docs: add soxr documentation
[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/channel_layout.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"                  , "set input channel count"     , OFFSET( in.ch_count   ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
45 {"in_channel_count"     , "set input channel count"     , OFFSET( in.ch_count   ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
46 {"och"                  , "set output channel count"    , OFFSET(out.ch_count   ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
47 {"out_channel_count"    , "set output channel count"    , OFFSET(out.ch_count   ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
48 {"uch"                  , "set used channel count"      , OFFSET(used_ch_count  ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
49 {"used_channel_count"   , "set used channel count"      , OFFSET(used_ch_count  ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
50 {"isr"                  , "set input sample rate"       , OFFSET( in_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
51 {"in_sample_rate"       , "set input sample rate"       , OFFSET( in_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
52 {"osr"                  , "set output sample rate"      , OFFSET(out_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
53 {"out_sample_rate"      , "set output sample rate"      , OFFSET(out_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
54 {"isf"                  , "set input sample format"     , OFFSET( in_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , AV_SAMPLE_FMT_NB-1, PARAM},
55 {"in_sample_fmt"        , "set input sample format"     , OFFSET( in_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , AV_SAMPLE_FMT_NB-1, PARAM},
56 {"osf"                  , "set output sample format"    , OFFSET(out_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , AV_SAMPLE_FMT_NB-1, PARAM},
57 {"out_sample_fmt"       , "set output sample format"    , OFFSET(out_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , AV_SAMPLE_FMT_NB-1, PARAM},
58 {"tsf"                  , "set internal sample format"  , OFFSET(int_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , AV_SAMPLE_FMT_NB-1, PARAM},
59 {"internal_sample_fmt"  , "set internal sample format"  , OFFSET(int_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , AV_SAMPLE_FMT_NB-1, PARAM},
60 {"icl"                  , "set input channel layout"    , OFFSET( in_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
61 {"in_channel_layout"    , "set input channel layout"    , OFFSET( in_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
62 {"ocl"                  , "set output channel layout"   , OFFSET(out_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
63 {"out_channel_layout"   , "set output channel layout"   , OFFSET(out_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
64 {"clev"                 , "set center mix level"        , OFFSET(clev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
65 {"center_mix_level"     , "set center mix level"        , OFFSET(clev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
66 {"slev"                 , "set surround mix level"      , OFFSET(slev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
67 {"surround_mix_level"   , "set surround mix Level"      , OFFSET(slev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
68 {"lfe_mix_level"        , "set LFE mix level"           , OFFSET(lfe_mix_level  ), AV_OPT_TYPE_FLOAT, {.dbl=0                     }, -32    , 32        , PARAM},
69 {"rmvol"                , "set rematrix volume"         , OFFSET(rematrix_volume), AV_OPT_TYPE_FLOAT, {.dbl=1.0                   }, -1000  , 1000      , PARAM},
70 {"rematrix_volume"      , "set rematrix volume"         , OFFSET(rematrix_volume), AV_OPT_TYPE_FLOAT, {.dbl=1.0                   }, -1000  , 1000      , PARAM},
71
72 {"flags"                , "set flags"                   , OFFSET(flags          ), AV_OPT_TYPE_FLAGS, {.i64=0                     }, 0      , UINT_MAX  , PARAM, "flags"},
73 {"swr_flags"            , "set flags"                   , OFFSET(flags          ), AV_OPT_TYPE_FLAGS, {.i64=0                     }, 0      , UINT_MAX  , PARAM, "flags"},
74 {"res"                  , "force resampling"            , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_FLAG_RESAMPLE     }, INT_MIN, INT_MAX   , PARAM, "flags"},
75
76 {"dither_scale"         , "set dither scale"            , OFFSET(dither_scale   ), AV_OPT_TYPE_FLOAT, {.dbl=1                     }, 0      , INT_MAX   , PARAM},
77
78 {"dither_method"        , "set dither method"           , OFFSET(dither_method  ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_DITHER_NB-1, PARAM, "dither_method"},
79 {"rectangular"          , "select rectangular dither"   , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_RECTANGULAR}, INT_MIN, INT_MAX   , PARAM, "dither_method"},
80 {"triangular"           , "select triangular dither"    , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_TRIANGULAR }, INT_MIN, INT_MAX   , PARAM, "dither_method"},
81 {"triangular_hp"        , "select triangular dither with high pass" , 0                 , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_TRIANGULAR_HIGHPASS }, INT_MIN, INT_MAX, PARAM, "dither_method"},
82
83 {"filter_size"          , "set swr resampling filter size", OFFSET(filter_size)  , AV_OPT_TYPE_INT  , {.i64=16                    }, 0      , INT_MAX   , PARAM },
