swr: resampling: add filter type and Kaiser window beta to AVOptions
[ffmpeg.git] / libswresample / swresample.c
1 /*
2  * Copyright (C) 2011-2012 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 #include <float.h>
28
29 #define  C30DB  M_SQRT2
30 #define  C15DB  1.189207115
31 #define C__0DB  1.0
32 #define C_15DB  0.840896415
33 #define C_30DB  M_SQRT1_2
34 #define C_45DB  0.594603558
35 #define C_60DB  0.5
36
37 #define ALIGN 32
38
39 //TODO split options array out?
40 #define OFFSET(x) offsetof(SwrContext,x)
41 #define PARAM AV_OPT_FLAG_AUDIO_PARAM
42
43 static const AVOption options[]={
44 {"ich"                  ,  "Input Channel Count"        , OFFSET( in.ch_count   ), AV_OPT_TYPE_INT  , {.dbl=2                     }, 0      , SWR_CH_MAX, PARAM},
45 {"in_channel_count"     ,  "Input Channel Count"        , OFFSET( in.ch_count   ), AV_OPT_TYPE_INT  , {.dbl=2                     }, 0      , SWR_CH_MAX, PARAM},
46 {"och"                  , "Output Channel Count"        , OFFSET(out.ch_count   ), AV_OPT_TYPE_INT  , {.dbl=2                     }, 0      , SWR_CH_MAX, PARAM},
47 {"out_channel_count"    , "Output Channel Count"        , OFFSET(out.ch_count   ), AV_OPT_TYPE_INT  , {.dbl=2                     }, 0      , SWR_CH_MAX, PARAM},
48 {"uch"                  ,   "Used Channel Count"        , OFFSET(used_ch_count  ), AV_OPT_TYPE_INT  , {.dbl=0                     }, 0      , SWR_CH_MAX, PARAM},
49 {"used_channel_count"   ,   "Used Channel Count"        , OFFSET(used_ch_count  ), AV_OPT_TYPE_INT  , {.dbl=0                     }, 0      , SWR_CH_MAX, PARAM},
50 {"isr"                  ,  "Input Sample Rate"          , OFFSET( in_sample_rate), AV_OPT_TYPE_INT  , {.dbl=0                     }, 0      , INT_MAX   , PARAM},
51 {"in_sample_rate"       ,  "Input Sample Rate"          , OFFSET( in_sample_rate), AV_OPT_TYPE_INT  , {.dbl=0                     }, 0      , INT_MAX   , PARAM},
52 {"osr"                  , "Output Sample Rate"          , OFFSET(out_sample_rate), AV_OPT_TYPE_INT  , {.dbl=0                     }, 0      , INT_MAX   , PARAM},
53 {"out_sample_rate"      , "Output Sample Rate"          , OFFSET(out_sample_rate), AV_OPT_TYPE_INT  , {.dbl=0                     }, 0      , INT_MAX   , PARAM},
54 {"isf"                  ,    "Input Sample Format"      , OFFSET( in_sample_fmt ), AV_OPT_TYPE_INT  , {.dbl=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_NB-1+256, PARAM},
55 {"in_sample_fmt"        ,    "Input Sample Format"      , OFFSET( in_sample_fmt ), AV_OPT_TYPE_INT  , {.dbl=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_NB-1+256, PARAM},
56 {"osf"                  ,   "Output Sample Format"      , OFFSET(out_sample_fmt ), AV_OPT_TYPE_INT  , {.dbl=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_NB-1+256, PARAM},
57 {"out_sample_fmt"       ,   "Output Sample Format"      , OFFSET(out_sample_fmt ), AV_OPT_TYPE_INT  , {.dbl=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_NB-1+256, PARAM},
58 {"tsf"                  , "Internal Sample Format"      , OFFSET(int_sample_fmt ), AV_OPT_TYPE_INT  , {.dbl=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_FLTP, PARAM},
59 {"internal_sample_fmt"  , "Internal Sample Format"      , OFFSET(int_sample_fmt ), AV_OPT_TYPE_INT  , {.dbl=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_FLTP, PARAM},
60 {"icl"                  ,   "Input Channel Layout"      , OFFSET( in_ch_layout  ), AV_OPT_TYPE_INT64, {.dbl=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
61 {"in_channel_layout"    ,   "Input Channel Layout"      , OFFSET( in_ch_layout  ), AV_OPT_TYPE_INT64, {.dbl=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
62 {"ocl"                  ,  "Output Channel Layout"      , OFFSET(out_ch_layout  ), AV_OPT_TYPE_INT64, {.dbl=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
