(s-> in_sample_fmt);
 208     s->int_bps= av_get_bytes_per_sample(s->int_sample_fmt);
 209     s->out.bps= av_get_bytes_per_sample(s->out_sample_fmt);
 210
 211     if(!s->resample && !s->rematrix && !s->channel_map){
 212         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
 213                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
 214         return 0;
 215     }
 216
 217     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
 218                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
 219     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
 220                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
 221
 222
 223     s->postin= s->in;
 224     s->preout= s->out;
 225     s->midbuf= s->in;
 226     s->in_buffer= s->in;
 227     if(s->channel_map){
 228         s->postin.ch_count=
 229         s->midbuf.ch_count=
 230         s->in_buffer.ch_count= s->used_ch_count;
 231     }
 232     if(!s->resample_first){
 233         s->midbuf.ch_count= s->out.ch_count;
 234         s->in_buffer.ch_count = s->out.ch_count;
 235     }
 236
 237     s->in_buffer.bps = s->postin.bps = s->midbuf.bps = s->preout.bps =  s->int_bps;
 238     s->in_buffer.planar = s->postin.planar = s->midbuf.planar = s->preout.planar =  1;
 239
 240
 241     if(s->rematrix && swr_rematrix_init(s)<0)
 242         return -1;
 243
 244     return 0;
 245 }
 246
 247 static int realloc_audio(AudioData *a, int count){
 248     int i, countb;
 249     AudioData old;
 250
 251     if(a->count >= count)
 252         return 0;
 253
 254     count*=2;
 255
 256     countb= FFALIGN(count*a->bps, 32);
 257     old= *a;
 258
 259     av_assert0(a->planar);
 260     av_assert0(a->bps);
 261     av_assert0(a->ch_count);
 262
 263     a->data= av_malloc(countb*a->ch_count);
 264     if(!a->data)
 265         return AVERROR(ENOMEM);
 266     for(i=0; i<a->ch_count; i++){
 267         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
 268         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
 269     }
 270     av_free(old.data);
 271     a->count= count;
 272
 273     return 1;
 274 }
 275
 276 static void copy(AudioData *out, AudioData *in,
 277                  int count){
 278     av_assert0(out->planar == in->planar);
 279     av_assert0(out->bps == in->bps);
 280     av_assert0(out->ch_count == in->ch_count);
 281     if(out->planar){
 282         int ch;
 283         for(ch=0; ch<out->ch_count; ch++)
 284             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
 285     }else
 286         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
 287 }
 288
 289 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
 290     int i;
 291     if(out->planar){
 292         for(i=0; i<out->ch_count; i++)
 293             out->ch[i]= in_arg[i];
 294     }else{
 295         for(i=0; i<out->ch_count; i++)
 296             out->ch[i]= in_arg[0] + i*out->bps;
 297     }
 298 }
 299
 300 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
 301                          const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
 302     AudioData *postin, *midbuf, *preout;
 303     int ret/*, in_max*/;
 304     AudioData * in= &s->in;
 305     AudioData *out= &s->out;
 306     AudioData preout_tmp, midbuf_tmp;
 307
 308     if(!s->resample){
 309         if(in_count > out_count)
 310             return -1;
 311         out_count = in_count;
 312     }
 313
 314     if(!in_arg){
 315         if(s->in_buffer_count){
 316             AudioData *a= &s->in_buffer;
 317             int i, j, ret;
 318             if((ret=realloc_audio(a, s->in_buffer_index + 2*s->in_buffer_count)) < 0)
 319                 return ret;
 320             av_assert0(a->planar);
 321             for(i=0; i<a->ch_count; i++){
 322                 for(j=0; j<s->in_buffer_count; j++){
 323                     memcpy(a->ch[i] + (s->in_buffer_index+s->in_buffer_count+j  )*a->bps,
 324                            a->ch[i] + (s->in_buffer_index+s->in_buffer_count-j-1)*a->bps, a->bps);
 325                 }
 326             }
 327             s->in_buffer_count += (s->in_buffer_count+1)/2;
 328             s->resample_in_constraint = 0;
 329         }else{
 330             return 0;
 331         }
 332     }else
 333         fill_audiodata(in ,  (void*)in_arg);
 334     fill_audiodata(out, out_arg);
 335
 336     if(s->full_convert){
 337         av_assert0(!s->resample);
 338         swri_audio_convert(s->full_convert, out, in, in_count);
 339         return out_count;
 340     }
 341
 342 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
 343 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
 344
 345     if((ret=realloc_audio(&s->postin, in_count))<0)
 346         return ret;
 347     if(s->resample_first){
 348         av_assert0(s->midbuf.ch_count == s->used_ch_count);
 349         if((ret=realloc_audio(&s->midbuf, out_count))<0)
 350             return ret;
 351     }else{
 352         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
 353         if((ret=realloc_audio(&s->midbuf,  in_count))<0)
 354             return ret;
 355     }
 356     if((ret=realloc_audio(&s->preout, out_count))<0)