84 {"phase_shift"          , "set swr resampling phase shift", OFFSET(phase_shift)  , AV_OPT_TYPE_INT  , {.i64=10                    }, 0      , 30        , PARAM },
85 {"linear_interp"        , "enable linear interpolation" , OFFSET(linear_interp)  , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , 1         , PARAM },
86 {"cutoff"               , "set cutoff frequency ratio"  , OFFSET(cutoff)         , AV_OPT_TYPE_DOUBLE,{.dbl=0.                    }, 0      , 1         , PARAM },
87 {"resampler"            , "set resampling Engine"       , OFFSET(engine)         , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_ENGINE_NB-1, PARAM, "resampler"},
88 {"swr"                  , "select SW Resampler"         , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_ENGINE_SWR        }, INT_MIN, INT_MAX   , PARAM, "resampler"},
89 {"soxr"                 , "select SoX Resampler"        , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_ENGINE_SOXR       }, INT_MIN, INT_MAX   , PARAM, "resampler"},
90 {"precision"            , "set soxr resampling precision (in bits)"
91                                                         , OFFSET(precision)      , AV_OPT_TYPE_DOUBLE,{.dbl=20.0                  }, 15.0   , 33.0      , PARAM },
92 {"cheby"                , "enable soxr Chebyshev passband & higher-precision irrational ratio approximation"
93                                                         , OFFSET(cheby)          , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , 1         , PARAM },
94 {"min_comp"             , "set minimum difference between timestamps and audio data (in seconds) below which no timestamp compensation of either kind is applied"
95                                                         , OFFSET(min_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=FLT_MAX               }, 0      , FLT_MAX   , PARAM },
96 {"min_hard_comp"        , "set minimum difference between timestamps and audio data (in seconds) to trigger padding/trimming the data."
97                                                         , OFFSET(min_hard_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0.1                   }, 0      , INT_MAX   , PARAM },
98 {"comp_duration"        , "set duration (in seconds) over which data is stretched/squeezed to make it match the timestamps."
99                                                         , OFFSET(soft_compensation_duration),AV_OPT_TYPE_FLOAT ,{.dbl=1                     }, 0      , INT_MAX   , PARAM },
100 {"max_soft_comp"        , "set maximum factor by which data is stretched/squeezed to make it match the timestamps."
101                                                         , OFFSET(max_soft_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0                     }, INT_MIN, INT_MAX   , PARAM },
102 {"async"                , "simplified 1 parameter audio timestamp matching, 0(disabled), 1(filling and trimming), >1(maximum stretch/squeeze in samples per second)"
103                                                         , OFFSET(async)          , AV_OPT_TYPE_FLOAT ,{.dbl=0                     }, INT_MIN, INT_MAX   , PARAM },
104
105 { "matrix_encoding"     , "set matrixed stereo encoding" , OFFSET(matrix_encoding), AV_OPT_TYPE_INT   ,{.i64 = AV_MATRIX_ENCODING_NONE}, AV_MATRIX_ENCODING_NONE,     AV_MATRIX_ENCODING_NB-1, PARAM, "matrix_encoding" },
106     { "none",  "select none",               0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_NONE  }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
107     { "dolby", "select Dolby",              0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DOLBY }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
108     { "dplii", "select Dolby Pro Logic II", 0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DPLII }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
109
110 { "filter_type"         , "select swr filter type"      , OFFSET(filter_type)    , AV_OPT_TYPE_INT  , { .i64 = SWR_FILTER_TYPE_KAISER }, SWR_FILTER_TYPE_CUBIC, SWR_FILTER_TYPE_KAISER, PARAM, "filter_type" },
111     { "cubic"           , "select cubic"                , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_CUBIC            }, INT_MIN, INT_MAX, PARAM, "filter_type" },
112     { "blackman_nuttall", "select Blackman Nuttall Windowed Sinc", 0             , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_BLACKMAN_NUTTALL }, INT_MIN, INT_MAX, PARAM, "filter_type" },
113     { "kaiser"          , "select Kaiser Windowed Sinc" , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_KAISER           }, INT_MIN, INT_MAX, PARAM, "filter_type" },
114
115 { "kaiser_beta"         , "set swr Kaiser Window Beta"  , OFFSET(kaiser_beta)    , AV_OPT_TYPE_INT  , {.i64=9                     }, 2      , 16        , PARAM },
116
117 {0}
118 };
119
120 static const char* context_to_name(void* ptr) {
121     return "SWR";
122 }
123
124 static const AVClass av_class = {
125     .class_name                = "SWResampler",
126     .item_name                 = context_to_name,
127     .option                    = options,
128     .version                   = LIBAVUTIL_VERSION_INT,
129     .log_level_offset_offset   = OFFSET(log_level_offset),
130     .parent_log_context_offset = OFFSET(log_ctx),
131     .category                  = AV_CLASS_CATEGORY_SWRESAMPLER,
132 };
133
134 unsigned swresample_version(void)
135 {
136     av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