63 {"out_channel_layout"   ,  "Output Channel Layout"      , OFFSET(out_ch_layout  ), AV_OPT_TYPE_INT64, {.dbl=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
64 {"clev"                 ,    "Center Mix Level"         , OFFSET(clev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
65 {"center_mix_level"     ,    "Center Mix Level"         , OFFSET(clev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
66 {"slev"                 , "Sourround Mix Level"         , OFFSET(slev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
67 {"surround_mix_level"   , "Sourround Mix Level"         , OFFSET(slev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
68 {"lfe_mix_level"        , "LFE Mix Level"               , OFFSET(lfe_mix_level  ), AV_OPT_TYPE_FLOAT, {.dbl=0                     }, -32    , 32        , PARAM},
69 {"rmvol"                , "Rematrix Volume"             , OFFSET(rematrix_volume), AV_OPT_TYPE_FLOAT, {.dbl=1.0                   }, -1000  , 1000      , PARAM},
70 {"rematrix_volume"      , "Rematrix Volume"             , OFFSET(rematrix_volume), AV_OPT_TYPE_FLOAT, {.dbl=1.0                   }, -1000  , 1000      , PARAM},
71 {"flags"                , NULL                          , OFFSET(flags          ), AV_OPT_TYPE_FLAGS, {.dbl=0                     }, 0      , UINT_MAX  , PARAM, "flags"},
72 {"swr_flags"            , NULL                          , OFFSET(flags          ), AV_OPT_TYPE_FLAGS, {.dbl=0                     }, 0      , UINT_MAX  , PARAM, "flags"},
73 {"res"                  , "Force Resampling"            , 0                      , AV_OPT_TYPE_CONST, {.dbl=SWR_FLAG_RESAMPLE     }, INT_MIN, INT_MAX   , PARAM, "flags"},
74 {"dither_scale"         , "Dither Scale"                , OFFSET(dither_scale   ), AV_OPT_TYPE_FLOAT, {.dbl=1                     }, 0      , INT_MAX   , PARAM},
75 {"dither_method"        , "Dither Method"               , OFFSET(dither_method  ), AV_OPT_TYPE_INT  , {.dbl=0                     }, 0      , SWR_DITHER_NB-1, PARAM, "dither_method"},
76 {"rectangular"          , "Rectangular Dither"          , 0                      , AV_OPT_TYPE_CONST, {.dbl=SWR_DITHER_RECTANGULAR}, INT_MIN, INT_MAX   , PARAM, "dither_method"},
77 {"triangular"           ,  "Triangular Dither"          , 0                      , AV_OPT_TYPE_CONST, {.dbl=SWR_DITHER_TRIANGULAR }, INT_MIN, INT_MAX   , PARAM, "dither_method"},
78 {"triangular_hp"        , "Triangular Dither With High Pass" , 0                 , AV_OPT_TYPE_CONST, {.dbl=SWR_DITHER_TRIANGULAR_HIGHPASS }, INT_MIN, INT_MAX, PARAM, "dither_method"},
79 {"filter_size"          , "Resampling Filter Size"      , OFFSET(filter_size)    , AV_OPT_TYPE_INT  , {.dbl=16                    }, 0      , INT_MAX   , PARAM },
80 {"phase_shift"          , "Resampling Phase Shift"      , OFFSET(phase_shift)    , AV_OPT_TYPE_INT  , {.dbl=10                    }, 0      , 30        , PARAM },
81 {"linear_interp"        , "Use Linear Interpolation"    , OFFSET(linear_interp)  , AV_OPT_TYPE_INT  , {.dbl=0                     }, 0      , 1         , PARAM },
82 {"cutoff"               , "Cutoff Frequency Ratio"      , OFFSET(cutoff)         , AV_OPT_TYPE_DOUBLE,{.dbl=0.8                   }, 0      , 1         , PARAM },
83 {"min_comp"             , "Minimum difference between timestamps and audio data (in seconds) below which no timestamp compensation of either kind is applied"
84                                                         , OFFSET(min_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=FLT_MAX               }, 0      , FLT_MAX   , PARAM },
85 {"min_hard_comp"        , "Minimum difference between timestamps and audio data (in seconds) to trigger padding/trimming the data."
