git.sesse.net Git - ffmpeg/blob - libswresample/swresample.c

   1 /*
   2  * Copyright (C) 2011-2013 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/channel_layout.h"
  26 #include "libavutil/internal.h"
  27
  28 #include <float.h>
  29
  30 #define ALIGN 32
  31
  32 #include "libavutil/ffversion.h"
  33 const char swr_ffversion[] = "FFmpeg version " FFMPEG_VERSION;
  34
  35 unsigned swresample_version(void)
  36 {
  37     av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
  38     return LIBSWRESAMPLE_VERSION_INT;
  39 }
  40
  41 const char *swresample_configuration(void)
  42 {
  43     return FFMPEG_CONFIGURATION;
  44 }
  45
  46 const char *swresample_license(void)
  47 {
  48 #define LICENSE_PREFIX "libswresample license: "
  49     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
  50 }
  51
  52 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
  53     if(!s || s->in_convert) // s needs to be allocated but not initialized
  54         return AVERROR(EINVAL);
  55     s->channel_map = channel_map;
  56     return 0;
  57 }
  58
  59 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
  60                                       int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
  61                                       int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
  62                                       int log_offset, void *log_ctx){
  63     if(!s) s= swr_alloc();
  64     if(!s) return NULL;
  65
  66     s->log_level_offset= log_offset;
  67     s->log_ctx= log_ctx;
  68
  69     if (av_opt_set_int(s, "ocl", out_ch_layout,   0) < 0)
  70         goto fail;
  71
  72     if (av_opt_set_int(s, "osf", out_sample_fmt,  0) < 0)
  73         goto fail;
  74
  75     if (av_opt_set_int(s, "osr", out_sample_rate, 0) < 0)
  76         goto fail;
  77
  78     if (av_opt_set_int(s, "icl", in_ch_layout,    0) < 0)
  79         goto fail;
  80
  81     if (av_opt_set_int(s, "isf", in_sample_fmt,   0) < 0)
  82         goto fail;
  83
  84     if (av_opt_set_int(s, "isr", in_sample_rate,  0) < 0)
  85         goto fail;
  86
  87     if (av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_NONE,   0) < 0)
  88         goto fail;
  89
  90     if (av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> user_in_ch_layout), 0) < 0)
  91         goto fail;
  92
  93     if (av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->user_out_ch_layout), 0) < 0)
  94         goto fail;
  95
  96     av_opt_set_int(s, "uch", 0, 0);
  97     return s;
  98 fail:
  99     av_log(s, AV_LOG_ERROR, "Failed to set option\n");
 100     swr_free(&s);
 101     return NULL;
 102 }
 103
 104 static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
 105     a->fmt   = fmt;
 106     a->bps   = av_get_bytes_per_sample(fmt);
 107     a->planar= av_sample_fmt_is_planar(fmt);
 108     if (a->ch_count == 1)
 109         a->planar = 1;
 110 }
 111
 112 static void free_temp(AudioData *a){
 113     av_free(a->data);
 114     memset(a, 0, sizeof(*a));
 115 }
 116
 117 static void clear_context(SwrContext *s){
 118     s->in_buffer_index= 0;
 119     s->in_buffer_count= 0;
 120     s->resample_in_constraint= 0;
 121     memset(s->in.ch, 0, sizeof(s->in.ch));
 122     memset(s->out.ch, 0, sizeof(s->out.ch));
 123     free_temp(&s->postin);
 124     free_temp(&s->midbuf);
 125     free_temp(&s->preout);
 126     free_temp(&s->in_buffer);
 127     free_temp(&s->silence);
 128     free_temp(&s->drop_temp);
 129     free_temp(&s->dither.noise);
 130     free_temp(&s->dither.temp);
 131     swri_audio_convert_free(&s-> in_convert);
 132     swri_audio_convert_free(&s->out_convert);
 133     swri_audio_convert_free(&s->full_convert);
 134     swri_rematrix_free(s);
 135
 136     s->delayed_samples_fixup = 0;
 137     s->flushed = 0;
 138 }
 139
 140 av_cold void swr_free(SwrContext **ss){
 141     SwrContext *s= *ss;
 142     if(s){
 143         clear_context(s);
 144         if (s->resampler)
 145             s->resampler->free(&s->resample);
 146     }
 147
 148     av_freep(ss);
 149 }
 150
 151 av_cold void swr_close(SwrContext *s){
 152     clear_context(s);
 153 }
 154
 155 av_cold int swr_init(struct SwrContext *s){
 156     int ret;
 157     char l1[1024], l2[1024];
