5caab4046f48a51e2e673b84c1e038df41aaad50
[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 resampling filter size"  , OFFSET(filter_size)    , AV_OPT_TYPE_INT  , {.i64=16                    }, 0      , INT_MAX   , PARAM },
84 {"phase_shift"          , "set 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 resampling precision"    , OFFSET(precision)      , AV_OPT_TYPE_DOUBLE,{.dbl=20.0                  }, 15.0   , 33.0      , PARAM },
91 {"cheby"                , "enable Chebyshev passband"   , OFFSET(cheby)          , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , 1         , PARAM },
92 {"min_comp"             , "set minimum difference between timestamps and audio data (in seconds) below which no timestamp compensation of either kind is applied"
93                                                         , OFFSET(min_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=FLT_MAX               }, 0      , FLT_MAX   , PARAM },
94 {"min_hard_comp"        , "set minimum difference between timestamps and audio data (in seconds) to trigger padding/trimming the data."
95                                                         , OFFSET(min_hard_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0.1                   }, 0      , INT_MAX   , PARAM },
96 {"comp_duration"        , "set duration (in seconds) over which data is stretched/squeezed to make it match the timestamps."
97                                                         , OFFSET(soft_compensation_duration),AV_OPT_TYPE_FLOAT ,{.dbl=1                     }, 0      , INT_MAX   , PARAM },
98 {"max_soft_comp"        , "set maximum factor by which data is stretched/squeezed to make it match the timestamps."
99                                                         , OFFSET(max_soft_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0                     }, INT_MIN, INT_MAX   , PARAM },
100 {"async"                , "simplified 1 parameter audio timestamp matching, 0(disabled), 1(filling and trimming), >1(maximum stretch/squeeze in samples per second)"
101                                                         , OFFSET(async)          , AV_OPT_TYPE_FLOAT ,{.dbl=0                     }, INT_MIN, INT_MAX   , PARAM },
102
103 { "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" },
104     { "none",  "select none",               0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_NONE  }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
105     { "dolby", "select Dolby",              0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DOLBY }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
106     { "dplii", "select Dolby Pro Logic II", 0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DPLII }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
107
108 { "filter_type"         , "select 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" },
109     { "cubic"           , "select cubic"                , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_CUBIC            }, INT_MIN, INT_MAX, PARAM, "filter_type" },
110     { "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" },
111     { "kaiser"          , "select Kaiser Windowed Sinc" , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_KAISER           }, INT_MIN, INT_MAX, PARAM, "filter_type" },
112
113 { "kaiser_beta"         , "set Kaiser Window Beta"      , OFFSET(kaiser_beta)    , AV_OPT_TYPE_INT  , {.i64=9                     }, 2      , 16        , PARAM },
114
115 {0}
116 };
117
118 static const char* context_to_name(void* ptr) {
119     return "SWR";
120 }
121
122 static const AVClass av_class = {
123     .class_name                = "SWResampler",
124     .item_name                 = context_to_name,
125     .option                    = options,
126     .version                   = LIBAVUTIL_VERSION_INT,
127     .log_level_offset_offset   = OFFSET(log_level_offset),
128     .parent_log_context_offset = OFFSET(log_ctx),
129     .category                  = AV_CLASS_CATEGORY_SWRESAMPLER,
130 };
131
132 unsigned swresample_version(void)
133 {
134     av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
135     return LIBSWRESAMPLE_VERSION_INT;
136 }
137
138 const char *swresample_configuration(void)
139 {
140     return FFMPEG_CONFIGURATION;
141 }
142
143 const char *swresample_license(void)
144 {
145 #define LICENSE_PREFIX "libswresample license: "
146     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
147 }
148
149 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
150     if(!s || s->in_convert) // s needs to be allocated but not initialized
151         return AVERROR(EINVAL);
152     s->channel_map = channel_map;
153     return 0;
154 }
155
156 const AVClass *swr_get_class(void)
157 {
158     return &av_class;
159 }
160
161 av_cold struct SwrContext *swr_alloc(void){
162     SwrContext *s= av_mallocz(sizeof(SwrContext));
163     if(s){
164         s->av_class= &av_class;
165         av_opt_set_defaults(s);
166     }
167     return s;
168 }
169
170 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