 357         return ret;
 358
 359     postin= &s->postin;
 360
 361     midbuf_tmp= s->midbuf;
 362     midbuf= &midbuf_tmp;
 363     preout_tmp= s->preout;
 364     preout= &preout_tmp;
 365
 366     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar)
 367         postin= in;
 368
 369     if(s->resample_first ? !s->resample : !s->rematrix)
 370         midbuf= postin;
 371
 372     if(s->resample_first ? !s->rematrix : !s->resample)
 373         preout= midbuf;
 374
 375     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
 376         if(preout==in){
 377             out_count= FFMIN(out_count, in_count); //TODO check at teh end if this is needed or redundant
 378             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
 379             copy(out, in, out_count);
 380             return out_count;
 381         }
 382         else if(preout==postin) preout= midbuf= postin= out;
 383         else if(preout==midbuf) preout= midbuf= out;
 384         else                    preout= out;
 385     }
 386
 387     if(in != postin){
 388         swri_audio_convert(s->in_convert, postin, in, in_count);
 389     }
 390
 391     if(s->resample_first){
 392         if(postin != midbuf)
 393             out_count= resample(s, midbuf, out_count, postin, in_count);
 394         if(midbuf != preout)
 395             swr_rematrix(s, preout, midbuf, out_count, preout==out);
 396     }else{
 397         if(postin != midbuf)
 398             swr_rematrix(s, midbuf, postin, in_count, midbuf==out);
 399         if(midbuf != preout)
 400             out_count= resample(s, preout, out_count, midbuf, in_count);
 401     }
 402
 403     if(preout != out){
 404 //FIXME packed doesnt need more than 1 chan here!
 405         swri_audio_convert(s->out_convert, out, preout, out_count);
 406     }
 407     if(!in_arg)
 408         s->in_buffer_count = 0;
 409     return out_count;
 410 }
 411
 412 /**
 413  *
 414  * out may be equal in.
 415  */
 416 static void buf_set(AudioData *out, AudioData *in, int count){
 417     if(in->planar){
 418         int ch;
 419         for(ch=0; ch<out->ch_count; ch++)
 420             out->ch[ch]= in->ch[ch] + count*out->bps;
 421     }else
 422         out->ch[0]= in->ch[0] + count*out->ch_count*out->bps;
 423 }
 424
 425 /**
 426  *
 427  * @return number of samples output per channel
 428  */
 429 static int resample(SwrContext *s, AudioData *out_param, int out_count,
 430                              const AudioData * in_param, int in_count){
 431     AudioData in, out, tmp;
 432     int ret_sum=0;
 433     int border=0;
 434
 435     tmp=out=*out_param;
 436     in =  *in_param;
 437
 438     do{
 439         int ret, size, consumed;
 440         if(!s->resample_in_constraint && s->in_buffer_count){
 441             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 442             ret= swr_multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
 443             out_count -= ret;
 444             ret_sum += ret;
 445             buf_set(&out, &out, ret);
 446             s->in_buffer_count -= consumed;
 447             s->in_buffer_index += consumed;
 448
 449             if(!in_count)
 450                 break;
 451             if(s->in_buffer_count <= border){
 452                 buf_set(&in, &in, -s->in_buffer_count);
 453                 in_count += s->in_buffer_count;
 454                 s->in_buffer_count=0;
 455                 s->in_buffer_index=0;
 456                 border = 0;
 457             }
 458         }
 459
 460         if(in_count && !s->in_buffer_count){
 461             s->in_buffer_index=0;
 462             ret= swr_multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
 463             out_count -= ret;
 464             ret_sum += ret;
 465             buf_set(&out, &out, ret);
 466             in_count -= consumed;
 467             buf_set(&in, &in, consumed);
 468         }
 469
 470         //TODO is this check sane considering the advanced copy avoidance below
 471         size= s->in_buffer_index + s->in_buffer_count + in_count;
 472         if(   size > s->in_buffer.count
 473            && s->in_buffer_count + in_count <= s->in_buffer_index){
 474             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 475             copy(&s->in_buffer, &tmp, s->in_buffer_count);
 476             s->in_buffer_index=0;
 477         }else
 478             if((ret=realloc_audio(&s->in_buffer, size)) < 0)
 479                 return ret;
 480
 481         if(in_count){
 482             int count= in_count;
 483             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
 484
 485             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 486             copy(&tmp, &in, /*in_*/count);
 487             s->in_buffer_count += count;
 488             in_count -= count;
 489             border += count;
 490             buf_set(&in, &in, count);
 491             s->resample_in_constraint= 0;
 492             if(s->in_buffer_count != count || in_count)
 493                 continue;
 494         }
 495         break;
 496     }while(1);
 497
 498     s->resample_in_constraint= !!out_count;
 499
 500     return ret_sum;
 501 }