 158
 159     clear_context(s);
 160
 161     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
 162         av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
 163         return AVERROR(EINVAL);
 164     }
 165     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
 166         av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
 167         return AVERROR(EINVAL);
 168     }
 169
 170     s->out.ch_count  = s-> user_out_ch_count;
 171     s-> in.ch_count  = s->  user_in_ch_count;
 172     s->used_ch_count = s->user_used_ch_count;
 173
 174     s-> in_ch_layout = s-> user_in_ch_layout;
 175     s->out_ch_layout = s->user_out_ch_layout;
 176
 177     s->int_sample_fmt= s->user_int_sample_fmt;
 178
 179     if(av_get_channel_layout_nb_channels(s-> in_ch_layout) > SWR_CH_MAX) {
 180         av_log(s, AV_LOG_WARNING, "Input channel layout 0x%"PRIx64" is invalid or unsupported.\n", s-> in_ch_layout);
 181         s->in_ch_layout = 0;
 182     }
 183
 184     if(av_get_channel_layout_nb_channels(s->out_ch_layout) > SWR_CH_MAX) {
 185         av_log(s, AV_LOG_WARNING, "Output channel layout 0x%"PRIx64" is invalid or unsupported.\n", s->out_ch_layout);
 186         s->out_ch_layout = 0;
 187     }
 188
 189     switch(s->engine){
 190 #if CONFIG_LIBSOXR
 191         case SWR_ENGINE_SOXR: s->resampler = &swri_soxr_resampler; break;
 192 #endif
 193         case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
 194         default:
 195             av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
 196             return AVERROR(EINVAL);
 197     }
 198
 199     if(!s->used_ch_count)
 200         s->used_ch_count= s->in.ch_count;
 201
 202     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
 203         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
 204         s-> in_ch_layout= 0;
 205     }
 206
 207     if(!s-> in_ch_layout)
 208         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
 209     if(!s->out_ch_layout)
 210         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
 211
 212     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
 213                  s->rematrix_custom;
 214
 215     if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
 216         if(   av_get_planar_sample_fmt(s-> in_sample_fmt) <= AV_SAMPLE_FMT_S16P
 217            && av_get_planar_sample_fmt(s->out_sample_fmt) <= AV_SAMPLE_FMT_S16P){
 218             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
 219         }else if(   av_get_planar_sample_fmt(s-> in_sample_fmt) <= AV_SAMPLE_FMT_S16P
 220            && !s->rematrix
 221            && s->out_sample_rate==s->in_sample_rate
 222            && !(s->flags & SWR_FLAG_RESAMPLE)){
 223             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
 224         }else if(   av_get_planar_sample_fmt(s-> in_sample_fmt) == AV_SAMPLE_FMT_S32P
 225                  && av_get_planar_sample_fmt(s->out_sample_fmt) == AV_SAMPLE_FMT_S32P
 226                  && !s->rematrix
 227                  && s->engine != SWR_ENGINE_SOXR){
 228             s->int_sample_fmt= AV_SAMPLE_FMT_S32P;
 229         }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
 230             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
 231         }else{
 232             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
 233         }
 234     }
 235     av_log(s, AV_LOG_DEBUG, "Using %s internally between filters\n", av_get_sample_fmt_name(s->int_sample_fmt));
 236
 237     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16P
 238         &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
 239         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
 240         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
 241         av_log(s, AV_LOG_ERROR, "Requested sample format %s is not supported internally, S16/S32/FLT/DBL is supported\n", av_get_sample_fmt_name(s->int_sample_fmt));
 242         return AVERROR(EINVAL);
 243     }
 244
 245     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
 246     set_audiodata_fmt(&s->out, s->out_sample_fmt);
 247
 248     if (s->firstpts_in_samples != AV_NOPTS_VALUE) {
 249         if (!s->async && s->min_compensation >= FLT_MAX/2)
 250             s->async = 1;