171                                       int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
172                                       int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
173                                       int log_offset, void *log_ctx){
174     if(!s) s= swr_alloc();
175     if(!s) return NULL;
176
177     s->log_level_offset= log_offset;
178     s->log_ctx= log_ctx;
179
180     av_opt_set_int(s, "ocl", out_ch_layout,   0);
181     av_opt_set_int(s, "osf", out_sample_fmt,  0);
182     av_opt_set_int(s, "osr", out_sample_rate, 0);
183     av_opt_set_int(s, "icl", in_ch_layout,    0);
184     av_opt_set_int(s, "isf", in_sample_fmt,   0);
185     av_opt_set_int(s, "isr", in_sample_rate,  0);
186     av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_NONE,   0);
187     av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> in_ch_layout), 0);
188     av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->out_ch_layout), 0);
189     av_opt_set_int(s, "uch", 0, 0);
190     return s;
191 }
192
193 static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
194     a->fmt   = fmt;
195     a->bps   = av_get_bytes_per_sample(fmt);
196     a->planar= av_sample_fmt_is_planar(fmt);
197 }
198
199 static void free_temp(AudioData *a){
200     av_free(a->data);
201     memset(a, 0, sizeof(*a));
202 }
203
204 av_cold void swr_free(SwrContext **ss){
205     SwrContext *s= *ss;
206     if(s){
207         free_temp(&s->postin);
208         free_temp(&s->midbuf);
209         free_temp(&s->preout);
210         free_temp(&s->in_buffer);
211         free_temp(&s->dither);
212         swri_audio_convert_free(&s-> in_convert);
213         swri_audio_convert_free(&s->out_convert);
214         swri_audio_convert_free(&s->full_convert);
215         if (s->resampler)
216             s->resampler->free(&s->resample);
217         swri_rematrix_free(s);
218     }
219
220     av_freep(ss);
221 }
222
223 av_cold int swr_init(struct SwrContext *s){
224     s->in_buffer_index= 0;
225     s->in_buffer_count= 0;
226     s->resample_in_constraint= 0;
227     free_temp(&s->postin);
228     free_temp(&s->midbuf);
229     free_temp(&s->preout);
230     free_temp(&s->in_buffer);
231     free_temp(&s->dither);
232     memset(s->in.ch, 0, sizeof(s->in.ch));
233     memset(s->out.ch, 0, sizeof(s->out.ch));
234     swri_audio_convert_free(&s-> in_convert);
235     swri_audio_convert_free(&s->out_convert);
236     swri_audio_convert_free(&s->full_convert);
237     swri_rematrix_free(s);
238
239     s->flushed = 0;
240
241     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
242         av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
243         return AVERROR(EINVAL);
244     }
245     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
246         av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
247         return AVERROR(EINVAL);
248     }
249
250     if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
251         if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
252             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
253         }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
254             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
255         }else{
256             av_log(s, AV_LOG_DEBUG, "Using double precision mode\n");
257             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
258         }
259     }
260
261     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16P
262         &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
263         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
264         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
265         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));
266         return AVERROR(EINVAL);
267     }
268
269     switch(s->engine){
270 #if CONFIG_LIBSOXR
271         extern struct Resampler const soxr_resampler;
272         case SWR_ENGINE_SOXR: s->resampler = &soxr_resampler; break;
273 #endif
274         case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
275         default:
276             av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
277             return AVERROR(EINVAL);
278     }
279
280     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
281     set_audiodata_fmt(&s->out, s->out_sample_fmt);
282
283     if (s->async) {
284         if (s->min_compensation >= FLT_MAX/2)
285             s->min_compensation = 0.001;
286         if (s->async > 1.0001) {
287             s->max_soft_compensation = s->async / (double) s->in_sample_rate;
288         }
289     }
290
291     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
292         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);
293     }else
294         s->resampler->free(&s->resample);
295     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
296         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
297         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