86                                                    , OFFSET(min_hard_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0.1                   }, 0      , INT_MAX   , PARAM },
87 {"comp_duration"        , "Duration (in seconds) over which data is stretched/squeezed to make it match the timestamps."
88                                               , OFFSET(soft_compensation_duration),AV_OPT_TYPE_FLOAT ,{.dbl=1                     }, 0      , INT_MAX   , PARAM },
89 {"max_soft_comp"        , "Maximum factor by which data is stretched/squeezed to make it match the timestamps."
90                                                    , OFFSET(max_soft_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0                     }, INT_MIN, INT_MAX   , PARAM },
91 { "filter_type"         , "Filter Type"                 , OFFSET(filter_type)    , AV_OPT_TYPE_INT  , { SWR_FILTER_TYPE_KAISER }, SWR_FILTER_TYPE_CUBIC, SWR_FILTER_TYPE_KAISER, PARAM, "filter_type" },
92     { "cubic"           , "Cubic"                       , 0                      , AV_OPT_TYPE_CONST, { SWR_FILTER_TYPE_CUBIC            }, INT_MIN, INT_MAX, PARAM, "filter_type" },
93     { "blackman_nuttall", "Blackman Nuttall Windowed Sinc", 0                    , AV_OPT_TYPE_CONST, { SWR_FILTER_TYPE_BLACKMAN_NUTTALL }, INT_MIN, INT_MAX, PARAM, "filter_type" },
94     { "kaiser"          , "Kaiser Windowed Sinc"        , 0                      , AV_OPT_TYPE_CONST, { SWR_FILTER_TYPE_KAISER           }, INT_MIN, INT_MAX, PARAM, "filter_type" },
95 { "kaiser_beta"         , "Kaiser Window Beta"          ,OFFSET(kaiser_beta)     , AV_OPT_TYPE_INT  , {.dbl=9                     }, 2      , 16        , PARAM },
96
97 {0}
98 };
99
100 static const char* context_to_name(void* ptr) {
101     return "SWR";
102 }
103
104 static const AVClass av_class = {
105     .class_name                = "SWResampler",
106     .item_name                 = context_to_name,
107     .option                    = options,
108     .version                   = LIBAVUTIL_VERSION_INT,
109     .log_level_offset_offset   = OFFSET(log_level_offset),
110     .parent_log_context_offset = OFFSET(log_ctx),
111     .category                  = AV_CLASS_CATEGORY_SWRESAMPLER,
112 };
113
114 unsigned swresample_version(void)
115 {
116     av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
117     return LIBSWRESAMPLE_VERSION_INT;
118 }
119
120 const char *swresample_configuration(void)
121 {
122     return FFMPEG_CONFIGURATION;
123 }
124
125 const char *swresample_license(void)
126 {
127 #define LICENSE_PREFIX "libswresample license: "
128     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
129 }
130
131 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
132     if(!s || s->in_convert) // s needs to be allocated but not initialized
133         return AVERROR(EINVAL);
134     s->channel_map = channel_map;
135     return 0;
136 }
137
138 const AVClass *swr_get_class(void)
139 {
140     return &av_class;
141 }
142
143 struct SwrContext *swr_alloc(void){
144     SwrContext *s= av_mallocz(sizeof(SwrContext));
145     if(s){
146         s->av_class= &av_class;
147         av_opt_set_defaults(s);
148     }
149     return s;
150 }
151
152 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
153                                       int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
154                                       int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
155                                       int log_offset, void *log_ctx){
156     if(!s) s= swr_alloc();
157     if(!s) return NULL;
158
159     s->log_level_offset= log_offset;
160     s->log_ctx= log_ctx;
161
162     av_opt_set_int(s, "ocl", out_ch_layout,   0);
163     av_opt_set_int(s, "osf", out_sample_fmt,  0);
164     av_opt_set_int(s, "osr", out_sample_rate, 0);
165     av_opt_set_int(s, "icl", in_ch_layout,    0);
166     av_opt_set_int(s, "isf", in_sample_fmt,   0);
167     av_opt_set_int(s, "isr", in_sample_rate,  0);
168     av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_NONE,   0);
169     av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> in_ch_layout), 0);