 251         s->firstpts =
 252         s->outpts   = s->firstpts_in_samples * s->out_sample_rate;
 253     } else
 254         s->firstpts = AV_NOPTS_VALUE;
 255
 256     if (s->async) {
 257         if (s->min_compensation >= FLT_MAX/2)
 258             s->min_compensation = 0.001;
 259         if (s->async > 1.0001) {
 260             s->max_soft_compensation = s->async / (double) s->in_sample_rate;
 261         }
 262     }
 263
 264     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
 265         s->resample = s->resampler->init(s->resample, s->out_sample_rate, s->in_sample_rate, s->filter_size, s->phase_shift, s->linear_interp, s->cutoff, s->int_sample_fmt, s->filter_type, s->kaiser_beta, s->precision, s->cheby);
 266         if (!s->resample) {
 267             av_log(s, AV_LOG_ERROR, "Failed to initialize resampler\n");
 268             return AVERROR(ENOMEM);
 269         }
 270     }else
 271         s->resampler->free(&s->resample);
 272     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
 273         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
 274         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
 275         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
 276         && s->resample){
 277         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
 278         ret = AVERROR(EINVAL);
 279         goto fail;
 280     }
 281
 282 #define RSC 1 //FIXME finetune
 283     if(!s-> in.ch_count)
 284         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
 285     if(!s->used_ch_count)
 286         s->used_ch_count= s->in.ch_count;
 287     if(!s->out.ch_count)
 288         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
 289
 290     if(!s-> in.ch_count){
 291         av_assert0(!s->in_ch_layout);
 292         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
 293         ret = AVERROR(EINVAL);
 294         goto fail;
 295     }
 296
 297     av_get_channel_layout_string(l1, sizeof(l1), s-> in.ch_count, s-> in_ch_layout);
 298     av_get_channel_layout_string(l2, sizeof(l2), s->out.ch_count, s->out_ch_layout);
 299     if (s->out_ch_layout && s->out.ch_count != av_get_channel_layout_nb_channels(s->out_ch_layout)) {
 300         av_log(s, AV_LOG_ERROR, "Output channel layout %s mismatches specified channel count %d\n", l2, s->out.ch_count);
 301         ret = AVERROR(EINVAL);
 302         goto fail;
 303     }
 304     if (s->in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s->in_ch_layout)) {
 305         av_log(s, AV_LOG_ERROR, "Input channel layout %s mismatches specified channel count %d\n", l1, s->used_ch_count);
 306         ret = AVERROR(EINVAL);
 307         goto fail;
 308     }
 309
 310     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
 311         av_log(s, AV_LOG_ERROR, "Rematrix is needed between %s and %s "
 312                "but there is not enough information to do it\n", l1, l2);
 313         ret = AVERROR(EINVAL);
 314         goto fail;
 315     }
 316
 317 av_assert0(s->used_ch_count);
 318 av_assert0(s->out.ch_count);
 319     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
 320
 321     s->in_buffer= s->in;
 322     s->silence  = s->in;
 323     s->drop_temp= s->out;
 324
 325     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither.method){
 326         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
 327                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
 328         return 0;
 329     }
 330
 331     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
 332                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
 333     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
 334                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
 335
 336     if (!s->in_convert || !s->out_convert) {
 337         ret = AVERROR(ENOMEM);
 338         goto fail;
 339     }
 340
 341     s->postin= s->in;
 342     s->preout= s->out;
 343     s->midbuf= s->in;
 344
 345     if(s->channel_map){
 346         s->postin.ch_count=
 347         s->midbuf.ch_count= s->used_ch_count;
 348         if(s->resample)
 349             s->in_buffer.ch_count= s->used_ch_count;
 350     }
 351     if(!s->resample_first){
 352         s->midbuf.ch_count= s->out.ch_count;