298         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
299         && s->resample){
300         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
301         return -1;
302     }
303
304     if(!s->used_ch_count)
305         s->used_ch_count= s->in.ch_count;
306
307     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
308         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
309         s-> in_ch_layout= 0;
310     }
311
312     if(!s-> in_ch_layout)
313         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
314     if(!s->out_ch_layout)
315         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
316
317     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
318                  s->rematrix_custom;
319
320 #define RSC 1 //FIXME finetune
321     if(!s-> in.ch_count)
322         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
323     if(!s->used_ch_count)
324         s->used_ch_count= s->in.ch_count;
325     if(!s->out.ch_count)
326         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
327
328     if(!s-> in.ch_count){
329         av_assert0(!s->in_ch_layout);
330         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
331         return -1;
332     }
333
334     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
335         av_log(s, AV_LOG_ERROR, "Rematrix is needed but there is not enough information to do it\n");
336         return -1;
337     }
338
339 av_assert0(s->used_ch_count);
340 av_assert0(s->out.ch_count);
341     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
342
343     s->in_buffer= s->in;
344
345     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither_method){
346         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
347                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
348         return 0;
349     }
350
351     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
352                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
353     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
354                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
355
356
357     s->postin= s->in;
358     s->preout= s->out;
359     s->midbuf= s->in;
360
361     if(s->channel_map){
362         s->postin.ch_count=
363         s->midbuf.ch_count= s->used_ch_count;
364         if(s->resample)
365             s->in_buffer.ch_count= s->used_ch_count;
366     }
367     if(!s->resample_first){
368         s->midbuf.ch_count= s->out.ch_count;
369         if(s->resample)
370             s->in_buffer.ch_count = s->out.ch_count;
371     }
372
373     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
374     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
375     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
376
377     if(s->resample){
378         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
379     }
380
381     s->dither = s->preout;
382
383     if(s->rematrix || s->dither_method)
384         return swri_rematrix_init(s);
385
386     return 0;
387 }
388
389 int swri_realloc_audio(AudioData *a, int count){
390     int i, countb;
391     AudioData old;
392
393     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
394         return AVERROR(EINVAL);
395
396     if(a->count >= count)
397         return 0;
398
399     count*=2;
400
401     countb= FFALIGN(count*a->bps, ALIGN);
402     old= *a;
403
404     av_assert0(a->bps);
405     av_assert0(a->ch_count);
406
407     a->data= av_mallocz(countb*a->ch_count);
408     if(!a->data)
409         return AVERROR(ENOMEM);
410     for(i=0; i<a->ch_count; i++){
411         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
412         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
413     }
414     if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
415     av_free(old.data);
416     a->count= count;
417
418     return 1;
419 }
420
421 static void copy(AudioData *out, AudioData *in,
422                  int count){
423     av_assert0(out->planar == in->planar);
424     av_assert0(out->bps == in->bps);
425     av_assert0(out->ch_count == in->ch_count);
426     if(out->planar){
427         int ch;
428         for(ch=0; ch<out->ch_count; ch++)
429             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
430     }else
431         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
432 }
433
434 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
435     int i;
436     if(!in_arg){
437         memset(out->ch, 0, sizeof(out->ch));
438     }else if(out->planar){
439         for(i=0; i<out->ch_count; i++)
440             out->ch[i]= in_arg[i];
441     }else{
442         for(i=0; i<out->ch_count; i++)
443             out->ch[i]= in_arg[0] + i*out->bps;
444     }
445 }
446
447 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
448     int i;
449     if(out->planar){
450         for(i=0; i<out->ch_count; i++)
451             in_arg[i]= out->ch[i];
452     }else{
453         in_arg[0]= out->ch[0];
454     }
455 }
456
457 /**
458  *
459  * out may be equal in.