170     av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->out_ch_layout), 0);
171     av_opt_set_int(s, "uch", 0, 0);
172     return s;
173 }
174
175 static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
176     a->fmt   = fmt;
177     a->bps   = av_get_bytes_per_sample(fmt);
178     a->planar= av_sample_fmt_is_planar(fmt);
179 }
180
181 static void free_temp(AudioData *a){
182     av_free(a->data);
183     memset(a, 0, sizeof(*a));
184 }
185
186 void swr_free(SwrContext **ss){
187     SwrContext *s= *ss;
188     if(s){
189         free_temp(&s->postin);
190         free_temp(&s->midbuf);
191         free_temp(&s->preout);
192         free_temp(&s->in_buffer);
193         free_temp(&s->dither);
194         swri_audio_convert_free(&s-> in_convert);
195         swri_audio_convert_free(&s->out_convert);
196         swri_audio_convert_free(&s->full_convert);
197         swri_resample_free(&s->resample);
198         swri_rematrix_free(s);
199     }
200
201     av_freep(ss);
202 }
203
204 int swr_init(struct SwrContext *s){
205     s->in_buffer_index= 0;
206     s->in_buffer_count= 0;
207     s->resample_in_constraint= 0;
208     free_temp(&s->postin);
209     free_temp(&s->midbuf);
210     free_temp(&s->preout);
211     free_temp(&s->in_buffer);
212     free_temp(&s->dither);
213     swri_audio_convert_free(&s-> in_convert);
214     swri_audio_convert_free(&s->out_convert);
215     swri_audio_convert_free(&s->full_convert);
216     swri_rematrix_free(s);
217
218     s->flushed = 0;
219
220     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
221         av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
222         return AVERROR(EINVAL);
223     }
224     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
225         av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
226         return AVERROR(EINVAL);
227     }
228
229     if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
230         if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
231             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
232         }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
233             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
234         }else{
235             av_log(s, AV_LOG_DEBUG, "Using double precision mode\n");
236             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
237         }
238     }
239
240     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16P
241         &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
242         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
243         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
244         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));
245         return AVERROR(EINVAL);
246     }
247
248     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
249     set_audiodata_fmt(&s->out, s->out_sample_fmt);
250
251     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
252         s->resample = swri_resample_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);
253     }else
254         swri_resample_free(&s->resample);
255     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
256         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
257         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
258         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
259         && s->resample){
260         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
261         return -1;
262     }
263
264     if(!s->used_ch_count)
265         s->used_ch_count= s->in.ch_count;
266
267     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
268         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
269         s-> in_ch_layout= 0;
270     }
271
272     if(!s-> in_ch_layout)
273         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
274     if(!s->out_ch_layout)
275         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
276
277     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
278                  s->rematrix_custom;
279
280 #define RSC 1 //FIXME finetune
281     if(!s-> in.ch_count)
282         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