137     return LIBSWRESAMPLE_VERSION_INT;
138 }
139
140 const char *swresample_configuration(void)
141 {
142     return FFMPEG_CONFIGURATION;
143 }
144
145 const char *swresample_license(void)
146 {
147 #define LICENSE_PREFIX "libswresample license: "
148     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
149 }
150
151 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
152     if(!s || s->in_convert) // s needs to be allocated but not initialized
153         return AVERROR(EINVAL);
154     s->channel_map = channel_map;
155     return 0;
156 }
157
158 const AVClass *swr_get_class(void)
159 {
160     return &av_class;
161 }
162
163 av_cold struct SwrContext *swr_alloc(void){
164     SwrContext *s= av_mallocz(sizeof(SwrContext));
165     if(s){
166         s->av_class= &av_class;
167         av_opt_set_defaults(s);
168     }
169     return s;
170 }
171
172 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
173                                       int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
174                                       int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
175                                       int log_offset, void *log_ctx){
176     if(!s) s= swr_alloc();
177     if(!s) return NULL;
178
179     s->log_level_offset= log_offset;
180     s->log_ctx= log_ctx;
181
182     av_opt_set_int(s, "ocl", out_ch_layout,   0);
183     av_opt_set_int(s, "osf", out_sample_fmt,  0);
184     av_opt_set_int(s, "osr", out_sample_rate, 0);
185     av_opt_set_int(s, "icl", in_ch_layout,    0);
186     av_opt_set_int(s, "isf", in_sample_fmt,   0);
187     av_opt_set_int(s, "isr", in_sample_rate,  0);
188     av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_NONE,   0);
189     av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> in_ch_layout), 0);
190     av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->out_ch_layout), 0);
191     av_opt_set_int(s, "uch", 0, 0);
192     return s;
193 }
194
195 static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
196     a->fmt   = fmt;
197     a->bps   = av_get_bytes_per_sample(fmt);
198     a->planar= av_sample_fmt_is_planar(fmt);
199 }
200
201 static void free_temp(AudioData *a){
202     av_free(a->data);
203     memset(a, 0, sizeof(*a));
204 }
205
206 av_cold void swr_free(SwrContext **ss){
207     SwrContext *s= *ss;
208     if(s){
209         free_temp(&s->postin);
210         free_temp(&s->midbuf);
211         free_temp(&s->preout);
212         free_temp(&s->in_buffer);
213         free_temp(&s->dither);
214         swri_audio_convert_free(&s-> in_convert);
215         swri_audio_convert_free(&s->out_convert);
216         swri_audio_convert_free(&s->full_convert);
217         if (s->resampler)
218             s->resampler->free(&s->resample);
219         swri_rematrix_free(s);
220     }
221
222     av_freep(ss);
223 }
224
225 av_cold int swr_init(struct SwrContext *s){
226     s->in_buffer_index= 0;
227     s->in_buffer_count= 0;
228     s->resample_in_constraint= 0;
229     free_temp(&s->postin);
230     free_temp(&s->midbuf);
231     free_temp(&s->preout);
232     free_temp(&s->in_buffer);
233     free_temp(&s->dither);
234     memset(s->in.ch, 0, sizeof(s->in.ch));
235     memset(s->out.ch, 0, sizeof(s->out.ch));
236     swri_audio_convert_free(&s-> in_convert);
237     swri_audio_convert_free(&s->out_convert);
238     swri_audio_convert_free(&s->full_convert);
239     swri_rematrix_free(s);
240
241     s->flushed = 0;
242
243     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
244         av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
245         return AVERROR(EINVAL);
246     }
247     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
248         av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
249         return AVERROR(EINVAL);
250     }
251
252     if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
253         if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
254             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
255         }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
256             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
257         }else{
258             av_log(s, AV_LOG_DEBUG, "Using double precision mode\n");
259             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
260         }
261     }
262
263     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16P
264         &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
265         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
266         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
267         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));
268         return AVERROR(EINVAL);
269     }
270
271     switch(s->engine){
272 #if CONFIG_LIBSOXR
273         extern struct Resampler const soxr_resampler;
274         case SWR_ENGINE_SOXR: s->resampler = &soxr_resampler; break;
275 #endif
276         case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
277         default:
278             av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
279             return AVERROR(EINVAL);
280     }
281
282     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
283     set_audiodata_fmt(&s->out, s->out_sample_fmt);
284
285     if (s->async) {
286         if (s->min_compensation >= FLT_MAX/2)
287             s->min_compensation = 0.001;
288         if (s->async > 1.0001) {
289             s->max_soft_compensation = s->async / (double) s->in_sample_rate;