460  */
461 static void buf_set(AudioData *out, AudioData *in, int count){
462     int ch;
463     if(in->planar){
464         for(ch=0; ch<out->ch_count; ch++)
465             out->ch[ch]= in->ch[ch] + count*out->bps;
466     }else{
467         for(ch=out->ch_count-1; ch>=0; ch--)
468             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
469     }
470 }
471
472 /**
473  *
474  * @return number of samples output per channel
475  */
476 static int resample(SwrContext *s, AudioData *out_param, int out_count,
477                              const AudioData * in_param, int in_count){
478     AudioData in, out, tmp;
479     int ret_sum=0;
480     int border=0;
481
482     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
483     av_assert1(s->in_buffer.planar   == in_param->planar);
484     av_assert1(s->in_buffer.fmt      == in_param->fmt);
485
486     tmp=out=*out_param;
487     in =  *in_param;
488
489     do{
490         int ret, size, consumed;
491         if(!s->resample_in_constraint && s->in_buffer_count){
492             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
493             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
494             out_count -= ret;
495             ret_sum += ret;
496             buf_set(&out, &out, ret);
497             s->in_buffer_count -= consumed;
498             s->in_buffer_index += consumed;
499
500             if(!in_count)
501                 break;
502             if(s->in_buffer_count <= border){
503                 buf_set(&in, &in, -s->in_buffer_count);
504                 in_count += s->in_buffer_count;
505                 s->in_buffer_count=0;
506                 s->in_buffer_index=0;
507                 border = 0;
508             }
509         }
510
511         if((s->flushed || in_count) && !s->in_buffer_count){
512             s->in_buffer_index=0;
513             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
514             out_count -= ret;
515             ret_sum += ret;
516             buf_set(&out, &out, ret);
517             in_count -= consumed;
518             buf_set(&in, &in, consumed);
519         }
520
521         //TODO is this check sane considering the advanced copy avoidance below
522         size= s->in_buffer_index + s->in_buffer_count + in_count;
523         if(   size > s->in_buffer.count
524            && s->in_buffer_count + in_count <= s->in_buffer_index){
525             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
526             copy(&s->in_buffer, &tmp, s->in_buffer_count);
527             s->in_buffer_index=0;
528         }else
529             if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
530                 return ret;
531
532         if(in_count){
533             int count= in_count;
534             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
535
536             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
537             copy(&tmp, &in, /*in_*/count);
538             s->in_buffer_count += count;
539             in_count -= count;
540             border += count;
541             buf_set(&in, &in, count);
542             s->resample_in_constraint= 0;
543             if(s->in_buffer_count != count || in_count)
544                 continue;
545         }
546         break;
547     }while(1);
548
549     s->resample_in_constraint= !!out_count;
550
551     return ret_sum;
552 }
553
554 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
555                                                       AudioData *in , int  in_count){
556     AudioData *postin, *midbuf, *preout;
557     int ret/*, in_max*/;
558     AudioData preout_tmp, midbuf_tmp;
559
560     if(s->full_convert){
561         av_assert0(!s->resample);
562         swri_audio_convert(s->full_convert, out, in, in_count);
563         return out_count;
564     }
565
566 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
567 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
568
569     if((ret=swri_realloc_audio(&s->postin, in_count))<0)
570         return ret;
571     if(s->resample_first){
572         av_assert0(s->midbuf.ch_count == s->used_ch_count);
573         if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
574             return ret;
575     }else{
576         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
577         if((ret=swri_realloc_audio(&s->midbuf,  in_count))<0)
578             return ret;
579     }
580     if((ret=swri_realloc_audio(&s->preout, out_count))<0)
581         return ret;
582
583     postin= &s->postin;
584
585     midbuf_tmp= s->midbuf;
586     midbuf= &midbuf_tmp;
587     preout_tmp= s->preout;
588     preout= &preout_tmp;
589
590     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