283     if(!s->used_ch_count)
284         s->used_ch_count= s->in.ch_count;
285     if(!s->out.ch_count)
286         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
287
288     if(!s-> in.ch_count){
289         av_assert0(!s->in_ch_layout);
290         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
291         return -1;
292     }
293
294     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
295         av_log(s, AV_LOG_ERROR, "Rematrix is needed but there is not enough information to do it\n");
296         return -1;
297     }
298
299 av_assert0(s->used_ch_count);
300 av_assert0(s->out.ch_count);
301     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
302
303     s->in_buffer= s->in;
304
305     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither_method){
306         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
307                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
308         return 0;
309     }
310
311     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
312                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
313     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
314                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
315
316
317     s->postin= s->in;
318     s->preout= s->out;
319     s->midbuf= s->in;
320
321     if(s->channel_map){
322         s->postin.ch_count=
323         s->midbuf.ch_count= s->used_ch_count;
324         if(s->resample)
325             s->in_buffer.ch_count= s->used_ch_count;
326     }
327     if(!s->resample_first){
328         s->midbuf.ch_count= s->out.ch_count;
329         if(s->resample)
330             s->in_buffer.ch_count = s->out.ch_count;
331     }
332
333     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
334     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
335     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
336
337     if(s->resample){
338         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
339     }
340
341     s->dither = s->preout;
342
343     if(s->rematrix || s->dither_method)
344         return swri_rematrix_init(s);
345
346     return 0;
347 }
348
349 static int realloc_audio(AudioData *a, int count){
350     int i, countb;
351     AudioData old;
352
353     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
354         return AVERROR(EINVAL);
355
356     if(a->count >= count)
357         return 0;
358
359     count*=2;
360
361     countb= FFALIGN(count*a->bps, ALIGN);
362     old= *a;
363
364     av_assert0(a->bps);
365     av_assert0(a->ch_count);
366
367     a->data= av_mallocz(countb*a->ch_count);
368     if(!a->data)
369         return AVERROR(ENOMEM);
370     for(i=0; i<a->ch_count; i++){
371         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
372         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
373     }
374     if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
375     av_free(old.data);
376     a->count= count;
377
378     return 1;
379 }
380
381 static void copy(AudioData *out, AudioData *in,
382                  int count){
383     av_assert0(out->planar == in->planar);
384     av_assert0(out->bps == in->bps);
385     av_assert0(out->ch_count == in->ch_count);
386     if(out->planar){
387         int ch;
388         for(ch=0; ch<out->ch_count; ch++)
389             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
390     }else
391         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
392 }
393
394 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
395     int i;
396     if(!in_arg){
397         memset(out->ch, 0, sizeof(out->ch));
398     }else if(out->planar){
399         for(i=0; i<out->ch_count; i++)
400             out->ch[i]= in_arg[i];
401     }else{
402         for(i=0; i<out->ch_count; i++)
403             out->ch[i]= in_arg[0] + i*out->bps;
404     }
405 }
406
407 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
408     int i;
409     if(out->planar){
410         for(i=0; i<out->ch_count; i++)
411             in_arg[i]= out->ch[i];
412     }else{
413         in_arg[0]= out->ch[0];
414     }
415 }
416
417 /**
418  *
419  * out may be equal in.