290         }
291     }
292
293     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
294         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);
295     }else
296         s->resampler->free(&s->resample);
297     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
298         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
299         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
300         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
301         && s->resample){
302         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
303         return -1;
304     }
305
306     if(!s->used_ch_count)
307         s->used_ch_count= s->in.ch_count;
308
309     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
310         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
311         s-> in_ch_layout= 0;
312     }
313
314     if(!s-> in_ch_layout)
315         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
316     if(!s->out_ch_layout)
317         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
318
319     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
320                  s->rematrix_custom;
321
322 #define RSC 1 //FIXME finetune
323     if(!s-> in.ch_count)
324         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
325     if(!s->used_ch_count)
326         s->used_ch_count= s->in.ch_count;
327     if(!s->out.ch_count)
328         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
329
330     if(!s-> in.ch_count){
331         av_assert0(!s->in_ch_layout);
332         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
333         return -1;
334     }
335
336     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
337         av_log(s, AV_LOG_ERROR, "Rematrix is needed but there is not enough information to do it\n");
338         return -1;
339     }
340
341 av_assert0(s->used_ch_count);
342 av_assert0(s->out.ch_count);
343     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
344
345     s->in_buffer= s->in;
346
347     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither_method){
348         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
349                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
350         return 0;
351     }
352
353     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
354                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
355     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
356                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
357
358
359     s->postin= s->in;
360     s->preout= s->out;
361     s->midbuf= s->in;
362
363     if(s->channel_map){
364         s->postin.ch_count=
365         s->midbuf.ch_count= s->used_ch_count;
366         if(s->resample)
367             s->in_buffer.ch_count= s->used_ch_count;
368     }
369     if(!s->resample_first){
370         s->midbuf.ch_count= s->out.ch_count;
371         if(s->resample)
372             s->in_buffer.ch_count = s->out.ch_count;
373     }
374
375     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
376     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
377     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
378
379     if(s->resample){
380         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
381     }
382
383     s->dither = s->preout;
384
385     if(s->rematrix || s->dither_method)
386         return swri_rematrix_init(s);
387
388     return 0;
389 }
390
391 int swri_realloc_audio(AudioData *a, int count){
392     int i, countb;
393     AudioData old;
394
395     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
396         return AVERROR(EINVAL);
397
398     if(a->count >= count)
399         return 0;
400
401     count*=2;
402
403     countb= FFALIGN(count*a->bps, ALIGN);
404     old= *a;
405
406     av_assert0(a->bps);
407     av_assert0(a->ch_count);
408
409     a->data= av_mallocz(countb*a->ch_count);
410     if(!a->data)
411         return AVERROR(ENOMEM);
412     for(i=0; i<a->ch_count; i++){
413         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
414         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
415     }
416     if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
417     av_free(old.data);
418     a->count= count;
419
420     return 1;
421 }
422
423 static void copy(AudioData *out, AudioData *in,
424                  int count){
425     av_assert0(out->planar == in->planar);
426     av_assert0(out->bps == in->bps);
427     av_assert0(out->ch_count == in->ch_count);
428     if(out->planar){
429         int ch;
430         for(ch=0; ch<out->ch_count; ch++)
431             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
432     }else
433         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
434 }
435
436 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
437     int i;
438     if(!in_arg){
439         memset(out->ch, 0, sizeof(out->ch));
440     }else if(out->planar){
441         for(i=0; i<out->ch_count; i++)
442             out->ch[i]= in_arg[i];
443     }else{
444         for(i=0; i<out->ch_count; i++)
445             out->ch[i]= in_arg[0] + i*out->bps;
446     }
447 }
448
449 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
450     int i;
451     if(out->planar){
452         for(i=0; i<out->ch_count; i++)
453             in_arg[i]= out->ch[i];
454     }else{
455         in_arg[0]= out->ch[0];
456     }
457 }
458
459 /**
460  *
461  * out may be equal in.