591         postin= in;
592
593     if(s->resample_first ? !s->resample : !s->rematrix)
594         midbuf= postin;
595
596     if(s->resample_first ? !s->rematrix : !s->resample)
597         preout= midbuf;
598
599     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
600         if(preout==in){
601             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
602             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
603             copy(out, in, out_count);
604             return out_count;
605         }
606         else if(preout==postin) preout= midbuf= postin= out;
607         else if(preout==midbuf) preout= midbuf= out;
608         else                    preout= out;
609     }
610
611     if(in != postin){
612         swri_audio_convert(s->in_convert, postin, in, in_count);
613     }
614
615     if(s->resample_first){
616         if(postin != midbuf)
617             out_count= resample(s, midbuf, out_count, postin, in_count);
618         if(midbuf != preout)
619             swri_rematrix(s, preout, midbuf, out_count, preout==out);
620     }else{
621         if(postin != midbuf)
622             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
623         if(midbuf != preout)
624             out_count= resample(s, preout, out_count, midbuf, in_count);
625     }
626
627     if(preout != out && out_count){
628         if(s->dither_method){
629             int ch;
630             int dither_count= FFMAX(out_count, 1<<16);
631             av_assert0(preout != in);
632
633             if((ret=swri_realloc_audio(&s->dither, dither_count))<0)
634                 return ret;
635             if(ret)
636                 for(ch=0; ch<s->dither.ch_count; ch++)
637                     swri_get_dither(s, s->dither.ch[ch], s->dither.count, 12345678913579<<ch, s->out_sample_fmt, s->int_sample_fmt);
638             av_assert0(s->dither.ch_count == preout->ch_count);
639
640             if(s->dither_pos + out_count > s->dither.count)
641                 s->dither_pos = 0;
642
643             for(ch=0; ch<preout->ch_count; ch++)
644                 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);
645
646             s->dither_pos += out_count;
647         }
648 //FIXME packed doesnt need more than 1 chan here!
649         swri_audio_convert(s->out_convert, out, preout, out_count);
650     }
651     return out_count;
652 }
653
654 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
655                                 const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
656     AudioData * in= &s->in;
657     AudioData *out= &s->out;
658
659     if(s->drop_output > 0){
660         int ret;
661         AudioData tmp = s->out;
662         uint8_t *tmp_arg[SWR_CH_MAX];
663         tmp.count = 0;
664         tmp.data  = NULL;
665         if((ret=swri_realloc_audio(&tmp, s->drop_output))<0)
666             return ret;
667
668         reversefill_audiodata(&tmp, tmp_arg);
669         s->drop_output *= -1; //FIXME find a less hackish solution
670         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
671         s->drop_output *= -1;
672         if(ret>0)
673             s->drop_output -= ret;
674
675         av_freep(&tmp.data);
676         if(s->drop_output || !out_arg)
677             return 0;
678         in_count = 0;
679     }
680
681     if(!in_arg){
682         if(s->resample){
683             if (!s->flushed)
684                 s->resampler->flush(s);
685             s->resample_in_constraint = 0;
686             s->flushed = 1;
687         }else if(!s->in_buffer_count){
688             return 0;
689         }
690     }else
691         fill_audiodata(in ,  (void*)in_arg);
692
693     fill_audiodata(out, out_arg);
694
695     if(s->resample){
696         int ret = swr_convert_internal(s, out, out_count, in, in_count);
697         if(ret>0 && !s->drop_output)
698             s->outpts += ret * (int64_t)s->in_sample_rate;
699         return ret;
700     }else{
701         AudioData tmp= *in;
702         int ret2=0;
703         int ret, size;
704         size = FFMIN(out_count, s->in_buffer_count);
705         if(size){
706             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
707             ret= swr_convert_internal(s, out, size, &tmp, size);
708             if(ret<0)
709                 return ret;
710             ret2= ret;
711             s->in_buffer_count -= ret;
712             s->in_buffer_index += ret;
713             buf_set(out, out, ret);
714             out_count -= ret;
715             if(!s->in_buffer_count)
716                 s->in_buffer_index = 0;
717         }
718
719         if(in_count){
720             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