 353         if(s->resample)
 354             s->in_buffer.ch_count = s->out.ch_count;
 355     }
 356
 357     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
 358     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
 359     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
 360
 361     if(s->resample){
 362         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
 363     }
 364
 365     if ((ret = swri_dither_init(s, s->out_sample_fmt, s->int_sample_fmt)) < 0)
 366         goto fail;
 367
 368     if(s->rematrix || s->dither.method) {
 369         ret = swri_rematrix_init(s);
 370         if (ret < 0)
 371             goto fail;
 372     }
 373
 374     return 0;
 375 fail:
 376     swr_close(s);
 377     return ret;
 378
 379 }
 380
 381 int swri_realloc_audio(AudioData *a, int count){
 382     int i, countb;
 383     AudioData old;
 384
 385     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
 386         return AVERROR(EINVAL);
 387
 388     if(a->count >= count)
 389         return 0;
 390
 391     count*=2;
 392
 393     countb= FFALIGN(count*a->bps, ALIGN);
 394     old= *a;
 395
 396     av_assert0(a->bps);
 397     av_assert0(a->ch_count);
 398
 399     a->data= av_mallocz_array(countb, a->ch_count);
 400     if(!a->data)
 401         return AVERROR(ENOMEM);
 402     for(i=0; i<a->ch_count; i++){
 403         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
 404         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
 405     }
 406     if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
 407     av_freep(&old.data);
 408     a->count= count;
 409
 410     return 1;
 411 }
 412
 413 static void copy(AudioData *out, AudioData *in,
 414                  int count){
 415     av_assert0(out->planar == in->planar);
 416     av_assert0(out->bps == in->bps);
 417     av_assert0(out->ch_count == in->ch_count);
 418     if(out->planar){
 419         int ch;
 420         for(ch=0; ch<out->ch_count; ch++)
 421             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
 422     }else
 423         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
 424 }
 425
 426 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
 427     int i;
 428     if(!in_arg){
 429         memset(out->ch, 0, sizeof(out->ch));
 430     }else if(out->planar){
 431         for(i=0; i<out->ch_count; i++)
 432             out->ch[i]= in_arg[i];
 433     }else{
 434         for(i=0; i<out->ch_count; i++)
 435             out->ch[i]= in_arg[0] + i*out->bps;
 436     }
 437 }
 438
 439 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
 440     int i;
 441     if(out->planar){
 442         for(i=0; i<out->ch_count; i++)
 443             in_arg[i]= out->ch[i];
 444     }else{
 445         in_arg[0]= out->ch[0];
 446     }
 447 }
 448
 449 /**
 450  *
 451  * out may be equal in.
 452  */
 453 static void buf_set(AudioData *out, AudioData *in, int count){
 454     int ch;
 455     if(in->planar){
 456         for(ch=0; ch<out->ch_count; ch++)
 457             out->ch[ch]= in->ch[ch] + count*out->bps;
 458     }else{
 459         for(ch=out->ch_count-1; ch>=0; ch--)
 460             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
 461     }
 462 }
 463
 464 /**
 465  *
 466  * @return number of samples output per channel
 467  */
 468 static int resample(SwrContext *s, AudioData *out_param, int out_count,
 469                              const AudioData * in_param, int in_count){
 470     AudioData in, out, tmp;
 471     int ret_sum=0;
 472     int border=0;
 473     int padless = ARCH_X86 && s->engine == SWR_ENGINE_SWR ? 7 : 0;
 474
 475     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
 476     av_assert1(s->in_buffer.planar   == in_param->planar);
 477     av_assert1(s->in_buffer.fmt      == in_param->fmt);
 478
 479     tmp=out=*out_param;
 480     in =  *in_param;
 481
 482     border = s->resampler->invert_initial_buffer(s->resample, &s->in_buffer,
 483                  &in, in_count, &s->in_buffer_index, &s->in_buffer_count);
 484     if (border == INT_MAX) {
 485         return 0;
 486     } else if (border < 0) {
 487         return border;
 488     } else if (border) {
 489         buf_set(&in, &in, border);