462  */
463 static void buf_set(AudioData *out, AudioData *in, int count){
464     int ch;
465     if(in->planar){
466         for(ch=0; ch<out->ch_count; ch++)
467             out->ch[ch]= in->ch[ch] + count*out->bps;
468     }else{
469         for(ch=out->ch_count-1; ch>=0; ch--)
470             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
471     }
472 }
473
474 /**
475  *
476  * @return number of samples output per channel
477  */
478 static int resample(SwrContext *s, AudioData *out_param, int out_count,
479                              const AudioData * in_param, int in_count){
480     AudioData in, out, tmp;
481     int ret_sum=0;
482     int border=0;
483
484     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
485     av_assert1(s->in_buffer.planar   == in_param->planar);
486     av_assert1(s->in_buffer.fmt      == in_param->fmt);
487
488     tmp=out=*out_param;
489     in =  *in_param;
490
491     do{
492         int ret, size, consumed;
493         if(!s->resample_in_constraint && s->in_buffer_count){
494             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
495             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
496             out_count -= ret;
497             ret_sum += ret;
498             buf_set(&out, &out, ret);
499             s->in_buffer_count -= consumed;
500             s->in_buffer_index += consumed;
501
502             if(!in_count)
503                 break;
504             if(s->in_buffer_count <= border){
505                 buf_set(&in, &in, -s->in_buffer_count);
506                 in_count += s->in_buffer_count;
507                 s->in_buffer_count=0;
508                 s->in_buffer_index=0;
509                 border = 0;
510             }
511         }
512
513         if((s->flushed || in_count) && !s->in_buffer_count){
514             s->in_buffer_index=0;
515             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
516             out_count -= ret;
517             ret_sum += ret;
518             buf_set(&out, &out, ret);
519             in_count -= consumed;
520             buf_set(&in, &in, consumed);
521         }
522
523         //TODO is this check sane considering the advanced copy avoidance below
524         size= s->in_buffer_index + s->in_buffer_count + in_count;
525         if(   size > s->in_buffer.count
526            && s->in_buffer_count + in_count <= s->in_buffer_index){
527             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
528             copy(&s->in_buffer, &tmp, s->in_buffer_count);
529             s->in_buffer_index=0;
530         }else
531             if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
532                 return ret;
533
534         if(in_count){
535             int count= in_count;
536             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
537
538             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
539             copy(&tmp, &in, /*in_*/count);
540             s->in_buffer_count += count;
541             in_count -= count;
542             border += count;
543             buf_set(&in, &in, count);
544             s->resample_in_constraint= 0;
545             if(s->in_buffer_count != count || in_count)
546                 continue;
547         }
548         break;
549     }while(1);
550
551     s->resample_in_constraint= !!out_count;
552
553     return ret_sum;
554 }
555
556 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
557                                                       AudioData *in , int  in_count){
558     AudioData *postin, *midbuf, *preout;
559     int ret/*, in_max*/;
560     AudioData preout_tmp, midbuf_tmp;
561
562     if(s->full_convert){
563         av_assert0(!s->resample);
564         swri_audio_convert(s->full_convert, out, in, in_count);
565         return out_count;
566     }
567
568 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
569 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
570
571     if((ret=swri_realloc_audio(&s->postin, in_count))<0)
572         return ret;
573     if(s->resample_first){
574         av_assert0(s->midbuf.ch_count == s->used_ch_count);
575         if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
576             return ret;
577     }else{
578         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
579         if((ret=swri_realloc_audio(&s->midbuf,  in_count))<0)
580             return ret;
581     }
582     if((ret=swri_realloc_audio(&s->preout, out_count))<0)
583         return ret;
584
585     postin= &s->postin;
586
587     midbuf_tmp= s->midbuf;
588     midbuf= &midbuf_tmp;
589     preout_tmp= s->preout;
590     preout= &preout_tmp;
591
592     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