420  */
421 static void buf_set(AudioData *out, AudioData *in, int count){
422     int ch;
423     if(in->planar){
424         for(ch=0; ch<out->ch_count; ch++)
425             out->ch[ch]= in->ch[ch] + count*out->bps;
426     }else{
427         for(ch=out->ch_count-1; ch>=0; ch--)
428             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
429     }
430 }
431
432 /**
433  *
434  * @return number of samples output per channel
435  */
436 static int resample(SwrContext *s, AudioData *out_param, int out_count,
437                              const AudioData * in_param, int in_count){
438     AudioData in, out, tmp;
439     int ret_sum=0;
440     int border=0;
441
442     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
443     av_assert1(s->in_buffer.planar   == in_param->planar);
444     av_assert1(s->in_buffer.fmt      == in_param->fmt);
445
446     tmp=out=*out_param;
447     in =  *in_param;
448
449     do{
450         int ret, size, consumed;
451         if(!s->resample_in_constraint && s->in_buffer_count){
452             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
453             ret= swri_multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
454             out_count -= ret;
455             ret_sum += ret;
456             buf_set(&out, &out, ret);
457             s->in_buffer_count -= consumed;
458             s->in_buffer_index += consumed;
459
460             if(!in_count)
461                 break;
462             if(s->in_buffer_count <= border){
463                 buf_set(&in, &in, -s->in_buffer_count);
464                 in_count += s->in_buffer_count;
465                 s->in_buffer_count=0;
466                 s->in_buffer_index=0;
467                 border = 0;
468             }
469         }
470
471         if(in_count && !s->in_buffer_count){
472             s->in_buffer_index=0;
473             ret= swri_multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
474             out_count -= ret;
475             ret_sum += ret;
476             buf_set(&out, &out, ret);
477             in_count -= consumed;
478             buf_set(&in, &in, consumed);
479         }
480
481         //TODO is this check sane considering the advanced copy avoidance below
482         size= s->in_buffer_index + s->in_buffer_count + in_count;
483         if(   size > s->in_buffer.count
484            && s->in_buffer_count + in_count <= s->in_buffer_index){
485             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
486             copy(&s->in_buffer, &tmp, s->in_buffer_count);
487             s->in_buffer_index=0;
488         }else
489             if((ret=realloc_audio(&s->in_buffer, size)) < 0)
490                 return ret;
491
492         if(in_count){
493             int count= in_count;
494             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
495
496             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
497             copy(&tmp, &in, /*in_*/count);
498             s->in_buffer_count += count;
499             in_count -= count;
500             border += count;
501             buf_set(&in, &in, count);
502             s->resample_in_constraint= 0;
503             if(s->in_buffer_count != count || in_count)
504                 continue;
505         }
506         break;
507     }while(1);
508
509     s->resample_in_constraint= !!out_count;
510
511     return ret_sum;
512 }
513
514 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
515                                                       AudioData *in , int  in_count){
516     AudioData *postin, *midbuf, *preout;
517     int ret/*, in_max*/;
518     AudioData preout_tmp, midbuf_tmp;
519
520     if(s->full_convert){
521         av_assert0(!s->resample);
522         swri_audio_convert(s->full_convert, out, in, in_count);
523         return out_count;
524     }
525
526 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
527 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
528
529     if((ret=realloc_audio(&s->postin, in_count))<0)
530         return ret;
531     if(s->resample_first){
532         av_assert0(s->midbuf.ch_count == s->used_ch_count);
533         if((ret=realloc_audio(&s->midbuf, out_count))<0)
534             return ret;
535     }else{
536         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
537         if((ret=realloc_audio(&s->midbuf,  in_count))<0)
538             return ret;
539     }
540     if((ret=realloc_audio(&s->preout, out_count))<0)
541         return ret;
542
543     postin= &s->postin;
544
545     midbuf_tmp= s->midbuf;
546     midbuf= &midbuf_tmp;
547     preout_tmp= s->preout;
548     preout= &preout_tmp;
549