721
722             if(in_count > out_count) { //FIXME move after swr_convert_internal
723                 if(   size > s->in_buffer.count
724                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
725                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
726                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
727                     s->in_buffer_index=0;
728                 }else
729                     if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
730                         return ret;
731             }
732
733             if(out_count){
734                 size = FFMIN(in_count, out_count);
735                 ret= swr_convert_internal(s, out, size, in, size);
736                 if(ret<0)
737                     return ret;
738                 buf_set(in, in, ret);
739                 in_count -= ret;
740                 ret2 += ret;
741             }
742             if(in_count){
743                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
744                 copy(&tmp, in, in_count);
745                 s->in_buffer_count += in_count;
746             }
747         }
748         if(ret2>0 && !s->drop_output)
749             s->outpts += ret2 * (int64_t)s->in_sample_rate;
750         return ret2;
751     }
752 }
753
754 int swr_drop_output(struct SwrContext *s, int count){
755     s->drop_output += count;
756
757     if(s->drop_output <= 0)
758         return 0;
759
760     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
761     return swr_convert(s, NULL, s->drop_output, NULL, 0);
762 }
763
764 int swr_inject_silence(struct SwrContext *s, int count){
765     int ret, i;
766     AudioData silence = s->in;
767     uint8_t *tmp_arg[SWR_CH_MAX];
768
769     if(count <= 0)
770         return 0;
771
772     silence.count = 0;
773     silence.data  = NULL;
774     if((ret=swri_realloc_audio(&silence, count))<0)
775         return ret;
776
777     if(silence.planar) for(i=0; i<silence.ch_count; i++) {
778         memset(silence.ch[i], silence.bps==1 ? 0x80 : 0, count*silence.bps);
779     } else
780         memset(silence.ch[0], silence.bps==1 ? 0x80 : 0, count*silence.bps*silence.ch_count);
781
782     reversefill_audiodata(&silence, tmp_arg);
783     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
784     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
785     av_freep(&silence.data);
786     return ret;
787 }
788
789 int64_t swr_get_delay(struct SwrContext *s, int64_t base){
790     if (s->resampler && s->resample){
791         return s->resampler->get_delay(s, base);
792     }else{
793         return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
794     }
795 }
796
797 int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
798     int ret;
799
800     if (!s || compensation_distance < 0)
801         return AVERROR(EINVAL);
802     if (!compensation_distance && sample_delta)
803         return AVERROR(EINVAL);
804     if (!s->resample) {
805         s->flags |= SWR_FLAG_RESAMPLE;
806         ret = swr_init(s);
807         if (ret < 0)
808             return ret;
809     }
810     if (!s->resampler->set_compensation){
811         return AVERROR(EINVAL);
812     }else{
813         return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
814     }
815 }
816
817 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
818     if(pts == INT64_MIN)
819         return s->outpts;
820     if(s->min_compensation >= FLT_MAX) {
821         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
822     } else {
823         int64_t delta = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate) - s->outpts;
824         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
825
826         if(fabs(fdelta) > s->min_compensation) {
827             if(!s->outpts || fabs(fdelta) > s->min_hard_compensation){
828                 int ret;
829                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
830                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
831                 if(ret<0){
832                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
833                 }
834             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
835                 int duration = s->out_sample_rate * s->soft_compensation_duration;
836                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
837                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
838                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
839                 swr_set_compensation(s, comp, duration);
840             }
841         }
842
843         return s->outpts;
844     }
845 }