593         postin= in;
594
595     if(s->resample_first ? !s->resample : !s->rematrix)
596         midbuf= postin;
597
598     if(s->resample_first ? !s->rematrix : !s->resample)
599         preout= midbuf;
600
601     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
602         if(preout==in){
603             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
604             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
605             copy(out, in, out_count);
606             return out_count;
607         }
608         else if(preout==postin) preout= midbuf= postin= out;
609         else if(preout==midbuf) preout= midbuf= out;
610         else                    preout= out;
611     }
612
613     if(in != postin){
614         swri_audio_convert(s->in_convert, postin, in, in_count);
615     }
616
617     if(s->resample_first){
618         if(postin != midbuf)
619             out_count= resample(s, midbuf, out_count, postin, in_count);
620         if(midbuf != preout)
621             swri_rematrix(s, preout, midbuf, out_count, preout==out);
622     }else{
623         if(postin != midbuf)
624             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
625         if(midbuf != preout)
626             out_count= resample(s, preout, out_count, midbuf, in_count);
627     }
628
629     if(preout != out && out_count){
630         if(s->dither_method){
631             int ch;
632             int dither_count= FFMAX(out_count, 1<<16);
633             av_assert0(preout != in);
634
635             if((ret=swri_realloc_audio(&s->dither, dither_count))<0)
636                 return ret;
637             if(ret)
638                 for(ch=0; ch<s->dither.ch_count; ch++)
639                     swri_get_dither(s, s->dither.ch[ch], s->dither.count, 12345678913579<<ch, s->out_sample_fmt, s->int_sample_fmt);
640             av_assert0(s->dither.ch_count == preout->ch_count);
641
642             if(s->dither_pos + out_count > s->dither.count)
643                 s->dither_pos = 0;
644
645             for(ch=0; ch<preout->ch_count; ch++)
646                 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);
647
648             s->dither_pos += out_count;
649         }
650 //FIXME packed doesnt need more than 1 chan here!
651         swri_audio_convert(s->out_convert, out, preout, out_count);
652     }
653     return out_count;
654 }
655
656 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
657                                 const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
658     AudioData * in= &s->in;
659     AudioData *out= &s->out;
660
661     if(s->drop_output > 0){
662         int ret;
663         AudioData tmp = s->out;
664         uint8_t *tmp_arg[SWR_CH_MAX];
665         tmp.count = 0;
666         tmp.data  = NULL;
667         if((ret=swri_realloc_audio(&tmp, s->drop_output))<0)
668             return ret;
669
670         reversefill_audiodata(&tmp, tmp_arg);
671         s->drop_output *= -1; //FIXME find a less hackish solution
672         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
673         s->drop_output *= -1;
674         if(ret>0)
675             s->drop_output -= ret;
676
677         av_freep(&tmp.data);
678         if(s->drop_output || !out_arg)
679             return 0;
680         in_count = 0;
681     }
682
683     if(!in_arg){
684         if(s->resample){
685             if (!s->flushed)
686                 s->resampler->flush(s);
687             s->resample_in_constraint = 0;
688             s->flushed = 1;
689         }else if(!s->in_buffer_count){
690             return 0;
691         }
692     }else
693         fill_audiodata(in ,  (void*)in_arg);
694
695     fill_audiodata(out, out_arg);
696
697     if(s->resample){
698         int ret = swr_convert_internal(s, out, out_count, in, in_count);
699         if(ret>0 && !s->drop_output)
700             s->outpts += ret * (int64_t)s->in_sample_rate;
701         return ret;
702     }else{
703         AudioData tmp= *in;
704         int ret2=0;
705         int ret, size;
706         size = FFMIN(out_count, s->in_buffer_count);
707         if(size){
708             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
709             ret= swr_convert_internal(s, out, size, &tmp, size);
710             if(ret<0)
711                 return ret;
712             ret2= ret;
713             s->in_buffer_count -= ret;
714             s->in_buffer_index += ret;
715             buf_set(out, out, ret);
716             out_count -= ret;
717             if(!s->in_buffer_count)
718                 s->in_buffer_index = 0;
719         }
720
721         if(in_count){
722             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