550     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
551         postin= in;
552
553     if(s->resample_first ? !s->resample : !s->rematrix)
554         midbuf= postin;
555
556     if(s->resample_first ? !s->rematrix : !s->resample)
557         preout= midbuf;
558
559     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
560         if(preout==in){
561             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
562             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
563             copy(out, in, out_count);
564             return out_count;
565         }
566         else if(preout==postin) preout= midbuf= postin= out;
567         else if(preout==midbuf) preout= midbuf= out;
568         else                    preout= out;
569     }
570
571     if(in != postin){
572         swri_audio_convert(s->in_convert, postin, in, in_count);
573     }
574
575     if(s->resample_first){
576         if(postin != midbuf)
577             out_count= resample(s, midbuf, out_count, postin, in_count);
578         if(midbuf != preout)
579             swri_rematrix(s, preout, midbuf, out_count, preout==out);
580     }else{
581         if(postin != midbuf)
582             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
583         if(midbuf != preout)
584             out_count= resample(s, preout, out_count, midbuf, in_count);
585     }
586
587     if(preout != out && out_count){
588         if(s->dither_method){
589             int ch;
590             int dither_count= FFMAX(out_count, 1<<16);
591             av_assert0(preout != in);
592
593             if((ret=realloc_audio(&s->dither, dither_count))<0)
594                 return ret;
595             if(ret)
596                 for(ch=0; ch<s->dither.ch_count; ch++)
597                     swri_get_dither(s, s->dither.ch[ch], s->dither.count, 12345678913579<<ch, s->out_sample_fmt, s->int_sample_fmt);
598             av_assert0(s->dither.ch_count == preout->ch_count);
599
600             if(s->dither_pos + out_count > s->dither.count)
601                 s->dither_pos = 0;
602
603             for(ch=0; ch<preout->ch_count; ch++)
604                 s->mix_2_1_f(preout->ch[ch], preout->ch[ch], s->dither.ch[ch] + s->dither.bps * s->dither_pos, s->native_one, 0, 0, out_count);
605
606             s->dither_pos += out_count;
607         }
608 //FIXME packed doesnt need more than 1 chan here!
609         swri_audio_convert(s->out_convert, out, preout, out_count);
610     }
611     return out_count;
612 }
613
614 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
615                                 const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
616     AudioData * in= &s->in;
617     AudioData *out= &s->out;
618
619     if(s->drop_output > 0){
620         int ret;
621         AudioData tmp = s->out;
622         uint8_t *tmp_arg[SWR_CH_MAX];
623         tmp.count = 0;
624         tmp.data  = NULL;
625         if((ret=realloc_audio(&tmp, s->drop_output))<0)
626             return ret;
627
628         reversefill_audiodata(&tmp, tmp_arg);
629         s->drop_output *= -1; //FIXME find a less hackish solution
630         ret = swr_convert(s, tmp_arg, -s->drop_output, in_arg, in_count); //FIXME optimize but this is as good as never called so maybe it doesnt matter
631         s->drop_output *= -1;
632         if(ret>0)
633             s->drop_output -= ret;
634
635         av_freep(&tmp.data);
636         if(s->drop_output || !out_arg)
637             return 0;
638         in_count = 0;
639     }
640
641     if(!in_arg){
642         if(s->in_buffer_count){
643             if (s->resample && !s->flushed) {
644                 AudioData *a= &s->in_buffer;
645                 int i, j, ret;
646                 if((ret=realloc_audio(a, s->in_buffer_index + 2*s->in_buffer_count)) < 0)
647                     return ret;
648                 av_assert0(a->planar);
649                 for(i=0; i<a->ch_count; i++){
650                     for(j=0; j<s->in_buffer_count; j++){
651                         memcpy(a->ch[i] + (s->in_buffer_index+s->in_buffer_count+j  )*a->bps,
652                             a->ch[i] + (s->in_buffer_index+s->in_buffer_count-j-1)*a->bps, a->bps);
653                     }
654                 }
655                 s->in_buffer_count += (s->in_buffer_count+1)/2;
656                 s->resample_in_constraint = 0;
657                 s->flushed = 1;
658             }
659         }else{
660             return 0;
661         }
662     }else
663         fill_audiodata(in ,  (void*)in_arg);
664
665     fill_audiodata(out, out_arg);
666
667     if(s->resample){
668         int ret = swr_convert_internal(s, out, out_count, in, in_count);