 490         in_count -= border;
 491         s->resample_in_constraint = 0;
 492     }
 493
 494     do{
 495         int ret, size, consumed;
 496         if(!s->resample_in_constraint && s->in_buffer_count){
 497             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 498             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
 499             out_count -= ret;
 500             ret_sum += ret;
 501             buf_set(&out, &out, ret);
 502             s->in_buffer_count -= consumed;
 503             s->in_buffer_index += consumed;
 504
 505             if(!in_count)
 506                 break;
 507             if(s->in_buffer_count <= border){
 508                 buf_set(&in, &in, -s->in_buffer_count);
 509                 in_count += s->in_buffer_count;
 510                 s->in_buffer_count=0;
 511                 s->in_buffer_index=0;
 512                 border = 0;
 513             }
 514         }
 515
 516         if((s->flushed || in_count > padless) && !s->in_buffer_count){
 517             s->in_buffer_index=0;
 518             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, FFMAX(in_count-padless, 0), &consumed);
 519             out_count -= ret;
 520             ret_sum += ret;
 521             buf_set(&out, &out, ret);
 522             in_count -= consumed;
 523             buf_set(&in, &in, consumed);
 524         }
 525
 526         //TODO is this check sane considering the advanced copy avoidance below
 527         size= s->in_buffer_index + s->in_buffer_count + in_count;
 528         if(   size > s->in_buffer.count
 529            && s->in_buffer_count + in_count <= s->in_buffer_index){
 530             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 531             copy(&s->in_buffer, &tmp, s->in_buffer_count);
 532             s->in_buffer_index=0;
 533         }else
 534             if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
 535                 return ret;
 536
 537         if(in_count){
 538             int count= in_count;
 539             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
 540
 541             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 542             copy(&tmp, &in, /*in_*/count);
 543             s->in_buffer_count += count;
 544             in_count -= count;
 545             border += count;
 546             buf_set(&in, &in, count);
 547             s->resample_in_constraint= 0;
 548             if(s->in_buffer_count != count || in_count)
 549                 continue;
 550             if (padless) {
 551                 padless = 0;
 552                 continue;
 553             }
 554         }
 555         break;
 556     }while(1);
 557
 558     s->resample_in_constraint= !!out_count;
 559
 560     return ret_sum;
 561 }
 562
 563 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
 564                                                       AudioData *in , int  in_count){
 565     AudioData *postin, *midbuf, *preout;
 566     int ret/*, in_max*/;
 567     AudioData preout_tmp, midbuf_tmp;
 568
 569     if(s->full_convert){
 570         av_assert0(!s->resample);
 571         swri_audio_convert(s->full_convert, out, in, in_count);
 572         return out_count;
 573     }
 574
 575 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
 576 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
 577
 578     if((ret=swri_realloc_audio(&s->postin, in_count))<0)
 579         return ret;
 580     if(s->resample_first){
 581         av_assert0(s->midbuf.ch_count == s->used_ch_count);
 582         if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
 583             return ret;
 584     }else{
 585         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
 586         if((ret=swri_realloc_audio(&s->midbuf,  in_count))<0)
 587             return ret;
 588     }
 589     if((ret=swri_realloc_audio(&s->preout, out_count))<0)
 590         return ret;
 591
 592     postin= &s->postin;
 593
 594     midbuf_tmp= s->midbuf;
 595     midbuf= &midbuf_tmp;
 596     preout_tmp= s->preout;
 597     preout= &preout_tmp;
 598
 599     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
 600         postin= in;
 601
 602     if(s->resample_first ? !s->resample : !s->rematrix)
 603         midbuf= postin;