723
724             if(in_count > out_count) { //FIXME move after swr_convert_internal
725                 if(   size > s->in_buffer.count
726                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
727                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
728                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
729                     s->in_buffer_index=0;
730                 }else
731                     if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
732                         return ret;
733             }
734
735             if(out_count){
736                 size = FFMIN(in_count, out_count);
737                 ret= swr_convert_internal(s, out, size, in, size);
738                 if(ret<0)
739                     return ret;
740                 buf_set(in, in, ret);
741                 in_count -= ret;
742                 ret2 += ret;
743             }
744             if(in_count){
745                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
746                 copy(&tmp, in, in_count);
747                 s->in_buffer_count += in_count;
748             }
749         }
750         if(ret2>0 && !s->drop_output)
751             s->outpts += ret2 * (int64_t)s->in_sample_rate;
752         return ret2;
753     }
754 }
755
756 int swr_drop_output(struct SwrContext *s, int count){
757     s->drop_output += count;
758
759     if(s->drop_output <= 0)
760         return 0;
761
762     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
763     return swr_convert(s, NULL, s->drop_output, NULL, 0);
764 }
765
766 int swr_inject_silence(struct SwrContext *s, int count){
767     int ret, i;
768     AudioData silence = s->in;
769     uint8_t *tmp_arg[SWR_CH_MAX];
770
771     if(count <= 0)
772         return 0;
773
774     silence.count = 0;
775     silence.data  = NULL;
776     if((ret=swri_realloc_audio(&silence, count))<0)
777         return ret;
778
779     if(silence.planar) for(i=0; i<silence.ch_count; i++) {
780         memset(silence.ch[i], silence.bps==1 ? 0x80 : 0, count*silence.bps);
781     } else
782         memset(silence.ch[0], silence.bps==1 ? 0x80 : 0, count*silence.bps*silence.ch_count);
783
784     reversefill_audiodata(&silence, tmp_arg);
785     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
786     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
787     av_freep(&silence.data);
788     return ret;
789 }
790
791 int64_t swr_get_delay(struct SwrContext *s, int64_t base){
792     if (s->resampler && s->resample){
793         return s->resampler->get_delay(s, base);
794     }else{
795         return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
796     }
797 }
798
799 int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
800     int ret;
801
802     if (!s || compensation_distance < 0)
803         return AVERROR(EINVAL);
804     if (!compensation_distance && sample_delta)
805         return AVERROR(EINVAL);
806     if (!s->resample) {
807         s->flags |= SWR_FLAG_RESAMPLE;
808         ret = swr_init(s);
809         if (ret < 0)
810             return ret;
811     }
812     if (!s->resampler->set_compensation){
813         return AVERROR(EINVAL);
814     }else{
815         return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
816     }
817 }
818
819 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
820     if(pts == INT64_MIN)
821         return s->outpts;
822     if(s->min_compensation >= FLT_MAX) {
823         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
824     } else {
825         int64_t delta = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate) - s->outpts;
826         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
827
828         if(fabs(fdelta) > s->min_compensation) {
829             if(!s->outpts || fabs(fdelta) > s->min_hard_compensation){
830                 int ret;
831                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
832                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
833                 if(ret<0){
834                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
835                 }
836             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
837                 int duration = s->out_sample_rate * s->soft_compensation_duration;
838                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
839                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
840                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
841                 swr_set_compensation(s, comp, duration);
842             }
843         }
844
845         return s->outpts;
846     }
847 }