669         if(ret>0 && !s->drop_output)
670             s->outpts += ret * (int64_t)s->in_sample_rate;
671         return ret;
672     }else{
673         AudioData tmp= *in;
674         int ret2=0;
675         int ret, size;
676         size = FFMIN(out_count, s->in_buffer_count);
677         if(size){
678             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
679             ret= swr_convert_internal(s, out, size, &tmp, size);
680             if(ret<0)
681                 return ret;
682             ret2= ret;
683             s->in_buffer_count -= ret;
684             s->in_buffer_index += ret;
685             buf_set(out, out, ret);
686             out_count -= ret;
687             if(!s->in_buffer_count)
688                 s->in_buffer_index = 0;
689         }
690
691         if(in_count){
692             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
693
694             if(in_count > out_count) { //FIXME move after swr_convert_internal
695                 if(   size > s->in_buffer.count
696                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
697                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
698                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
699                     s->in_buffer_index=0;
700                 }else
701                     if((ret=realloc_audio(&s->in_buffer, size)) < 0)
702                         return ret;
703             }
704
705             if(out_count){
706                 size = FFMIN(in_count, out_count);
707                 ret= swr_convert_internal(s, out, size, in, size);
708                 if(ret<0)
709                     return ret;
710                 buf_set(in, in, ret);
711                 in_count -= ret;
712                 ret2 += ret;
713             }
714             if(in_count){
715                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
716                 copy(&tmp, in, in_count);
717                 s->in_buffer_count += in_count;
718             }
719         }
720         if(ret2>0 && !s->drop_output)
721             s->outpts += ret2 * (int64_t)s->in_sample_rate;
722         return ret2;
723     }
724 }
725
726 int swr_drop_output(struct SwrContext *s, int count){
727     s->drop_output += count;
728
729     if(s->drop_output <= 0)
730         return 0;
731
732     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
733     return swr_convert(s, NULL, s->drop_output, NULL, 0);
734 }
735
736 int swr_inject_silence(struct SwrContext *s, int count){
737     int ret, i;
738     AudioData silence = s->out;
739     uint8_t *tmp_arg[SWR_CH_MAX];
740
741     if(count <= 0)
742         return 0;
743
744     silence.count = 0;
745     silence.data  = NULL;
746     if((ret=realloc_audio(&silence, count))<0)
747         return ret;
748
749     if(silence.planar) for(i=0; i<silence.ch_count; i++) {
750         memset(silence.ch[i], silence.bps==1 ? 0x80 : 0, count*silence.bps);
751     } else
752         memset(silence.ch[0], silence.bps==1 ? 0x80 : 0, count*silence.bps*silence.ch_count);
753
754     reversefill_audiodata(&silence, tmp_arg);
755     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
756     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
757     av_freep(&silence.data);
758     return ret;
759 }
760
761 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
762     if(pts == INT64_MIN)
763         return s->outpts;
764     if(s->min_compensation >= FLT_MAX) {
765         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
766     } else {
767         int64_t delta = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate) - s->outpts;
768         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
769
770         if(fabs(fdelta) > s->min_compensation) {
771             if(!s->outpts || fabs(fdelta) > s->min_hard_compensation){
772                 int ret;
773                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
774                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
775                 if(ret<0){
776                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
777                 }
778             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
779                 int duration = s->out_sample_rate * s->soft_compensation_duration;
780                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
781                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
782                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
783                 swr_set_compensation(s, comp, duration);
784             }
785         }
786
787         return s->outpts;
788     }
789 }