 604
 605     if(s->resample_first ? !s->rematrix : !s->resample)
 606         preout= midbuf;
 607
 608     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar
 609        && !(s->out_sample_fmt==AV_SAMPLE_FMT_S32P && (s->dither.output_sample_bits&31))){
 610         if(preout==in){
 611             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
 612             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
 613             copy(out, in, out_count);
 614             return out_count;
 615         }
 616         else if(preout==postin) preout= midbuf= postin= out;
 617         else if(preout==midbuf) preout= midbuf= out;
 618         else                    preout= out;
 619     }
 620
 621     if(in != postin){
 622         swri_audio_convert(s->in_convert, postin, in, in_count);
 623     }
 624
 625     if(s->resample_first){
 626         if(postin != midbuf)
 627             out_count= resample(s, midbuf, out_count, postin, in_count);
 628         if(midbuf != preout)
 629             swri_rematrix(s, preout, midbuf, out_count, preout==out);
 630     }else{
 631         if(postin != midbuf)
 632             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
 633         if(midbuf != preout)
 634             out_count= resample(s, preout, out_count, midbuf, in_count);
 635     }
 636
 637     if(preout != out && out_count){
 638         AudioData *conv_src = preout;
 639         if(s->dither.method){
 640             int ch;
 641             int dither_count= FFMAX(out_count, 1<<16);
 642
 643             if (preout == in) {
 644                 conv_src = &s->dither.temp;
 645                 if((ret=swri_realloc_audio(&s->dither.temp, dither_count))<0)
 646                     return ret;
 647             }
 648
 649             if((ret=swri_realloc_audio(&s->dither.noise, dither_count))<0)
 650                 return ret;
 651             if(ret)
 652                 for(ch=0; ch<s->dither.noise.ch_count; ch++)
 653                     if((ret=swri_get_dither(s, s->dither.noise.ch[ch], s->dither.noise.count, (12345678913579ULL*ch + 3141592) % 2718281828U, s->dither.noise.fmt))<0)
 654                         return ret;
 655             av_assert0(s->dither.noise.ch_count == preout->ch_count);
 656
 657             if(s->dither.noise_pos + out_count > s->dither.noise.count)
 658                 s->dither.noise_pos = 0;
 659
 660             if (s->dither.method < SWR_DITHER_NS){
 661                 if (s->mix_2_1_simd) {
 662                     int len1= out_count&~15;
 663                     int off = len1 * preout->bps;
 664
 665                     if(len1)
 666                         for(ch=0; ch<preout->ch_count; ch++)
 667                             s->mix_2_1_simd(conv_src->ch[ch], preout->ch[ch], s->dither.noise.ch[ch] + s->dither.noise.bps * s->dither.noise_pos, s->native_simd_one, 0, 0, len1);
 668                     if(out_count != len1)
 669                         for(ch=0; ch<preout->ch_count; ch++)
 670                             s->mix_2_1_f(conv_src->ch[ch] + off, preout->ch[ch] + off, s->dither.noise.ch[ch] + s->dither.noise.bps * s->dither.noise_pos + off + len1, s->native_one, 0, 0, out_count - len1);
 671                 } else {
 672                     for(ch=0; ch<preout->ch_count; ch++)
 673                         s->mix_2_1_f(conv_src->ch[ch], preout->ch[ch], s->dither.noise.ch[ch] + s->dither.noise.bps * s->dither.noise_pos, s->native_one, 0, 0, out_count);
 674                 }
 675             } else {
 676                 switch(s->int_sample_fmt) {
 677                 case AV_SAMPLE_FMT_S16P :swri_noise_shaping_int16(s, conv_src, preout, &s->dither.noise, out_count); break;
 678                 case AV_SAMPLE_FMT_S32P :swri_noise_shaping_int32(s, conv_src, preout, &s->dither.noise, out_count); break;
 679                 case AV_SAMPLE_FMT_FLTP :swri_noise_shaping_float(s, conv_src, preout, &s->dither.noise, out_count); break;
 680                 case AV_SAMPLE_FMT_DBLP :swri_noise_shaping_double(s,conv_src, preout, &s->dither.noise, out_count); break;
 681                 }
 682             }
 683             s->dither.noise_pos += out_count;
 684         }
 685 //FIXME packed doesn't need more than 1 chan here!
 686         swri_audio_convert(s->out_convert, out, conv_src, out_count);
 687     }
 688     return out_count;
 689 }
 690
 691 int swr_is_initialized(struct SwrContext *s) {
 692     return !!s->in_buffer.ch_count;
 693 }
 694
 695 int attribute_align_arg swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
 696                                                     const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
 697     AudioData * in= &s->in;
 698     AudioData *out= &s->out;
 699     int av_unused max_output;
 700
 701     if (!swr_is_initialized(s)) {
 702         av_log(s, AV_LOG_ERROR, "Context has not been initialized\n");
 703         return AVERROR(EINVAL);
 704     }
 705 #if defined(ASSERT_LEVEL) && ASSERT_LEVEL >1
 706     max_output = swr_get_out_samples(s, in_count);
 707 #endif
 708
 709     while(s->drop_output > 0){
 710         int ret;
 711         uint8_t *tmp_arg[SWR_CH_MAX];
 712 #define MAX_DROP_STEP 16384
 713         if((ret=swri_realloc_audio(&s->drop_temp, FFMIN(s->drop_output, MAX_DROP_STEP)))<0)
 714             return ret;
 715
 716         reversefill_audiodata(&s->drop_temp, tmp_arg);
 717         s->drop_output *= -1; //FIXME find a less hackish solution
 718         ret = swr_convert(s, tmp_arg, FFMIN(-s->drop_output, MAX_DROP_STEP), in_arg, in_count); //FIXME optimize but this is as good as never called so maybe it doesn't matter
 719         s->drop_output *= -1;
 720         in_count = 0;
 721         if(ret>0) {
 722             s->drop_output -= ret;
 723             if (!s->drop_output && !out_arg)
 724                 return 0;
 725             continue;
 726         }
 727
 728         av_assert0(s->drop_output);
 729         return 0;
 730     }
 731
 732     if(!in_arg){
 733         if(s->resample){
 734             if (!s->flushed)
 735                 s->resampler->flush(s);
 736             s->resample_in_constraint = 0;
 737             s->flushed = 1;
 738         }else if(!s->in_buffer_count){
 739             return 0;
 740         }
 741     }else
 742         fill_audiodata(in ,  (void*)in_arg);
 743
 744     fill_audiodata(out, out_arg);
 745
 746     if(s->resample){
 747         int ret = swr_convert_internal(s, out, out_count, in, in_count);
 748         if(ret>0 && !s->drop_output)
 749             s->outpts += ret * (int64_t)s->in_sample_rate;
 750
 751         av_assert2(max_output < 0 || ret < 0 || ret <= max_output);
 752
 753         return ret;
 754     }else{
 755         AudioData tmp= *in;
 756         int ret2=0;
 757         int ret, size;
 758         size = FFMIN(out_count, s->in_buffer_count);
 759         if(size){
 760             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 761             ret= swr_convert_internal(s, out, size, &tmp, size);
 762             if(ret<0)
 763                 return ret;
 764             ret2= ret;
 765             s->in_buffer_count -= ret;
 766             s->in_buffer_index += ret;
 767             buf_set(out, out, ret);
 768             out_count -= ret;
 769             if(!s->in_buffer_count)
 770                 s->in_buffer_index = 0;
 771         }
 772
 773         if(in_count){
 774             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
 775
 776             if(in_count > out_count) { //FIXME move after swr_convert_internal
 777                 if(   size > s->in_buffer.count
 778                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
 779                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 780                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
 781                     s->in_buffer_index=0;
 782                 }else
 783                     if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
 784                         return ret;
 785             }
 786
 787             if(out_count){
 788                 size = FFMIN(in_count, out_count);
 789                 ret= swr_convert_internal(s, out, size, in, size);
 790                 if(ret<0)
 791                     return ret;
 792                 buf_set(in, in, ret);
 793                 in_count -= ret;
 794                 ret2 += ret;
 795             }
 796             if(in_count){
 797                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 798                 copy(&tmp, in, in_count);
 799                 s->in_buffer_count += in_count;
 800             }
 801         }
 802         if(ret2>0 && !s->drop_output)
 803             s->outpts += ret2 * (int64_t)s->in_sample_rate;
 804         av_assert2(max_output < 0 || ret2 < 0 || ret2 <= max_output);
 805         return ret2;
 806     }
 807 }
 808
 809 int swr_drop_output(struct SwrContext *s, int count){
 810     const uint8_t *tmp_arg[SWR_CH_MAX];
 811     s->drop_output += count;
 812
 813     if(s->drop_output <= 0)
 814         return 0;
 815
 816     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
 817     return swr_convert(s, NULL, s->drop_output, tmp_arg, 0);
 818 }
 819
 820 int swr_inject_silence(struct SwrContext *s, int count){
 821     int ret, i;
 822     uint8_t *tmp_arg[SWR_CH_MAX];
 823
 824     if(count <= 0)
 825         return 0;
 826
 827 #define MAX_SILENCE_STEP 16384
 828     while (count > MAX_SILENCE_STEP) {
 829         if ((ret = swr_inject_silence(s, MAX_SILENCE_STEP)) < 0)
 830             return ret;
 831         count -= MAX_SILENCE_STEP;
 832     }
 833
 834     if((ret=swri_realloc_audio(&s->silence, count))<0)
 835         return ret;
 836
 837     if(s->silence.planar) for(i=0; i<s->silence.ch_count; i++) {
 838         memset(s->silence.ch[i], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps);
 839     } else
 840         memset(s->silence.ch[0], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps*s->silence.ch_count);
 841
 842     reversefill_audiodata(&s->silence, tmp_arg);
 843     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
 844     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
 845     return ret;
 846 }
 847
 848 int64_t swr_get_delay(struct SwrContext *s, int64_t base){
 849     if (s->resampler && s->resample){
 850         return s->resampler->get_delay(s, base);
 851     }else{
 852         return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
 853     }
 854 }
 855
 856 int swr_get_out_samples(struct SwrContext *s, int in_samples)
 857 {
 858     int64_t out_samples;
 859
 860     if (in_samples < 0)
 861         return AVERROR(EINVAL);
 862
 863     if (s->resampler && s->resample) {
 864         if (!s->resampler->get_out_samples)
 865             return AVERROR(ENOSYS);
 866         out_samples = s->resampler->get_out_samples(s, in_samples);
 867     } else {
 868         out_samples = s->in_buffer_count + in_samples;
 869         av_assert0(s->out_sample_rate == s->in_sample_rate);
 870     }
 871
 872     if (out_samples > INT_MAX)
 873         return AVERROR(EINVAL);
 874
 875     return out_samples;
 876 }
 877
 878 int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
 879     int ret;
 880
 881     if (!s || compensation_distance < 0)
 882         return AVERROR(EINVAL);
 883     if (!compensation_distance && sample_delta)
 884         return AVERROR(EINVAL);
 885     if (!s->resample) {
 886         s->flags |= SWR_FLAG_RESAMPLE;
 887         ret = swr_init(s);
 888         if (ret < 0)
 889             return ret;
 890     }
 891     if (!s->resampler->set_compensation){
 892         return AVERROR(EINVAL);
 893     }else{
 894         return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
 895     }
 896 }
 897
 898 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
 899     if(pts == INT64_MIN)
 900         return s->outpts;
 901
 902     if (s->firstpts == AV_NOPTS_VALUE)
 903         s->outpts = s->firstpts = pts;
 904
 905     if(s->min_compensation >= FLT_MAX) {
 906         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
 907     } else {
 908         int64_t delta = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate) - s->outpts + s->drop_output*(int64_t)s->in_sample_rate;
 909         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
 910
 911         if(fabs(fdelta) > s->min_compensation) {
 912             if(s->outpts == s->firstpts || fabs(fdelta) > s->min_hard_compensation){
 913                 int ret;
 914                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
 915                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
 916                 if(ret<0){
 917                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
 918                 }
 919             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
 920                 int duration = s->out_sample_rate * s->soft_compensation_duration;
 921                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
 922                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
 923                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
 924                 swr_set_compensation(s, comp, duration);
 925             }
 926         }
 927
 928         return s->outpts;
 929     }
 930 }